maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/clang/test/OpenMP/single_codegen.cpp

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

5241 lines
367 KiB
C++
Raw Normal View History

[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" "reduction_size[.].+[.]" "pl_cond[.].+[.|,]" --prefix-filecheck-ir-name _
// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=50 -fnoopenmp-use-tls -x c++ -std=c++11 -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s --check-prefix=CHECK1
// RUN: %clang_cc1 -verify -fopenmp -fopenmp-version=50 -fnoopenmp-use-tls -x c++ -std=c++11 -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -fopenmp-version=45 -o - | FileCheck %s --check-prefix=CHECK2
[OpenMP] Upgrade default version of OpenMP to 5.0 Summary: When -fopenmp option is specified then version 5.0 will be set as default. Reviewers: gregrodgers, jdoerfert, ABataev Reviewed By: ABataev Subscribers: pdhaliwal, yaxunl, guansong, sstefan1, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D81098
2020-06-25 15:01:15 +08:00
// RUN: %clang_cc1 -fopenmp -fopenmp-version=50 -fnoopenmp-use-tls -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// RUN: %clang_cc1 -fopenmp -fopenmp-version=50 -fnoopenmp-use-tls -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK3
[OpenMP] Upgrade default version of OpenMP to 5.0 Summary: When -fopenmp option is specified then version 5.0 will be set as default. Reviewers: gregrodgers, jdoerfert, ABataev Reviewed By: ABataev Subscribers: pdhaliwal, yaxunl, guansong, sstefan1, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D81098
2020-06-25 15:01:15 +08:00
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fnoopenmp-use-tls -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -fnoopenmp-use-tls -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK4
[OpenMP] Upgrade default version of OpenMP to 5.0 Summary: When -fopenmp option is specified then version 5.0 will be set as default. Reviewers: gregrodgers, jdoerfert, ABataev Reviewed By: ABataev Subscribers: pdhaliwal, yaxunl, guansong, sstefan1, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D81098
2020-06-25 15:01:15 +08:00
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -std=c++11 -fopenmp -fnoopenmp-use-tls -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK5
// RUN: %clang_cc1 -verify -fopenmp -fnoopenmp-use-tls -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK6
[OPENMP] Support for -fopenmp-simd option with compilation of simd loops only. Added support for -fopenmp-simd option that allows compilation of simd-based constructs without emission of OpenMP runtime calls. llvm-svn: 321560
2017-12-30 02:07:07 +08:00
[OpenMP][NFC] Remove SIMD check lines for non-simd tests If a test does not contain an " simd" but -fopenmp-simd RUN lines we can just check that we do not create __kmpc|__tgt calls. Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101973
2021-05-19 10:52:53 +08:00
// RUN: %clang_cc1 -verify -fopenmp-simd -fnoopenmp-use-tls -x c++ -std=c++11 -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
[OPENMP] Support for -fopenmp-simd option with compilation of simd loops only. Added support for -fopenmp-simd option that allows compilation of simd-based constructs without emission of OpenMP runtime calls. llvm-svn: 321560
2017-12-30 02:07:07 +08:00
// RUN: %clang_cc1 -fopenmp-simd -fnoopenmp-use-tls -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s
[OpenMP][NFC] Remove SIMD check lines for non-simd tests If a test does not contain an " simd" but -fopenmp-simd RUN lines we can just check that we do not create __kmpc|__tgt calls. Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101973
2021-05-19 10:52:53 +08:00
// RUN: %clang_cc1 -fopenmp-simd -fnoopenmp-use-tls -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -std=c++11 -fopenmp-simd -fnoopenmp-use-tls -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
// RUN: %clang_cc1 -verify -fopenmp-simd -fnoopenmp-use-tls -x c++ -std=c++11 -DARRAY -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
[OPENMP] Initial codegen for 'single' directive. This patch emits the following code for the single directive: #pragma omp single <body> <----> if(__kmpc_single(...)) { <body> __kmpc_end_single(...); } Differential Revision: http://reviews.llvm.org/D7045 llvm-svn: 228275
2015-02-05 14:35:41 +08:00
// expected-no-diagnostics
[OPENMP] Fixed codegen for copying/initialization of array variables/parameters. This modification generates proper copyin/initialization sequences for array variables/parameters. Before they were considered as pointers, not arrays. llvm-svn: 237691
2015-05-19 20:31:28 +08:00
#ifndef ARRAY
[OPENMP] Initial codegen for 'single' directive. This patch emits the following code for the single directive: #pragma omp single <body> <----> if(__kmpc_single(...)) { <body> __kmpc_end_single(...); } Differential Revision: http://reviews.llvm.org/D7045 llvm-svn: 228275
2015-02-05 14:35:41 +08:00
#ifndef HEADER
#define HEADER
[OPENMP] Codegen for 'copyprivate' clause ('single' directive). If there is at least one 'copyprivate' clause is associated with the single directive, the following code is generated: ``` i32 did_it = 0; \\ for 'copyprivate' clause if(__kmpc_single(ident_t *, gtid)) { SingleOpGen(); __kmpc_end_single(ident_t *, gtid); did_it = 1; \\ for 'copyprivate' clause } <copyprivate_list>[0] = &var0; ... <copyprivate_list>[n] = &varn; call __kmpc_copyprivate(ident_t *, gtid, <copyprivate_list_size>, <copyprivate_list>, <copy_func>, did_it); ... void<copy_func>(void *LHSArg, void *RHSArg) { Dst = (void * [n])(LHSArg); Src = (void * [n])(RHSArg); Dst[0] = Src[0]; ... Dst[n] = Src[n]; } ``` All list items from all 'copyprivate' clauses are gathered into single <copyprivate list> (<copyprivate_list_size> is a size in bytes of this list) and <copy_func> is used to propagate values of private or threadprivate variables from the 'single' region to other implicit threads from outer 'parallel' region. Differential Revision: http://reviews.llvm.org/D8410 llvm-svn: 232932
2015-03-23 14:18:07 +08:00
class TestClass {
public:
int a;
TestClass() : a(0) {}
TestClass(const TestClass &C) : a(C.a) {}
TestClass &operator=(const TestClass &) { return *this;}
~TestClass(){};
};
[OPENMP] Initial codegen for 'single' directive. This patch emits the following code for the single directive: #pragma omp single <body> <----> if(__kmpc_single(...)) { <body> __kmpc_end_single(...); } Differential Revision: http://reviews.llvm.org/D7045 llvm-svn: 228275
2015-02-05 14:35:41 +08:00
[OPENMP] Codegen for 'copyprivate' clause ('single' directive). If there is at least one 'copyprivate' clause is associated with the single directive, the following code is generated: ``` i32 did_it = 0; \\ for 'copyprivate' clause if(__kmpc_single(ident_t *, gtid)) { SingleOpGen(); __kmpc_end_single(ident_t *, gtid); did_it = 1; \\ for 'copyprivate' clause } <copyprivate_list>[0] = &var0; ... <copyprivate_list>[n] = &varn; call __kmpc_copyprivate(ident_t *, gtid, <copyprivate_list_size>, <copyprivate_list>, <copy_func>, did_it); ... void<copy_func>(void *LHSArg, void *RHSArg) { Dst = (void * [n])(LHSArg); Src = (void * [n])(RHSArg); Dst[0] = Src[0]; ... Dst[n] = Src[n]; } ``` All list items from all 'copyprivate' clauses are gathered into single <copyprivate list> (<copyprivate_list_size> is a size in bytes of this list) and <copy_func> is used to propagate values of private or threadprivate variables from the 'single' region to other implicit threads from outer 'parallel' region. Differential Revision: http://reviews.llvm.org/D8410 llvm-svn: 232932
2015-03-23 14:18:07 +08:00
TestClass tc;
[OPENMP] Fixed codegen for arrays in 'copyprivate' clause. Fixed a bug with codegen of variables with array types specified in 'copyprivate' clause of 'single' directive. Differential Revision: http://reviews.llvm.org/D8914 llvm-svn: 234856
2015-04-14 13:11:24 +08:00
TestClass tc2[2];
#pragma omp threadprivate(tc, tc2)
[OPENMP] Codegen for 'copyprivate' clause ('single' directive). If there is at least one 'copyprivate' clause is associated with the single directive, the following code is generated: ``` i32 did_it = 0; \\ for 'copyprivate' clause if(__kmpc_single(ident_t *, gtid)) { SingleOpGen(); __kmpc_end_single(ident_t *, gtid); did_it = 1; \\ for 'copyprivate' clause } <copyprivate_list>[0] = &var0; ... <copyprivate_list>[n] = &varn; call __kmpc_copyprivate(ident_t *, gtid, <copyprivate_list_size>, <copyprivate_list>, <copy_func>, did_it); ... void<copy_func>(void *LHSArg, void *RHSArg) { Dst = (void * [n])(LHSArg); Src = (void * [n])(RHSArg); Dst[0] = Src[0]; ... Dst[n] = Src[n]; } ``` All list items from all 'copyprivate' clauses are gathered into single <copyprivate list> (<copyprivate_list_size> is a size in bytes of this list) and <copy_func> is used to propagate values of private or threadprivate variables from the 'single' region to other implicit threads from outer 'parallel' region. Differential Revision: http://reviews.llvm.org/D8410 llvm-svn: 232932
2015-03-23 14:18:07 +08:00
[CodeGen] Emit a call instruction instead of an invoke if the called llvm function is marked nounwind This fixes cases where an invoke is emitted, despite the called llvm function being marked nounwind, because ConstructAttributeList failed to add the attribute to the attribute list. llvm optimization passes turn invokes into calls and optimize away the exception handling code, but it's better to avoid emitting the code in the front-end if the called function is known not to raise an exception. Differential Revision: https://reviews.llvm.org/D83906
2020-07-16 05:47:45 +08:00
void foo() { extern void mayThrow(); mayThrow(); }
[OPENMP] Initial codegen for 'single' directive. This patch emits the following code for the single directive: #pragma omp single <body> <----> if(__kmpc_single(...)) { <body> __kmpc_end_single(...); } Differential Revision: http://reviews.llvm.org/D7045 llvm-svn: 228275
2015-02-05 14:35:41 +08:00
[OPENMP 4.5] Allow to use private data members in 'copyprivate' clause. OpenMP 4.5 allows privatization of non-static data members in non-static member functions. This patch adds support of private data members in 'copyprivate' clauses. llvm-svn: 263706
2016-03-17 18:50:17 +08:00
struct SS {
int a;
int b : 4;
int &c;
SS(int &d) : a(0), b(0), c(d) {
#pragma omp parallel firstprivate(a, b, c)
#pragma omp single copyprivate(a, this->b, (this)->c)
[&]() {
++this->a, --b, (this)->c /= 1;
#pragma omp parallel firstprivate(a, b, c)
#pragma omp single copyprivate(a, this->b, (this)->c)
++(this)->a, --b, this->c /= 1;
}();
}
};
template<typename T>
struct SST {
T a;
SST() : a(T()) {
#pragma omp parallel firstprivate(a)
#pragma omp single copyprivate(this->a)
[&]() {
[&]() {
++this->a;
#pragma omp parallel firstprivate(a)
#pragma omp single copyprivate((this)->a)
++(this)->a;
}();
}();
}
};
[OPENMP] Initial codegen for 'single' directive. This patch emits the following code for the single directive: #pragma omp single <body> <----> if(__kmpc_single(...)) { <body> __kmpc_end_single(...); } Differential Revision: http://reviews.llvm.org/D7045 llvm-svn: 228275
2015-02-05 14:35:41 +08:00
int main() {
char a;
[OPENMP] Fixed codegen for arrays in 'copyprivate' clause. Fixed a bug with codegen of variables with array types specified in 'copyprivate' clause of 'single' directive. Differential Revision: http://reviews.llvm.org/D8914 llvm-svn: 234856
2015-04-14 13:11:24 +08:00
char a2[2];
[OPENMP] Fix for http://llvm.org/PR24674: assertion failed and and abort trap Fix processing of shared variables with reference types in OpenMP constructs. Previously, if the variable was not marked in one of the private clauses, the reference to this variable was emitted incorrectly and caused an assertion later. llvm-svn: 246846
2015-09-04 19:26:21 +08:00
TestClass &c = tc;
[OPENMP 4.5] Allow to use private data members in 'copyprivate' clause. OpenMP 4.5 allows privatization of non-static data members in non-static member functions. This patch adds support of private data members in 'copyprivate' clauses. llvm-svn: 263706
2016-03-17 18:50:17 +08:00
SST<double> sst;
SS ss(c.a);
[OPENMP] Initial codegen for 'single' directive. This patch emits the following code for the single directive: #pragma omp single <body> <----> if(__kmpc_single(...)) { <body> __kmpc_end_single(...); } Differential Revision: http://reviews.llvm.org/D7045 llvm-svn: 228275
2015-02-05 14:35:41 +08:00
[OPENMP] Codegen for 'copyprivate' clause ('single' directive). If there is at least one 'copyprivate' clause is associated with the single directive, the following code is generated: ``` i32 did_it = 0; \\ for 'copyprivate' clause if(__kmpc_single(ident_t *, gtid)) { SingleOpGen(); __kmpc_end_single(ident_t *, gtid); did_it = 1; \\ for 'copyprivate' clause } <copyprivate_list>[0] = &var0; ... <copyprivate_list>[n] = &varn; call __kmpc_copyprivate(ident_t *, gtid, <copyprivate_list_size>, <copyprivate_list>, <copy_func>, did_it); ... void<copy_func>(void *LHSArg, void *RHSArg) { Dst = (void * [n])(LHSArg); Src = (void * [n])(RHSArg); Dst[0] = Src[0]; ... Dst[n] = Src[n]; } ``` All list items from all 'copyprivate' clauses are gathered into single <copyprivate list> (<copyprivate_list_size> is a size in bytes of this list) and <copy_func> is used to propagate values of private or threadprivate variables from the 'single' region to other implicit threads from outer 'parallel' region. Differential Revision: http://reviews.llvm.org/D8410 llvm-svn: 232932
2015-03-23 14:18:07 +08:00
#pragma omp single nowait
[OPENMP] Initial codegen for 'single' directive. This patch emits the following code for the single directive: #pragma omp single <body> <----> if(__kmpc_single(...)) { <body> __kmpc_end_single(...); } Differential Revision: http://reviews.llvm.org/D7045 llvm-svn: 228275
2015-02-05 14:35:41 +08:00
a = 2;
[OPENMP] Do not emit implicit barrier for single directive with 'copyprivate' clause(s). Runtime function for 'copyprivate' directive generates implicit barriers, so no need to emit it. Differential Revision: http://reviews.llvm.org/D9215 llvm-svn: 235692
2015-04-24 12:00:39 +08:00
#pragma omp single
a = 2;
[OPENMP] Fixed codegen for arrays in 'copyprivate' clause. Fixed a bug with codegen of variables with array types specified in 'copyprivate' clause of 'single' directive. Differential Revision: http://reviews.llvm.org/D8914 llvm-svn: 234856
2015-04-14 13:11:24 +08:00
#pragma omp single copyprivate(a, c, tc, a2, tc2)
[OPENMP] Initial codegen for 'single' directive. This patch emits the following code for the single directive: #pragma omp single <body> <----> if(__kmpc_single(...)) { <body> __kmpc_end_single(...); } Differential Revision: http://reviews.llvm.org/D7045 llvm-svn: 228275
2015-02-05 14:35:41 +08:00
foo();
return a;
}
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
[OPENMP] Improved code for generating debug info + generation of all OpenMP regions in termination scope Patch adds proper generation of debug info for all OpenMP regions. Also, all OpenMP regions are generated in a termination scope, because standard does not allow to throw exceptions out of structured blocks, associated with the OpenMP regions Differential Revision: http://reviews.llvm.org/D7935 llvm-svn: 231757
2015-03-10 13:15:26 +08:00
void parallel_single() {
[OPENMP] Fixed codegen for directives without function outlining. Fixed crash on codegen for directives like 'omp for', 'omp single' etc. inside of the 'omp parallel', 'omp task' etc. regions. llvm-svn: 230621
2015-02-26 18:27:34 +08:00
#pragma omp parallel
#pragma omp single
[OPENMP] Improved code for generating debug info + generation of all OpenMP regions in termination scope Patch adds proper generation of debug info for all OpenMP regions. Also, all OpenMP regions are generated in a termination scope, because standard does not allow to throw exceptions out of structured blocks, associated with the OpenMP regions Differential Revision: http://reviews.llvm.org/D7935 llvm-svn: 231757
2015-03-10 13:15:26 +08:00
foo();
[OPENMP] Fixed codegen for directives without function outlining. Fixed crash on codegen for directives like 'omp for', 'omp single' etc. inside of the 'omp parallel', 'omp task' etc. regions. llvm-svn: 230621
2015-02-26 18:27:34 +08:00
}
[OPENMP] Fixed codegen for copying/initialization of array variables/parameters. This modification generates proper copyin/initialization sequences for array variables/parameters. Before they were considered as pointers, not arrays. llvm-svn: 237691
2015-05-19 20:31:28 +08:00
#endif
#else
struct St {
int a, b;
St() : a(0), b(0) {}
St &operator=(const St &) { return *this; };
~St() {}
};
[OPENMP] Fixed codegen for directives without function outlining. Fixed crash on codegen for directives like 'omp for', 'omp single' etc. inside of the 'omp parallel', 'omp task' etc. regions. llvm-svn: 230621
2015-02-26 18:27:34 +08:00
[OPENMP] Fixed codegen for parameters privatization. For parameters we shall take a derived type of parameters, not the original one. llvm-svn: 237882
2015-05-21 17:47:46 +08:00
void array_func(int n, int a[n], St s[2]) {
[OPENMP] Fixed codegen for copying/initialization of array variables/parameters. This modification generates proper copyin/initialization sequences for array variables/parameters. Before they were considered as pointers, not arrays. llvm-svn: 237691
2015-05-19 20:31:28 +08:00
#pragma omp single copyprivate(a, s)
;
}
[OPENMP] Initial codegen for 'single' directive. This patch emits the following code for the single directive: #pragma omp single <body> <----> if(__kmpc_single(...)) { <body> __kmpc_end_single(...); } Differential Revision: http://reviews.llvm.org/D7045 llvm-svn: 228275
2015-02-05 14:35:41 +08:00
#endif
[OPENMP 4.5] Allow to use private data members in 'copyprivate' clause. OpenMP 4.5 allows privatization of non-static data members in non-static member functions. This patch adds support of private data members in 'copyprivate' clauses. llvm-svn: 263706
2016-03-17 18:50:17 +08:00
// Private a
// Private b
// Private c
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
[OPENMP 4.5] Allow to use private data members in 'copyprivate' clause. OpenMP 4.5 allows privatization of non-static data members in non-static member functions. This patch adds support of private data members in 'copyprivate' clauses. llvm-svn: 263706
2016-03-17 18:50:17 +08:00
// Private a
// Private b
// Private c
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_.
// CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0:[0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: call void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP2]])
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK1-NEXT: ret i8* [[TMP3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN9TestClassC1Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK1-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: call void @_ZN9TestClassC2Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-LABEL: define {{[^@]+}}@.__kmpc_global_dtor_.
// CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP2]]) #[[ATTR3:[0-9]+]]
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: ret void
//
//
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-LABEL: define {{[^@]+}}@_ZN9TestClassD1Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2:[0-9]+]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK1-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: call void @_ZN9TestClassD2Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-LABEL: define {{[^@]+}}@.__omp_threadprivate_init_.
// CHECK1-SAME: () #[[ATTR0]] {
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
// CHECK1-NEXT: call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast (%class.TestClass* @tc to i8*), i8* (i8*)* @.__kmpc_global_ctor_., i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_.)
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_..1
// CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to [2 x %class.TestClass]*
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %class.TestClass], [2 x %class.TestClass]* [[TMP2]], i32 0, i32 0
// CHECK1-NEXT: [[ARRAYCTOR_END:%.*]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAY_BEGIN]], i64 2
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: br label [[ARRAYCTOR_LOOP:%.*]]
// CHECK1: arrayctor.loop:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[ARRAYCTOR_CUR:%.*]] = phi %class.TestClass* [ [[ARRAY_BEGIN]], [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[INVOKE_CONT:%.*]] ]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: invoke void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: to label [[INVOKE_CONT]] unwind label [[LPAD:%.*]]
// CHECK1: invoke.cont:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYCTOR_CUR]], i64 1
// CHECK1-NEXT: [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], [[ARRAYCTOR_END]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
// CHECK1: arrayctor.cont:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK1-NEXT: ret i8* [[TMP3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1: lpad:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[TMP4:%.*]] = landingpad { i8*, i32 }
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: cleanup
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[TMP5:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 0
// CHECK1-NEXT: store i8* [[TMP5]], i8** [[EXN_SLOT]], align 8
// CHECK1-NEXT: [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 1
// CHECK1-NEXT: store i32 [[TMP6]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK1-NEXT: [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %class.TestClass* [[ARRAY_BEGIN]], [[ARRAYCTOR_CUR]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ [[ARRAYCTOR_CUR]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done1:
// CHECK1-NEXT: br label [[EH_RESUME:%.*]]
// CHECK1: eh.resume:
// CHECK1-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK1-NEXT: [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4
// CHECK1-NEXT: [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0
// CHECK1-NEXT: [[LPAD_VAL2:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1
// CHECK1-NEXT: resume { i8*, i32 } [[LPAD_VAL2]]
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-LABEL: define {{[^@]+}}@.__kmpc_global_dtor_..2
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = bitcast i8* [[TMP1]] to %class.TestClass*
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAY_BEGIN]], i64 2
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ [[TMP2]], [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done1:
// CHECK1-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-LABEL: define {{[^@]+}}@.__omp_threadprivate_init_..3
// CHECK1-SAME: () #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-NEXT: call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast ([2 x %class.TestClass]* @tc2 to i8*), i8* (i8*)* @.__kmpc_global_ctor_..1, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..2)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-LABEL: define {{[^@]+}}@__cxx_global_var_init
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-SAME: () #[[ATTR0]] {
// CHECK1-NEXT: entry:
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: call void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) @tc)
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%class.TestClass*)* @_ZN9TestClassD1Ev to void (i8*)*), i8* bitcast (%class.TestClass* @tc to i8*), i8* @__dso_handle) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-LABEL: define {{[^@]+}}@__cxx_global_var_init.4
// CHECK1-SAME: () #[[ATTR0]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK1-NEXT: br label [[ARRAYCTOR_LOOP:%.*]]
// CHECK1: arrayctor.loop:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[ARRAYCTOR_CUR:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[INVOKE_CONT:%.*]] ]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: invoke void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: to label [[INVOKE_CONT]] unwind label [[LPAD:%.*]]
// CHECK1: invoke.cont:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAYCTOR_CUR]], i64 1
// CHECK1-NEXT: [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([[CLASS_TESTCLASS]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
// CHECK1: arrayctor.cont:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]]
// CHECK1-NEXT: ret void
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1: lpad:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[TMP1:%.*]] = landingpad { i8*, i32 }
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: cleanup
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[TMP2:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 0
// CHECK1-NEXT: store i8* [[TMP2]], i8** [[EXN_SLOT]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 1
// CHECK1-NEXT: store i32 [[TMP3]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK1-NEXT: [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), [[ARRAYCTOR_CUR]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ [[ARRAYCTOR_CUR]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done1:
// CHECK1-NEXT: br label [[EH_RESUME:%.*]]
// CHECK1: eh.resume:
// CHECK1-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK1-NEXT: [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4
// CHECK1-NEXT: [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0
// CHECK1-NEXT: [[LPAD_VAL2:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1
// CHECK1-NEXT: resume { i8*, i32 } [[LPAD_VAL2]]
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([[CLASS_TESTCLASS:%.*]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done1:
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN9TestClassC2Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK1-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: store i32 0, i32* [[A]], align 4
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN9TestClassD2Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK1-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_Z3foov
// CHECK1-SAME: () #[[ATTR4:[0-9]+]] {
// CHECK1-NEXT: entry:
// CHECK1-NEXT: call void @_Z8mayThrowv()
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@main
// CHECK1-SAME: () #[[ATTR6:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[RETVAL:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[A:%.*]] = alloca i8, align 1
// CHECK1-NEXT: [[A2:%.*]] = alloca [2 x i8], align 1
// CHECK1-NEXT: [[C:%.*]] = alloca %class.TestClass*, align 8
// CHECK1-NEXT: [[SST:%.*]] = alloca [[STRUCT_SST:%.*]], align 8
// CHECK1-NEXT: [[SS:%.*]] = alloca [[STRUCT_SS:%.*]], align 8
// CHECK1-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [5 x i8*], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK1-NEXT: store i32 0, i32* [[RETVAL]], align 4
// CHECK1-NEXT: store %class.TestClass* @tc, %class.TestClass** [[C]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: call void @_ZN3SSTIdEC1Ev(%struct.SST* nonnull align 8 dereferenceable(8) [[SST]])
// CHECK1-NEXT: call void @_ZN2SSC1ERi(%struct.SS* nonnull align 8 dereferenceable(16) [[SS]], i32* nonnull align 4 dereferenceable(4) getelementptr inbounds ([[CLASS_TESTCLASS:%.*]], %class.TestClass* @tc, i32 0, i32 0))
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK1-NEXT: [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK1-NEXT: br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
// CHECK1: omp_if.then:
// CHECK1-NEXT: store i8 2, i8* [[A]], align 1
// CHECK1-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
// CHECK1-NEXT: [[TMP3:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK1-NEXT: [[TMP4:%.*]] = icmp ne i32 [[TMP3]], 0
// CHECK1-NEXT: br i1 [[TMP4]], label [[OMP_IF_THEN1:%.*]], label [[OMP_IF_END2:%.*]]
// CHECK1: omp_if.then1:
// CHECK1-NEXT: store i8 2, i8* [[A]], align 1
// CHECK1-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK1-NEXT: br label [[OMP_IF_END2]]
// CHECK1: omp_if.end2:
// CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP0]])
// CHECK1-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK1-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK1-NEXT: [[TMP6:%.*]] = icmp ne i32 [[TMP5]], 0
// CHECK1-NEXT: br i1 [[TMP6]], label [[OMP_IF_THEN3:%.*]], label [[OMP_IF_END4:%.*]]
// CHECK1: omp_if.then3:
// CHECK1-NEXT: invoke void @_Z3foov()
// CHECK1-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK1: invoke.cont:
// CHECK1-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK1-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK1-NEXT: br label [[OMP_IF_END4]]
// CHECK1: lpad:
// CHECK1-NEXT: [[TMP7:%.*]] = landingpad { i8*, i32 }
// CHECK1-NEXT: catch i8* null
// CHECK1-NEXT: [[TMP8:%.*]] = extractvalue { i8*, i32 } [[TMP7]], 0
// CHECK1-NEXT: store i8* [[TMP8]], i8** [[EXN_SLOT]], align 8
// CHECK1-NEXT: [[TMP9:%.*]] = extractvalue { i8*, i32 } [[TMP7]], 1
// CHECK1-NEXT: store i32 [[TMP9]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK1-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK1-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK1: omp_if.end4:
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK1-NEXT: store i8* [[A]], i8** [[TMP10]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1
// CHECK1-NEXT: store i8* bitcast (%class.TestClass* @tc to i8*), i8** [[TMP11]], align 8
// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 2
// CHECK1-NEXT: [[TMP13:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i8* bitcast (%class.TestClass* @tc to i8*), i64 4, i8*** @tc.cache.)
// CHECK1-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to %class.TestClass*
// CHECK1-NEXT: [[TMP15:%.*]] = bitcast %class.TestClass* [[TMP14]] to i8*
// CHECK1-NEXT: store i8* [[TMP15]], i8** [[TMP12]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 3
// CHECK1-NEXT: [[TMP17:%.*]] = bitcast [2 x i8]* [[A2]] to i8*
// CHECK1-NEXT: store i8* [[TMP17]], i8** [[TMP16]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 4
// CHECK1-NEXT: [[TMP19:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i8* bitcast ([2 x %class.TestClass]* @tc2 to i8*), i64 8, i8*** @tc2.cache.)
// CHECK1-NEXT: [[TMP20:%.*]] = bitcast i8* [[TMP19]] to [2 x %class.TestClass]*
// CHECK1-NEXT: [[TMP21:%.*]] = bitcast [2 x %class.TestClass]* [[TMP20]] to i8*
// CHECK1-NEXT: store i8* [[TMP21]], i8** [[TMP18]], align 8
// CHECK1-NEXT: [[TMP22:%.*]] = bitcast [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK1-NEXT: [[TMP23:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK1-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i64 40, i8* [[TMP22]], void (i8*, i8*)* @.omp.copyprivate.copy_func, i32 [[TMP23]])
// CHECK1-NEXT: [[TMP24:%.*]] = load i8, i8* [[A]], align 1
// CHECK1-NEXT: [[CONV:%.*]] = sext i8 [[TMP24]] to i32
// CHECK1-NEXT: ret i32 [[CONV]]
// CHECK1: terminate.handler:
// CHECK1-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK1-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13:[0-9]+]]
// CHECK1-NEXT: unreachable
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN3SSTIdEC1Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-SAME: (%struct.SST* nonnull align 8 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK1-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: call void @_ZN3SSTIdEC2Ev(%struct.SST* nonnull align 8 dereferenceable(8) [[THIS1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN2SSC1ERi
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-SAME: (%struct.SS* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32* nonnull align 4 dereferenceable(4) [[D:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK1-NEXT: [[D_ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK1-NEXT: store i32* [[D]], i32** [[D_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load i32*, i32** [[D_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: call void @_ZN2SSC2ERi(%struct.SS* nonnull align 8 dereferenceable(16) [[THIS1]], i32* nonnull align 4 dereferenceable(4) [[TMP0]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@__clang_call_terminate
// CHECK1-SAME: (i8* [[TMP0:%.*]]) #[[ATTR8:[0-9]+]] comdat {
// CHECK1-NEXT: [[TMP2:%.*]] = call i8* @__cxa_begin_catch(i8* [[TMP0]]) #[[ATTR3]]
// CHECK1-NEXT: call void @_ZSt9terminatev() #[[ATTR13]]
// CHECK1-NEXT: unreachable
//
//
// CHECK1-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func
// CHECK1-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9:[0-9]+]] {
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK1-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [5 x i8*]*
// CHECK1-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [5 x i8*]*
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK1-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK1-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8
// CHECK1-NEXT: [[TMP10:%.*]] = load i8, i8* [[TMP9]], align 1
// CHECK1-NEXT: store i8 [[TMP10]], i8* [[TMP7]], align 1
// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 1
// CHECK1-NEXT: [[TMP12:%.*]] = load i8*, i8** [[TMP11]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = bitcast i8* [[TMP12]] to %class.TestClass*
// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 1
// CHECK1-NEXT: [[TMP15:%.*]] = load i8*, i8** [[TMP14]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = bitcast i8* [[TMP15]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %class.TestClass* @_ZN9TestClassaSERKS_(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP13]], %class.TestClass* nonnull align 4 dereferenceable(4) [[TMP16]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 2
// CHECK1-NEXT: [[TMP18:%.*]] = load i8*, i8** [[TMP17]], align 8
// CHECK1-NEXT: [[TMP19:%.*]] = bitcast i8* [[TMP18]] to %class.TestClass*
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 2
// CHECK1-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8
// CHECK1-NEXT: [[TMP22:%.*]] = bitcast i8* [[TMP21]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: [[CALL2:%.*]] = call nonnull align 4 dereferenceable(4) %class.TestClass* @_ZN9TestClassaSERKS_(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP19]], %class.TestClass* nonnull align 4 dereferenceable(4) [[TMP22]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 3
// CHECK1-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8
// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 3
// CHECK1-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 8
// CHECK1-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 1 [[TMP24]], i8* align 1 [[TMP26]], i64 2, i1 false)
// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 4
// CHECK1-NEXT: [[TMP28:%.*]] = load i8*, i8** [[TMP27]], align 8
// CHECK1-NEXT: [[TMP29:%.*]] = bitcast i8* [[TMP28]] to %class.TestClass*
// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 4
// CHECK1-NEXT: [[TMP31:%.*]] = load i8*, i8** [[TMP30]], align 8
// CHECK1-NEXT: [[TMP32:%.*]] = bitcast i8* [[TMP31]] to %class.TestClass*
// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[TMP29]], i64 2
// CHECK1-NEXT: [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq %class.TestClass* [[TMP29]], [[TMP33]]
// CHECK1-NEXT: br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
// CHECK1: omp.arraycpy.body:
// CHECK1-NEXT: [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %class.TestClass* [ [[TMP32]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
// CHECK1-NEXT: [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %class.TestClass* [ [[TMP29]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: [[CALL3:%.*]] = call nonnull align 4 dereferenceable(4) %class.TestClass* @_ZN9TestClassaSERKS_(%class.TestClass* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %class.TestClass* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[CLASS_TESTCLASS]], %class.TestClass* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
// CHECK1-NEXT: [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[CLASS_TESTCLASS]], %class.TestClass* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
// CHECK1-NEXT: [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %class.TestClass* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP33]]
// CHECK1-NEXT: br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE4]], label [[OMP_ARRAYCPY_BODY]]
// CHECK1: omp.arraycpy.done4:
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN9TestClassaSERKS_
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]], %class.TestClass* nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) #[[ATTR10:[0-9]+]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK1-NEXT: [[DOTADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK1-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK1-NEXT: store %class.TestClass* [[TMP0]], %class.TestClass** [[DOTADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
// CHECK1-NEXT: ret %class.TestClass* [[THIS1]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN3SSTIdEC2Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-SAME: (%struct.SST* nonnull align 8 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK1-NEXT: [[A2:%.*]] = alloca double*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SST:%.*]], %struct.SST* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: store double 0.000000e+00, double* [[A]], align 8
// CHECK1-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SST]], %struct.SST* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: store double* [[A3]], double** [[A2]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP0:%.*]] = load double*, double** [[A2]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load double, double* [[TMP0]], align 8
// CHECK1-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to double*
// CHECK1-NEXT: store double [[TMP1]], double* [[CONV]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK1-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SST*, i64)* @.omp_outlined. to void (i32*, i32*, ...)*), %struct.SST* [[THIS1]], i64 [[TMP2]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@.omp_outlined.
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SST* [[THIS:%.*]], i64 [[A:%.*]]) #[[ATTR12:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[TMP:%.*]] = alloca double*, align 8
// CHECK1-NEXT: [[_TMP1:%.*]] = alloca double*, align 8
// CHECK1-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON:%.*]], align 8
// CHECK1-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK1-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK1-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to double*
// CHECK1-NEXT: store double* [[CONV]], double** [[TMP]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load double*, double** [[TMP]], align 8
// CHECK1-NEXT: store double* [[TMP1]], double** [[_TMP1]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP2:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4
// CHECK1-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
// CHECK1-NEXT: [[TMP5:%.*]] = icmp ne i32 [[TMP4]], 0
// CHECK1-NEXT: br i1 [[TMP5]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 0
// CHECK1-NEXT: store %struct.SST* [[TMP0]], %struct.SST** [[TMP6]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 1
// CHECK1-NEXT: [[TMP8:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK1-NEXT: store double* [[TMP8]], double** [[TMP7]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: invoke void @_ZZN3SSTIdEC1EvENKUlvE_clEv(%class.anon* nonnull align 8 dereferenceable(16) [[REF_TMP]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK1: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: lpad:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP9:%.*]] = landingpad { i8*, i32 }
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: catch i8* null
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP10:%.*]] = extractvalue { i8*, i32 } [[TMP9]], 0
// CHECK1-NEXT: store i8* [[TMP10]], i8** [[EXN_SLOT]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = extractvalue { i8*, i32 } [[TMP9]], 1
// CHECK1-NEXT: store i32 [[TMP11]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK1-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK1: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK1-NEXT: [[TMP13:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK1-NEXT: [[TMP14:%.*]] = bitcast double* [[TMP13]] to i8*
// CHECK1-NEXT: store i8* [[TMP14]], i8** [[TMP12]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = bitcast [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK1-NEXT: [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK1-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]], i64 8, i8* [[TMP15]], void (i8*, i8*)* @.omp.copyprivate.copy_func.5, i32 [[TMP16]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
// CHECK1: terminate.handler:
// CHECK1-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK1-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13]]
// CHECK1-NEXT: unreachable
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZZN3SSTIdEC1EvENKUlvE_clEv
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-SAME: (%class.anon* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR4]] align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %class.anon*, align 8
// CHECK1-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON_0:%.*]], align 8
// CHECK1-NEXT: store %class.anon* [[THIS]], %class.anon** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %class.anon*, %class.anon** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON:%.*]], %class.anon* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[TMP1:%.*]] = load %struct.SST*, %struct.SST** [[TMP0]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 0
// CHECK1-NEXT: store %struct.SST* [[TMP1]], %struct.SST** [[TMP2]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 1
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[THIS1]], i32 0, i32 1
// CHECK1-NEXT: [[TMP5:%.*]] = load double*, double** [[TMP4]], align 8
// CHECK1-NEXT: store double* [[TMP5]], double** [[TMP3]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-NEXT: call void @_ZZZN3SSTIdEC1EvENKUlvE_clEvENKUlvE_clEv(%class.anon.0* nonnull align 8 dereferenceable(16) [[REF_TMP]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.5
// CHECK1-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK1-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [1 x i8*]*
// CHECK1-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK1-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to double*
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK1-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to double*
// CHECK1-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 8
// CHECK1-NEXT: store double [[TMP12]], double* [[TMP8]], align 8
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZZZN3SSTIdEC1EvENKUlvE_clEvENKUlvE_clEv
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-SAME: (%class.anon.0* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR10]] align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %class.anon.0*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: store %class.anon.0* [[THIS]], %class.anon.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %class.anon.0*, %class.anon.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON_0:%.*]], %class.anon.0* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[TMP1:%.*]] = load %struct.SST*, %struct.SST** [[TMP0]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[THIS1]], i32 0, i32 1
// CHECK1-NEXT: [[TMP3:%.*]] = load double*, double** [[TMP2]], align 8
// CHECK1-NEXT: [[TMP4:%.*]] = load double, double* [[TMP3]], align 8
// CHECK1-NEXT: [[INC:%.*]] = fadd double [[TMP4]], 1.000000e+00
// CHECK1-NEXT: store double [[INC]], double* [[TMP3]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[THIS1]], i32 0, i32 1
// CHECK1-NEXT: [[TMP6:%.*]] = load double*, double** [[TMP5]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = load double, double* [[TMP6]], align 8
// CHECK1-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to double*
// CHECK1-NEXT: store double [[TMP7]], double* [[CONV]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK1-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SST*, i64)* @.omp_outlined..6 to void (i32*, i32*, ...)*), %struct.SST* [[TMP1]], i64 [[TMP8]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@.omp_outlined..6
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SST* [[THIS:%.*]], i64 [[A:%.*]]) #[[ATTR12]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[TMP:%.*]] = alloca double*, align 8
// CHECK1-NEXT: [[_TMP1:%.*]] = alloca double*, align 8
// CHECK1-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK1-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK1-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to double*
// CHECK1-NEXT: store double* [[CONV]], double** [[TMP]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load double*, double** [[TMP]], align 8
// CHECK1-NEXT: store double* [[TMP1]], double** [[_TMP1]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP2:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4
// CHECK1-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
// CHECK1-NEXT: [[TMP5:%.*]] = icmp ne i32 [[TMP4]], 0
// CHECK1-NEXT: br i1 [[TMP5]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP6:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = load double, double* [[TMP6]], align 8
// CHECK1-NEXT: [[INC:%.*]] = fadd double [[TMP7]], 1.000000e+00
// CHECK1-NEXT: store double [[INC]], double* [[TMP6]], align 8
// CHECK1-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK1-NEXT: [[TMP9:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK1-NEXT: [[TMP10:%.*]] = bitcast double* [[TMP9]] to i8*
// CHECK1-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = bitcast [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK1-NEXT: [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK1-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]], i64 8, i8* [[TMP11]], void (i8*, i8*)* @.omp.copyprivate.copy_func.7, i32 [[TMP12]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.7
// CHECK1-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK1-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [1 x i8*]*
// CHECK1-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK1-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to double*
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK1-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to double*
// CHECK1-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 8
// CHECK1-NEXT: store double [[TMP12]], double* [[TMP8]], align 8
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN2SSC2ERi
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK1-SAME: (%struct.SS* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32* nonnull align 4 dereferenceable(4) [[D:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK1-NEXT: [[D_ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT: [[A2:%.*]] = alloca i32*, align 8
// CHECK1-NEXT: [[B4:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[C7:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[B_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[C_CASTED:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK1-NEXT: store i32* [[D]], i32** [[D_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], %struct.SS* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: store i32 0, i32* [[A]], align 8
// CHECK1-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 1
// CHECK1-NEXT: [[BF_LOAD:%.*]] = load i8, i8* [[B]], align 4
// CHECK1-NEXT: [[BF_CLEAR:%.*]] = and i8 [[BF_LOAD]], -16
// CHECK1-NEXT: store i8 [[BF_CLEAR]], i8* [[B]], align 4
// CHECK1-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 2
// CHECK1-NEXT: [[TMP0:%.*]] = load i32*, i32** [[D_ADDR]], align 8
// CHECK1-NEXT: store i32* [[TMP0]], i32** [[C]], align 8
// CHECK1-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: store i32* [[A3]], i32** [[A2]], align 8
// CHECK1-NEXT: [[B5:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 1
// CHECK1-NEXT: [[BF_LOAD6:%.*]] = load i8, i8* [[B5]], align 4
// CHECK1-NEXT: [[BF_SHL:%.*]] = shl i8 [[BF_LOAD6]], 4
// CHECK1-NEXT: [[BF_ASHR:%.*]] = ashr i8 [[BF_SHL]], 4
// CHECK1-NEXT: [[BF_CAST:%.*]] = sext i8 [[BF_ASHR]] to i32
// CHECK1-NEXT: store i32 [[BF_CAST]], i32* [[B4]], align 4
// CHECK1-NEXT: [[C8:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 2
// CHECK1-NEXT: [[TMP1:%.*]] = load i32*, i32** [[C8]], align 8
// CHECK1-NEXT: store i32* [[TMP1]], i32** [[C7]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP2:%.*]] = load i32*, i32** [[A2]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4
// CHECK1-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to i32*
// CHECK1-NEXT: store i32 [[TMP3]], i32* [[CONV]], align 4
// CHECK1-NEXT: [[TMP4:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = load i32, i32* [[B4]], align 4
// CHECK1-NEXT: [[CONV9:%.*]] = bitcast i64* [[B_CASTED]] to i32*
// CHECK1-NEXT: store i32 [[TMP5]], i32* [[CONV9]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = load i64, i64* [[B_CASTED]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = load i32*, i32** [[C7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK1-NEXT: [[CONV10:%.*]] = bitcast i64* [[C_CASTED]] to i32*
// CHECK1-NEXT: store i32 [[TMP8]], i32* [[CONV10]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = load i64, i64* [[C_CASTED]], align 8
// CHECK1-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SS*, i64, i64, i64)* @.omp_outlined..8 to void (i32*, i32*, ...)*), %struct.SS* [[THIS1]], i64 [[TMP4]], i64 [[TMP6]], i64 [[TMP9]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@.omp_outlined..8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SS* [[THIS:%.*]], i64 [[A:%.*]], i64 [[B:%.*]], i64 [[C:%.*]]) #[[ATTR12]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[C_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[TMP:%.*]] = alloca i32*, align 8
// CHECK1-NEXT: [[_TMP3:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[_TMP4:%.*]] = alloca i32*, align 8
// CHECK1-NEXT: [[_TMP5:%.*]] = alloca i32*, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON_1:%.*]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [3 x i8*], align 8
// CHECK1-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK1-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK1-NEXT: store i64 [[B]], i64* [[B_ADDR]], align 8
// CHECK1-NEXT: store i64 [[C]], i64* [[C_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to i32*
// CHECK1-NEXT: [[CONV1:%.*]] = bitcast i64* [[B_ADDR]] to i32*
// CHECK1-NEXT: [[CONV2:%.*]] = bitcast i64* [[C_ADDR]] to i32*
// CHECK1-NEXT: store i32* [[CONV]], i32** [[TMP]], align 8
// CHECK1-NEXT: store i32* [[CONV2]], i32** [[_TMP3]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load i32*, i32** [[TMP]], align 8
// CHECK1-NEXT: store i32* [[TMP1]], i32** [[_TMP4]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load i32*, i32** [[_TMP3]], align 8
// CHECK1-NEXT: store i32* [[TMP2]], i32** [[_TMP5]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK1-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK1-NEXT: [[TMP6:%.*]] = icmp ne i32 [[TMP5]], 0
// CHECK1-NEXT: br i1 [[TMP6]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 0
// CHECK1-NEXT: store %struct.SS* [[TMP0]], %struct.SS** [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 1
// CHECK1-NEXT: [[TMP9:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK1-NEXT: store i32* [[TMP9]], i32** [[TMP8]], align 8
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 2
// CHECK1-NEXT: store i32* [[CONV1]], i32** [[TMP10]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 3
// CHECK1-NEXT: [[TMP12:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK1-NEXT: store i32* [[TMP12]], i32** [[TMP11]], align 8
// CHECK1-NEXT: invoke void @_ZZN2SSC1ERiENKUlvE_clEv(%class.anon.1* nonnull align 8 dereferenceable(32) [[REF_TMP]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK1: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: lpad:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP13:%.*]] = landingpad { i8*, i32 }
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: catch i8* null
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP14:%.*]] = extractvalue { i8*, i32 } [[TMP13]], 0
// CHECK1-NEXT: store i8* [[TMP14]], i8** [[EXN_SLOT]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = extractvalue { i8*, i32 } [[TMP13]], 1
// CHECK1-NEXT: store i32 [[TMP15]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK1-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK1: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK1-NEXT: [[TMP17:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = bitcast i32* [[TMP17]] to i8*
// CHECK1-NEXT: store i8* [[TMP18]], i8** [[TMP16]], align 8
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1
// CHECK1-NEXT: [[TMP20:%.*]] = bitcast i32* [[CONV1]] to i8*
// CHECK1-NEXT: store i8* [[TMP20]], i8** [[TMP19]], align 8
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 2
// CHECK1-NEXT: [[TMP22:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK1-NEXT: [[TMP23:%.*]] = bitcast i32* [[TMP22]] to i8*
// CHECK1-NEXT: store i8* [[TMP23]], i8** [[TMP21]], align 8
// CHECK1-NEXT: [[TMP24:%.*]] = bitcast [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK1-NEXT: [[TMP25:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK1-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]], i64 24, i8* [[TMP24]], void (i8*, i8*)* @.omp.copyprivate.copy_func.9, i32 [[TMP25]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
// CHECK1: terminate.handler:
// CHECK1-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK1-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13]]
// CHECK1-NEXT: unreachable
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZZN2SSC1ERiENKUlvE_clEv
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-SAME: (%class.anon.1* nonnull align 8 dereferenceable(32) [[THIS:%.*]]) #[[ATTR10]] align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %class.anon.1*, align 8
// CHECK1-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[B_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[C_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT: store %class.anon.1* [[THIS]], %class.anon.1** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %class.anon.1*, %class.anon.1** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON_1:%.*]], %class.anon.1* [[THIS1]], i32 0, i32 0
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[TMP1:%.*]] = load %struct.SS*, %struct.SS** [[TMP0]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 1
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[TMP3:%.*]] = load i32*, i32** [[TMP2]], align 8
// CHECK1-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK1-NEXT: [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// CHECK1-NEXT: store i32 [[INC]], i32* [[TMP3]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 2
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[TMP6:%.*]] = load i32*, i32** [[TMP5]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4
// CHECK1-NEXT: [[DEC:%.*]] = add nsw i32 [[TMP7]], -1
// CHECK1-NEXT: store i32 [[DEC]], i32* [[TMP6]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 3
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[TMP9:%.*]] = load i32*, i32** [[TMP8]], align 8
// CHECK1-NEXT: [[TMP10:%.*]] = load i32, i32* [[TMP9]], align 4
// CHECK1-NEXT: [[DIV:%.*]] = sdiv i32 [[TMP10]], 1
// CHECK1-NEXT: store i32 [[DIV]], i32* [[TMP9]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 1
// CHECK1-NEXT: [[TMP12:%.*]] = load i32*, i32** [[TMP11]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = load i32, i32* [[TMP12]], align 4
// CHECK1-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to i32*
// CHECK1-NEXT: store i32 [[TMP13]], i32* [[CONV]], align 4
// CHECK1-NEXT: [[TMP14:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 2
// CHECK1-NEXT: [[TMP16:%.*]] = load i32*, i32** [[TMP15]], align 8
// CHECK1-NEXT: [[TMP17:%.*]] = load i32, i32* [[TMP16]], align 4
// CHECK1-NEXT: [[CONV2:%.*]] = bitcast i64* [[B_CASTED]] to i32*
// CHECK1-NEXT: store i32 [[TMP17]], i32* [[CONV2]], align 4
// CHECK1-NEXT: [[TMP18:%.*]] = load i64, i64* [[B_CASTED]], align 8
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 3
// CHECK1-NEXT: [[TMP20:%.*]] = load i32*, i32** [[TMP19]], align 8
// CHECK1-NEXT: [[TMP21:%.*]] = load i32, i32* [[TMP20]], align 4
// CHECK1-NEXT: [[CONV3:%.*]] = bitcast i64* [[C_CASTED]] to i32*
// CHECK1-NEXT: store i32 [[TMP21]], i32* [[CONV3]], align 4
// CHECK1-NEXT: [[TMP22:%.*]] = load i64, i64* [[C_CASTED]], align 8
// CHECK1-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SS*, i64, i64, i64)* @.omp_outlined..10 to void (i32*, i32*, ...)*), %struct.SS* [[TMP1]], i64 [[TMP14]], i64 [[TMP18]], i64 [[TMP22]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.9
// CHECK1-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK1-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [3 x i8*]*
// CHECK1-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [3 x i8*]*
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK1-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to i32*
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK1-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to i32*
// CHECK1-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
// CHECK1-NEXT: store i32 [[TMP12]], i32* [[TMP8]], align 4
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 1
// CHECK1-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i32*
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 1
// CHECK1-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32*
// CHECK1-NEXT: [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
// CHECK1-NEXT: store i32 [[TMP19]], i32* [[TMP15]], align 4
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 2
// CHECK1-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8
// CHECK1-NEXT: [[TMP22:%.*]] = bitcast i8* [[TMP21]] to i32*
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 2
// CHECK1-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8
// CHECK1-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32*
// CHECK1-NEXT: [[TMP26:%.*]] = load i32, i32* [[TMP25]], align 4
// CHECK1-NEXT: store i32 [[TMP26]], i32* [[TMP22]], align 4
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@.omp_outlined..10
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SS* [[THIS:%.*]], i64 [[A:%.*]], i64 [[B:%.*]], i64 [[C:%.*]]) #[[ATTR12]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[C_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[TMP:%.*]] = alloca i32*, align 8
// CHECK1-NEXT: [[_TMP3:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[_TMP4:%.*]] = alloca i32*, align 8
// CHECK1-NEXT: [[_TMP5:%.*]] = alloca i32*, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [3 x i8*], align 8
// CHECK1-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK1-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK1-NEXT: store i64 [[B]], i64* [[B_ADDR]], align 8
// CHECK1-NEXT: store i64 [[C]], i64* [[C_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to i32*
// CHECK1-NEXT: [[CONV1:%.*]] = bitcast i64* [[B_ADDR]] to i32*
// CHECK1-NEXT: [[CONV2:%.*]] = bitcast i64* [[C_ADDR]] to i32*
// CHECK1-NEXT: store i32* [[CONV]], i32** [[TMP]], align 8
// CHECK1-NEXT: store i32* [[CONV2]], i32** [[_TMP3]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load i32*, i32** [[TMP]], align 8
// CHECK1-NEXT: store i32* [[TMP1]], i32** [[_TMP4]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load i32*, i32** [[_TMP3]], align 8
// CHECK1-NEXT: store i32* [[TMP2]], i32** [[_TMP5]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK1-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK1-NEXT: [[TMP6:%.*]] = icmp ne i32 [[TMP5]], 0
// CHECK1-NEXT: br i1 [[TMP6]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP7:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK1-NEXT: [[INC:%.*]] = add nsw i32 [[TMP8]], 1
// CHECK1-NEXT: store i32 [[INC]], i32* [[TMP7]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK1-NEXT: [[DEC:%.*]] = add nsw i32 [[TMP9]], -1
// CHECK1-NEXT: store i32 [[DEC]], i32* [[CONV1]], align 8
// CHECK1-NEXT: [[TMP10:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK1-NEXT: [[DIV:%.*]] = sdiv i32 [[TMP11]], 1
// CHECK1-NEXT: store i32 [[DIV]], i32* [[TMP10]], align 4
// CHECK1-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK1-NEXT: [[TMP13:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK1-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8*
// CHECK1-NEXT: store i8* [[TMP14]], i8** [[TMP12]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1
// CHECK1-NEXT: [[TMP16:%.*]] = bitcast i32* [[CONV1]] to i8*
// CHECK1-NEXT: store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 2
// CHECK1-NEXT: [[TMP18:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK1-NEXT: [[TMP19:%.*]] = bitcast i32* [[TMP18]] to i8*
// CHECK1-NEXT: store i8* [[TMP19]], i8** [[TMP17]], align 8
// CHECK1-NEXT: [[TMP20:%.*]] = bitcast [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK1-NEXT: [[TMP21:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK1-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]], i64 24, i8* [[TMP20]], void (i8*, i8*)* @.omp.copyprivate.copy_func.11, i32 [[TMP21]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.11
// CHECK1-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK1-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [3 x i8*]*
// CHECK1-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [3 x i8*]*
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK1-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to i32*
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK1-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to i32*
// CHECK1-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
// CHECK1-NEXT: store i32 [[TMP12]], i32* [[TMP8]], align 4
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 1
// CHECK1-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i32*
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 1
// CHECK1-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32*
// CHECK1-NEXT: [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
// CHECK1-NEXT: store i32 [[TMP19]], i32* [[TMP15]], align 4
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 2
// CHECK1-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8
// CHECK1-NEXT: [[TMP22:%.*]] = bitcast i8* [[TMP21]] to i32*
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 2
// CHECK1-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8
// CHECK1-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32*
// CHECK1-NEXT: [[TMP26:%.*]] = load i32, i32* [[TMP25]], align 4
// CHECK1-NEXT: store i32 [[TMP26]], i32* [[TMP22]], align 4
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_Z15parallel_singlev
// CHECK1-SAME: () #[[ATTR10]] {
// CHECK1-NEXT: entry:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* @.omp_outlined..12 to void (i32*, i32*, ...)*))
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@.omp_outlined..12
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]]) #[[ATTR12]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK1-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP0:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
// CHECK1-NEXT: [[TMP3:%.*]] = icmp ne i32 [[TMP2]], 0
// CHECK1-NEXT: br i1 [[TMP3]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1: omp_if.then:
// CHECK1-NEXT: invoke void @_Z3foov()
// CHECK1-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK1: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: lpad:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP4:%.*]] = landingpad { i8*, i32 }
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: catch i8* null
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: [[TMP5:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 0
// CHECK1-NEXT: store i8* [[TMP5]], i8** [[EXN_SLOT]], align 8
// CHECK1-NEXT: [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 1
// CHECK1-NEXT: store i32 [[TMP6]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK1-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK1: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
// CHECK1: terminate.handler:
// CHECK1-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK1-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13]]
// CHECK1-NEXT: unreachable
//
//
// CHECK1-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_single_codegen.cpp
// CHECK1-SAME: () #[[ATTR0]] {
// CHECK1-NEXT: entry:
// CHECK1-NEXT: call void @__cxx_global_var_init()
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: call void @__cxx_global_var_init.4()
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: call void @.__omp_threadprivate_init_.()
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK1-NEXT: call void @.__omp_threadprivate_init_..3()
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK1-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__cxx_global_var_init
// CHECK2-SAME: () #[[ATTR0:[0-9]+]] {
// CHECK2-NEXT: entry:
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: call void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) @tc)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%class.TestClass*)* @_ZN9TestClassD1Ev to void (i8*)*), i8* bitcast (%class.TestClass* @tc to i8*), i8* @__dso_handle) #[[ATTR3:[0-9]+]]
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN9TestClassC1Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK2-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: call void @_ZN9TestClassC2Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN9TestClassD1Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2:[0-9]+]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK2-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: call void @_ZN9TestClassD2Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__cxx_global_var_init.1
// CHECK2-SAME: () #[[ATTR0]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK2-NEXT: br label [[ARRAYCTOR_LOOP:%.*]]
// CHECK2: arrayctor.loop:
// CHECK2-NEXT: [[ARRAYCTOR_CUR:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[INVOKE_CONT:%.*]] ]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: invoke void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: to label [[INVOKE_CONT]] unwind label [[LPAD:%.*]]
// CHECK2: invoke.cont:
// CHECK2-NEXT: [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAYCTOR_CUR]], i64 1
// CHECK2-NEXT: [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([[CLASS_TESTCLASS]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2)
// CHECK2-NEXT: br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
// CHECK2: arrayctor.cont:
// CHECK2-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]]
// CHECK2-NEXT: ret void
// CHECK2: lpad:
// CHECK2-NEXT: [[TMP1:%.*]] = landingpad { i8*, i32 }
// CHECK2-NEXT: cleanup
// CHECK2-NEXT: [[TMP2:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 0
// CHECK2-NEXT: store i8* [[TMP2]], i8** [[EXN_SLOT]], align 8
// CHECK2-NEXT: [[TMP3:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 1
// CHECK2-NEXT: store i32 [[TMP3]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK2-NEXT: [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), [[ARRAYCTOR_CUR]]
// CHECK2-NEXT: br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY:%.*]]
// CHECK2: arraydestroy.body:
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ [[ARRAYCTOR_CUR]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0)
// CHECK2-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1]], label [[ARRAYDESTROY_BODY]]
// CHECK2: arraydestroy.done1:
// CHECK2-NEXT: br label [[EH_RESUME:%.*]]
// CHECK2: eh.resume:
// CHECK2-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK2-NEXT: [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4
// CHECK2-NEXT: [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0
// CHECK2-NEXT: [[LPAD_VAL2:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1
// CHECK2-NEXT: resume { i8*, i32 } [[LPAD_VAL2]]
//
//
// CHECK2-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
// CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK2-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK2: arraydestroy.body:
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([[CLASS_TESTCLASS:%.*]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0)
// CHECK2-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK2: arraydestroy.done1:
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_.
// CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK2-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK2-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: call void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP2]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK2-NEXT: ret i8* [[TMP3]]
//
//
// CHECK2-LABEL: define {{[^@]+}}@.__kmpc_global_dtor_.
// CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK2-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK2-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP2]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@.__omp_threadprivate_init_.
// CHECK2-SAME: () #[[ATTR0]] {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
// CHECK2-NEXT: call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast (%class.TestClass* @tc to i8*), i8* (i8*)* @.__kmpc_global_ctor_., i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_.)
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_..2
// CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK2-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK2-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to [2 x %class.TestClass]*
// CHECK2-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %class.TestClass], [2 x %class.TestClass]* [[TMP2]], i32 0, i32 0
// CHECK2-NEXT: [[ARRAYCTOR_END:%.*]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAY_BEGIN]], i64 2
// CHECK2-NEXT: br label [[ARRAYCTOR_LOOP:%.*]]
// CHECK2: arrayctor.loop:
// CHECK2-NEXT: [[ARRAYCTOR_CUR:%.*]] = phi %class.TestClass* [ [[ARRAY_BEGIN]], [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[INVOKE_CONT:%.*]] ]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: invoke void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: to label [[INVOKE_CONT]] unwind label [[LPAD:%.*]]
// CHECK2: invoke.cont:
// CHECK2-NEXT: [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYCTOR_CUR]], i64 1
// CHECK2-NEXT: [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], [[ARRAYCTOR_END]]
// CHECK2-NEXT: br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
// CHECK2: arrayctor.cont:
// CHECK2-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK2-NEXT: ret i8* [[TMP3]]
// CHECK2: lpad:
// CHECK2-NEXT: [[TMP4:%.*]] = landingpad { i8*, i32 }
// CHECK2-NEXT: cleanup
// CHECK2-NEXT: [[TMP5:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 0
// CHECK2-NEXT: store i8* [[TMP5]], i8** [[EXN_SLOT]], align 8
// CHECK2-NEXT: [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 1
// CHECK2-NEXT: store i32 [[TMP6]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK2-NEXT: [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %class.TestClass* [[ARRAY_BEGIN]], [[ARRAYCTOR_CUR]]
// CHECK2-NEXT: br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY:%.*]]
// CHECK2: arraydestroy.body:
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ [[ARRAYCTOR_CUR]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK2-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1]], label [[ARRAYDESTROY_BODY]]
// CHECK2: arraydestroy.done1:
// CHECK2-NEXT: br label [[EH_RESUME:%.*]]
// CHECK2: eh.resume:
// CHECK2-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK2-NEXT: [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4
// CHECK2-NEXT: [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0
// CHECK2-NEXT: [[LPAD_VAL2:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1
// CHECK2-NEXT: resume { i8*, i32 } [[LPAD_VAL2]]
//
//
// CHECK2-LABEL: define {{[^@]+}}@.__kmpc_global_dtor_..3
// CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK2-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK2-NEXT: [[ARRAY_BEGIN:%.*]] = bitcast i8* [[TMP1]] to %class.TestClass*
// CHECK2-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAY_BEGIN]], i64 2
// CHECK2-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK2: arraydestroy.body:
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ [[TMP2]], [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK2-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK2: arraydestroy.done1:
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@.__omp_threadprivate_init_..4
// CHECK2-SAME: () #[[ATTR0]] {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-NEXT: call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast ([2 x %class.TestClass]* @tc2 to i8*), i8* (i8*)* @.__kmpc_global_ctor_..2, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..3)
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_Z3foov
// CHECK2-SAME: () #[[ATTR4:[0-9]+]] {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: call void @_Z8mayThrowv()
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@main
// CHECK2-SAME: () #[[ATTR6:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[RETVAL:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[A:%.*]] = alloca i8, align 1
// CHECK2-NEXT: [[A2:%.*]] = alloca [2 x i8], align 1
// CHECK2-NEXT: [[C:%.*]] = alloca %class.TestClass*, align 8
// CHECK2-NEXT: [[SST:%.*]] = alloca [[STRUCT_SST:%.*]], align 8
// CHECK2-NEXT: [[SS:%.*]] = alloca [[STRUCT_SS:%.*]], align 8
// CHECK2-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [5 x i8*], align 8
// CHECK2-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK2-NEXT: store i32 0, i32* [[RETVAL]], align 4
// CHECK2-NEXT: store %class.TestClass* @tc, %class.TestClass** [[C]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: call void @_ZN3SSTIdEC1Ev(%struct.SST* nonnull align 8 dereferenceable(8) [[SST]])
// CHECK2-NEXT: call void @_ZN2SSC1ERi(%struct.SS* nonnull align 8 dereferenceable(16) [[SS]], i32* nonnull align 4 dereferenceable(4) getelementptr inbounds ([[CLASS_TESTCLASS:%.*]], %class.TestClass* @tc, i32 0, i32 0))
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT: [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK2-NEXT: br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
// CHECK2: omp_if.then:
// CHECK2-NEXT: store i8 2, i8* [[A]], align 1
// CHECK2-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT: br label [[OMP_IF_END]]
// CHECK2: omp_if.end:
// CHECK2-NEXT: [[TMP3:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT: [[TMP4:%.*]] = icmp ne i32 [[TMP3]], 0
// CHECK2-NEXT: br i1 [[TMP4]], label [[OMP_IF_THEN1:%.*]], label [[OMP_IF_END2:%.*]]
// CHECK2: omp_if.then1:
// CHECK2-NEXT: store i8 2, i8* [[A]], align 1
// CHECK2-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT: br label [[OMP_IF_END2]]
// CHECK2: omp_if.end2:
// CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP0]])
// CHECK2-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK2-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT: [[TMP6:%.*]] = icmp ne i32 [[TMP5]], 0
// CHECK2-NEXT: br i1 [[TMP6]], label [[OMP_IF_THEN3:%.*]], label [[OMP_IF_END4:%.*]]
// CHECK2: omp_if.then3:
// CHECK2-NEXT: invoke void @_Z3foov()
// CHECK2-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK2: invoke.cont:
// CHECK2-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK2-NEXT: br label [[OMP_IF_END4]]
// CHECK2: lpad:
// CHECK2-NEXT: [[TMP7:%.*]] = landingpad { i8*, i32 }
// CHECK2-NEXT: catch i8* null
// CHECK2-NEXT: [[TMP8:%.*]] = extractvalue { i8*, i32 } [[TMP7]], 0
// CHECK2-NEXT: store i8* [[TMP8]], i8** [[EXN_SLOT]], align 8
// CHECK2-NEXT: [[TMP9:%.*]] = extractvalue { i8*, i32 } [[TMP7]], 1
// CHECK2-NEXT: store i32 [[TMP9]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK2-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK2-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK2: omp_if.end4:
// CHECK2-NEXT: [[TMP10:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK2-NEXT: store i8* [[A]], i8** [[TMP10]], align 8
// CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1
// CHECK2-NEXT: store i8* bitcast (%class.TestClass* @tc to i8*), i8** [[TMP11]], align 8
// CHECK2-NEXT: [[TMP12:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 2
// CHECK2-NEXT: [[TMP13:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i8* bitcast (%class.TestClass* @tc to i8*), i64 4, i8*** @tc.cache.)
// CHECK2-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to %class.TestClass*
// CHECK2-NEXT: [[TMP15:%.*]] = bitcast %class.TestClass* [[TMP14]] to i8*
// CHECK2-NEXT: store i8* [[TMP15]], i8** [[TMP12]], align 8
// CHECK2-NEXT: [[TMP16:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 3
// CHECK2-NEXT: [[TMP17:%.*]] = bitcast [2 x i8]* [[A2]] to i8*
// CHECK2-NEXT: store i8* [[TMP17]], i8** [[TMP16]], align 8
// CHECK2-NEXT: [[TMP18:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 4
// CHECK2-NEXT: [[TMP19:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i8* bitcast ([2 x %class.TestClass]* @tc2 to i8*), i64 8, i8*** @tc2.cache.)
// CHECK2-NEXT: [[TMP20:%.*]] = bitcast i8* [[TMP19]] to [2 x %class.TestClass]*
// CHECK2-NEXT: [[TMP21:%.*]] = bitcast [2 x %class.TestClass]* [[TMP20]] to i8*
// CHECK2-NEXT: store i8* [[TMP21]], i8** [[TMP18]], align 8
// CHECK2-NEXT: [[TMP22:%.*]] = bitcast [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK2-NEXT: [[TMP23:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK2-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i64 40, i8* [[TMP22]], void (i8*, i8*)* @.omp.copyprivate.copy_func, i32 [[TMP23]])
// CHECK2-NEXT: [[TMP24:%.*]] = load i8, i8* [[A]], align 1
// CHECK2-NEXT: [[CONV:%.*]] = sext i8 [[TMP24]] to i32
// CHECK2-NEXT: ret i32 [[CONV]]
// CHECK2: terminate.handler:
// CHECK2-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK2-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13:[0-9]+]]
// CHECK2-NEXT: unreachable
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN3SSTIdEC1Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-SAME: (%struct.SST* nonnull align 8 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK2-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: call void @_ZN3SSTIdEC2Ev(%struct.SST* nonnull align 8 dereferenceable(8) [[THIS1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN2SSC1ERi
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-SAME: (%struct.SS* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32* nonnull align 4 dereferenceable(4) [[D:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK2-NEXT: [[D_ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK2-NEXT: store i32* [[D]], i32** [[D_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[TMP0:%.*]] = load i32*, i32** [[D_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: call void @_ZN2SSC2ERi(%struct.SS* nonnull align 8 dereferenceable(16) [[THIS1]], i32* nonnull align 4 dereferenceable(4) [[TMP0]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__clang_call_terminate
// CHECK2-SAME: (i8* [[TMP0:%.*]]) #[[ATTR8:[0-9]+]] comdat {
// CHECK2-NEXT: [[TMP2:%.*]] = call i8* @__cxa_begin_catch(i8* [[TMP0]]) #[[ATTR3]]
// CHECK2-NEXT: call void @_ZSt9terminatev() #[[ATTR13]]
// CHECK2-NEXT: unreachable
//
//
// CHECK2-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func
// CHECK2-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9:[0-9]+]] {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK2-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK2-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK2-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [5 x i8*]*
// CHECK2-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK2-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [5 x i8*]*
// CHECK2-NEXT: [[TMP6:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK2-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK2-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8
// CHECK2-NEXT: [[TMP10:%.*]] = load i8, i8* [[TMP9]], align 1
// CHECK2-NEXT: store i8 [[TMP10]], i8* [[TMP7]], align 1
// CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 1
// CHECK2-NEXT: [[TMP12:%.*]] = load i8*, i8** [[TMP11]], align 8
// CHECK2-NEXT: [[TMP13:%.*]] = bitcast i8* [[TMP12]] to %class.TestClass*
// CHECK2-NEXT: [[TMP14:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 1
// CHECK2-NEXT: [[TMP15:%.*]] = load i8*, i8** [[TMP14]], align 8
// CHECK2-NEXT: [[TMP16:%.*]] = bitcast i8* [[TMP15]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %class.TestClass* @_ZN9TestClassaSERKS_(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP13]], %class.TestClass* nonnull align 4 dereferenceable(4) [[TMP16]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[TMP17:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 2
// CHECK2-NEXT: [[TMP18:%.*]] = load i8*, i8** [[TMP17]], align 8
// CHECK2-NEXT: [[TMP19:%.*]] = bitcast i8* [[TMP18]] to %class.TestClass*
// CHECK2-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 2
// CHECK2-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8
// CHECK2-NEXT: [[TMP22:%.*]] = bitcast i8* [[TMP21]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: [[CALL2:%.*]] = call nonnull align 4 dereferenceable(4) %class.TestClass* @_ZN9TestClassaSERKS_(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP19]], %class.TestClass* nonnull align 4 dereferenceable(4) [[TMP22]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 3
// CHECK2-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8
// CHECK2-NEXT: [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 3
// CHECK2-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 8
// CHECK2-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 1 [[TMP24]], i8* align 1 [[TMP26]], i64 2, i1 false)
// CHECK2-NEXT: [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 4
// CHECK2-NEXT: [[TMP28:%.*]] = load i8*, i8** [[TMP27]], align 8
// CHECK2-NEXT: [[TMP29:%.*]] = bitcast i8* [[TMP28]] to %class.TestClass*
// CHECK2-NEXT: [[TMP30:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 4
// CHECK2-NEXT: [[TMP31:%.*]] = load i8*, i8** [[TMP30]], align 8
// CHECK2-NEXT: [[TMP32:%.*]] = bitcast i8* [[TMP31]] to %class.TestClass*
// CHECK2-NEXT: [[TMP33:%.*]] = getelementptr [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[TMP29]], i64 2
// CHECK2-NEXT: [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq %class.TestClass* [[TMP29]], [[TMP33]]
// CHECK2-NEXT: br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
// CHECK2: omp.arraycpy.body:
// CHECK2-NEXT: [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %class.TestClass* [ [[TMP32]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
// CHECK2-NEXT: [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %class.TestClass* [ [[TMP29]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: [[CALL3:%.*]] = call nonnull align 4 dereferenceable(4) %class.TestClass* @_ZN9TestClassaSERKS_(%class.TestClass* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %class.TestClass* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[CLASS_TESTCLASS]], %class.TestClass* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
// CHECK2-NEXT: [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[CLASS_TESTCLASS]], %class.TestClass* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
// CHECK2-NEXT: [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %class.TestClass* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP33]]
// CHECK2-NEXT: br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE4]], label [[OMP_ARRAYCPY_BODY]]
// CHECK2: omp.arraycpy.done4:
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN9TestClassaSERKS_
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]], %class.TestClass* nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) #[[ATTR10:[0-9]+]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK2-NEXT: [[DOTADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK2-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK2-NEXT: store %class.TestClass* [[TMP0]], %class.TestClass** [[DOTADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
// CHECK2-NEXT: ret %class.TestClass* [[THIS1]]
//
//
// CHECK2-LABEL: define {{[^@]+}}@_Z15parallel_singlev
// CHECK2-SAME: () #[[ATTR10]] {
// CHECK2-NEXT: entry:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* @.omp_outlined. to void (i32*, i32*, ...)*))
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@.omp_outlined.
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]]) #[[ATTR12:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK2-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP0:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT: [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK2-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
// CHECK2-NEXT: [[TMP3:%.*]] = icmp ne i32 [[TMP2]], 0
// CHECK2-NEXT: br i1 [[TMP3]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2: omp_if.then:
// CHECK2-NEXT: invoke void @_Z3foov()
// CHECK2-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK2: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: br label [[OMP_IF_END]]
// CHECK2: lpad:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP4:%.*]] = landingpad { i8*, i32 }
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: catch i8* null
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP5:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 0
// CHECK2-NEXT: store i8* [[TMP5]], i8** [[EXN_SLOT]], align 8
// CHECK2-NEXT: [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 1
// CHECK2-NEXT: store i32 [[TMP6]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK2-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK2: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
// CHECK2: terminate.handler:
// CHECK2-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK2-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13]]
// CHECK2-NEXT: unreachable
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN9TestClassC2Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK2-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[A:%.*]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT: store i32 0, i32* [[A]], align 4
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN9TestClassD2Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK2-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN2SSC2ERi
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-SAME: (%struct.SS* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32* nonnull align 4 dereferenceable(4) [[D:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK2-NEXT: [[D_ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT: [[A2:%.*]] = alloca i32*, align 8
// CHECK2-NEXT: [[B4:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[C7:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT: [[B_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT: [[C_CASTED:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK2-NEXT: store i32* [[D]], i32** [[D_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], %struct.SS* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT: store i32 0, i32* [[A]], align 8
// CHECK2-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 1
// CHECK2-NEXT: [[BF_LOAD:%.*]] = load i8, i8* [[B]], align 4
// CHECK2-NEXT: [[BF_CLEAR:%.*]] = and i8 [[BF_LOAD]], -16
// CHECK2-NEXT: store i8 [[BF_CLEAR]], i8* [[B]], align 4
// CHECK2-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 2
// CHECK2-NEXT: [[TMP0:%.*]] = load i32*, i32** [[D_ADDR]], align 8
// CHECK2-NEXT: store i32* [[TMP0]], i32** [[C]], align 8
// CHECK2-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT: store i32* [[A3]], i32** [[A2]], align 8
// CHECK2-NEXT: [[B5:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 1
// CHECK2-NEXT: [[BF_LOAD6:%.*]] = load i8, i8* [[B5]], align 4
// CHECK2-NEXT: [[BF_SHL:%.*]] = shl i8 [[BF_LOAD6]], 4
// CHECK2-NEXT: [[BF_ASHR:%.*]] = ashr i8 [[BF_SHL]], 4
// CHECK2-NEXT: [[BF_CAST:%.*]] = sext i8 [[BF_ASHR]] to i32
// CHECK2-NEXT: store i32 [[BF_CAST]], i32* [[B4]], align 4
// CHECK2-NEXT: [[C8:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 2
// CHECK2-NEXT: [[TMP1:%.*]] = load i32*, i32** [[C8]], align 8
// CHECK2-NEXT: store i32* [[TMP1]], i32** [[C7]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP2:%.*]] = load i32*, i32** [[A2]], align 8
// CHECK2-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4
// CHECK2-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to i32*
// CHECK2-NEXT: store i32 [[TMP3]], i32* [[CONV]], align 4
// CHECK2-NEXT: [[TMP4:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK2-NEXT: [[TMP5:%.*]] = load i32, i32* [[B4]], align 4
// CHECK2-NEXT: [[CONV9:%.*]] = bitcast i64* [[B_CASTED]] to i32*
// CHECK2-NEXT: store i32 [[TMP5]], i32* [[CONV9]], align 4
// CHECK2-NEXT: [[TMP6:%.*]] = load i64, i64* [[B_CASTED]], align 8
// CHECK2-NEXT: [[TMP7:%.*]] = load i32*, i32** [[C7]], align 8
// CHECK2-NEXT: [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK2-NEXT: [[CONV10:%.*]] = bitcast i64* [[C_CASTED]] to i32*
// CHECK2-NEXT: store i32 [[TMP8]], i32* [[CONV10]], align 4
// CHECK2-NEXT: [[TMP9:%.*]] = load i64, i64* [[C_CASTED]], align 8
// CHECK2-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SS*, i64, i64, i64)* @.omp_outlined..5 to void (i32*, i32*, ...)*), %struct.SS* [[THIS1]], i64 [[TMP4]], i64 [[TMP6]], i64 [[TMP9]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@.omp_outlined..5
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SS* [[THIS:%.*]], i64 [[A:%.*]], i64 [[B:%.*]], i64 [[C:%.*]]) #[[ATTR12]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK2-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT: [[C_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[TMP:%.*]] = alloca i32*, align 8
// CHECK2-NEXT: [[_TMP3:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[_TMP4:%.*]] = alloca i32*, align 8
// CHECK2-NEXT: [[_TMP5:%.*]] = alloca i32*, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON:%.*]], align 8
// CHECK2-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [3 x i8*], align 8
// CHECK2-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK2-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK2-NEXT: store i64 [[B]], i64* [[B_ADDR]], align 8
// CHECK2-NEXT: store i64 [[C]], i64* [[C_ADDR]], align 8
// CHECK2-NEXT: [[TMP0:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to i32*
// CHECK2-NEXT: [[CONV1:%.*]] = bitcast i64* [[B_ADDR]] to i32*
// CHECK2-NEXT: [[CONV2:%.*]] = bitcast i64* [[C_ADDR]] to i32*
// CHECK2-NEXT: store i32* [[CONV]], i32** [[TMP]], align 8
// CHECK2-NEXT: store i32* [[CONV2]], i32** [[_TMP3]], align 8
// CHECK2-NEXT: [[TMP1:%.*]] = load i32*, i32** [[TMP]], align 8
// CHECK2-NEXT: store i32* [[TMP1]], i32** [[_TMP4]], align 8
// CHECK2-NEXT: [[TMP2:%.*]] = load i32*, i32** [[_TMP3]], align 8
// CHECK2-NEXT: store i32* [[TMP2]], i32** [[_TMP5]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK2-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK2-NEXT: [[TMP6:%.*]] = icmp ne i32 [[TMP5]], 0
// CHECK2-NEXT: br i1 [[TMP6]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 0
// CHECK2-NEXT: store %struct.SS* [[TMP0]], %struct.SS** [[TMP7]], align 8
// CHECK2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 1
// CHECK2-NEXT: [[TMP9:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK2-NEXT: store i32* [[TMP9]], i32** [[TMP8]], align 8
// CHECK2-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 2
// CHECK2-NEXT: store i32* [[CONV1]], i32** [[TMP10]], align 8
// CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 3
// CHECK2-NEXT: [[TMP12:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK2-NEXT: store i32* [[TMP12]], i32** [[TMP11]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-NEXT: invoke void @_ZZN2SSC1ERiENKUlvE_clEv(%class.anon* nonnull align 8 dereferenceable(32) [[REF_TMP]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK2: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK2-NEXT: br label [[OMP_IF_END]]
// CHECK2: lpad:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP13:%.*]] = landingpad { i8*, i32 }
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: catch i8* null
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP14:%.*]] = extractvalue { i8*, i32 } [[TMP13]], 0
// CHECK2-NEXT: store i8* [[TMP14]], i8** [[EXN_SLOT]], align 8
// CHECK2-NEXT: [[TMP15:%.*]] = extractvalue { i8*, i32 } [[TMP13]], 1
// CHECK2-NEXT: store i32 [[TMP15]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK2-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK2: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK2-NEXT: [[TMP17:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK2-NEXT: [[TMP18:%.*]] = bitcast i32* [[TMP17]] to i8*
// CHECK2-NEXT: store i8* [[TMP18]], i8** [[TMP16]], align 8
// CHECK2-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1
// CHECK2-NEXT: [[TMP20:%.*]] = bitcast i32* [[CONV1]] to i8*
// CHECK2-NEXT: store i8* [[TMP20]], i8** [[TMP19]], align 8
// CHECK2-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 2
// CHECK2-NEXT: [[TMP22:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK2-NEXT: [[TMP23:%.*]] = bitcast i32* [[TMP22]] to i8*
// CHECK2-NEXT: store i8* [[TMP23]], i8** [[TMP21]], align 8
// CHECK2-NEXT: [[TMP24:%.*]] = bitcast [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK2-NEXT: [[TMP25:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK2-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]], i64 24, i8* [[TMP24]], void (i8*, i8*)* @.omp.copyprivate.copy_func.6, i32 [[TMP25]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
// CHECK2: terminate.handler:
// CHECK2-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK2-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13]]
// CHECK2-NEXT: unreachable
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZZN2SSC1ERiENKUlvE_clEv
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-SAME: (%class.anon* nonnull align 8 dereferenceable(32) [[THIS:%.*]]) #[[ATTR10]] align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %class.anon*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT: [[B_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT: [[C_CASTED:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: store %class.anon* [[THIS]], %class.anon** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %class.anon*, %class.anon** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON:%.*]], %class.anon* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT: [[TMP1:%.*]] = load %struct.SS*, %struct.SS** [[TMP0]], align 8
// CHECK2-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[THIS1]], i32 0, i32 1
// CHECK2-NEXT: [[TMP3:%.*]] = load i32*, i32** [[TMP2]], align 8
// CHECK2-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK2-NEXT: [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// CHECK2-NEXT: store i32 [[INC]], i32* [[TMP3]], align 4
// CHECK2-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[THIS1]], i32 0, i32 2
// CHECK2-NEXT: [[TMP6:%.*]] = load i32*, i32** [[TMP5]], align 8
// CHECK2-NEXT: [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4
// CHECK2-NEXT: [[DEC:%.*]] = add nsw i32 [[TMP7]], -1
// CHECK2-NEXT: store i32 [[DEC]], i32* [[TMP6]], align 4
// CHECK2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[THIS1]], i32 0, i32 3
// CHECK2-NEXT: [[TMP9:%.*]] = load i32*, i32** [[TMP8]], align 8
// CHECK2-NEXT: [[TMP10:%.*]] = load i32, i32* [[TMP9]], align 4
// CHECK2-NEXT: [[DIV:%.*]] = sdiv i32 [[TMP10]], 1
// CHECK2-NEXT: store i32 [[DIV]], i32* [[TMP9]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[THIS1]], i32 0, i32 1
// CHECK2-NEXT: [[TMP12:%.*]] = load i32*, i32** [[TMP11]], align 8
// CHECK2-NEXT: [[TMP13:%.*]] = load i32, i32* [[TMP12]], align 4
// CHECK2-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to i32*
// CHECK2-NEXT: store i32 [[TMP13]], i32* [[CONV]], align 4
// CHECK2-NEXT: [[TMP14:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK2-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[THIS1]], i32 0, i32 2
// CHECK2-NEXT: [[TMP16:%.*]] = load i32*, i32** [[TMP15]], align 8
// CHECK2-NEXT: [[TMP17:%.*]] = load i32, i32* [[TMP16]], align 4
// CHECK2-NEXT: [[CONV2:%.*]] = bitcast i64* [[B_CASTED]] to i32*
// CHECK2-NEXT: store i32 [[TMP17]], i32* [[CONV2]], align 4
// CHECK2-NEXT: [[TMP18:%.*]] = load i64, i64* [[B_CASTED]], align 8
// CHECK2-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[THIS1]], i32 0, i32 3
// CHECK2-NEXT: [[TMP20:%.*]] = load i32*, i32** [[TMP19]], align 8
// CHECK2-NEXT: [[TMP21:%.*]] = load i32, i32* [[TMP20]], align 4
// CHECK2-NEXT: [[CONV3:%.*]] = bitcast i64* [[C_CASTED]] to i32*
// CHECK2-NEXT: store i32 [[TMP21]], i32* [[CONV3]], align 4
// CHECK2-NEXT: [[TMP22:%.*]] = load i64, i64* [[C_CASTED]], align 8
// CHECK2-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SS*, i64, i64, i64)* @.omp_outlined..7 to void (i32*, i32*, ...)*), %struct.SS* [[TMP1]], i64 [[TMP14]], i64 [[TMP18]], i64 [[TMP22]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.6
// CHECK2-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK2-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK2-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK2-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [3 x i8*]*
// CHECK2-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK2-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [3 x i8*]*
// CHECK2-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK2-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK2-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to i32*
// CHECK2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK2-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK2-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to i32*
// CHECK2-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
// CHECK2-NEXT: store i32 [[TMP12]], i32* [[TMP8]], align 4
// CHECK2-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 1
// CHECK2-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8
// CHECK2-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i32*
// CHECK2-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 1
// CHECK2-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8
// CHECK2-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32*
// CHECK2-NEXT: [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
// CHECK2-NEXT: store i32 [[TMP19]], i32* [[TMP15]], align 4
// CHECK2-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 2
// CHECK2-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8
// CHECK2-NEXT: [[TMP22:%.*]] = bitcast i8* [[TMP21]] to i32*
// CHECK2-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 2
// CHECK2-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8
// CHECK2-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32*
// CHECK2-NEXT: [[TMP26:%.*]] = load i32, i32* [[TMP25]], align 4
// CHECK2-NEXT: store i32 [[TMP26]], i32* [[TMP22]], align 4
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@.omp_outlined..7
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SS* [[THIS:%.*]], i64 [[A:%.*]], i64 [[B:%.*]], i64 [[C:%.*]]) #[[ATTR12]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK2-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8
// CHECK2-NEXT: [[C_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[TMP:%.*]] = alloca i32*, align 8
// CHECK2-NEXT: [[_TMP3:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[_TMP4:%.*]] = alloca i32*, align 8
// CHECK2-NEXT: [[_TMP5:%.*]] = alloca i32*, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [3 x i8*], align 8
// CHECK2-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK2-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK2-NEXT: store i64 [[B]], i64* [[B_ADDR]], align 8
// CHECK2-NEXT: store i64 [[C]], i64* [[C_ADDR]], align 8
// CHECK2-NEXT: [[TMP0:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to i32*
// CHECK2-NEXT: [[CONV1:%.*]] = bitcast i64* [[B_ADDR]] to i32*
// CHECK2-NEXT: [[CONV2:%.*]] = bitcast i64* [[C_ADDR]] to i32*
// CHECK2-NEXT: store i32* [[CONV]], i32** [[TMP]], align 8
// CHECK2-NEXT: store i32* [[CONV2]], i32** [[_TMP3]], align 8
// CHECK2-NEXT: [[TMP1:%.*]] = load i32*, i32** [[TMP]], align 8
// CHECK2-NEXT: store i32* [[TMP1]], i32** [[_TMP4]], align 8
// CHECK2-NEXT: [[TMP2:%.*]] = load i32*, i32** [[_TMP3]], align 8
// CHECK2-NEXT: store i32* [[TMP2]], i32** [[_TMP5]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK2-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK2-NEXT: [[TMP6:%.*]] = icmp ne i32 [[TMP5]], 0
// CHECK2-NEXT: br i1 [[TMP6]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP7:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK2-NEXT: [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK2-NEXT: [[INC:%.*]] = add nsw i32 [[TMP8]], 1
// CHECK2-NEXT: store i32 [[INC]], i32* [[TMP7]], align 4
// CHECK2-NEXT: [[TMP9:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK2-NEXT: [[DEC:%.*]] = add nsw i32 [[TMP9]], -1
// CHECK2-NEXT: store i32 [[DEC]], i32* [[CONV1]], align 8
// CHECK2-NEXT: [[TMP10:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK2-NEXT: [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK2-NEXT: [[DIV:%.*]] = sdiv i32 [[TMP11]], 1
// CHECK2-NEXT: store i32 [[DIV]], i32* [[TMP10]], align 4
// CHECK2-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK2-NEXT: br label [[OMP_IF_END]]
// CHECK2: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK2-NEXT: [[TMP13:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK2-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8*
// CHECK2-NEXT: store i8* [[TMP14]], i8** [[TMP12]], align 8
// CHECK2-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1
// CHECK2-NEXT: [[TMP16:%.*]] = bitcast i32* [[CONV1]] to i8*
// CHECK2-NEXT: store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK2-NEXT: [[TMP17:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 2
// CHECK2-NEXT: [[TMP18:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK2-NEXT: [[TMP19:%.*]] = bitcast i32* [[TMP18]] to i8*
// CHECK2-NEXT: store i8* [[TMP19]], i8** [[TMP17]], align 8
// CHECK2-NEXT: [[TMP20:%.*]] = bitcast [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK2-NEXT: [[TMP21:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK2-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]], i64 24, i8* [[TMP20]], void (i8*, i8*)* @.omp.copyprivate.copy_func.8, i32 [[TMP21]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.8
// CHECK2-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK2-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK2-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK2-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [3 x i8*]*
// CHECK2-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK2-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [3 x i8*]*
// CHECK2-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK2-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK2-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to i32*
// CHECK2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK2-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK2-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to i32*
// CHECK2-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
// CHECK2-NEXT: store i32 [[TMP12]], i32* [[TMP8]], align 4
// CHECK2-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 1
// CHECK2-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8
// CHECK2-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i32*
// CHECK2-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 1
// CHECK2-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8
// CHECK2-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32*
// CHECK2-NEXT: [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
// CHECK2-NEXT: store i32 [[TMP19]], i32* [[TMP15]], align 4
// CHECK2-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 2
// CHECK2-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8
// CHECK2-NEXT: [[TMP22:%.*]] = bitcast i8* [[TMP21]] to i32*
// CHECK2-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 2
// CHECK2-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8
// CHECK2-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32*
// CHECK2-NEXT: [[TMP26:%.*]] = load i32, i32* [[TMP25]], align 4
// CHECK2-NEXT: store i32 [[TMP26]], i32* [[TMP22]], align 4
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN3SSTIdEC2Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK2-SAME: (%struct.SST* nonnull align 8 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK2-NEXT: [[A2:%.*]] = alloca double*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SST:%.*]], %struct.SST* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT: store double 0.000000e+00, double* [[A]], align 8
// CHECK2-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SST]], %struct.SST* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT: store double* [[A3]], double** [[A2]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP0:%.*]] = load double*, double** [[A2]], align 8
// CHECK2-NEXT: [[TMP1:%.*]] = load double, double* [[TMP0]], align 8
// CHECK2-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to double*
// CHECK2-NEXT: store double [[TMP1]], double* [[CONV]], align 8
// CHECK2-NEXT: [[TMP2:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK2-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SST*, i64)* @.omp_outlined..9 to void (i32*, i32*, ...)*), %struct.SST* [[THIS1]], i64 [[TMP2]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@.omp_outlined..9
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SST* [[THIS:%.*]], i64 [[A:%.*]]) #[[ATTR12]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK2-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[TMP:%.*]] = alloca double*, align 8
// CHECK2-NEXT: [[_TMP1:%.*]] = alloca double*, align 8
// CHECK2-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON_0:%.*]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK2-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK2-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK2-NEXT: [[TMP0:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to double*
// CHECK2-NEXT: store double* [[CONV]], double** [[TMP]], align 8
// CHECK2-NEXT: [[TMP1:%.*]] = load double*, double** [[TMP]], align 8
// CHECK2-NEXT: store double* [[TMP1]], double** [[_TMP1]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP2:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4
// CHECK2-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
// CHECK2-NEXT: [[TMP5:%.*]] = icmp ne i32 [[TMP4]], 0
// CHECK2-NEXT: br i1 [[TMP5]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 0
// CHECK2-NEXT: store %struct.SST* [[TMP0]], %struct.SST** [[TMP6]], align 8
// CHECK2-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 1
// CHECK2-NEXT: [[TMP8:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK2-NEXT: store double* [[TMP8]], double** [[TMP7]], align 8
// CHECK2-NEXT: invoke void @_ZZN3SSTIdEC1EvENKUlvE_clEv(%class.anon.0* nonnull align 8 dereferenceable(16) [[REF_TMP]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK2: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK2-NEXT: br label [[OMP_IF_END]]
// CHECK2: lpad:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP9:%.*]] = landingpad { i8*, i32 }
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: catch i8* null
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP10:%.*]] = extractvalue { i8*, i32 } [[TMP9]], 0
// CHECK2-NEXT: store i8* [[TMP10]], i8** [[EXN_SLOT]], align 8
// CHECK2-NEXT: [[TMP11:%.*]] = extractvalue { i8*, i32 } [[TMP9]], 1
// CHECK2-NEXT: store i32 [[TMP11]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK2-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK2: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP12:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK2-NEXT: [[TMP13:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK2-NEXT: [[TMP14:%.*]] = bitcast double* [[TMP13]] to i8*
// CHECK2-NEXT: store i8* [[TMP14]], i8** [[TMP12]], align 8
// CHECK2-NEXT: [[TMP15:%.*]] = bitcast [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK2-NEXT: [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK2-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]], i64 8, i8* [[TMP15]], void (i8*, i8*)* @.omp.copyprivate.copy_func.10, i32 [[TMP16]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
// CHECK2: terminate.handler:
// CHECK2-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK2-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13]]
// CHECK2-NEXT: unreachable
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZZN3SSTIdEC1EvENKUlvE_clEv
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-SAME: (%class.anon.0* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR4]] align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %class.anon.0*, align 8
// CHECK2-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON_1:%.*]], align 8
// CHECK2-NEXT: store %class.anon.0* [[THIS]], %class.anon.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %class.anon.0*, %class.anon.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON_0:%.*]], %class.anon.0* [[THIS1]], i32 0, i32 0
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[TMP1:%.*]] = load %struct.SST*, %struct.SST** [[TMP0]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 0
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: store %struct.SST* [[TMP1]], %struct.SST** [[TMP2]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP3:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 1
// CHECK2-NEXT: [[TMP4:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[THIS1]], i32 0, i32 1
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[TMP5:%.*]] = load double*, double** [[TMP4]], align 8
// CHECK2-NEXT: store double* [[TMP5]], double** [[TMP3]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: call void @_ZZZN3SSTIdEC1EvENKUlvE_clEvENKUlvE_clEv(%class.anon.1* nonnull align 8 dereferenceable(16) [[REF_TMP]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.10
// CHECK2-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK2-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK2-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK2-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [1 x i8*]*
// CHECK2-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK2-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*
// CHECK2-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK2-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK2-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to double*
// CHECK2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK2-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK2-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to double*
// CHECK2-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 8
// CHECK2-NEXT: store double [[TMP12]], double* [[TMP8]], align 8
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZZZN3SSTIdEC1EvENKUlvE_clEvENKUlvE_clEv
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-SAME: (%class.anon.1* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR10]] align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %class.anon.1*, align 8
// CHECK2-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK2-NEXT: store %class.anon.1* [[THIS]], %class.anon.1** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %class.anon.1*, %class.anon.1** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON_1:%.*]], %class.anon.1* [[THIS1]], i32 0, i32 0
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[TMP1:%.*]] = load %struct.SST*, %struct.SST** [[TMP0]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 1
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[TMP3:%.*]] = load double*, double** [[TMP2]], align 8
// CHECK2-NEXT: [[TMP4:%.*]] = load double, double* [[TMP3]], align 8
// CHECK2-NEXT: [[INC:%.*]] = fadd double [[TMP4]], 1.000000e+00
// CHECK2-NEXT: store double [[INC]], double* [[TMP3]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 1
// CHECK2-NEXT: [[TMP6:%.*]] = load double*, double** [[TMP5]], align 8
// CHECK2-NEXT: [[TMP7:%.*]] = load double, double* [[TMP6]], align 8
// CHECK2-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to double*
// CHECK2-NEXT: store double [[TMP7]], double* [[CONV]], align 8
// CHECK2-NEXT: [[TMP8:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK2-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SST*, i64)* @.omp_outlined..11 to void (i32*, i32*, ...)*), %struct.SST* [[TMP1]], i64 [[TMP8]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@.omp_outlined..11
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SST* [[THIS:%.*]], i64 [[A:%.*]]) #[[ATTR12]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK2-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: [[TMP:%.*]] = alloca double*, align 8
// CHECK2-NEXT: [[_TMP1:%.*]] = alloca double*, align 8
// CHECK2-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK2-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK2-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK2-NEXT: [[TMP0:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to double*
// CHECK2-NEXT: store double* [[CONV]], double** [[TMP]], align 8
// CHECK2-NEXT: [[TMP1:%.*]] = load double*, double** [[TMP]], align 8
// CHECK2-NEXT: store double* [[TMP1]], double** [[_TMP1]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP2:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4
// CHECK2-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
// CHECK2-NEXT: [[TMP5:%.*]] = icmp ne i32 [[TMP4]], 0
// CHECK2-NEXT: br i1 [[TMP5]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP6:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK2-NEXT: [[TMP7:%.*]] = load double, double* [[TMP6]], align 8
// CHECK2-NEXT: [[INC:%.*]] = fadd double [[TMP7]], 1.000000e+00
// CHECK2-NEXT: store double [[INC]], double* [[TMP6]], align 8
// CHECK2-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK2-NEXT: br label [[OMP_IF_END]]
// CHECK2: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK2-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK2-NEXT: [[TMP9:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK2-NEXT: [[TMP10:%.*]] = bitcast double* [[TMP9]] to i8*
// CHECK2-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 8
// CHECK2-NEXT: [[TMP11:%.*]] = bitcast [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK2-NEXT: [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK2-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]], i64 8, i8* [[TMP11]], void (i8*, i8*)* @.omp.copyprivate.copy_func.12, i32 [[TMP12]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.12
// CHECK2-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK2-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK2-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK2-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [1 x i8*]*
// CHECK2-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK2-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*
// CHECK2-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK2-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK2-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to double*
// CHECK2-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK2-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK2-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to double*
// CHECK2-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 8
// CHECK2-NEXT: store double [[TMP12]], double* [[TMP8]], align 8
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_single_codegen.cpp
// CHECK2-SAME: () #[[ATTR0]] {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: call void @__cxx_global_var_init()
// CHECK2-NEXT: call void @__cxx_global_var_init.1()
// CHECK2-NEXT: call void @.__omp_threadprivate_init_.()
// CHECK2-NEXT: call void @.__omp_threadprivate_init_..4()
// CHECK2-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_.
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0:[0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK3-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK3-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: call void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP2]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK3-NEXT: ret i8* [[TMP3]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN9TestClassC1Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK3-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: call void @_ZN9TestClassC2Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@.__kmpc_global_dtor_.
// CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK3-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK3-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP2]]) #[[ATTR3:[0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN9TestClassD1Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2:[0-9]+]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK3-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: call void @_ZN9TestClassD2Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-LABEL: define {{[^@]+}}@.__omp_threadprivate_init_.
// CHECK3-SAME: () #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
// CHECK3-NEXT: call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast (%class.TestClass* @tc to i8*), i8* (i8*)* @.__kmpc_global_ctor_., i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_.)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_..1
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK3-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK3-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to [2 x %class.TestClass]*
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %class.TestClass], [2 x %class.TestClass]* [[TMP2]], i32 0, i32 0
// CHECK3-NEXT: [[ARRAYCTOR_END:%.*]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAY_BEGIN]], i64 2
// CHECK3-NEXT: br label [[ARRAYCTOR_LOOP:%.*]]
// CHECK3: arrayctor.loop:
// CHECK3-NEXT: [[ARRAYCTOR_CUR:%.*]] = phi %class.TestClass* [ [[ARRAY_BEGIN]], [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[INVOKE_CONT:%.*]] ]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: invoke void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: to label [[INVOKE_CONT]] unwind label [[LPAD:%.*]]
// CHECK3: invoke.cont:
// CHECK3-NEXT: [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYCTOR_CUR]], i64 1
// CHECK3-NEXT: [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], [[ARRAYCTOR_END]]
// CHECK3-NEXT: br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
// CHECK3: arrayctor.cont:
// CHECK3-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK3-NEXT: ret i8* [[TMP3]]
// CHECK3: lpad:
// CHECK3-NEXT: [[TMP4:%.*]] = landingpad { i8*, i32 }
// CHECK3-NEXT: cleanup
// CHECK3-NEXT: [[TMP5:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 0
// CHECK3-NEXT: store i8* [[TMP5]], i8** [[EXN_SLOT]], align 8
// CHECK3-NEXT: [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 1
// CHECK3-NEXT: store i32 [[TMP6]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK3-NEXT: [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %class.TestClass* [[ARRAY_BEGIN]], [[ARRAYCTOR_CUR]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ [[ARRAYCTOR_CUR]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done1:
// CHECK3-NEXT: br label [[EH_RESUME:%.*]]
// CHECK3: eh.resume:
// CHECK3-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK3-NEXT: [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4
// CHECK3-NEXT: [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0
// CHECK3-NEXT: [[LPAD_VAL2:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1
// CHECK3-NEXT: resume { i8*, i32 } [[LPAD_VAL2]]
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-LABEL: define {{[^@]+}}@.__kmpc_global_dtor_..2
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK3-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK3-NEXT: [[ARRAY_BEGIN:%.*]] = bitcast i8* [[TMP1]] to %class.TestClass*
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAY_BEGIN]], i64 2
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ [[TMP2]], [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done1:
// CHECK3-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-LABEL: define {{[^@]+}}@.__omp_threadprivate_init_..3
// CHECK3-SAME: () #[[ATTR0]] {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-NEXT: call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast ([2 x %class.TestClass]* @tc2 to i8*), i8* (i8*)* @.__kmpc_global_ctor_..1, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..2)
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__cxx_global_var_init
// CHECK3-SAME: () #[[ATTR0]] {
// CHECK3-NEXT: entry:
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: call void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) @tc)
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%class.TestClass*)* @_ZN9TestClassD1Ev to void (i8*)*), i8* bitcast (%class.TestClass* @tc to i8*), i8* @__dso_handle) #[[ATTR3]]
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__cxx_global_var_init.4
// CHECK3-SAME: () #[[ATTR0]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK3-NEXT: br label [[ARRAYCTOR_LOOP:%.*]]
// CHECK3: arrayctor.loop:
// CHECK3-NEXT: [[ARRAYCTOR_CUR:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[INVOKE_CONT:%.*]] ]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: invoke void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-NEXT: to label [[INVOKE_CONT]] unwind label [[LPAD:%.*]]
// CHECK3: invoke.cont:
// CHECK3-NEXT: [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAYCTOR_CUR]], i64 1
// CHECK3-NEXT: [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([[CLASS_TESTCLASS]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2)
// CHECK3-NEXT: br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
// CHECK3: arrayctor.cont:
// CHECK3-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]]
// CHECK3-NEXT: ret void
// CHECK3: lpad:
// CHECK3-NEXT: [[TMP1:%.*]] = landingpad { i8*, i32 }
// CHECK3-NEXT: cleanup
// CHECK3-NEXT: [[TMP2:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 0
// CHECK3-NEXT: store i8* [[TMP2]], i8** [[EXN_SLOT]], align 8
// CHECK3-NEXT: [[TMP3:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 1
// CHECK3-NEXT: store i32 [[TMP3]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK3-NEXT: [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), [[ARRAYCTOR_CUR]]
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ [[ARRAYCTOR_CUR]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0)
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done1:
// CHECK3-NEXT: br label [[EH_RESUME:%.*]]
// CHECK3: eh.resume:
// CHECK3-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK3-NEXT: [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4
// CHECK3-NEXT: [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0
// CHECK3-NEXT: [[LPAD_VAL2:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1
// CHECK3-NEXT: resume { i8*, i32 } [[LPAD_VAL2]]
//
//
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
// CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK3-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([[CLASS_TESTCLASS:%.*]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0)
// CHECK3-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK3: arraydestroy.done1:
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN9TestClassC2Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK3-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[A:%.*]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: store i32 0, i32* [[A]], align 4
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN9TestClassD2Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK3-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
// CHECK3-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-LABEL: define {{[^@]+}}@_Z3foov
// CHECK3-SAME: () #[[ATTR4:[0-9]+]] {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: call void @_Z8mayThrowv()
// CHECK3-NEXT: ret void
//
//
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-LABEL: define {{[^@]+}}@main
// CHECK3-SAME: () #[[ATTR6:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[RETVAL:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[A:%.*]] = alloca i8, align 1
// CHECK3-NEXT: [[A2:%.*]] = alloca [2 x i8], align 1
// CHECK3-NEXT: [[C:%.*]] = alloca %class.TestClass*, align 8
// CHECK3-NEXT: [[SST:%.*]] = alloca [[STRUCT_SST:%.*]], align 8
// CHECK3-NEXT: [[SS:%.*]] = alloca [[STRUCT_SS:%.*]], align 8
// CHECK3-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [5 x i8*], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK3-NEXT: store i32 0, i32* [[RETVAL]], align 4
// CHECK3-NEXT: store %class.TestClass* @tc, %class.TestClass** [[C]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: call void @_ZN3SSTIdEC1Ev(%struct.SST* nonnull align 8 dereferenceable(8) [[SST]])
// CHECK3-NEXT: call void @_ZN2SSC1ERi(%struct.SS* nonnull align 8 dereferenceable(16) [[SS]], i32* nonnull align 4 dereferenceable(4) getelementptr inbounds ([[CLASS_TESTCLASS:%.*]], %class.TestClass* @tc, i32 0, i32 0))
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK3-NEXT: [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK3-NEXT: br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
// CHECK3: omp_if.then:
// CHECK3-NEXT: store i8 2, i8* [[A]], align 1
// CHECK3-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
// CHECK3-NEXT: [[TMP3:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK3-NEXT: [[TMP4:%.*]] = icmp ne i32 [[TMP3]], 0
// CHECK3-NEXT: br i1 [[TMP4]], label [[OMP_IF_THEN1:%.*]], label [[OMP_IF_END2:%.*]]
// CHECK3: omp_if.then1:
// CHECK3-NEXT: store i8 2, i8* [[A]], align 1
// CHECK3-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK3-NEXT: br label [[OMP_IF_END2]]
// CHECK3: omp_if.end2:
// CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP0]])
// CHECK3-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK3-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK3-NEXT: [[TMP6:%.*]] = icmp ne i32 [[TMP5]], 0
// CHECK3-NEXT: br i1 [[TMP6]], label [[OMP_IF_THEN3:%.*]], label [[OMP_IF_END4:%.*]]
// CHECK3: omp_if.then3:
// CHECK3-NEXT: invoke void @_Z3foov()
// CHECK3-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK3: invoke.cont:
// CHECK3-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK3-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK3-NEXT: br label [[OMP_IF_END4]]
// CHECK3: lpad:
// CHECK3-NEXT: [[TMP7:%.*]] = landingpad { i8*, i32 }
// CHECK3-NEXT: catch i8* null
// CHECK3-NEXT: [[TMP8:%.*]] = extractvalue { i8*, i32 } [[TMP7]], 0
// CHECK3-NEXT: store i8* [[TMP8]], i8** [[EXN_SLOT]], align 8
// CHECK3-NEXT: [[TMP9:%.*]] = extractvalue { i8*, i32 } [[TMP7]], 1
// CHECK3-NEXT: store i32 [[TMP9]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK3-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK3-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK3: omp_if.end4:
// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK3-NEXT: store i8* [[A]], i8** [[TMP10]], align 8
// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1
// CHECK3-NEXT: store i8* bitcast (%class.TestClass* @tc to i8*), i8** [[TMP11]], align 8
// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 2
// CHECK3-NEXT: [[TMP13:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i8* bitcast (%class.TestClass* @tc to i8*), i64 4, i8*** @tc.cache.)
// CHECK3-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to %class.TestClass*
// CHECK3-NEXT: [[TMP15:%.*]] = bitcast %class.TestClass* [[TMP14]] to i8*
// CHECK3-NEXT: store i8* [[TMP15]], i8** [[TMP12]], align 8
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 3
// CHECK3-NEXT: [[TMP17:%.*]] = bitcast [2 x i8]* [[A2]] to i8*
// CHECK3-NEXT: store i8* [[TMP17]], i8** [[TMP16]], align 8
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 4
// CHECK3-NEXT: [[TMP19:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i8* bitcast ([2 x %class.TestClass]* @tc2 to i8*), i64 8, i8*** @tc2.cache.)
// CHECK3-NEXT: [[TMP20:%.*]] = bitcast i8* [[TMP19]] to [2 x %class.TestClass]*
// CHECK3-NEXT: [[TMP21:%.*]] = bitcast [2 x %class.TestClass]* [[TMP20]] to i8*
// CHECK3-NEXT: store i8* [[TMP21]], i8** [[TMP18]], align 8
// CHECK3-NEXT: [[TMP22:%.*]] = bitcast [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK3-NEXT: [[TMP23:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK3-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i64 40, i8* [[TMP22]], void (i8*, i8*)* @.omp.copyprivate.copy_func, i32 [[TMP23]])
// CHECK3-NEXT: [[TMP24:%.*]] = load i8, i8* [[A]], align 1
// CHECK3-NEXT: [[CONV:%.*]] = sext i8 [[TMP24]] to i32
// CHECK3-NEXT: ret i32 [[CONV]]
// CHECK3: terminate.handler:
// CHECK3-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK3-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13:[0-9]+]]
// CHECK3-NEXT: unreachable
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN3SSTIdEC1Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-SAME: (%struct.SST* nonnull align 8 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK3-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: call void @_ZN3SSTIdEC2Ev(%struct.SST* nonnull align 8 dereferenceable(8) [[THIS1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN2SSC1ERi
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-SAME: (%struct.SS* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32* nonnull align 4 dereferenceable(4) [[D:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK3-NEXT: [[D_ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK3-NEXT: store i32* [[D]], i32** [[D_ADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = load i32*, i32** [[D_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: call void @_ZN2SSC2ERi(%struct.SS* nonnull align 8 dereferenceable(16) [[THIS1]], i32* nonnull align 4 dereferenceable(4) [[TMP0]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__clang_call_terminate
// CHECK3-SAME: (i8* [[TMP0:%.*]]) #[[ATTR8:[0-9]+]] comdat {
// CHECK3-NEXT: [[TMP2:%.*]] = call i8* @__cxa_begin_catch(i8* [[TMP0]]) #[[ATTR3]]
// CHECK3-NEXT: call void @_ZSt9terminatev() #[[ATTR13]]
// CHECK3-NEXT: unreachable
//
//
// CHECK3-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func
// CHECK3-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9:[0-9]+]] {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK3-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK3-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK3-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [5 x i8*]*
// CHECK3-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [5 x i8*]*
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK3-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK3-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8
// CHECK3-NEXT: [[TMP10:%.*]] = load i8, i8* [[TMP9]], align 1
// CHECK3-NEXT: store i8 [[TMP10]], i8* [[TMP7]], align 1
// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 1
// CHECK3-NEXT: [[TMP12:%.*]] = load i8*, i8** [[TMP11]], align 8
// CHECK3-NEXT: [[TMP13:%.*]] = bitcast i8* [[TMP12]] to %class.TestClass*
// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 1
// CHECK3-NEXT: [[TMP15:%.*]] = load i8*, i8** [[TMP14]], align 8
// CHECK3-NEXT: [[TMP16:%.*]] = bitcast i8* [[TMP15]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %class.TestClass* @_ZN9TestClassaSERKS_(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP13]], %class.TestClass* nonnull align 4 dereferenceable(4) [[TMP16]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 2
// CHECK3-NEXT: [[TMP18:%.*]] = load i8*, i8** [[TMP17]], align 8
// CHECK3-NEXT: [[TMP19:%.*]] = bitcast i8* [[TMP18]] to %class.TestClass*
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 2
// CHECK3-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8
// CHECK3-NEXT: [[TMP22:%.*]] = bitcast i8* [[TMP21]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: [[CALL2:%.*]] = call nonnull align 4 dereferenceable(4) %class.TestClass* @_ZN9TestClassaSERKS_(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP19]], %class.TestClass* nonnull align 4 dereferenceable(4) [[TMP22]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 3
// CHECK3-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8
// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 3
// CHECK3-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 8
// CHECK3-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 1 [[TMP24]], i8* align 1 [[TMP26]], i64 2, i1 false)
// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 4
// CHECK3-NEXT: [[TMP28:%.*]] = load i8*, i8** [[TMP27]], align 8
// CHECK3-NEXT: [[TMP29:%.*]] = bitcast i8* [[TMP28]] to %class.TestClass*
// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 4
// CHECK3-NEXT: [[TMP31:%.*]] = load i8*, i8** [[TMP30]], align 8
// CHECK3-NEXT: [[TMP32:%.*]] = bitcast i8* [[TMP31]] to %class.TestClass*
// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[TMP29]], i64 2
// CHECK3-NEXT: [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq %class.TestClass* [[TMP29]], [[TMP33]]
// CHECK3-NEXT: br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
// CHECK3: omp.arraycpy.body:
// CHECK3-NEXT: [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %class.TestClass* [ [[TMP32]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
// CHECK3-NEXT: [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %class.TestClass* [ [[TMP29]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: [[CALL3:%.*]] = call nonnull align 4 dereferenceable(4) %class.TestClass* @_ZN9TestClassaSERKS_(%class.TestClass* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %class.TestClass* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[CLASS_TESTCLASS]], %class.TestClass* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
// CHECK3-NEXT: [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[CLASS_TESTCLASS]], %class.TestClass* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
// CHECK3-NEXT: [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %class.TestClass* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP33]]
// CHECK3-NEXT: br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE4]], label [[OMP_ARRAYCPY_BODY]]
// CHECK3: omp.arraycpy.done4:
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN9TestClassaSERKS_
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]], %class.TestClass* nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) #[[ATTR10:[0-9]+]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK3-NEXT: [[DOTADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK3-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK3-NEXT: store %class.TestClass* [[TMP0]], %class.TestClass** [[DOTADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
// CHECK3-NEXT: ret %class.TestClass* [[THIS1]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN3SSTIdEC2Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-SAME: (%struct.SST* nonnull align 8 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK3-NEXT: [[A2:%.*]] = alloca double*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SST:%.*]], %struct.SST* [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: store double 0.000000e+00, double* [[A]], align 8
// CHECK3-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SST]], %struct.SST* [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: store double* [[A3]], double** [[A2]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP0:%.*]] = load double*, double** [[A2]], align 8
// CHECK3-NEXT: [[TMP1:%.*]] = load double, double* [[TMP0]], align 8
// CHECK3-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to double*
// CHECK3-NEXT: store double [[TMP1]], double* [[CONV]], align 8
// CHECK3-NEXT: [[TMP2:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK3-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SST*, i64)* @.omp_outlined. to void (i32*, i32*, ...)*), %struct.SST* [[THIS1]], i64 [[TMP2]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@.omp_outlined.
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SST* [[THIS:%.*]], i64 [[A:%.*]]) #[[ATTR12:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[TMP:%.*]] = alloca double*, align 8
// CHECK3-NEXT: [[_TMP1:%.*]] = alloca double*, align 8
// CHECK3-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON:%.*]], align 8
// CHECK3-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK3-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK3-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to double*
// CHECK3-NEXT: store double* [[CONV]], double** [[TMP]], align 8
// CHECK3-NEXT: [[TMP1:%.*]] = load double*, double** [[TMP]], align 8
// CHECK3-NEXT: store double* [[TMP1]], double** [[_TMP1]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP2:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4
// CHECK3-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
// CHECK3-NEXT: [[TMP5:%.*]] = icmp ne i32 [[TMP4]], 0
// CHECK3-NEXT: br i1 [[TMP5]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 0
// CHECK3-NEXT: store %struct.SST* [[TMP0]], %struct.SST** [[TMP6]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 1
// CHECK3-NEXT: [[TMP8:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK3-NEXT: store double* [[TMP8]], double** [[TMP7]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: invoke void @_ZZN3SSTIdEC1EvENKUlvE_clEv(%class.anon* nonnull align 8 dereferenceable(16) [[REF_TMP]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK3: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: lpad:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP9:%.*]] = landingpad { i8*, i32 }
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: catch i8* null
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP10:%.*]] = extractvalue { i8*, i32 } [[TMP9]], 0
// CHECK3-NEXT: store i8* [[TMP10]], i8** [[EXN_SLOT]], align 8
// CHECK3-NEXT: [[TMP11:%.*]] = extractvalue { i8*, i32 } [[TMP9]], 1
// CHECK3-NEXT: store i32 [[TMP11]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK3-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK3: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK3-NEXT: [[TMP13:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK3-NEXT: [[TMP14:%.*]] = bitcast double* [[TMP13]] to i8*
// CHECK3-NEXT: store i8* [[TMP14]], i8** [[TMP12]], align 8
// CHECK3-NEXT: [[TMP15:%.*]] = bitcast [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK3-NEXT: [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK3-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]], i64 8, i8* [[TMP15]], void (i8*, i8*)* @.omp.copyprivate.copy_func.5, i32 [[TMP16]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
// CHECK3: terminate.handler:
// CHECK3-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK3-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13]]
// CHECK3-NEXT: unreachable
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZZN3SSTIdEC1EvENKUlvE_clEv
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-SAME: (%class.anon* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR4]] align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %class.anon*, align 8
// CHECK3-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON_0:%.*]], align 8
// CHECK3-NEXT: store %class.anon* [[THIS]], %class.anon** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %class.anon*, %class.anon** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON:%.*]], %class.anon* [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[TMP1:%.*]] = load %struct.SST*, %struct.SST** [[TMP0]], align 8
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 0
// CHECK3-NEXT: store %struct.SST* [[TMP1]], %struct.SST** [[TMP2]], align 8
// CHECK3-NEXT: [[TMP3:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 1
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[THIS1]], i32 0, i32 1
// CHECK3-NEXT: [[TMP5:%.*]] = load double*, double** [[TMP4]], align 8
// CHECK3-NEXT: store double* [[TMP5]], double** [[TMP3]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-NEXT: call void @_ZZZN3SSTIdEC1EvENKUlvE_clEvENKUlvE_clEv(%class.anon.0* nonnull align 8 dereferenceable(16) [[REF_TMP]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.5
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK3-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK3-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK3-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [1 x i8*]*
// CHECK3-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK3-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to double*
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK3-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK3-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to double*
// CHECK3-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 8
// CHECK3-NEXT: store double [[TMP12]], double* [[TMP8]], align 8
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZZZN3SSTIdEC1EvENKUlvE_clEvENKUlvE_clEv
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-SAME: (%class.anon.0* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR10]] align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %class.anon.0*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: store %class.anon.0* [[THIS]], %class.anon.0** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %class.anon.0*, %class.anon.0** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON_0:%.*]], %class.anon.0* [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[TMP1:%.*]] = load %struct.SST*, %struct.SST** [[TMP0]], align 8
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[THIS1]], i32 0, i32 1
// CHECK3-NEXT: [[TMP3:%.*]] = load double*, double** [[TMP2]], align 8
// CHECK3-NEXT: [[TMP4:%.*]] = load double, double* [[TMP3]], align 8
// CHECK3-NEXT: [[INC:%.*]] = fadd double [[TMP4]], 1.000000e+00
// CHECK3-NEXT: store double [[INC]], double* [[TMP3]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[THIS1]], i32 0, i32 1
// CHECK3-NEXT: [[TMP6:%.*]] = load double*, double** [[TMP5]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = load double, double* [[TMP6]], align 8
// CHECK3-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to double*
// CHECK3-NEXT: store double [[TMP7]], double* [[CONV]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK3-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SST*, i64)* @.omp_outlined..6 to void (i32*, i32*, ...)*), %struct.SST* [[TMP1]], i64 [[TMP8]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-LABEL: define {{[^@]+}}@.omp_outlined..6
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SST* [[THIS:%.*]], i64 [[A:%.*]]) #[[ATTR12]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[TMP:%.*]] = alloca double*, align 8
// CHECK3-NEXT: [[_TMP1:%.*]] = alloca double*, align 8
// CHECK3-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK3-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK3-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to double*
// CHECK3-NEXT: store double* [[CONV]], double** [[TMP]], align 8
// CHECK3-NEXT: [[TMP1:%.*]] = load double*, double** [[TMP]], align 8
// CHECK3-NEXT: store double* [[TMP1]], double** [[_TMP1]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP2:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4
// CHECK3-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
// CHECK3-NEXT: [[TMP5:%.*]] = icmp ne i32 [[TMP4]], 0
// CHECK3-NEXT: br i1 [[TMP5]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP6:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = load double, double* [[TMP6]], align 8
// CHECK3-NEXT: [[INC:%.*]] = fadd double [[TMP7]], 1.000000e+00
// CHECK3-NEXT: store double [[INC]], double* [[TMP6]], align 8
// CHECK3-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK3-NEXT: [[TMP9:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK3-NEXT: [[TMP10:%.*]] = bitcast double* [[TMP9]] to i8*
// CHECK3-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 8
// CHECK3-NEXT: [[TMP11:%.*]] = bitcast [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK3-NEXT: [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK3-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]], i64 8, i8* [[TMP11]], void (i8*, i8*)* @.omp.copyprivate.copy_func.7, i32 [[TMP12]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.7
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK3-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK3-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK3-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [1 x i8*]*
// CHECK3-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK3-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to double*
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK3-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK3-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to double*
// CHECK3-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 8
// CHECK3-NEXT: store double [[TMP12]], double* [[TMP8]], align 8
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN2SSC2ERi
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK3-SAME: (%struct.SS* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32* nonnull align 4 dereferenceable(4) [[D:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK3-NEXT: [[D_ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT: [[A2:%.*]] = alloca i32*, align 8
// CHECK3-NEXT: [[B4:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[C7:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK3-NEXT: [[B_CASTED:%.*]] = alloca i64, align 8
// CHECK3-NEXT: [[C_CASTED:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK3-NEXT: store i32* [[D]], i32** [[D_ADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], %struct.SS* [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: store i32 0, i32* [[A]], align 8
// CHECK3-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 1
// CHECK3-NEXT: [[BF_LOAD:%.*]] = load i8, i8* [[B]], align 4
// CHECK3-NEXT: [[BF_CLEAR:%.*]] = and i8 [[BF_LOAD]], -16
// CHECK3-NEXT: store i8 [[BF_CLEAR]], i8* [[B]], align 4
// CHECK3-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 2
// CHECK3-NEXT: [[TMP0:%.*]] = load i32*, i32** [[D_ADDR]], align 8
// CHECK3-NEXT: store i32* [[TMP0]], i32** [[C]], align 8
// CHECK3-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: store i32* [[A3]], i32** [[A2]], align 8
// CHECK3-NEXT: [[B5:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 1
// CHECK3-NEXT: [[BF_LOAD6:%.*]] = load i8, i8* [[B5]], align 4
// CHECK3-NEXT: [[BF_SHL:%.*]] = shl i8 [[BF_LOAD6]], 4
// CHECK3-NEXT: [[BF_ASHR:%.*]] = ashr i8 [[BF_SHL]], 4
// CHECK3-NEXT: [[BF_CAST:%.*]] = sext i8 [[BF_ASHR]] to i32
// CHECK3-NEXT: store i32 [[BF_CAST]], i32* [[B4]], align 4
// CHECK3-NEXT: [[C8:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 2
// CHECK3-NEXT: [[TMP1:%.*]] = load i32*, i32** [[C8]], align 8
// CHECK3-NEXT: store i32* [[TMP1]], i32** [[C7]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP2:%.*]] = load i32*, i32** [[A2]], align 8
// CHECK3-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4
// CHECK3-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to i32*
// CHECK3-NEXT: store i32 [[TMP3]], i32* [[CONV]], align 4
// CHECK3-NEXT: [[TMP4:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = load i32, i32* [[B4]], align 4
// CHECK3-NEXT: [[CONV9:%.*]] = bitcast i64* [[B_CASTED]] to i32*
// CHECK3-NEXT: store i32 [[TMP5]], i32* [[CONV9]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = load i64, i64* [[B_CASTED]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = load i32*, i32** [[C7]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK3-NEXT: [[CONV10:%.*]] = bitcast i64* [[C_CASTED]] to i32*
// CHECK3-NEXT: store i32 [[TMP8]], i32* [[CONV10]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = load i64, i64* [[C_CASTED]], align 8
// CHECK3-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SS*, i64, i64, i64)* @.omp_outlined..8 to void (i32*, i32*, ...)*), %struct.SS* [[THIS1]], i64 [[TMP4]], i64 [[TMP6]], i64 [[TMP9]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-LABEL: define {{[^@]+}}@.omp_outlined..8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SS* [[THIS:%.*]], i64 [[A:%.*]], i64 [[B:%.*]], i64 [[C:%.*]]) #[[ATTR12]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK3-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8
// CHECK3-NEXT: [[C_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[TMP:%.*]] = alloca i32*, align 8
// CHECK3-NEXT: [[_TMP3:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[_TMP4:%.*]] = alloca i32*, align 8
// CHECK3-NEXT: [[_TMP5:%.*]] = alloca i32*, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON_1:%.*]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [3 x i8*], align 8
// CHECK3-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK3-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK3-NEXT: store i64 [[B]], i64* [[B_ADDR]], align 8
// CHECK3-NEXT: store i64 [[C]], i64* [[C_ADDR]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to i32*
// CHECK3-NEXT: [[CONV1:%.*]] = bitcast i64* [[B_ADDR]] to i32*
// CHECK3-NEXT: [[CONV2:%.*]] = bitcast i64* [[C_ADDR]] to i32*
// CHECK3-NEXT: store i32* [[CONV]], i32** [[TMP]], align 8
// CHECK3-NEXT: store i32* [[CONV2]], i32** [[_TMP3]], align 8
// CHECK3-NEXT: [[TMP1:%.*]] = load i32*, i32** [[TMP]], align 8
// CHECK3-NEXT: store i32* [[TMP1]], i32** [[_TMP4]], align 8
// CHECK3-NEXT: [[TMP2:%.*]] = load i32*, i32** [[_TMP3]], align 8
// CHECK3-NEXT: store i32* [[TMP2]], i32** [[_TMP5]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK3-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK3-NEXT: [[TMP6:%.*]] = icmp ne i32 [[TMP5]], 0
// CHECK3-NEXT: br i1 [[TMP6]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 0
// CHECK3-NEXT: store %struct.SS* [[TMP0]], %struct.SS** [[TMP7]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 1
// CHECK3-NEXT: [[TMP9:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK3-NEXT: store i32* [[TMP9]], i32** [[TMP8]], align 8
// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 2
// CHECK3-NEXT: store i32* [[CONV1]], i32** [[TMP10]], align 8
// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 3
// CHECK3-NEXT: [[TMP12:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK3-NEXT: store i32* [[TMP12]], i32** [[TMP11]], align 8
// CHECK3-NEXT: invoke void @_ZZN2SSC1ERiENKUlvE_clEv(%class.anon.1* nonnull align 8 dereferenceable(32) [[REF_TMP]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK3: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: lpad:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP13:%.*]] = landingpad { i8*, i32 }
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: catch i8* null
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP14:%.*]] = extractvalue { i8*, i32 } [[TMP13]], 0
// CHECK3-NEXT: store i8* [[TMP14]], i8** [[EXN_SLOT]], align 8
// CHECK3-NEXT: [[TMP15:%.*]] = extractvalue { i8*, i32 } [[TMP13]], 1
// CHECK3-NEXT: store i32 [[TMP15]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK3-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK3: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK3-NEXT: [[TMP17:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK3-NEXT: [[TMP18:%.*]] = bitcast i32* [[TMP17]] to i8*
// CHECK3-NEXT: store i8* [[TMP18]], i8** [[TMP16]], align 8
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1
// CHECK3-NEXT: [[TMP20:%.*]] = bitcast i32* [[CONV1]] to i8*
// CHECK3-NEXT: store i8* [[TMP20]], i8** [[TMP19]], align 8
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 2
// CHECK3-NEXT: [[TMP22:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK3-NEXT: [[TMP23:%.*]] = bitcast i32* [[TMP22]] to i8*
// CHECK3-NEXT: store i8* [[TMP23]], i8** [[TMP21]], align 8
// CHECK3-NEXT: [[TMP24:%.*]] = bitcast [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK3-NEXT: [[TMP25:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK3-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]], i64 24, i8* [[TMP24]], void (i8*, i8*)* @.omp.copyprivate.copy_func.9, i32 [[TMP25]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
// CHECK3: terminate.handler:
// CHECK3-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK3-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13]]
// CHECK3-NEXT: unreachable
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZZN2SSC1ERiENKUlvE_clEv
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-SAME: (%class.anon.1* nonnull align 8 dereferenceable(32) [[THIS:%.*]]) #[[ATTR10]] align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %class.anon.1*, align 8
// CHECK3-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK3-NEXT: [[B_CASTED:%.*]] = alloca i64, align 8
// CHECK3-NEXT: [[C_CASTED:%.*]] = alloca i64, align 8
// CHECK3-NEXT: store %class.anon.1* [[THIS]], %class.anon.1** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %class.anon.1*, %class.anon.1** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON_1:%.*]], %class.anon.1* [[THIS1]], i32 0, i32 0
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[TMP1:%.*]] = load %struct.SS*, %struct.SS** [[TMP0]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 1
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[TMP3:%.*]] = load i32*, i32** [[TMP2]], align 8
// CHECK3-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK3-NEXT: [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// CHECK3-NEXT: store i32 [[INC]], i32* [[TMP3]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 2
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[TMP6:%.*]] = load i32*, i32** [[TMP5]], align 8
// CHECK3-NEXT: [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4
// CHECK3-NEXT: [[DEC:%.*]] = add nsw i32 [[TMP7]], -1
// CHECK3-NEXT: store i32 [[DEC]], i32* [[TMP6]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 3
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[TMP9:%.*]] = load i32*, i32** [[TMP8]], align 8
// CHECK3-NEXT: [[TMP10:%.*]] = load i32, i32* [[TMP9]], align 4
// CHECK3-NEXT: [[DIV:%.*]] = sdiv i32 [[TMP10]], 1
// CHECK3-NEXT: store i32 [[DIV]], i32* [[TMP9]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 1
// CHECK3-NEXT: [[TMP12:%.*]] = load i32*, i32** [[TMP11]], align 8
// CHECK3-NEXT: [[TMP13:%.*]] = load i32, i32* [[TMP12]], align 4
// CHECK3-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to i32*
// CHECK3-NEXT: store i32 [[TMP13]], i32* [[CONV]], align 4
// CHECK3-NEXT: [[TMP14:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 2
// CHECK3-NEXT: [[TMP16:%.*]] = load i32*, i32** [[TMP15]], align 8
// CHECK3-NEXT: [[TMP17:%.*]] = load i32, i32* [[TMP16]], align 4
// CHECK3-NEXT: [[CONV2:%.*]] = bitcast i64* [[B_CASTED]] to i32*
// CHECK3-NEXT: store i32 [[TMP17]], i32* [[CONV2]], align 4
// CHECK3-NEXT: [[TMP18:%.*]] = load i64, i64* [[B_CASTED]], align 8
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 3
// CHECK3-NEXT: [[TMP20:%.*]] = load i32*, i32** [[TMP19]], align 8
// CHECK3-NEXT: [[TMP21:%.*]] = load i32, i32* [[TMP20]], align 4
// CHECK3-NEXT: [[CONV3:%.*]] = bitcast i64* [[C_CASTED]] to i32*
// CHECK3-NEXT: store i32 [[TMP21]], i32* [[CONV3]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = load i64, i64* [[C_CASTED]], align 8
// CHECK3-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SS*, i64, i64, i64)* @.omp_outlined..10 to void (i32*, i32*, ...)*), %struct.SS* [[TMP1]], i64 [[TMP14]], i64 [[TMP18]], i64 [[TMP22]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.9
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK3-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK3-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK3-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [3 x i8*]*
// CHECK3-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [3 x i8*]*
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK3-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to i32*
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK3-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK3-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to i32*
// CHECK3-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
// CHECK3-NEXT: store i32 [[TMP12]], i32* [[TMP8]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 1
// CHECK3-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8
// CHECK3-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i32*
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 1
// CHECK3-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8
// CHECK3-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32*
// CHECK3-NEXT: [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
// CHECK3-NEXT: store i32 [[TMP19]], i32* [[TMP15]], align 4
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 2
// CHECK3-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8
// CHECK3-NEXT: [[TMP22:%.*]] = bitcast i8* [[TMP21]] to i32*
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 2
// CHECK3-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8
// CHECK3-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32*
// CHECK3-NEXT: [[TMP26:%.*]] = load i32, i32* [[TMP25]], align 4
// CHECK3-NEXT: store i32 [[TMP26]], i32* [[TMP22]], align 4
// CHECK3-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-LABEL: define {{[^@]+}}@.omp_outlined..10
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SS* [[THIS:%.*]], i64 [[A:%.*]], i64 [[B:%.*]], i64 [[C:%.*]]) #[[ATTR12]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK3-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8
// CHECK3-NEXT: [[C_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[TMP:%.*]] = alloca i32*, align 8
// CHECK3-NEXT: [[_TMP3:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[_TMP4:%.*]] = alloca i32*, align 8
// CHECK3-NEXT: [[_TMP5:%.*]] = alloca i32*, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [3 x i8*], align 8
// CHECK3-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK3-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK3-NEXT: store i64 [[B]], i64* [[B_ADDR]], align 8
// CHECK3-NEXT: store i64 [[C]], i64* [[C_ADDR]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to i32*
// CHECK3-NEXT: [[CONV1:%.*]] = bitcast i64* [[B_ADDR]] to i32*
// CHECK3-NEXT: [[CONV2:%.*]] = bitcast i64* [[C_ADDR]] to i32*
// CHECK3-NEXT: store i32* [[CONV]], i32** [[TMP]], align 8
// CHECK3-NEXT: store i32* [[CONV2]], i32** [[_TMP3]], align 8
// CHECK3-NEXT: [[TMP1:%.*]] = load i32*, i32** [[TMP]], align 8
// CHECK3-NEXT: store i32* [[TMP1]], i32** [[_TMP4]], align 8
// CHECK3-NEXT: [[TMP2:%.*]] = load i32*, i32** [[_TMP3]], align 8
// CHECK3-NEXT: store i32* [[TMP2]], i32** [[_TMP5]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK3-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK3-NEXT: [[TMP6:%.*]] = icmp ne i32 [[TMP5]], 0
// CHECK3-NEXT: br i1 [[TMP6]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP7:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK3-NEXT: [[INC:%.*]] = add nsw i32 [[TMP8]], 1
// CHECK3-NEXT: store i32 [[INC]], i32* [[TMP7]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK3-NEXT: [[DEC:%.*]] = add nsw i32 [[TMP9]], -1
// CHECK3-NEXT: store i32 [[DEC]], i32* [[CONV1]], align 8
// CHECK3-NEXT: [[TMP10:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK3-NEXT: [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK3-NEXT: [[DIV:%.*]] = sdiv i32 [[TMP11]], 1
// CHECK3-NEXT: store i32 [[DIV]], i32* [[TMP10]], align 4
// CHECK3-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK3-NEXT: [[TMP13:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK3-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8*
// CHECK3-NEXT: store i8* [[TMP14]], i8** [[TMP12]], align 8
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1
// CHECK3-NEXT: [[TMP16:%.*]] = bitcast i32* [[CONV1]] to i8*
// CHECK3-NEXT: store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 2
// CHECK3-NEXT: [[TMP18:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK3-NEXT: [[TMP19:%.*]] = bitcast i32* [[TMP18]] to i8*
// CHECK3-NEXT: store i8* [[TMP19]], i8** [[TMP17]], align 8
// CHECK3-NEXT: [[TMP20:%.*]] = bitcast [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK3-NEXT: [[TMP21:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK3-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]], i64 24, i8* [[TMP20]], void (i8*, i8*)* @.omp.copyprivate.copy_func.11, i32 [[TMP21]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.11
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK3-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK3-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK3-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [3 x i8*]*
// CHECK3-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK3-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [3 x i8*]*
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK3-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK3-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to i32*
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK3-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK3-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to i32*
// CHECK3-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
// CHECK3-NEXT: store i32 [[TMP12]], i32* [[TMP8]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 1
// CHECK3-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8
// CHECK3-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i32*
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 1
// CHECK3-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8
// CHECK3-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32*
// CHECK3-NEXT: [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
// CHECK3-NEXT: store i32 [[TMP19]], i32* [[TMP15]], align 4
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 2
// CHECK3-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8
// CHECK3-NEXT: [[TMP22:%.*]] = bitcast i8* [[TMP21]] to i32*
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 2
// CHECK3-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8
// CHECK3-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32*
// CHECK3-NEXT: [[TMP26:%.*]] = load i32, i32* [[TMP25]], align 4
// CHECK3-NEXT: store i32 [[TMP26]], i32* [[TMP22]], align 4
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_Z15parallel_singlev
// CHECK3-SAME: () #[[ATTR10]] {
// CHECK3-NEXT: entry:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* @.omp_outlined..12 to void (i32*, i32*, ...)*))
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-LABEL: define {{[^@]+}}@.omp_outlined..12
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]]) #[[ATTR12]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK3-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK3-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK3-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP0:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
// CHECK3-NEXT: [[TMP3:%.*]] = icmp ne i32 [[TMP2]], 0
// CHECK3-NEXT: br i1 [[TMP3]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3: omp_if.then:
// CHECK3-NEXT: invoke void @_Z3foov()
// CHECK3-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK3: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: lpad:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP4:%.*]] = landingpad { i8*, i32 }
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: catch i8* null
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: [[TMP5:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 0
// CHECK3-NEXT: store i8* [[TMP5]], i8** [[EXN_SLOT]], align 8
// CHECK3-NEXT: [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 1
// CHECK3-NEXT: store i32 [[TMP6]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK3-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK3: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
// CHECK3: terminate.handler:
// CHECK3-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK3-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13]]
// CHECK3-NEXT: unreachable
//
//
// CHECK3-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_single_codegen.cpp
// CHECK3-SAME: () #[[ATTR0]] {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: call void @__cxx_global_var_init()
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK3-NEXT: call void @__cxx_global_var_init.4()
// CHECK3-NEXT: call void @.__omp_threadprivate_init_.()
// CHECK3-NEXT: call void @.__omp_threadprivate_init_..3()
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK3-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__cxx_global_var_init
// CHECK4-SAME: () #[[ATTR0:[0-9]+]] {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
// CHECK4-NEXT: call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast (%class.TestClass* @tc to i8*), i8* (i8*)* @.__kmpc_global_ctor_., i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_.)
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: call void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) @tc)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[TMP1:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%class.TestClass*)* @_ZN9TestClassD1Ev to void (i8*)*), i8* bitcast (%class.TestClass* @tc to i8*), i8* @__dso_handle) #[[ATTR3:[0-9]+]]
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_.
// CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK4-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK4-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: call void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP2]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK4-NEXT: ret i8* [[TMP3]]
//
//
// CHECK4-LABEL: define {{[^@]+}}@_ZN9TestClassC1Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK4-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: call void @_ZN9TestClassC2Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@.__kmpc_global_dtor_.
// CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK4-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK4-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP2]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_ZN9TestClassD1Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2:[0-9]+]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK4-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: call void @_ZN9TestClassD2Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__cxx_global_var_init.1
// CHECK4-SAME: () #[[ATTR0]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK4-NEXT: call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast ([2 x %class.TestClass]* @tc2 to i8*), i8* (i8*)* @.__kmpc_global_ctor_..2, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..3)
// CHECK4-NEXT: br label [[ARRAYCTOR_LOOP:%.*]]
// CHECK4: arrayctor.loop:
// CHECK4-NEXT: [[ARRAYCTOR_CUR:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[INVOKE_CONT:%.*]] ]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: invoke void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: to label [[INVOKE_CONT]] unwind label [[LPAD:%.*]]
// CHECK4: invoke.cont:
// CHECK4-NEXT: [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAYCTOR_CUR]], i64 1
// CHECK4-NEXT: [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([[CLASS_TESTCLASS]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2)
// CHECK4-NEXT: br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
// CHECK4: arrayctor.cont:
// CHECK4-NEXT: [[TMP1:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]]
// CHECK4-NEXT: ret void
// CHECK4: lpad:
// CHECK4-NEXT: [[TMP2:%.*]] = landingpad { i8*, i32 }
// CHECK4-NEXT: cleanup
// CHECK4-NEXT: [[TMP3:%.*]] = extractvalue { i8*, i32 } [[TMP2]], 0
// CHECK4-NEXT: store i8* [[TMP3]], i8** [[EXN_SLOT]], align 8
// CHECK4-NEXT: [[TMP4:%.*]] = extractvalue { i8*, i32 } [[TMP2]], 1
// CHECK4-NEXT: store i32 [[TMP4]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK4-NEXT: [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), [[ARRAYCTOR_CUR]]
// CHECK4-NEXT: br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY:%.*]]
// CHECK4: arraydestroy.body:
// CHECK4-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ [[ARRAYCTOR_CUR]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK4-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0)
// CHECK4-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1]], label [[ARRAYDESTROY_BODY]]
// CHECK4: arraydestroy.done1:
// CHECK4-NEXT: br label [[EH_RESUME:%.*]]
// CHECK4: eh.resume:
// CHECK4-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK4-NEXT: [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4
// CHECK4-NEXT: [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0
// CHECK4-NEXT: [[LPAD_VAL2:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1
// CHECK4-NEXT: resume { i8*, i32 } [[LPAD_VAL2]]
//
//
// CHECK4-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_..2
// CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK4-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK4-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to [2 x %class.TestClass]*
// CHECK4-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %class.TestClass], [2 x %class.TestClass]* [[TMP2]], i32 0, i32 0
// CHECK4-NEXT: [[ARRAYCTOR_END:%.*]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAY_BEGIN]], i64 2
// CHECK4-NEXT: br label [[ARRAYCTOR_LOOP:%.*]]
// CHECK4: arrayctor.loop:
// CHECK4-NEXT: [[ARRAYCTOR_CUR:%.*]] = phi %class.TestClass* [ [[ARRAY_BEGIN]], [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[INVOKE_CONT:%.*]] ]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: invoke void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: to label [[INVOKE_CONT]] unwind label [[LPAD:%.*]]
// CHECK4: invoke.cont:
// CHECK4-NEXT: [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYCTOR_CUR]], i64 1
// CHECK4-NEXT: [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], [[ARRAYCTOR_END]]
// CHECK4-NEXT: br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]]
// CHECK4: arrayctor.cont:
// CHECK4-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK4-NEXT: ret i8* [[TMP3]]
// CHECK4: lpad:
// CHECK4-NEXT: [[TMP4:%.*]] = landingpad { i8*, i32 }
// CHECK4-NEXT: cleanup
// CHECK4-NEXT: [[TMP5:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 0
// CHECK4-NEXT: store i8* [[TMP5]], i8** [[EXN_SLOT]], align 8
// CHECK4-NEXT: [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 1
// CHECK4-NEXT: store i32 [[TMP6]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK4-NEXT: [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %class.TestClass* [[ARRAY_BEGIN]], [[ARRAYCTOR_CUR]]
// CHECK4-NEXT: br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY:%.*]]
// CHECK4: arraydestroy.body:
// CHECK4-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ [[ARRAYCTOR_CUR]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK4-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK4-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1]], label [[ARRAYDESTROY_BODY]]
// CHECK4: arraydestroy.done1:
// CHECK4-NEXT: br label [[EH_RESUME:%.*]]
// CHECK4: eh.resume:
// CHECK4-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK4-NEXT: [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4
// CHECK4-NEXT: [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0
// CHECK4-NEXT: [[LPAD_VAL2:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1
// CHECK4-NEXT: resume { i8*, i32 } [[LPAD_VAL2]]
//
//
// CHECK4-LABEL: define {{[^@]+}}@.__kmpc_global_dtor_..3
// CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK4-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK4-NEXT: [[ARRAY_BEGIN:%.*]] = bitcast i8* [[TMP1]] to %class.TestClass*
// CHECK4-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAY_BEGIN]], i64 2
// CHECK4-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK4: arraydestroy.body:
// CHECK4-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ [[TMP2]], [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK4-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK4-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK4: arraydestroy.done1:
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
// CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK4-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK4: arraydestroy.body:
// CHECK4-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([[CLASS_TESTCLASS:%.*]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK4-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0)
// CHECK4-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK4: arraydestroy.done1:
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_Z3foov
// CHECK4-SAME: () #[[ATTR4:[0-9]+]] {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: call void @_Z8mayThrowv()
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@main
// CHECK4-SAME: () #[[ATTR6:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[RETVAL:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[A:%.*]] = alloca i8, align 1
// CHECK4-NEXT: [[A2:%.*]] = alloca [2 x i8], align 1
// CHECK4-NEXT: [[C:%.*]] = alloca %class.TestClass*, align 8
// CHECK4-NEXT: [[SST:%.*]] = alloca [[STRUCT_SST:%.*]], align 8
// CHECK4-NEXT: [[SS:%.*]] = alloca [[STRUCT_SS:%.*]], align 8
// CHECK4-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [5 x i8*], align 8
// CHECK4-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1]])
// CHECK4-NEXT: store i32 0, i32* [[RETVAL]], align 4
// CHECK4-NEXT: store %class.TestClass* @tc, %class.TestClass** [[C]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: call void @_ZN3SSTIdEC1Ev(%struct.SST* nonnull align 8 dereferenceable(8) [[SST]])
// CHECK4-NEXT: call void @_ZN2SSC1ERi(%struct.SS* nonnull align 8 dereferenceable(16) [[SS]], i32* nonnull align 4 dereferenceable(4) getelementptr inbounds ([[CLASS_TESTCLASS:%.*]], %class.TestClass* @tc, i32 0, i32 0))
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT: [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0
// CHECK4-NEXT: br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
// CHECK4: omp_if.then:
// CHECK4-NEXT: store i8 2, i8* [[A]], align 1
// CHECK4-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT: br label [[OMP_IF_END]]
// CHECK4: omp_if.end:
// CHECK4-NEXT: [[TMP3:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT: [[TMP4:%.*]] = icmp ne i32 [[TMP3]], 0
// CHECK4-NEXT: br i1 [[TMP4]], label [[OMP_IF_THEN1:%.*]], label [[OMP_IF_END2:%.*]]
// CHECK4: omp_if.then1:
// CHECK4-NEXT: store i8 2, i8* [[A]], align 1
// CHECK4-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT: br label [[OMP_IF_END2]]
// CHECK4: omp_if.end2:
// CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], i32 [[TMP0]])
// CHECK4-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK4-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT: [[TMP6:%.*]] = icmp ne i32 [[TMP5]], 0
// CHECK4-NEXT: br i1 [[TMP6]], label [[OMP_IF_THEN3:%.*]], label [[OMP_IF_END4:%.*]]
// CHECK4: omp_if.then3:
// CHECK4-NEXT: invoke void @_Z3foov()
// CHECK4-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK4: invoke.cont:
// CHECK4-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK4-NEXT: br label [[OMP_IF_END4]]
// CHECK4: lpad:
// CHECK4-NEXT: [[TMP7:%.*]] = landingpad { i8*, i32 }
// CHECK4-NEXT: catch i8* null
// CHECK4-NEXT: [[TMP8:%.*]] = extractvalue { i8*, i32 } [[TMP7]], 0
// CHECK4-NEXT: store i8* [[TMP8]], i8** [[EXN_SLOT]], align 8
// CHECK4-NEXT: [[TMP9:%.*]] = extractvalue { i8*, i32 } [[TMP7]], 1
// CHECK4-NEXT: store i32 [[TMP9]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK4-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK4-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK4: omp_if.end4:
// CHECK4-NEXT: [[TMP10:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK4-NEXT: store i8* [[A]], i8** [[TMP10]], align 8
// CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1
// CHECK4-NEXT: store i8* bitcast (%class.TestClass* @tc to i8*), i8** [[TMP11]], align 8
// CHECK4-NEXT: [[TMP12:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 2
// CHECK4-NEXT: [[TMP13:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i8* bitcast (%class.TestClass* @tc to i8*), i64 4, i8*** @tc.cache.)
// CHECK4-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to %class.TestClass*
// CHECK4-NEXT: [[TMP15:%.*]] = bitcast %class.TestClass* [[TMP14]] to i8*
// CHECK4-NEXT: store i8* [[TMP15]], i8** [[TMP12]], align 8
// CHECK4-NEXT: [[TMP16:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 3
// CHECK4-NEXT: [[TMP17:%.*]] = bitcast [2 x i8]* [[A2]] to i8*
// CHECK4-NEXT: store i8* [[TMP17]], i8** [[TMP16]], align 8
// CHECK4-NEXT: [[TMP18:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 4
// CHECK4-NEXT: [[TMP19:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i8* bitcast ([2 x %class.TestClass]* @tc2 to i8*), i64 8, i8*** @tc2.cache.)
// CHECK4-NEXT: [[TMP20:%.*]] = bitcast i8* [[TMP19]] to [2 x %class.TestClass]*
// CHECK4-NEXT: [[TMP21:%.*]] = bitcast [2 x %class.TestClass]* [[TMP20]] to i8*
// CHECK4-NEXT: store i8* [[TMP21]], i8** [[TMP18]], align 8
// CHECK4-NEXT: [[TMP22:%.*]] = bitcast [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK4-NEXT: [[TMP23:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK4-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i64 40, i8* [[TMP22]], void (i8*, i8*)* @.omp.copyprivate.copy_func, i32 [[TMP23]])
// CHECK4-NEXT: [[TMP24:%.*]] = load i8, i8* [[A]], align 1
// CHECK4-NEXT: [[CONV:%.*]] = sext i8 [[TMP24]] to i32
// CHECK4-NEXT: ret i32 [[CONV]]
// CHECK4: terminate.handler:
// CHECK4-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK4-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13:[0-9]+]]
// CHECK4-NEXT: unreachable
//
//
// CHECK4-LABEL: define {{[^@]+}}@_ZN3SSTIdEC1Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-SAME: (%struct.SST* nonnull align 8 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK4-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: call void @_ZN3SSTIdEC2Ev(%struct.SST* nonnull align 8 dereferenceable(8) [[THIS1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_ZN2SSC1ERi
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-SAME: (%struct.SS* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32* nonnull align 4 dereferenceable(4) [[D:%.*]]) unnamed_addr #[[ATTR1]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK4-NEXT: [[D_ADDR:%.*]] = alloca i32*, align 8
// CHECK4-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK4-NEXT: store i32* [[D]], i32** [[D_ADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[TMP0:%.*]] = load i32*, i32** [[D_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: call void @_ZN2SSC2ERi(%struct.SS* nonnull align 8 dereferenceable(16) [[THIS1]], i32* nonnull align 4 dereferenceable(4) [[TMP0]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__clang_call_terminate
// CHECK4-SAME: (i8* [[TMP0:%.*]]) #[[ATTR8:[0-9]+]] comdat {
// CHECK4-NEXT: [[TMP2:%.*]] = call i8* @__cxa_begin_catch(i8* [[TMP0]]) #[[ATTR3]]
// CHECK4-NEXT: call void @_ZSt9terminatev() #[[ATTR13]]
// CHECK4-NEXT: unreachable
//
//
// CHECK4-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func
// CHECK4-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9:[0-9]+]] {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK4-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK4-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK4-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [5 x i8*]*
// CHECK4-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK4-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [5 x i8*]*
// CHECK4-NEXT: [[TMP6:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK4-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK4-NEXT: [[TMP8:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK4-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8
// CHECK4-NEXT: [[TMP10:%.*]] = load i8, i8* [[TMP9]], align 1
// CHECK4-NEXT: store i8 [[TMP10]], i8* [[TMP7]], align 1
// CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 1
// CHECK4-NEXT: [[TMP12:%.*]] = load i8*, i8** [[TMP11]], align 8
// CHECK4-NEXT: [[TMP13:%.*]] = bitcast i8* [[TMP12]] to %class.TestClass*
// CHECK4-NEXT: [[TMP14:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 1
// CHECK4-NEXT: [[TMP15:%.*]] = load i8*, i8** [[TMP14]], align 8
// CHECK4-NEXT: [[TMP16:%.*]] = bitcast i8* [[TMP15]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %class.TestClass* @_ZN9TestClassaSERKS_(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP13]], %class.TestClass* nonnull align 4 dereferenceable(4) [[TMP16]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[TMP17:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 2
// CHECK4-NEXT: [[TMP18:%.*]] = load i8*, i8** [[TMP17]], align 8
// CHECK4-NEXT: [[TMP19:%.*]] = bitcast i8* [[TMP18]] to %class.TestClass*
// CHECK4-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 2
// CHECK4-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8
// CHECK4-NEXT: [[TMP22:%.*]] = bitcast i8* [[TMP21]] to %class.TestClass*
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: [[CALL2:%.*]] = call nonnull align 4 dereferenceable(4) %class.TestClass* @_ZN9TestClassaSERKS_(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP19]], %class.TestClass* nonnull align 4 dereferenceable(4) [[TMP22]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 3
// CHECK4-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8
// CHECK4-NEXT: [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 3
// CHECK4-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 8
// CHECK4-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 1 [[TMP24]], i8* align 1 [[TMP26]], i64 2, i1 false)
// CHECK4-NEXT: [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 4
// CHECK4-NEXT: [[TMP28:%.*]] = load i8*, i8** [[TMP27]], align 8
// CHECK4-NEXT: [[TMP29:%.*]] = bitcast i8* [[TMP28]] to %class.TestClass*
// CHECK4-NEXT: [[TMP30:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 4
// CHECK4-NEXT: [[TMP31:%.*]] = load i8*, i8** [[TMP30]], align 8
// CHECK4-NEXT: [[TMP32:%.*]] = bitcast i8* [[TMP31]] to %class.TestClass*
// CHECK4-NEXT: [[TMP33:%.*]] = getelementptr [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[TMP29]], i64 2
// CHECK4-NEXT: [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq %class.TestClass* [[TMP29]], [[TMP33]]
// CHECK4-NEXT: br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
// CHECK4: omp.arraycpy.body:
// CHECK4-NEXT: [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %class.TestClass* [ [[TMP32]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
// CHECK4-NEXT: [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %class.TestClass* [ [[TMP29]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: [[CALL3:%.*]] = call nonnull align 4 dereferenceable(4) %class.TestClass* @_ZN9TestClassaSERKS_(%class.TestClass* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %class.TestClass* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[CLASS_TESTCLASS]], %class.TestClass* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
// CHECK4-NEXT: [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[CLASS_TESTCLASS]], %class.TestClass* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
// CHECK4-NEXT: [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %class.TestClass* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP33]]
// CHECK4-NEXT: br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE4]], label [[OMP_ARRAYCPY_BODY]]
// CHECK4: omp.arraycpy.done4:
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_ZN9TestClassaSERKS_
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]], %class.TestClass* nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) #[[ATTR10:[0-9]+]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK4-NEXT: [[DOTADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK4-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK4-NEXT: store %class.TestClass* [[TMP0]], %class.TestClass** [[DOTADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
// CHECK4-NEXT: ret %class.TestClass* [[THIS1]]
//
//
// CHECK4-LABEL: define {{[^@]+}}@_Z15parallel_singlev
// CHECK4-SAME: () #[[ATTR10]] {
// CHECK4-NEXT: entry:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* @.omp_outlined. to void (i32*, i32*, ...)*))
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@.omp_outlined.
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]]) #[[ATTR12:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK4-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK4-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK4-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK4-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP0:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK4-NEXT: [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK4-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
// CHECK4-NEXT: [[TMP3:%.*]] = icmp ne i32 [[TMP2]], 0
// CHECK4-NEXT: br i1 [[TMP3]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4: omp_if.then:
// CHECK4-NEXT: invoke void @_Z3foov()
// CHECK4-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK4: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: br label [[OMP_IF_END]]
// CHECK4: lpad:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP4:%.*]] = landingpad { i8*, i32 }
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: catch i8* null
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP5:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 0
// CHECK4-NEXT: store i8* [[TMP5]], i8** [[EXN_SLOT]], align 8
// CHECK4-NEXT: [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 1
// CHECK4-NEXT: store i32 [[TMP6]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK4-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK4: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2]], i32 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
// CHECK4: terminate.handler:
// CHECK4-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK4-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13]]
// CHECK4-NEXT: unreachable
//
//
// CHECK4-LABEL: define {{[^@]+}}@_ZN9TestClassC2Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK4-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[A:%.*]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[THIS1]], i32 0, i32 0
// CHECK4-NEXT: store i32 0, i32* [[A]], align 4
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_ZN9TestClassD2Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK4-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_ZN2SSC2ERi
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-SAME: (%struct.SS* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32* nonnull align 4 dereferenceable(4) [[D:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK4-NEXT: [[D_ADDR:%.*]] = alloca i32*, align 8
// CHECK4-NEXT: [[A2:%.*]] = alloca i32*, align 8
// CHECK4-NEXT: [[B4:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[C7:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK4-NEXT: [[B_CASTED:%.*]] = alloca i64, align 8
// CHECK4-NEXT: [[C_CASTED:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK4-NEXT: store i32* [[D]], i32** [[D_ADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], %struct.SS* [[THIS1]], i32 0, i32 0
// CHECK4-NEXT: store i32 0, i32* [[A]], align 8
// CHECK4-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 1
// CHECK4-NEXT: [[BF_LOAD:%.*]] = load i8, i8* [[B]], align 4
// CHECK4-NEXT: [[BF_CLEAR:%.*]] = and i8 [[BF_LOAD]], -16
// CHECK4-NEXT: store i8 [[BF_CLEAR]], i8* [[B]], align 4
// CHECK4-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 2
// CHECK4-NEXT: [[TMP0:%.*]] = load i32*, i32** [[D_ADDR]], align 8
// CHECK4-NEXT: store i32* [[TMP0]], i32** [[C]], align 8
// CHECK4-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 0
// CHECK4-NEXT: store i32* [[A3]], i32** [[A2]], align 8
// CHECK4-NEXT: [[B5:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 1
// CHECK4-NEXT: [[BF_LOAD6:%.*]] = load i8, i8* [[B5]], align 4
// CHECK4-NEXT: [[BF_SHL:%.*]] = shl i8 [[BF_LOAD6]], 4
// CHECK4-NEXT: [[BF_ASHR:%.*]] = ashr i8 [[BF_SHL]], 4
// CHECK4-NEXT: [[BF_CAST:%.*]] = sext i8 [[BF_ASHR]] to i32
// CHECK4-NEXT: store i32 [[BF_CAST]], i32* [[B4]], align 4
// CHECK4-NEXT: [[C8:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 2
// CHECK4-NEXT: [[TMP1:%.*]] = load i32*, i32** [[C8]], align 8
// CHECK4-NEXT: store i32* [[TMP1]], i32** [[C7]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP2:%.*]] = load i32*, i32** [[A2]], align 8
// CHECK4-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4
// CHECK4-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to i32*
// CHECK4-NEXT: store i32 [[TMP3]], i32* [[CONV]], align 4
// CHECK4-NEXT: [[TMP4:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK4-NEXT: [[TMP5:%.*]] = load i32, i32* [[B4]], align 4
// CHECK4-NEXT: [[CONV9:%.*]] = bitcast i64* [[B_CASTED]] to i32*
// CHECK4-NEXT: store i32 [[TMP5]], i32* [[CONV9]], align 4
// CHECK4-NEXT: [[TMP6:%.*]] = load i64, i64* [[B_CASTED]], align 8
// CHECK4-NEXT: [[TMP7:%.*]] = load i32*, i32** [[C7]], align 8
// CHECK4-NEXT: [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK4-NEXT: [[CONV10:%.*]] = bitcast i64* [[C_CASTED]] to i32*
// CHECK4-NEXT: store i32 [[TMP8]], i32* [[CONV10]], align 4
// CHECK4-NEXT: [[TMP9:%.*]] = load i64, i64* [[C_CASTED]], align 8
// CHECK4-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SS*, i64, i64, i64)* @.omp_outlined..4 to void (i32*, i32*, ...)*), %struct.SS* [[THIS1]], i64 [[TMP4]], i64 [[TMP6]], i64 [[TMP9]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@.omp_outlined..4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SS* [[THIS:%.*]], i64 [[A:%.*]], i64 [[B:%.*]], i64 [[C:%.*]]) #[[ATTR12]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK4-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK4-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK4-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8
// CHECK4-NEXT: [[C_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[TMP:%.*]] = alloca i32*, align 8
// CHECK4-NEXT: [[_TMP3:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[_TMP4:%.*]] = alloca i32*, align 8
// CHECK4-NEXT: [[_TMP5:%.*]] = alloca i32*, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON:%.*]], align 8
// CHECK4-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [3 x i8*], align 8
// CHECK4-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK4-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK4-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK4-NEXT: store i64 [[B]], i64* [[B_ADDR]], align 8
// CHECK4-NEXT: store i64 [[C]], i64* [[C_ADDR]], align 8
// CHECK4-NEXT: [[TMP0:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to i32*
// CHECK4-NEXT: [[CONV1:%.*]] = bitcast i64* [[B_ADDR]] to i32*
// CHECK4-NEXT: [[CONV2:%.*]] = bitcast i64* [[C_ADDR]] to i32*
// CHECK4-NEXT: store i32* [[CONV]], i32** [[TMP]], align 8
// CHECK4-NEXT: store i32* [[CONV2]], i32** [[_TMP3]], align 8
// CHECK4-NEXT: [[TMP1:%.*]] = load i32*, i32** [[TMP]], align 8
// CHECK4-NEXT: store i32* [[TMP1]], i32** [[_TMP4]], align 8
// CHECK4-NEXT: [[TMP2:%.*]] = load i32*, i32** [[_TMP3]], align 8
// CHECK4-NEXT: store i32* [[TMP2]], i32** [[_TMP5]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK4-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK4-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK4-NEXT: [[TMP6:%.*]] = icmp ne i32 [[TMP5]], 0
// CHECK4-NEXT: br i1 [[TMP6]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 0
// CHECK4-NEXT: store %struct.SS* [[TMP0]], %struct.SS** [[TMP7]], align 8
// CHECK4-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 1
// CHECK4-NEXT: [[TMP9:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK4-NEXT: store i32* [[TMP9]], i32** [[TMP8]], align 8
// CHECK4-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 2
// CHECK4-NEXT: store i32* [[CONV1]], i32** [[TMP10]], align 8
// CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 3
// CHECK4-NEXT: [[TMP12:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK4-NEXT: store i32* [[TMP12]], i32** [[TMP11]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-NEXT: invoke void @_ZZN2SSC1ERiENKUlvE_clEv(%class.anon* nonnull align 8 dereferenceable(32) [[REF_TMP]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK4: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK4-NEXT: br label [[OMP_IF_END]]
// CHECK4: lpad:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP13:%.*]] = landingpad { i8*, i32 }
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: catch i8* null
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP14:%.*]] = extractvalue { i8*, i32 } [[TMP13]], 0
// CHECK4-NEXT: store i8* [[TMP14]], i8** [[EXN_SLOT]], align 8
// CHECK4-NEXT: [[TMP15:%.*]] = extractvalue { i8*, i32 } [[TMP13]], 1
// CHECK4-NEXT: store i32 [[TMP15]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK4-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK4: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK4-NEXT: [[TMP17:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK4-NEXT: [[TMP18:%.*]] = bitcast i32* [[TMP17]] to i8*
// CHECK4-NEXT: store i8* [[TMP18]], i8** [[TMP16]], align 8
// CHECK4-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1
// CHECK4-NEXT: [[TMP20:%.*]] = bitcast i32* [[CONV1]] to i8*
// CHECK4-NEXT: store i8* [[TMP20]], i8** [[TMP19]], align 8
// CHECK4-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 2
// CHECK4-NEXT: [[TMP22:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK4-NEXT: [[TMP23:%.*]] = bitcast i32* [[TMP22]] to i8*
// CHECK4-NEXT: store i8* [[TMP23]], i8** [[TMP21]], align 8
// CHECK4-NEXT: [[TMP24:%.*]] = bitcast [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK4-NEXT: [[TMP25:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK4-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]], i64 24, i8* [[TMP24]], void (i8*, i8*)* @.omp.copyprivate.copy_func.5, i32 [[TMP25]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
// CHECK4: terminate.handler:
// CHECK4-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK4-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13]]
// CHECK4-NEXT: unreachable
//
//
// CHECK4-LABEL: define {{[^@]+}}@_ZZN2SSC1ERiENKUlvE_clEv
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-SAME: (%class.anon* nonnull align 8 dereferenceable(32) [[THIS:%.*]]) #[[ATTR10]] align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %class.anon*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK4-NEXT: [[B_CASTED:%.*]] = alloca i64, align 8
// CHECK4-NEXT: [[C_CASTED:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: store %class.anon* [[THIS]], %class.anon** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %class.anon*, %class.anon** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON:%.*]], %class.anon* [[THIS1]], i32 0, i32 0
// CHECK4-NEXT: [[TMP1:%.*]] = load %struct.SS*, %struct.SS** [[TMP0]], align 8
// CHECK4-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[THIS1]], i32 0, i32 1
// CHECK4-NEXT: [[TMP3:%.*]] = load i32*, i32** [[TMP2]], align 8
// CHECK4-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK4-NEXT: [[INC:%.*]] = add nsw i32 [[TMP4]], 1
// CHECK4-NEXT: store i32 [[INC]], i32* [[TMP3]], align 4
// CHECK4-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[THIS1]], i32 0, i32 2
// CHECK4-NEXT: [[TMP6:%.*]] = load i32*, i32** [[TMP5]], align 8
// CHECK4-NEXT: [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4
// CHECK4-NEXT: [[DEC:%.*]] = add nsw i32 [[TMP7]], -1
// CHECK4-NEXT: store i32 [[DEC]], i32* [[TMP6]], align 4
// CHECK4-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[THIS1]], i32 0, i32 3
// CHECK4-NEXT: [[TMP9:%.*]] = load i32*, i32** [[TMP8]], align 8
// CHECK4-NEXT: [[TMP10:%.*]] = load i32, i32* [[TMP9]], align 4
// CHECK4-NEXT: [[DIV:%.*]] = sdiv i32 [[TMP10]], 1
// CHECK4-NEXT: store i32 [[DIV]], i32* [[TMP9]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[THIS1]], i32 0, i32 1
// CHECK4-NEXT: [[TMP12:%.*]] = load i32*, i32** [[TMP11]], align 8
// CHECK4-NEXT: [[TMP13:%.*]] = load i32, i32* [[TMP12]], align 4
// CHECK4-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to i32*
// CHECK4-NEXT: store i32 [[TMP13]], i32* [[CONV]], align 4
// CHECK4-NEXT: [[TMP14:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK4-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[THIS1]], i32 0, i32 2
// CHECK4-NEXT: [[TMP16:%.*]] = load i32*, i32** [[TMP15]], align 8
// CHECK4-NEXT: [[TMP17:%.*]] = load i32, i32* [[TMP16]], align 4
// CHECK4-NEXT: [[CONV2:%.*]] = bitcast i64* [[B_CASTED]] to i32*
// CHECK4-NEXT: store i32 [[TMP17]], i32* [[CONV2]], align 4
// CHECK4-NEXT: [[TMP18:%.*]] = load i64, i64* [[B_CASTED]], align 8
// CHECK4-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[THIS1]], i32 0, i32 3
// CHECK4-NEXT: [[TMP20:%.*]] = load i32*, i32** [[TMP19]], align 8
// CHECK4-NEXT: [[TMP21:%.*]] = load i32, i32* [[TMP20]], align 4
// CHECK4-NEXT: [[CONV3:%.*]] = bitcast i64* [[C_CASTED]] to i32*
// CHECK4-NEXT: store i32 [[TMP21]], i32* [[CONV3]], align 4
// CHECK4-NEXT: [[TMP22:%.*]] = load i64, i64* [[C_CASTED]], align 8
// CHECK4-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SS*, i64, i64, i64)* @.omp_outlined..6 to void (i32*, i32*, ...)*), %struct.SS* [[TMP1]], i64 [[TMP14]], i64 [[TMP18]], i64 [[TMP22]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.5
// CHECK4-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK4-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK4-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK4-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [3 x i8*]*
// CHECK4-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK4-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [3 x i8*]*
// CHECK4-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK4-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK4-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to i32*
// CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK4-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK4-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to i32*
// CHECK4-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
// CHECK4-NEXT: store i32 [[TMP12]], i32* [[TMP8]], align 4
// CHECK4-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 1
// CHECK4-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8
// CHECK4-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i32*
// CHECK4-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 1
// CHECK4-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8
// CHECK4-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32*
// CHECK4-NEXT: [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
// CHECK4-NEXT: store i32 [[TMP19]], i32* [[TMP15]], align 4
// CHECK4-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 2
// CHECK4-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8
// CHECK4-NEXT: [[TMP22:%.*]] = bitcast i8* [[TMP21]] to i32*
// CHECK4-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 2
// CHECK4-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8
// CHECK4-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32*
// CHECK4-NEXT: [[TMP26:%.*]] = load i32, i32* [[TMP25]], align 4
// CHECK4-NEXT: store i32 [[TMP26]], i32* [[TMP22]], align 4
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@.omp_outlined..6
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SS* [[THIS:%.*]], i64 [[A:%.*]], i64 [[B:%.*]], i64 [[C:%.*]]) #[[ATTR12]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK4-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK4-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK4-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8
// CHECK4-NEXT: [[C_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[TMP:%.*]] = alloca i32*, align 8
// CHECK4-NEXT: [[_TMP3:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[_TMP4:%.*]] = alloca i32*, align 8
// CHECK4-NEXT: [[_TMP5:%.*]] = alloca i32*, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [3 x i8*], align 8
// CHECK4-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK4-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK4-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK4-NEXT: store i64 [[B]], i64* [[B_ADDR]], align 8
// CHECK4-NEXT: store i64 [[C]], i64* [[C_ADDR]], align 8
// CHECK4-NEXT: [[TMP0:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to i32*
// CHECK4-NEXT: [[CONV1:%.*]] = bitcast i64* [[B_ADDR]] to i32*
// CHECK4-NEXT: [[CONV2:%.*]] = bitcast i64* [[C_ADDR]] to i32*
// CHECK4-NEXT: store i32* [[CONV]], i32** [[TMP]], align 8
// CHECK4-NEXT: store i32* [[CONV2]], i32** [[_TMP3]], align 8
// CHECK4-NEXT: [[TMP1:%.*]] = load i32*, i32** [[TMP]], align 8
// CHECK4-NEXT: store i32* [[TMP1]], i32** [[_TMP4]], align 8
// CHECK4-NEXT: [[TMP2:%.*]] = load i32*, i32** [[_TMP3]], align 8
// CHECK4-NEXT: store i32* [[TMP2]], i32** [[_TMP5]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK4-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK4-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK4-NEXT: [[TMP6:%.*]] = icmp ne i32 [[TMP5]], 0
// CHECK4-NEXT: br i1 [[TMP6]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP7:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK4-NEXT: [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4
// CHECK4-NEXT: [[INC:%.*]] = add nsw i32 [[TMP8]], 1
// CHECK4-NEXT: store i32 [[INC]], i32* [[TMP7]], align 4
// CHECK4-NEXT: [[TMP9:%.*]] = load i32, i32* [[CONV1]], align 8
// CHECK4-NEXT: [[DEC:%.*]] = add nsw i32 [[TMP9]], -1
// CHECK4-NEXT: store i32 [[DEC]], i32* [[CONV1]], align 8
// CHECK4-NEXT: [[TMP10:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK4-NEXT: [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4
// CHECK4-NEXT: [[DIV:%.*]] = sdiv i32 [[TMP11]], 1
// CHECK4-NEXT: store i32 [[DIV]], i32* [[TMP10]], align 4
// CHECK4-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK4-NEXT: br label [[OMP_IF_END]]
// CHECK4: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK4-NEXT: [[TMP13:%.*]] = load i32*, i32** [[_TMP4]], align 8
// CHECK4-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8*
// CHECK4-NEXT: store i8* [[TMP14]], i8** [[TMP12]], align 8
// CHECK4-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1
// CHECK4-NEXT: [[TMP16:%.*]] = bitcast i32* [[CONV1]] to i8*
// CHECK4-NEXT: store i8* [[TMP16]], i8** [[TMP15]], align 8
// CHECK4-NEXT: [[TMP17:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 2
// CHECK4-NEXT: [[TMP18:%.*]] = load i32*, i32** [[_TMP5]], align 8
// CHECK4-NEXT: [[TMP19:%.*]] = bitcast i32* [[TMP18]] to i8*
// CHECK4-NEXT: store i8* [[TMP19]], i8** [[TMP17]], align 8
// CHECK4-NEXT: [[TMP20:%.*]] = bitcast [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK4-NEXT: [[TMP21:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK4-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]], i64 24, i8* [[TMP20]], void (i8*, i8*)* @.omp.copyprivate.copy_func.7, i32 [[TMP21]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.7
// CHECK4-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK4-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK4-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK4-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [3 x i8*]*
// CHECK4-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK4-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [3 x i8*]*
// CHECK4-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK4-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK4-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to i32*
// CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK4-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK4-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to i32*
// CHECK4-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4
// CHECK4-NEXT: store i32 [[TMP12]], i32* [[TMP8]], align 4
// CHECK4-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 1
// CHECK4-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8
// CHECK4-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i32*
// CHECK4-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 1
// CHECK4-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8
// CHECK4-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32*
// CHECK4-NEXT: [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4
// CHECK4-NEXT: store i32 [[TMP19]], i32* [[TMP15]], align 4
// CHECK4-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 2
// CHECK4-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8
// CHECK4-NEXT: [[TMP22:%.*]] = bitcast i8* [[TMP21]] to i32*
// CHECK4-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 2
// CHECK4-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8
// CHECK4-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32*
// CHECK4-NEXT: [[TMP26:%.*]] = load i32, i32* [[TMP25]], align 4
// CHECK4-NEXT: store i32 [[TMP26]], i32* [[TMP22]], align 4
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_ZN3SSTIdEC2Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK4-SAME: (%struct.SST* nonnull align 8 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK4-NEXT: [[A2:%.*]] = alloca double*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SST:%.*]], %struct.SST* [[THIS1]], i32 0, i32 0
// CHECK4-NEXT: store double 0.000000e+00, double* [[A]], align 8
// CHECK4-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SST]], %struct.SST* [[THIS1]], i32 0, i32 0
// CHECK4-NEXT: store double* [[A3]], double** [[A2]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP0:%.*]] = load double*, double** [[A2]], align 8
// CHECK4-NEXT: [[TMP1:%.*]] = load double, double* [[TMP0]], align 8
// CHECK4-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to double*
// CHECK4-NEXT: store double [[TMP1]], double* [[CONV]], align 8
// CHECK4-NEXT: [[TMP2:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK4-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SST*, i64)* @.omp_outlined..8 to void (i32*, i32*, ...)*), %struct.SST* [[THIS1]], i64 [[TMP2]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@.omp_outlined..8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SST* [[THIS:%.*]], i64 [[A:%.*]]) #[[ATTR12]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK4-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK4-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[TMP:%.*]] = alloca double*, align 8
// CHECK4-NEXT: [[_TMP1:%.*]] = alloca double*, align 8
// CHECK4-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON_0:%.*]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK4-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK4-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK4-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK4-NEXT: [[TMP0:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to double*
// CHECK4-NEXT: store double* [[CONV]], double** [[TMP]], align 8
// CHECK4-NEXT: [[TMP1:%.*]] = load double*, double** [[TMP]], align 8
// CHECK4-NEXT: store double* [[TMP1]], double** [[_TMP1]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP2:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK4-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4
// CHECK4-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
// CHECK4-NEXT: [[TMP5:%.*]] = icmp ne i32 [[TMP4]], 0
// CHECK4-NEXT: br i1 [[TMP5]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 0
// CHECK4-NEXT: store %struct.SST* [[TMP0]], %struct.SST** [[TMP6]], align 8
// CHECK4-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 1
// CHECK4-NEXT: [[TMP8:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK4-NEXT: store double* [[TMP8]], double** [[TMP7]], align 8
// CHECK4-NEXT: invoke void @_ZZN3SSTIdEC1EvENKUlvE_clEv(%class.anon.0* nonnull align 8 dereferenceable(16) [[REF_TMP]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]]
// CHECK4: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK4-NEXT: br label [[OMP_IF_END]]
// CHECK4: lpad:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP9:%.*]] = landingpad { i8*, i32 }
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: catch i8* null
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP10:%.*]] = extractvalue { i8*, i32 } [[TMP9]], 0
// CHECK4-NEXT: store i8* [[TMP10]], i8** [[EXN_SLOT]], align 8
// CHECK4-NEXT: [[TMP11:%.*]] = extractvalue { i8*, i32 } [[TMP9]], 1
// CHECK4-NEXT: store i32 [[TMP11]], i32* [[EHSELECTOR_SLOT]], align 4
// CHECK4-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: br label [[TERMINATE_HANDLER:%.*]]
// CHECK4: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP12:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK4-NEXT: [[TMP13:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK4-NEXT: [[TMP14:%.*]] = bitcast double* [[TMP13]] to i8*
// CHECK4-NEXT: store i8* [[TMP14]], i8** [[TMP12]], align 8
// CHECK4-NEXT: [[TMP15:%.*]] = bitcast [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK4-NEXT: [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK4-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]], i64 8, i8* [[TMP15]], void (i8*, i8*)* @.omp.copyprivate.copy_func.9, i32 [[TMP16]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
// CHECK4: terminate.handler:
// CHECK4-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8
// CHECK4-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13]]
// CHECK4-NEXT: unreachable
//
//
// CHECK4-LABEL: define {{[^@]+}}@_ZZN3SSTIdEC1EvENKUlvE_clEv
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-SAME: (%class.anon.0* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR4]] align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %class.anon.0*, align 8
// CHECK4-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON_1:%.*]], align 8
// CHECK4-NEXT: store %class.anon.0* [[THIS]], %class.anon.0** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %class.anon.0*, %class.anon.0** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON_0:%.*]], %class.anon.0* [[THIS1]], i32 0, i32 0
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[TMP1:%.*]] = load %struct.SST*, %struct.SST** [[TMP0]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 0
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: store %struct.SST* [[TMP1]], %struct.SST** [[TMP2]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP3:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 1
// CHECK4-NEXT: [[TMP4:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[THIS1]], i32 0, i32 1
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[TMP5:%.*]] = load double*, double** [[TMP4]], align 8
// CHECK4-NEXT: store double* [[TMP5]], double** [[TMP3]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: call void @_ZZZN3SSTIdEC1EvENKUlvE_clEvENKUlvE_clEv(%class.anon.1* nonnull align 8 dereferenceable(16) [[REF_TMP]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.9
// CHECK4-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK4-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK4-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK4-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [1 x i8*]*
// CHECK4-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK4-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*
// CHECK4-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK4-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK4-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to double*
// CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK4-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK4-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to double*
// CHECK4-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 8
// CHECK4-NEXT: store double [[TMP12]], double* [[TMP8]], align 8
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_ZZZN3SSTIdEC1EvENKUlvE_clEvENKUlvE_clEv
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-SAME: (%class.anon.1* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR10]] align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %class.anon.1*, align 8
// CHECK4-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK4-NEXT: store %class.anon.1* [[THIS]], %class.anon.1** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %class.anon.1*, %class.anon.1** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON_1:%.*]], %class.anon.1* [[THIS1]], i32 0, i32 0
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[TMP1:%.*]] = load %struct.SST*, %struct.SST** [[TMP0]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 1
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[TMP3:%.*]] = load double*, double** [[TMP2]], align 8
// CHECK4-NEXT: [[TMP4:%.*]] = load double, double* [[TMP3]], align 8
// CHECK4-NEXT: [[INC:%.*]] = fadd double [[TMP4]], 1.000000e+00
// CHECK4-NEXT: store double [[INC]], double* [[TMP3]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 1
// CHECK4-NEXT: [[TMP6:%.*]] = load double*, double** [[TMP5]], align 8
// CHECK4-NEXT: [[TMP7:%.*]] = load double, double* [[TMP6]], align 8
// CHECK4-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to double*
// CHECK4-NEXT: store double [[TMP7]], double* [[CONV]], align 8
// CHECK4-NEXT: [[TMP8:%.*]] = load i64, i64* [[A_CASTED]], align 8
// CHECK4-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB1]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SST*, i64)* @.omp_outlined..10 to void (i32*, i32*, ...)*), %struct.SST* [[TMP1]], i64 [[TMP8]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@.omp_outlined..10
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SST* [[THIS:%.*]], i64 [[A:%.*]]) #[[ATTR12]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK4-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK4-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: [[TMP:%.*]] = alloca double*, align 8
// CHECK4-NEXT: [[_TMP1:%.*]] = alloca double*, align 8
// CHECK4-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK4-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK4-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK4-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK4-NEXT: [[TMP0:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to double*
// CHECK4-NEXT: store double* [[CONV]], double** [[TMP]], align 8
// CHECK4-NEXT: [[TMP1:%.*]] = load double*, double** [[TMP]], align 8
// CHECK4-NEXT: store double* [[TMP1]], double** [[_TMP1]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP2:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK4-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4
// CHECK4-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
// CHECK4-NEXT: [[TMP5:%.*]] = icmp ne i32 [[TMP4]], 0
// CHECK4-NEXT: br i1 [[TMP5]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP6:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK4-NEXT: [[TMP7:%.*]] = load double, double* [[TMP6]], align 8
// CHECK4-NEXT: [[INC:%.*]] = fadd double [[TMP7]], 1.000000e+00
// CHECK4-NEXT: store double [[INC]], double* [[TMP6]], align 8
// CHECK4-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK4-NEXT: br label [[OMP_IF_END]]
// CHECK4: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK4-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK4-NEXT: [[TMP9:%.*]] = load double*, double** [[_TMP1]], align 8
// CHECK4-NEXT: [[TMP10:%.*]] = bitcast double* [[TMP9]] to i8*
// CHECK4-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 8
// CHECK4-NEXT: [[TMP11:%.*]] = bitcast [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK4-NEXT: [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK4-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP3]], i64 8, i8* [[TMP11]], void (i8*, i8*)* @.omp.copyprivate.copy_func.11, i32 [[TMP12]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.11
// CHECK4-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK4-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK4-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK4-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [1 x i8*]*
// CHECK4-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK4-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*
// CHECK4-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK4-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK4-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to double*
// CHECK4-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK4-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK4-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to double*
// CHECK4-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 8
// CHECK4-NEXT: store double [[TMP12]], double* [[TMP8]], align 8
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_single_codegen.cpp
// CHECK4-SAME: () #[[ATTR0]] {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: call void @__cxx_global_var_init()
// CHECK4-NEXT: call void @__cxx_global_var_init.1()
// CHECK4-NEXT: ret void
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_.
// CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0:[0-9]+]] section "__TEXT,__StaticInit,regular,pure_instructions" !dbg [[DBG6:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK5-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG9:![0-9]+]]
// CHECK5-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %class.TestClass*, !dbg [[DBG9]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: call void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP2]]), !dbg [[DBG10:![0-9]+]]
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG9]]
// CHECK5-NEXT: ret i8* [[TMP3]], !dbg [[DBG9]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZN9TestClassC1Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] align 2 !dbg [[DBG11:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK5-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK5-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: call void @_ZN9TestClassC2Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS1]]), !dbg [[DBG12:![0-9]+]]
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: ret void, !dbg [[DBG13:![0-9]+]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@.__kmpc_global_dtor_.
// CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" !dbg [[DBG14:![0-9]+]] {
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK5-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG15:![0-9]+]]
// CHECK5-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to %class.TestClass*, !dbg [[DBG15]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP2]]) #[[ATTR3:[0-9]+]], !dbg [[DBG15]]
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: ret void, !dbg [[DBG16:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZN9TestClassD1Ev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2:[0-9]+]] align 2 !dbg [[DBG17:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK5-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK5-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: call void @_ZN9TestClassD2Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR3]], !dbg [[DBG18:![0-9]+]]
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: ret void, !dbg [[DBG19:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-LABEL: define {{[^@]+}}@.__omp_threadprivate_init_.
// CHECK5-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" !dbg [[DBG20:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]]), !dbg [[DBG21:![0-9]+]]
// CHECK5-NEXT: call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB1]], i8* bitcast (%class.TestClass* @tc to i8*), i8* (i8*)* @.__kmpc_global_ctor_., i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_.), !dbg [[DBG21]]
// CHECK5-NEXT: ret void, !dbg [[DBG21]]
//
//
// CHECK5-LABEL: define {{[^@]+}}@.__kmpc_global_ctor_..1
// CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG22:![0-9]+]] {
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK5-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG23:![0-9]+]]
// CHECK5-NEXT: [[TMP2:%.*]] = bitcast i8* [[TMP1]] to [2 x %class.TestClass]*, !dbg [[DBG23]]
// CHECK5-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %class.TestClass], [2 x %class.TestClass]* [[TMP2]], i32 0, i32 0, !dbg [[DBG24:![0-9]+]]
// CHECK5-NEXT: [[ARRAYCTOR_END:%.*]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAY_BEGIN]], i64 2, !dbg [[DBG24]]
// CHECK5-NEXT: br label [[ARRAYCTOR_LOOP:%.*]], !dbg [[DBG24]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: arrayctor.loop:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: [[ARRAYCTOR_CUR:%.*]] = phi %class.TestClass* [ [[ARRAY_BEGIN]], [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[INVOKE_CONT:%.*]] ], !dbg [[DBG24]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: invoke void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: to label [[INVOKE_CONT]] unwind label [[LPAD:%.*]], !dbg [[DBG24]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: invoke.cont:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYCTOR_CUR]], i64 1, !dbg [[DBG24]]
// CHECK5-NEXT: [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], [[ARRAYCTOR_END]], !dbg [[DBG24]]
// CHECK5-NEXT: br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]], !dbg [[DBG24]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: arrayctor.cont:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: [[TMP3:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG23]]
// CHECK5-NEXT: ret i8* [[TMP3]], !dbg [[DBG23]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: lpad:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: [[TMP4:%.*]] = landingpad { i8*, i32 }
// CHECK5-NEXT: cleanup, !dbg [[DBG25:![0-9]+]]
// CHECK5-NEXT: [[TMP5:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 0, !dbg [[DBG25]]
// CHECK5-NEXT: store i8* [[TMP5]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG25]]
// CHECK5-NEXT: [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 1, !dbg [[DBG25]]
// CHECK5-NEXT: store i32 [[TMP6]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG25]]
// CHECK5-NEXT: [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %class.TestClass* [[ARRAY_BEGIN]], [[ARRAYCTOR_CUR]], !dbg [[DBG24]]
// CHECK5-NEXT: br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG24]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: arraydestroy.body:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ [[ARRAYCTOR_CUR]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG24]]
// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG24]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]], !dbg [[DBG24]]
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]], !dbg [[DBG24]]
// CHECK5-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG24]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: arraydestroy.done1:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: br label [[EH_RESUME:%.*]], !dbg [[DBG24]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: eh.resume:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG24]]
// CHECK5-NEXT: [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG24]]
// CHECK5-NEXT: [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0, !dbg [[DBG24]]
// CHECK5-NEXT: [[LPAD_VAL2:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1, !dbg [[DBG24]]
// CHECK5-NEXT: resume { i8*, i32 } [[LPAD_VAL2]], !dbg [[DBG24]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-LABEL: define {{[^@]+}}@.__kmpc_global_dtor_..2
// CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" !dbg [[DBG26:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: [[TMP1:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG27:![0-9]+]]
// CHECK5-NEXT: [[ARRAY_BEGIN:%.*]] = bitcast i8* [[TMP1]] to %class.TestClass*, !dbg [[DBG27]]
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAY_BEGIN]], i64 2, !dbg [[DBG27]]
// CHECK5-NEXT: br label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG27]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: arraydestroy.body:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ [[TMP2]], [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG27]]
// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG27]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]], !dbg [[DBG27]]
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]], !dbg [[DBG27]]
// CHECK5-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG27]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: arraydestroy.done1:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: ret void, !dbg [[DBG28:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-LABEL: define {{[^@]+}}@.__omp_threadprivate_init_..3
// CHECK5-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" !dbg [[DBG29:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB3:[0-9]+]]), !dbg [[DBG30:![0-9]+]]
// CHECK5-NEXT: call void @__kmpc_threadprivate_register(%struct.ident_t* @[[GLOB3]], i8* bitcast ([2 x %class.TestClass]* @tc2 to i8*), i8* (i8*)* @.__kmpc_global_ctor_..1, i8* (i8*, i8*)* null, void (i8*)* @.__kmpc_global_dtor_..2), !dbg [[DBG30]]
// CHECK5-NEXT: ret void, !dbg [[DBG30]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-LABEL: define {{[^@]+}}@__cxx_global_var_init
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" !dbg [[DBG31:![0-9]+]] {
// CHECK5-NEXT: entry:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: call void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) @tc), !dbg [[DBG33:![0-9]+]]
// CHECK5-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%class.TestClass*)* @_ZN9TestClassD1Ev to void (i8*)*), i8* bitcast (%class.TestClass* @tc to i8*), i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG35:![0-9]+]]
// CHECK5-NEXT: ret void, !dbg [[DBG33]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-LABEL: define {{[^@]+}}@__cxx_global_var_init.4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG36:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: br label [[ARRAYCTOR_LOOP:%.*]], !dbg [[DBG37:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: arrayctor.loop:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[ARRAYCTOR_CUR:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), [[ENTRY:%.*]] ], [ [[ARRAYCTOR_NEXT:%.*]], [[INVOKE_CONT:%.*]] ], !dbg [[DBG37]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: invoke void @_ZN9TestClassC1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYCTOR_CUR]])
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: to label [[INVOKE_CONT]] unwind label [[LPAD:%.*]], !dbg [[DBG37]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[ARRAYCTOR_NEXT]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[ARRAYCTOR_CUR]], i64 1, !dbg [[DBG37]]
// CHECK5-NEXT: [[ARRAYCTOR_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYCTOR_NEXT]], getelementptr inbounds ([[CLASS_TESTCLASS]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2), !dbg [[DBG37]]
// CHECK5-NEXT: br i1 [[ARRAYCTOR_DONE]], label [[ARRAYCTOR_CONT:%.*]], label [[ARRAYCTOR_LOOP]], !dbg [[DBG37]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: arrayctor.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR3]], !dbg [[DBG39:![0-9]+]]
// CHECK5-NEXT: ret void, !dbg [[DBG39]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: lpad:
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-NEXT: [[TMP1:%.*]] = landingpad { i8*, i32 }
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: cleanup, !dbg [[DBG40:![0-9]+]]
// CHECK5-NEXT: [[TMP2:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 0, !dbg [[DBG40]]
// CHECK5-NEXT: store i8* [[TMP2]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG40]]
// CHECK5-NEXT: [[TMP3:%.*]] = extractvalue { i8*, i32 } [[TMP1]], 1, !dbg [[DBG40]]
// CHECK5-NEXT: store i32 [[TMP3]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG40]]
// CHECK5-NEXT: [[ARRAYDESTROY_ISEMPTY:%.*]] = icmp eq %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), [[ARRAYCTOR_CUR]], !dbg [[DBG37]]
// CHECK5-NEXT: br i1 [[ARRAYDESTROY_ISEMPTY]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG37]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: arraydestroy.body:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ [[ARRAYCTOR_CUR]], [[LPAD]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG37]]
// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG37]]
// CHECK5-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]], !dbg [[DBG37]]
// CHECK5-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), !dbg [[DBG37]]
// CHECK5-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG37]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: arraydestroy.done1:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: br label [[EH_RESUME:%.*]], !dbg [[DBG37]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: eh.resume:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG37]]
// CHECK5-NEXT: [[SEL:%.*]] = load i32, i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG37]]
// CHECK5-NEXT: [[LPAD_VAL:%.*]] = insertvalue { i8*, i32 } undef, i8* [[EXN]], 0, !dbg [[DBG37]]
// CHECK5-NEXT: [[LPAD_VAL2:%.*]] = insertvalue { i8*, i32 } [[LPAD_VAL]], i32 [[SEL]], 1, !dbg [[DBG37]]
// CHECK5-NEXT: resume { i8*, i32 } [[LPAD_VAL2]], !dbg [[DBG37]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
[OpenMP] Overhaul `declare target` handling This patch fixes various issues with our prior `declare target` handling and extends it to support `omp begin declare target` as well. This started with PR49649 in mind, trying to provide a way for users to avoid the "ref" global use introduced for globals with internal linkage. From there it went down the rabbit hole, e.g., all variables, even `nohost` ones, were emitted into the device code so it was impossible to determine if "ref" was needed late in the game (based on the name only). To make it really useful, `begin declare target` was needed as it can carry the `device_type`. Not emitting variables eagerly had a ripple effect. Finally, the precedence of the (explicit) declare target list items needed to be taken into account, that meant we cannot just look for any declare target attribute to make a decision. This caused the handling of functions to require fixup as well. I tried to clean up things while I was at it, e.g., we should not "parse declarations and defintions" as part of OpenMP parsing, this will always break at some point. Instead, we keep track what region we are in and act on definitions and declarations instead, this is what we do for declare variant and other begin/end directives already. Highlights: - new diagnosis for restrictions specificed in the standard, - delayed emission of globals not mentioned in an explicit list of a declare target, - omission of `nohost` globals on the host and `host` globals on the device, - no explicit parsing of declarations in-between `omp [begin] declare variant` and the corresponding end anymore, regular parsing instead, - precedence for explicit mentions in `declare target` lists over implicit mentions in the declaration-definition-seq, and - `omp allocate` declarations will now replace an earlier emitted global, if necessary. --- Notes: The patch is larger than I hoped but it turns out that most changes do on their own lead to "inconsistent states", which seem less desirable overall. After working through this I feel the standard should remove the explicit declare target forms as the delayed emission is horrible. That said, while we delay things anyway, it seems to me we check too often for the current status even though that is often not sufficient to act upon. There seems to be a lot of duplication that can probably be trimmed down. Eagerly emitting some things seems pretty weak as an argument to keep so much logic around. --- Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D101030
2021-04-22 13:57:28 +08:00
// CHECK5-LABEL: define {{[^@]+}}@__cxx_global_array_dtor
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" !dbg [[DBG41:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: br label [[ARRAYDESTROY_BODY:%.*]], !dbg [[DBG42:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: arraydestroy.body:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %class.TestClass* [ getelementptr inbounds ([[CLASS_TESTCLASS:%.*]], %class.TestClass* getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), i64 2), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ], !dbg [[DBG42]]
// CHECK5-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[CLASS_TESTCLASS]], %class.TestClass* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1, !dbg [[DBG42]]
// CHECK5-NEXT: call void @_ZN9TestClassD1Ev(%class.TestClass* nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR3]], !dbg [[DBG42]]
// CHECK5-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %class.TestClass* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %class.TestClass], [2 x %class.TestClass]* @tc2, i32 0, i32 0), !dbg [[DBG42]]
// CHECK5-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]], !dbg [[DBG42]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: arraydestroy.done1:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: ret void, !dbg [[DBG42]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZN9TestClassC2Ev
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG43:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK5-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK5-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[A:%.*]] = getelementptr inbounds [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[THIS1]], i32 0, i32 0, !dbg [[DBG44:![0-9]+]]
// CHECK5-NEXT: store i32 0, i32* [[A]], align 4, !dbg [[DBG44]]
// CHECK5-NEXT: ret void, !dbg [[DBG45:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZN9TestClassD2Ev
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG46:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK5-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK5-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: ret void, !dbg [[DBG47:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@_Z3foov
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: () #[[ATTR4:[0-9]+]] !dbg [[DBG48:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: call void @_Z8mayThrowv(), !dbg [[DBG49:![0-9]+]]
// CHECK5-NEXT: ret void, !dbg [[DBG50:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@main
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: () #[[ATTR6:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG51:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[RETVAL:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[A:%.*]] = alloca i8, align 1
// CHECK5-NEXT: [[A2:%.*]] = alloca [2 x i8], align 1
// CHECK5-NEXT: [[C:%.*]] = alloca %class.TestClass*, align 8
// CHECK5-NEXT: [[SST:%.*]] = alloca [[STRUCT_SST:%.*]], align 8
// CHECK5-NEXT: [[SS:%.*]] = alloca [[STRUCT_SS:%.*]], align 8
// CHECK5-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [5 x i8*], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB5:[0-9]+]])
// CHECK5-NEXT: store i32 0, i32* [[RETVAL]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: store %class.TestClass* @tc, %class.TestClass** [[C]], align 8, !dbg [[DBG52:![0-9]+]]
// CHECK5-NEXT: call void @_ZN3SSTIdEC1Ev(%struct.SST* nonnull align 8 dereferenceable(8) [[SST]]), !dbg [[DBG53:![0-9]+]]
// CHECK5-NEXT: call void @_ZN2SSC1ERi(%struct.SS* nonnull align 8 dereferenceable(16) [[SS]], i32* nonnull align 4 dereferenceable(4) getelementptr inbounds ([[CLASS_TESTCLASS:%.*]], %class.TestClass* @tc, i32 0, i32 0)), !dbg [[DBG54:![0-9]+]]
// CHECK5-NEXT: [[TMP1:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB5]], i32 [[TMP0]]), !dbg [[DBG55:![0-9]+]]
// CHECK5-NEXT: [[TMP2:%.*]] = icmp ne i32 [[TMP1]], 0, !dbg [[DBG55]]
// CHECK5-NEXT: br i1 [[TMP2]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]], !dbg [[DBG55]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: store i8 2, i8* [[A]], align 1, !dbg [[DBG56:![0-9]+]]
// CHECK5-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB5]], i32 [[TMP0]]), !dbg [[DBG57:![0-9]+]]
// CHECK5-NEXT: br label [[OMP_IF_END]], !dbg [[DBG57]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP3:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB7:[0-9]+]], i32 [[TMP0]]), !dbg [[DBG58:![0-9]+]]
// CHECK5-NEXT: [[TMP4:%.*]] = icmp ne i32 [[TMP3]], 0, !dbg [[DBG58]]
// CHECK5-NEXT: br i1 [[TMP4]], label [[OMP_IF_THEN1:%.*]], label [[OMP_IF_END2:%.*]], !dbg [[DBG58]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.then1:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: store i8 2, i8* [[A]], align 1, !dbg [[DBG59:![0-9]+]]
// CHECK5-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB7]], i32 [[TMP0]]), !dbg [[DBG60:![0-9]+]]
// CHECK5-NEXT: br label [[OMP_IF_END2]], !dbg [[DBG60]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.end2:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB8:[0-9]+]], i32 [[TMP0]]), !dbg [[DBG61:![0-9]+]]
// CHECK5-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4, !dbg [[DBG62:![0-9]+]]
// CHECK5-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB10:[0-9]+]], i32 [[TMP0]]), !dbg [[DBG62]]
// CHECK5-NEXT: [[TMP6:%.*]] = icmp ne i32 [[TMP5]], 0, !dbg [[DBG62]]
// CHECK5-NEXT: br i1 [[TMP6]], label [[OMP_IF_THEN3:%.*]], label [[OMP_IF_END4:%.*]], !dbg [[DBG62]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.then3:
// CHECK5-NEXT: invoke void @_Z3foov()
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]], !dbg [[DBG63:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB10]], i32 [[TMP0]]), !dbg [[DBG63]]
// CHECK5-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4, !dbg [[DBG63]]
// CHECK5-NEXT: br label [[OMP_IF_END4]], !dbg [[DBG63]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: lpad:
// CHECK5-NEXT: [[TMP7:%.*]] = landingpad { i8*, i32 }
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: catch i8* null, !dbg [[DBG64:![0-9]+]]
// CHECK5-NEXT: [[TMP8:%.*]] = extractvalue { i8*, i32 } [[TMP7]], 0, !dbg [[DBG64]]
// CHECK5-NEXT: store i8* [[TMP8]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG64]]
// CHECK5-NEXT: [[TMP9:%.*]] = extractvalue { i8*, i32 } [[TMP7]], 1, !dbg [[DBG64]]
// CHECK5-NEXT: store i32 [[TMP9]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG64]]
// CHECK5-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB10]], i32 [[TMP0]]), !dbg [[DBG63]]
// CHECK5-NEXT: br label [[TERMINATE_HANDLER:%.*]], !dbg [[DBG63]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.end4:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0, !dbg [[DBG63]]
// CHECK5-NEXT: store i8* [[A]], i8** [[TMP10]], align 8, !dbg [[DBG63]]
// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1, !dbg [[DBG63]]
// CHECK5-NEXT: store i8* bitcast (%class.TestClass* @tc to i8*), i8** [[TMP11]], align 8, !dbg [[DBG63]]
// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 2, !dbg [[DBG63]]
// CHECK5-NEXT: [[TMP13:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB12:[0-9]+]], i32 [[TMP0]], i8* bitcast (%class.TestClass* @tc to i8*), i64 4, i8*** @tc.cache.), !dbg [[DBG65:![0-9]+]]
// CHECK5-NEXT: [[TMP14:%.*]] = bitcast i8* [[TMP13]] to %class.TestClass*, !dbg [[DBG65]]
// CHECK5-NEXT: [[TMP15:%.*]] = bitcast %class.TestClass* [[TMP14]] to i8*, !dbg [[DBG63]]
// CHECK5-NEXT: store i8* [[TMP15]], i8** [[TMP12]], align 8, !dbg [[DBG63]]
// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 3, !dbg [[DBG63]]
// CHECK5-NEXT: [[TMP17:%.*]] = bitcast [2 x i8]* [[A2]] to i8*, !dbg [[DBG63]]
// CHECK5-NEXT: store i8* [[TMP17]], i8** [[TMP16]], align 8, !dbg [[DBG63]]
// CHECK5-NEXT: [[TMP18:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 4, !dbg [[DBG63]]
// CHECK5-NEXT: [[TMP19:%.*]] = call i8* @__kmpc_threadprivate_cached(%struct.ident_t* @[[GLOB14:[0-9]+]], i32 [[TMP0]], i8* bitcast ([2 x %class.TestClass]* @tc2 to i8*), i64 8, i8*** @tc2.cache.), !dbg [[DBG66:![0-9]+]]
// CHECK5-NEXT: [[TMP20:%.*]] = bitcast i8* [[TMP19]] to [2 x %class.TestClass]*, !dbg [[DBG66]]
// CHECK5-NEXT: [[TMP21:%.*]] = bitcast [2 x %class.TestClass]* [[TMP20]] to i8*, !dbg [[DBG63]]
// CHECK5-NEXT: store i8* [[TMP21]], i8** [[TMP18]], align 8, !dbg [[DBG63]]
// CHECK5-NEXT: [[TMP22:%.*]] = bitcast [5 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*, !dbg [[DBG63]]
// CHECK5-NEXT: [[TMP23:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4, !dbg [[DBG63]]
// CHECK5-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB10]], i32 [[TMP0]], i64 40, i8* [[TMP22]], void (i8*, i8*)* @.omp.copyprivate.copy_func, i32 [[TMP23]]), !dbg [[DBG63]]
// CHECK5-NEXT: [[TMP24:%.*]] = load i8, i8* [[A]], align 1, !dbg [[DBG67:![0-9]+]]
// CHECK5-NEXT: [[CONV:%.*]] = sext i8 [[TMP24]] to i32, !dbg [[DBG67]]
// CHECK5-NEXT: ret i32 [[CONV]], !dbg [[DBG68:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: terminate.handler:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG63]]
// CHECK5-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13:[0-9]+]], !dbg [[DBG63]]
// CHECK5-NEXT: unreachable, !dbg [[DBG63]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZN3SSTIdEC1Ev
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (%struct.SST* nonnull align 8 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 !dbg [[DBG69:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK5-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK5-NEXT: [[THIS1:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: call void @_ZN3SSTIdEC2Ev(%struct.SST* nonnull align 8 dereferenceable(8) [[THIS1]]), !dbg [[DBG70:![0-9]+]]
// CHECK5-NEXT: ret void, !dbg [[DBG71:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZN2SSC1ERi
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (%struct.SS* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32* nonnull align 4 dereferenceable(4) [[D:%.*]]) unnamed_addr #[[ATTR1]] align 2 !dbg [[DBG72:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK5-NEXT: [[D_ADDR:%.*]] = alloca i32*, align 8
// CHECK5-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK5-NEXT: store i32* [[D]], i32** [[D_ADDR]], align 8
// CHECK5-NEXT: [[THIS1:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP0:%.*]] = load i32*, i32** [[D_ADDR]], align 8, !dbg [[DBG73:![0-9]+]]
// CHECK5-NEXT: call void @_ZN2SSC2ERi(%struct.SS* nonnull align 8 dereferenceable(16) [[THIS1]], i32* nonnull align 4 dereferenceable(4) [[TMP0]]), !dbg [[DBG73]]
// CHECK5-NEXT: ret void, !dbg [[DBG74:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@__clang_call_terminate
// CHECK5-SAME: (i8* [[TMP0:%.*]]) #[[ATTR8:[0-9]+]] {
// CHECK5-NEXT: [[TMP2:%.*]] = call i8* @__cxa_begin_catch(i8* [[TMP0]]) #[[ATTR3]]
// CHECK5-NEXT: call void @_ZSt9terminatev() #[[ATTR13]]
// CHECK5-NEXT: unreachable
//
//
// CHECK5-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9:[0-9]+]] !dbg [[DBG75:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK5-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG76:![0-9]+]]
// CHECK5-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [5 x i8*]*, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [5 x i8*]*, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 0, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 0, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP9:%.*]] = load i8*, i8** [[TMP8]], align 8, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP10:%.*]] = load i8, i8* [[TMP9]], align 1, !dbg [[DBG77:![0-9]+]]
// CHECK5-NEXT: store i8 [[TMP10]], i8* [[TMP7]], align 1, !dbg [[DBG77]]
// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 1, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP12:%.*]] = load i8*, i8** [[TMP11]], align 8, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP13:%.*]] = bitcast i8* [[TMP12]] to %class.TestClass*, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP14:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 1, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP15:%.*]] = load i8*, i8** [[TMP14]], align 8, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP16:%.*]] = bitcast i8* [[TMP15]] to %class.TestClass*, !dbg [[DBG76]]
// CHECK5-NEXT: [[CALL:%.*]] = call nonnull align 4 dereferenceable(4) %class.TestClass* @_ZN9TestClassaSERKS_(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP13]], %class.TestClass* nonnull align 4 dereferenceable(4) [[TMP16]]), !dbg [[DBG78:![0-9]+]]
// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 2, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP18:%.*]] = load i8*, i8** [[TMP17]], align 8, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP19:%.*]] = bitcast i8* [[TMP18]] to %class.TestClass*, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 2, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP22:%.*]] = bitcast i8* [[TMP21]] to %class.TestClass*, !dbg [[DBG76]]
// CHECK5-NEXT: [[CALL2:%.*]] = call nonnull align 4 dereferenceable(4) %class.TestClass* @_ZN9TestClassaSERKS_(%class.TestClass* nonnull align 4 dereferenceable(4) [[TMP19]], %class.TestClass* nonnull align 4 dereferenceable(4) [[TMP22]]), !dbg [[DBG79:![0-9]+]]
// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 3, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP25:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 3, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP26:%.*]] = load i8*, i8** [[TMP25]], align 8, !dbg [[DBG76]]
// CHECK5-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 1 [[TMP24]], i8* align 1 [[TMP26]], i64 2, i1 false), !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP27:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP3]], i64 0, i64 4, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP28:%.*]] = load i8*, i8** [[TMP27]], align 8, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP29:%.*]] = bitcast i8* [[TMP28]] to %class.TestClass*, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP30:%.*]] = getelementptr inbounds [5 x i8*], [5 x i8*]* [[TMP5]], i64 0, i64 4, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP31:%.*]] = load i8*, i8** [[TMP30]], align 8, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP32:%.*]] = bitcast i8* [[TMP31]] to %class.TestClass*, !dbg [[DBG76]]
// CHECK5-NEXT: [[TMP33:%.*]] = getelementptr [[CLASS_TESTCLASS:%.*]], %class.TestClass* [[TMP29]], i64 2, !dbg [[DBG76]]
// CHECK5-NEXT: [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq %class.TestClass* [[TMP29]], [[TMP33]], !dbg [[DBG76]]
// CHECK5-NEXT: br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]], !dbg [[DBG76]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp.arraycpy.body:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %class.TestClass* [ [[TMP32]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ], !dbg [[DBG76]]
// CHECK5-NEXT: [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %class.TestClass* [ [[TMP29]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ], !dbg [[DBG76]]
// CHECK5-NEXT: [[CALL3:%.*]] = call nonnull align 4 dereferenceable(4) %class.TestClass* @_ZN9TestClassaSERKS_(%class.TestClass* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %class.TestClass* nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]]), !dbg [[DBG80:![0-9]+]]
// CHECK5-NEXT: [[OMP_ARRAYCPY_DEST_ELEMENT]] = getelementptr [[CLASS_TESTCLASS]], %class.TestClass* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1, !dbg [[DBG76]]
// CHECK5-NEXT: [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[CLASS_TESTCLASS]], %class.TestClass* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1, !dbg [[DBG76]]
// CHECK5-NEXT: [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %class.TestClass* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP33]], !dbg [[DBG76]]
// CHECK5-NEXT: br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE4]], label [[OMP_ARRAYCPY_BODY]], !dbg [[DBG76]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp.arraycpy.done4:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: ret void, !dbg [[DBG80]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZN9TestClassaSERKS_
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (%class.TestClass* nonnull align 4 dereferenceable(4) [[THIS:%.*]], %class.TestClass* nonnull align 4 dereferenceable(4) [[TMP0:%.*]]) #[[ATTR10:[0-9]+]] align 2 !dbg [[DBG81:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK5-NEXT: [[DOTADDR:%.*]] = alloca %class.TestClass*, align 8
// CHECK5-NEXT: store %class.TestClass* [[THIS]], %class.TestClass** [[THIS_ADDR]], align 8
// CHECK5-NEXT: store %class.TestClass* [[TMP0]], %class.TestClass** [[DOTADDR]], align 8
// CHECK5-NEXT: [[THIS1:%.*]] = load %class.TestClass*, %class.TestClass** [[THIS_ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: ret %class.TestClass* [[THIS1]], !dbg [[DBG82:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZN3SSTIdEC2Ev
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (%struct.SST* nonnull align 8 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG83:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK5-NEXT: [[A2:%.*]] = alloca double*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK5-NEXT: [[THIS1:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SST:%.*]], %struct.SST* [[THIS1]], i32 0, i32 0, !dbg [[DBG84:![0-9]+]]
// CHECK5-NEXT: store double 0.000000e+00, double* [[A]], align 8, !dbg [[DBG84]]
// CHECK5-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SST]], %struct.SST* [[THIS1]], i32 0, i32 0, !dbg [[DBG85:![0-9]+]]
// CHECK5-NEXT: store double* [[A3]], double** [[A2]], align 8, !dbg [[DBG85]]
// CHECK5-NEXT: [[TMP0:%.*]] = load double*, double** [[A2]], align 8, !dbg [[DBG86:![0-9]+]]
// CHECK5-NEXT: [[TMP1:%.*]] = load double, double* [[TMP0]], align 8, !dbg [[DBG87:![0-9]+]]
// CHECK5-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to double*, !dbg [[DBG87]]
// CHECK5-NEXT: store double [[TMP1]], double* [[CONV]], align 8, !dbg [[DBG87]]
// CHECK5-NEXT: [[TMP2:%.*]] = load i64, i64* [[A_CASTED]], align 8, !dbg [[DBG87]]
// CHECK5-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB18:[0-9]+]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SST*, i64)* @.omp_outlined. to void (i32*, i32*, ...)*), %struct.SST* [[THIS1]], i64 [[TMP2]]), !dbg [[DBG87]]
// CHECK5-NEXT: ret void, !dbg [[DBG88:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@.omp_outlined.
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SST* [[THIS:%.*]], i64 [[A:%.*]]) #[[ATTR12:[0-9]+]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG89:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK5-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: [[TMP:%.*]] = alloca double*, align 8
// CHECK5-NEXT: [[_TMP1:%.*]] = alloca double*, align 8
// CHECK5-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON:%.*]], align 8
// CHECK5-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK5-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK5-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK5-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8, !dbg [[DBG90:![0-9]+]]
// CHECK5-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to double*, !dbg [[DBG90]]
// CHECK5-NEXT: store double* [[CONV]], double** [[TMP]], align 8, !dbg [[DBG90]]
// CHECK5-NEXT: [[TMP1:%.*]] = load double*, double** [[TMP]], align 8, !dbg [[DBG91:![0-9]+]]
// CHECK5-NEXT: store double* [[TMP1]], double** [[_TMP1]], align 8, !dbg [[DBG92:![0-9]+]]
// CHECK5-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4, !dbg [[DBG92]]
// CHECK5-NEXT: [[TMP2:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8, !dbg [[DBG92]]
// CHECK5-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4, !dbg [[DBG92]]
// CHECK5-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB16:[0-9]+]], i32 [[TMP3]]), !dbg [[DBG92]]
// CHECK5-NEXT: [[TMP5:%.*]] = icmp ne i32 [[TMP4]], 0, !dbg [[DBG92]]
// CHECK5-NEXT: br i1 [[TMP5]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]], !dbg [[DBG92]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 0, !dbg [[DBG93:![0-9]+]]
// CHECK5-NEXT: store %struct.SST* [[TMP0]], %struct.SST** [[TMP6]], align 8, !dbg [[DBG93]]
// CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 1, !dbg [[DBG93]]
// CHECK5-NEXT: [[TMP8:%.*]] = load double*, double** [[_TMP1]], align 8, !dbg [[DBG94:![0-9]+]]
// CHECK5-NEXT: store double* [[TMP8]], double** [[TMP7]], align 8, !dbg [[DBG93]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: invoke void @_ZZN3SSTIdEC1EvENKUlvE_clEv(%class.anon* nonnull align 8 dereferenceable(16) [[REF_TMP]])
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]], !dbg [[DBG93]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB16]], i32 [[TMP3]]), !dbg [[DBG93]]
// CHECK5-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4, !dbg [[DBG93]]
// CHECK5-NEXT: br label [[OMP_IF_END]], !dbg [[DBG93]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: lpad:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP9:%.*]] = landingpad { i8*, i32 }
// CHECK5-NEXT: catch i8* null, !dbg [[DBG95:![0-9]+]]
// CHECK5-NEXT: [[TMP10:%.*]] = extractvalue { i8*, i32 } [[TMP9]], 0, !dbg [[DBG95]]
// CHECK5-NEXT: store i8* [[TMP10]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG95]]
// CHECK5-NEXT: [[TMP11:%.*]] = extractvalue { i8*, i32 } [[TMP9]], 1, !dbg [[DBG95]]
// CHECK5-NEXT: store i32 [[TMP11]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG95]]
// CHECK5-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB16]], i32 [[TMP3]]), !dbg [[DBG93]]
// CHECK5-NEXT: br label [[TERMINATE_HANDLER:%.*]], !dbg [[DBG93]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0, !dbg [[DBG93]]
// CHECK5-NEXT: [[TMP13:%.*]] = load double*, double** [[_TMP1]], align 8, !dbg [[DBG96:![0-9]+]]
// CHECK5-NEXT: [[TMP14:%.*]] = bitcast double* [[TMP13]] to i8*, !dbg [[DBG93]]
// CHECK5-NEXT: store i8* [[TMP14]], i8** [[TMP12]], align 8, !dbg [[DBG93]]
// CHECK5-NEXT: [[TMP15:%.*]] = bitcast [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*, !dbg [[DBG93]]
// CHECK5-NEXT: [[TMP16:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4, !dbg [[DBG93]]
// CHECK5-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB16]], i32 [[TMP3]], i64 8, i8* [[TMP15]], void (i8*, i8*)* @.omp.copyprivate.copy_func.5, i32 [[TMP16]]), !dbg [[DBG93]]
// CHECK5-NEXT: ret void, !dbg [[DBG97:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: terminate.handler:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG93]]
// CHECK5-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13]], !dbg [[DBG93]]
// CHECK5-NEXT: unreachable, !dbg [[DBG93]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZZN3SSTIdEC1EvENKUlvE_clEv
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (%class.anon* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR4]] align 2 !dbg [[DBG98:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %class.anon*, align 8
// CHECK5-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON_0:%.*]], align 8
// CHECK5-NEXT: store %class.anon* [[THIS]], %class.anon** [[THIS_ADDR]], align 8
// CHECK5-NEXT: [[THIS1:%.*]] = load %class.anon*, %class.anon** [[THIS_ADDR]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON:%.*]], %class.anon* [[THIS1]], i32 0, i32 0
// CHECK5-NEXT: [[TMP1:%.*]] = load %struct.SST*, %struct.SST** [[TMP0]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 0, !dbg [[DBG99:![0-9]+]]
// CHECK5-NEXT: store %struct.SST* [[TMP1]], %struct.SST** [[TMP2]], align 8, !dbg [[DBG99]]
// CHECK5-NEXT: [[TMP3:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 1, !dbg [[DBG99]]
// CHECK5-NEXT: [[TMP4:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[THIS1]], i32 0, i32 1, !dbg [[DBG100:![0-9]+]]
// CHECK5-NEXT: [[TMP5:%.*]] = load double*, double** [[TMP4]], align 8, !dbg [[DBG100]]
// CHECK5-NEXT: store double* [[TMP5]], double** [[TMP3]], align 8, !dbg [[DBG99]]
// CHECK5-NEXT: call void @_ZZZN3SSTIdEC1EvENKUlvE_clEvENKUlvE_clEv(%class.anon.0* nonnull align 8 dereferenceable(16) [[REF_TMP]]), !dbg [[DBG99]]
// CHECK5-NEXT: ret void, !dbg [[DBG101:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.5
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] !dbg [[DBG102:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK5-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG103:![0-9]+]]
// CHECK5-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [1 x i8*]*, !dbg [[DBG103]]
// CHECK5-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8, !dbg [[DBG103]]
// CHECK5-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*, !dbg [[DBG103]]
// CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP3]], i64 0, i64 0, !dbg [[DBG103]]
// CHECK5-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8, !dbg [[DBG103]]
// CHECK5-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to double*, !dbg [[DBG103]]
// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0, !dbg [[DBG103]]
// CHECK5-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8, !dbg [[DBG103]]
// CHECK5-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to double*, !dbg [[DBG103]]
// CHECK5-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 8, !dbg [[DBG104:![0-9]+]]
// CHECK5-NEXT: store double [[TMP12]], double* [[TMP8]], align 8, !dbg [[DBG104]]
// CHECK5-NEXT: ret void, !dbg [[DBG104]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZZZN3SSTIdEC1EvENKUlvE_clEvENKUlvE_clEv
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (%class.anon.0* nonnull align 8 dereferenceable(16) [[THIS:%.*]]) #[[ATTR10]] align 2 !dbg [[DBG107:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %class.anon.0*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: store %class.anon.0* [[THIS]], %class.anon.0** [[THIS_ADDR]], align 8
// CHECK5-NEXT: [[THIS1:%.*]] = load %class.anon.0*, %class.anon.0** [[THIS_ADDR]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON_0:%.*]], %class.anon.0* [[THIS1]], i32 0, i32 0
// CHECK5-NEXT: [[TMP1:%.*]] = load %struct.SST*, %struct.SST** [[TMP0]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[THIS1]], i32 0, i32 1, !dbg [[DBG108:![0-9]+]]
// CHECK5-NEXT: [[TMP3:%.*]] = load double*, double** [[TMP2]], align 8, !dbg [[DBG108]]
// CHECK5-NEXT: [[TMP4:%.*]] = load double, double* [[TMP3]], align 8, !dbg [[DBG109:![0-9]+]]
// CHECK5-NEXT: [[INC:%.*]] = fadd double [[TMP4]], 1.000000e+00, !dbg [[DBG109]]
// CHECK5-NEXT: store double [[INC]], double* [[TMP3]], align 8, !dbg [[DBG109]]
// CHECK5-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[THIS1]], i32 0, i32 1, !dbg [[DBG110:![0-9]+]]
// CHECK5-NEXT: [[TMP6:%.*]] = load double*, double** [[TMP5]], align 8, !dbg [[DBG110]]
// CHECK5-NEXT: [[TMP7:%.*]] = load double, double* [[TMP6]], align 8, !dbg [[DBG111:![0-9]+]]
// CHECK5-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to double*, !dbg [[DBG111]]
// CHECK5-NEXT: store double [[TMP7]], double* [[CONV]], align 8, !dbg [[DBG111]]
// CHECK5-NEXT: [[TMP8:%.*]] = load i64, i64* [[A_CASTED]], align 8, !dbg [[DBG111]]
// CHECK5-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB22:[0-9]+]], i32 2, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SST*, i64)* @.omp_outlined..6 to void (i32*, i32*, ...)*), %struct.SST* [[TMP1]], i64 [[TMP8]]), !dbg [[DBG111]]
// CHECK5-NEXT: ret void, !dbg [[DBG112:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@.omp_outlined..6
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SST* [[THIS:%.*]], i64 [[A:%.*]]) #[[ATTR12]] !dbg [[DBG113:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SST*, align 8
// CHECK5-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: [[TMP:%.*]] = alloca double*, align 8
// CHECK5-NEXT: [[_TMP1:%.*]] = alloca double*, align 8
// CHECK5-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [1 x i8*], align 8
// CHECK5-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK5-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: store %struct.SST* [[THIS]], %struct.SST** [[THIS_ADDR]], align 8
// CHECK5-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = load %struct.SST*, %struct.SST** [[THIS_ADDR]], align 8, !dbg [[DBG114:![0-9]+]]
// CHECK5-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to double*, !dbg [[DBG114]]
// CHECK5-NEXT: store double* [[CONV]], double** [[TMP]], align 8, !dbg [[DBG114]]
// CHECK5-NEXT: [[TMP1:%.*]] = load double*, double** [[TMP]], align 8, !dbg [[DBG115:![0-9]+]]
// CHECK5-NEXT: store double* [[TMP1]], double** [[_TMP1]], align 8, !dbg [[DBG116:![0-9]+]]
// CHECK5-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4, !dbg [[DBG116]]
// CHECK5-NEXT: [[TMP2:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8, !dbg [[DBG116]]
// CHECK5-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4, !dbg [[DBG116]]
// CHECK5-NEXT: [[TMP4:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB20:[0-9]+]], i32 [[TMP3]]), !dbg [[DBG116]]
// CHECK5-NEXT: [[TMP5:%.*]] = icmp ne i32 [[TMP4]], 0, !dbg [[DBG116]]
// CHECK5-NEXT: br i1 [[TMP5]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]], !dbg [[DBG116]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP6:%.*]] = load double*, double** [[_TMP1]], align 8, !dbg [[DBG115]]
// CHECK5-NEXT: [[TMP7:%.*]] = load double, double* [[TMP6]], align 8, !dbg [[DBG117:![0-9]+]]
// CHECK5-NEXT: [[INC:%.*]] = fadd double [[TMP7]], 1.000000e+00, !dbg [[DBG117]]
// CHECK5-NEXT: store double [[INC]], double* [[TMP6]], align 8, !dbg [[DBG117]]
// CHECK5-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB20]], i32 [[TMP3]]), !dbg [[DBG117]]
// CHECK5-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4, !dbg [[DBG117]]
// CHECK5-NEXT: br label [[OMP_IF_END]], !dbg [[DBG117]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0, !dbg [[DBG117]]
// CHECK5-NEXT: [[TMP9:%.*]] = load double*, double** [[_TMP1]], align 8, !dbg [[DBG118:![0-9]+]]
// CHECK5-NEXT: [[TMP10:%.*]] = bitcast double* [[TMP9]] to i8*, !dbg [[DBG117]]
// CHECK5-NEXT: store i8* [[TMP10]], i8** [[TMP8]], align 8, !dbg [[DBG117]]
// CHECK5-NEXT: [[TMP11:%.*]] = bitcast [1 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*, !dbg [[DBG117]]
// CHECK5-NEXT: [[TMP12:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4, !dbg [[DBG117]]
// CHECK5-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB20]], i32 [[TMP3]], i64 8, i8* [[TMP11]], void (i8*, i8*)* @.omp.copyprivate.copy_func.7, i32 [[TMP12]]), !dbg [[DBG117]]
// CHECK5-NEXT: ret void, !dbg [[DBG119:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.7
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] !dbg [[DBG120:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK5-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG121:![0-9]+]]
// CHECK5-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [1 x i8*]*, !dbg [[DBG121]]
// CHECK5-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8, !dbg [[DBG121]]
// CHECK5-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [1 x i8*]*, !dbg [[DBG121]]
// CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP3]], i64 0, i64 0, !dbg [[DBG121]]
// CHECK5-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8, !dbg [[DBG121]]
// CHECK5-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to double*, !dbg [[DBG121]]
// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[TMP5]], i64 0, i64 0, !dbg [[DBG121]]
// CHECK5-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8, !dbg [[DBG121]]
// CHECK5-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to double*, !dbg [[DBG121]]
// CHECK5-NEXT: [[TMP12:%.*]] = load double, double* [[TMP11]], align 8, !dbg [[DBG122:![0-9]+]]
// CHECK5-NEXT: store double [[TMP12]], double* [[TMP8]], align 8, !dbg [[DBG122]]
// CHECK5-NEXT: ret void, !dbg [[DBG122]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZN2SSC2ERi
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (%struct.SS* nonnull align 8 dereferenceable(16) [[THIS:%.*]], i32* nonnull align 4 dereferenceable(4) [[D:%.*]]) unnamed_addr #[[ATTR2]] align 2 !dbg [[DBG123:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK5-NEXT: [[D_ADDR:%.*]] = alloca i32*, align 8
// CHECK5-NEXT: [[A2:%.*]] = alloca i32*, align 8
// CHECK5-NEXT: [[B4:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[C7:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK5-NEXT: [[B_CASTED:%.*]] = alloca i64, align 8
// CHECK5-NEXT: [[C_CASTED:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK5-NEXT: store i32* [[D]], i32** [[D_ADDR]], align 8
// CHECK5-NEXT: [[THIS1:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_SS:%.*]], %struct.SS* [[THIS1]], i32 0, i32 0, !dbg [[DBG124:![0-9]+]]
// CHECK5-NEXT: store i32 0, i32* [[A]], align 8, !dbg [[DBG124]]
// CHECK5-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 1, !dbg [[DBG125:![0-9]+]]
// CHECK5-NEXT: [[BF_LOAD:%.*]] = load i8, i8* [[B]], align 4, !dbg [[DBG125]]
// CHECK5-NEXT: [[BF_CLEAR:%.*]] = and i8 [[BF_LOAD]], -16, !dbg [[DBG125]]
// CHECK5-NEXT: store i8 [[BF_CLEAR]], i8* [[B]], align 4, !dbg [[DBG125]]
// CHECK5-NEXT: [[C:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 2, !dbg [[DBG126:![0-9]+]]
// CHECK5-NEXT: [[TMP0:%.*]] = load i32*, i32** [[D_ADDR]], align 8, !dbg [[DBG127:![0-9]+]]
// CHECK5-NEXT: store i32* [[TMP0]], i32** [[C]], align 8, !dbg [[DBG126]]
// CHECK5-NEXT: [[A3:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 0, !dbg [[DBG128:![0-9]+]]
// CHECK5-NEXT: store i32* [[A3]], i32** [[A2]], align 8, !dbg [[DBG128]]
// CHECK5-NEXT: [[B5:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 1, !dbg [[DBG129:![0-9]+]]
// CHECK5-NEXT: [[BF_LOAD6:%.*]] = load i8, i8* [[B5]], align 4, !dbg [[DBG129]]
// CHECK5-NEXT: [[BF_SHL:%.*]] = shl i8 [[BF_LOAD6]], 4, !dbg [[DBG129]]
// CHECK5-NEXT: [[BF_ASHR:%.*]] = ashr i8 [[BF_SHL]], 4, !dbg [[DBG129]]
// CHECK5-NEXT: [[BF_CAST:%.*]] = sext i8 [[BF_ASHR]] to i32, !dbg [[DBG129]]
// CHECK5-NEXT: store i32 [[BF_CAST]], i32* [[B4]], align 4, !dbg [[DBG129]]
// CHECK5-NEXT: [[C8:%.*]] = getelementptr inbounds [[STRUCT_SS]], %struct.SS* [[THIS1]], i32 0, i32 2, !dbg [[DBG130:![0-9]+]]
// CHECK5-NEXT: [[TMP1:%.*]] = load i32*, i32** [[C8]], align 8, !dbg [[DBG130]]
// CHECK5-NEXT: store i32* [[TMP1]], i32** [[C7]], align 8, !dbg [[DBG130]]
// CHECK5-NEXT: [[TMP2:%.*]] = load i32*, i32** [[A2]], align 8, !dbg [[DBG131:![0-9]+]]
// CHECK5-NEXT: [[TMP3:%.*]] = load i32, i32* [[TMP2]], align 4, !dbg [[DBG132:![0-9]+]]
// CHECK5-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to i32*, !dbg [[DBG132]]
// CHECK5-NEXT: store i32 [[TMP3]], i32* [[CONV]], align 4, !dbg [[DBG132]]
// CHECK5-NEXT: [[TMP4:%.*]] = load i64, i64* [[A_CASTED]], align 8, !dbg [[DBG132]]
// CHECK5-NEXT: [[TMP5:%.*]] = load i32, i32* [[B4]], align 4, !dbg [[DBG132]]
// CHECK5-NEXT: [[CONV9:%.*]] = bitcast i64* [[B_CASTED]] to i32*, !dbg [[DBG132]]
// CHECK5-NEXT: store i32 [[TMP5]], i32* [[CONV9]], align 4, !dbg [[DBG132]]
// CHECK5-NEXT: [[TMP6:%.*]] = load i64, i64* [[B_CASTED]], align 8, !dbg [[DBG132]]
// CHECK5-NEXT: [[TMP7:%.*]] = load i32*, i32** [[C7]], align 8, !dbg [[DBG133:![0-9]+]]
// CHECK5-NEXT: [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4, !dbg [[DBG132]]
// CHECK5-NEXT: [[CONV10:%.*]] = bitcast i64* [[C_CASTED]] to i32*, !dbg [[DBG132]]
// CHECK5-NEXT: store i32 [[TMP8]], i32* [[CONV10]], align 4, !dbg [[DBG132]]
// CHECK5-NEXT: [[TMP9:%.*]] = load i64, i64* [[C_CASTED]], align 8, !dbg [[DBG132]]
// CHECK5-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB26:[0-9]+]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SS*, i64, i64, i64)* @.omp_outlined..8 to void (i32*, i32*, ...)*), %struct.SS* [[THIS1]], i64 [[TMP4]], i64 [[TMP6]], i64 [[TMP9]]), !dbg [[DBG132]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: ret void, !dbg [[DBG134:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@.omp_outlined..8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SS* [[THIS:%.*]], i64 [[A:%.*]], i64 [[B:%.*]], i64 [[C:%.*]]) #[[ATTR12]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG135:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK5-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK5-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8
// CHECK5-NEXT: [[C_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: [[TMP:%.*]] = alloca i32*, align 8
// CHECK5-NEXT: [[_TMP3:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[_TMP4:%.*]] = alloca i32*, align 8
// CHECK5-NEXT: [[_TMP5:%.*]] = alloca i32*, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON_1:%.*]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [3 x i8*], align 8
// CHECK5-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK5-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK5-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK5-NEXT: store i64 [[B]], i64* [[B_ADDR]], align 8
// CHECK5-NEXT: store i64 [[C]], i64* [[C_ADDR]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8, !dbg [[DBG136:![0-9]+]]
// CHECK5-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to i32*, !dbg [[DBG136]]
// CHECK5-NEXT: [[CONV1:%.*]] = bitcast i64* [[B_ADDR]] to i32*, !dbg [[DBG136]]
// CHECK5-NEXT: [[CONV2:%.*]] = bitcast i64* [[C_ADDR]] to i32*, !dbg [[DBG136]]
// CHECK5-NEXT: store i32* [[CONV]], i32** [[TMP]], align 8, !dbg [[DBG136]]
// CHECK5-NEXT: store i32* [[CONV2]], i32** [[_TMP3]], align 8, !dbg [[DBG136]]
// CHECK5-NEXT: [[TMP1:%.*]] = load i32*, i32** [[TMP]], align 8, !dbg [[DBG137:![0-9]+]]
// CHECK5-NEXT: store i32* [[TMP1]], i32** [[_TMP4]], align 8, !dbg [[DBG138:![0-9]+]]
// CHECK5-NEXT: [[TMP2:%.*]] = load i32*, i32** [[_TMP3]], align 8, !dbg [[DBG139:![0-9]+]]
// CHECK5-NEXT: store i32* [[TMP2]], i32** [[_TMP5]], align 8, !dbg [[DBG138]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4, !dbg [[DBG138]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8, !dbg [[DBG138]]
// CHECK5-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4, !dbg [[DBG138]]
// CHECK5-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB24:[0-9]+]], i32 [[TMP4]]), !dbg [[DBG138]]
// CHECK5-NEXT: [[TMP6:%.*]] = icmp ne i32 [[TMP5]], 0, !dbg [[DBG138]]
// CHECK5-NEXT: br i1 [[TMP6]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]], !dbg [[DBG138]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP7:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 0, !dbg [[DBG140:![0-9]+]]
// CHECK5-NEXT: store %struct.SS* [[TMP0]], %struct.SS** [[TMP7]], align 8, !dbg [[DBG140]]
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 1, !dbg [[DBG140]]
// CHECK5-NEXT: [[TMP9:%.*]] = load i32*, i32** [[_TMP4]], align 8, !dbg [[DBG141:![0-9]+]]
// CHECK5-NEXT: store i32* [[TMP9]], i32** [[TMP8]], align 8, !dbg [[DBG140]]
// CHECK5-NEXT: [[TMP10:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 2, !dbg [[DBG140]]
// CHECK5-NEXT: store i32* [[CONV1]], i32** [[TMP10]], align 8, !dbg [[DBG140]]
// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[REF_TMP]], i32 0, i32 3, !dbg [[DBG140]]
// CHECK5-NEXT: [[TMP12:%.*]] = load i32*, i32** [[_TMP5]], align 8, !dbg [[DBG141]]
// CHECK5-NEXT: store i32* [[TMP12]], i32** [[TMP11]], align 8, !dbg [[DBG140]]
// CHECK5-NEXT: invoke void @_ZZN2SSC1ERiENKUlvE_clEv(%class.anon.1* nonnull align 8 dereferenceable(32) [[REF_TMP]])
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]], !dbg [[DBG140]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB24]], i32 [[TMP4]]), !dbg [[DBG140]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4, !dbg [[DBG140]]
// CHECK5-NEXT: br label [[OMP_IF_END]], !dbg [[DBG140]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: lpad:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP13:%.*]] = landingpad { i8*, i32 }
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: catch i8* null, !dbg [[DBG142:![0-9]+]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP14:%.*]] = extractvalue { i8*, i32 } [[TMP13]], 0, !dbg [[DBG142]]
// CHECK5-NEXT: store i8* [[TMP14]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG142]]
// CHECK5-NEXT: [[TMP15:%.*]] = extractvalue { i8*, i32 } [[TMP13]], 1, !dbg [[DBG142]]
// CHECK5-NEXT: store i32 [[TMP15]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG142]]
// CHECK5-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB24]], i32 [[TMP4]]), !dbg [[DBG140]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: br label [[TERMINATE_HANDLER:%.*]], !dbg [[DBG140]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0, !dbg [[DBG140]]
// CHECK5-NEXT: [[TMP17:%.*]] = load i32*, i32** [[_TMP4]], align 8, !dbg [[DBG143:![0-9]+]]
// CHECK5-NEXT: [[TMP18:%.*]] = bitcast i32* [[TMP17]] to i8*, !dbg [[DBG140]]
// CHECK5-NEXT: store i8* [[TMP18]], i8** [[TMP16]], align 8, !dbg [[DBG140]]
// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1, !dbg [[DBG140]]
// CHECK5-NEXT: [[TMP20:%.*]] = bitcast i32* [[CONV1]] to i8*, !dbg [[DBG140]]
// CHECK5-NEXT: store i8* [[TMP20]], i8** [[TMP19]], align 8, !dbg [[DBG140]]
// CHECK5-NEXT: [[TMP21:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 2, !dbg [[DBG140]]
// CHECK5-NEXT: [[TMP22:%.*]] = load i32*, i32** [[_TMP5]], align 8, !dbg [[DBG144:![0-9]+]]
// CHECK5-NEXT: [[TMP23:%.*]] = bitcast i32* [[TMP22]] to i8*, !dbg [[DBG140]]
// CHECK5-NEXT: store i8* [[TMP23]], i8** [[TMP21]], align 8, !dbg [[DBG140]]
// CHECK5-NEXT: [[TMP24:%.*]] = bitcast [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*, !dbg [[DBG140]]
// CHECK5-NEXT: [[TMP25:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4, !dbg [[DBG140]]
// CHECK5-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB24]], i32 [[TMP4]], i64 24, i8* [[TMP24]], void (i8*, i8*)* @.omp.copyprivate.copy_func.9, i32 [[TMP25]]), !dbg [[DBG140]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: ret void, !dbg [[DBG145:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: terminate.handler:
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG140]]
// CHECK5-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13]], !dbg [[DBG140]]
// CHECK5-NEXT: unreachable, !dbg [[DBG140]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@_ZZN2SSC1ERiENKUlvE_clEv
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (%class.anon.1* nonnull align 8 dereferenceable(32) [[THIS:%.*]]) #[[ATTR10]] align 2 !dbg [[DBG146:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %class.anon.1*, align 8
// CHECK5-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK5-NEXT: [[B_CASTED:%.*]] = alloca i64, align 8
// CHECK5-NEXT: [[C_CASTED:%.*]] = alloca i64, align 8
// CHECK5-NEXT: store %class.anon.1* [[THIS]], %class.anon.1** [[THIS_ADDR]], align 8
// CHECK5-NEXT: [[THIS1:%.*]] = load %class.anon.1*, %class.anon.1** [[THIS_ADDR]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON_1:%.*]], %class.anon.1* [[THIS1]], i32 0, i32 0
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: [[TMP1:%.*]] = load %struct.SS*, %struct.SS** [[TMP0]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 1, !dbg [[DBG147:![0-9]+]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: [[TMP3:%.*]] = load i32*, i32** [[TMP2]], align 8, !dbg [[DBG147]]
// CHECK5-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4, !dbg [[DBG148:![0-9]+]]
// CHECK5-NEXT: [[INC:%.*]] = add nsw i32 [[TMP4]], 1, !dbg [[DBG148]]
// CHECK5-NEXT: store i32 [[INC]], i32* [[TMP3]], align 4, !dbg [[DBG148]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP5:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 2, !dbg [[DBG149:![0-9]+]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: [[TMP6:%.*]] = load i32*, i32** [[TMP5]], align 8, !dbg [[DBG149]]
// CHECK5-NEXT: [[TMP7:%.*]] = load i32, i32* [[TMP6]], align 4, !dbg [[DBG150:![0-9]+]]
// CHECK5-NEXT: [[DEC:%.*]] = add nsw i32 [[TMP7]], -1, !dbg [[DBG150]]
// CHECK5-NEXT: store i32 [[DEC]], i32* [[TMP6]], align 4, !dbg [[DBG150]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP8:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 3, !dbg [[DBG151:![0-9]+]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: [[TMP9:%.*]] = load i32*, i32** [[TMP8]], align 8, !dbg [[DBG151]]
// CHECK5-NEXT: [[TMP10:%.*]] = load i32, i32* [[TMP9]], align 4, !dbg [[DBG152:![0-9]+]]
// CHECK5-NEXT: [[DIV:%.*]] = sdiv i32 [[TMP10]], 1, !dbg [[DBG152]]
// CHECK5-NEXT: store i32 [[DIV]], i32* [[TMP9]], align 4, !dbg [[DBG152]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP11:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 1, !dbg [[DBG153:![0-9]+]]
// CHECK5-NEXT: [[TMP12:%.*]] = load i32*, i32** [[TMP11]], align 8, !dbg [[DBG153]]
// CHECK5-NEXT: [[TMP13:%.*]] = load i32, i32* [[TMP12]], align 4, !dbg [[DBG154:![0-9]+]]
// CHECK5-NEXT: [[CONV:%.*]] = bitcast i64* [[A_CASTED]] to i32*, !dbg [[DBG154]]
// CHECK5-NEXT: store i32 [[TMP13]], i32* [[CONV]], align 4, !dbg [[DBG154]]
// CHECK5-NEXT: [[TMP14:%.*]] = load i64, i64* [[A_CASTED]], align 8, !dbg [[DBG154]]
// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 2, !dbg [[DBG155:![0-9]+]]
// CHECK5-NEXT: [[TMP16:%.*]] = load i32*, i32** [[TMP15]], align 8, !dbg [[DBG155]]
// CHECK5-NEXT: [[TMP17:%.*]] = load i32, i32* [[TMP16]], align 4, !dbg [[DBG154]]
// CHECK5-NEXT: [[CONV2:%.*]] = bitcast i64* [[B_CASTED]] to i32*, !dbg [[DBG154]]
// CHECK5-NEXT: store i32 [[TMP17]], i32* [[CONV2]], align 4, !dbg [[DBG154]]
// CHECK5-NEXT: [[TMP18:%.*]] = load i64, i64* [[B_CASTED]], align 8, !dbg [[DBG154]]
// CHECK5-NEXT: [[TMP19:%.*]] = getelementptr inbounds [[CLASS_ANON_1]], %class.anon.1* [[THIS1]], i32 0, i32 3, !dbg [[DBG156:![0-9]+]]
// CHECK5-NEXT: [[TMP20:%.*]] = load i32*, i32** [[TMP19]], align 8, !dbg [[DBG156]]
// CHECK5-NEXT: [[TMP21:%.*]] = load i32, i32* [[TMP20]], align 4, !dbg [[DBG154]]
// CHECK5-NEXT: [[CONV3:%.*]] = bitcast i64* [[C_CASTED]] to i32*, !dbg [[DBG154]]
// CHECK5-NEXT: store i32 [[TMP21]], i32* [[CONV3]], align 4, !dbg [[DBG154]]
// CHECK5-NEXT: [[TMP22:%.*]] = load i64, i64* [[C_CASTED]], align 8, !dbg [[DBG154]]
// CHECK5-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB30:[0-9]+]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, %struct.SS*, i64, i64, i64)* @.omp_outlined..10 to void (i32*, i32*, ...)*), %struct.SS* [[TMP1]], i64 [[TMP14]], i64 [[TMP18]], i64 [[TMP22]]), !dbg [[DBG154]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: ret void, !dbg [[DBG157:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.9
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] !dbg [[DBG158:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK5-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG159:![0-9]+]]
// CHECK5-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [3 x i8*]*, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [3 x i8*]*, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 0, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to i32*, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 0, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to i32*, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4, !dbg [[DBG160:![0-9]+]]
// CHECK5-NEXT: store i32 [[TMP12]], i32* [[TMP8]], align 4, !dbg [[DBG160]]
// CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 1, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i32*, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 1, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32*, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4, !dbg [[DBG161:![0-9]+]]
// CHECK5-NEXT: store i32 [[TMP19]], i32* [[TMP15]], align 4, !dbg [[DBG161]]
// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 2, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP22:%.*]] = bitcast i8* [[TMP21]] to i32*, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 2, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32*, !dbg [[DBG159]]
// CHECK5-NEXT: [[TMP26:%.*]] = load i32, i32* [[TMP25]], align 4, !dbg [[DBG162:![0-9]+]]
// CHECK5-NEXT: store i32 [[TMP26]], i32* [[TMP22]], align 4, !dbg [[DBG162]]
// CHECK5-NEXT: ret void, !dbg [[DBG162]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@.omp_outlined..10
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]], %struct.SS* [[THIS:%.*]], i64 [[A:%.*]], i64 [[B:%.*]], i64 [[C:%.*]]) #[[ATTR12]] !dbg [[DBG163:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.SS*, align 8
// CHECK5-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK5-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8
// CHECK5-NEXT: [[C_ADDR:%.*]] = alloca i64, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: [[TMP:%.*]] = alloca i32*, align 8
// CHECK5-NEXT: [[_TMP3:%.*]] = alloca i32*, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[_TMP4:%.*]] = alloca i32*, align 8
// CHECK5-NEXT: [[_TMP5:%.*]] = alloca i32*, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK5-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [3 x i8*], align 8
// CHECK5-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK5-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: store %struct.SS* [[THIS]], %struct.SS** [[THIS_ADDR]], align 8
// CHECK5-NEXT: store i64 [[A]], i64* [[A_ADDR]], align 8
// CHECK5-NEXT: store i64 [[B]], i64* [[B_ADDR]], align 8
// CHECK5-NEXT: store i64 [[C]], i64* [[C_ADDR]], align 8
// CHECK5-NEXT: [[TMP0:%.*]] = load %struct.SS*, %struct.SS** [[THIS_ADDR]], align 8, !dbg [[DBG164:![0-9]+]]
// CHECK5-NEXT: [[CONV:%.*]] = bitcast i64* [[A_ADDR]] to i32*, !dbg [[DBG164]]
// CHECK5-NEXT: [[CONV1:%.*]] = bitcast i64* [[B_ADDR]] to i32*, !dbg [[DBG164]]
// CHECK5-NEXT: [[CONV2:%.*]] = bitcast i64* [[C_ADDR]] to i32*, !dbg [[DBG164]]
// CHECK5-NEXT: store i32* [[CONV]], i32** [[TMP]], align 8, !dbg [[DBG164]]
// CHECK5-NEXT: store i32* [[CONV2]], i32** [[_TMP3]], align 8, !dbg [[DBG164]]
// CHECK5-NEXT: [[TMP1:%.*]] = load i32*, i32** [[TMP]], align 8, !dbg [[DBG165:![0-9]+]]
// CHECK5-NEXT: store i32* [[TMP1]], i32** [[_TMP4]], align 8, !dbg [[DBG166:![0-9]+]]
// CHECK5-NEXT: [[TMP2:%.*]] = load i32*, i32** [[_TMP3]], align 8, !dbg [[DBG167:![0-9]+]]
// CHECK5-NEXT: store i32* [[TMP2]], i32** [[_TMP5]], align 8, !dbg [[DBG166]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4, !dbg [[DBG166]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8, !dbg [[DBG166]]
// CHECK5-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4, !dbg [[DBG166]]
// CHECK5-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB28:[0-9]+]], i32 [[TMP4]]), !dbg [[DBG166]]
// CHECK5-NEXT: [[TMP6:%.*]] = icmp ne i32 [[TMP5]], 0, !dbg [[DBG166]]
// CHECK5-NEXT: br i1 [[TMP6]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]], !dbg [[DBG166]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.then:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP7:%.*]] = load i32*, i32** [[_TMP4]], align 8, !dbg [[DBG165]]
// CHECK5-NEXT: [[TMP8:%.*]] = load i32, i32* [[TMP7]], align 4, !dbg [[DBG168:![0-9]+]]
// CHECK5-NEXT: [[INC:%.*]] = add nsw i32 [[TMP8]], 1, !dbg [[DBG168]]
// CHECK5-NEXT: store i32 [[INC]], i32* [[TMP7]], align 4, !dbg [[DBG168]]
// CHECK5-NEXT: [[TMP9:%.*]] = load i32, i32* [[CONV1]], align 8, !dbg [[DBG169:![0-9]+]]
// CHECK5-NEXT: [[DEC:%.*]] = add nsw i32 [[TMP9]], -1, !dbg [[DBG169]]
// CHECK5-NEXT: store i32 [[DEC]], i32* [[CONV1]], align 8, !dbg [[DBG169]]
// CHECK5-NEXT: [[TMP10:%.*]] = load i32*, i32** [[_TMP5]], align 8, !dbg [[DBG167]]
// CHECK5-NEXT: [[TMP11:%.*]] = load i32, i32* [[TMP10]], align 4, !dbg [[DBG170:![0-9]+]]
// CHECK5-NEXT: [[DIV:%.*]] = sdiv i32 [[TMP11]], 1, !dbg [[DBG170]]
// CHECK5-NEXT: store i32 [[DIV]], i32* [[TMP10]], align 4, !dbg [[DBG170]]
// CHECK5-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB28]], i32 [[TMP4]]), !dbg [[DBG168]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4, !dbg [[DBG168]]
// CHECK5-NEXT: br label [[OMP_IF_END]], !dbg [[DBG168]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0, !dbg [[DBG168]]
// CHECK5-NEXT: [[TMP13:%.*]] = load i32*, i32** [[_TMP4]], align 8, !dbg [[DBG171:![0-9]+]]
// CHECK5-NEXT: [[TMP14:%.*]] = bitcast i32* [[TMP13]] to i8*, !dbg [[DBG168]]
// CHECK5-NEXT: store i8* [[TMP14]], i8** [[TMP12]], align 8, !dbg [[DBG168]]
// CHECK5-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1, !dbg [[DBG168]]
// CHECK5-NEXT: [[TMP16:%.*]] = bitcast i32* [[CONV1]] to i8*, !dbg [[DBG168]]
// CHECK5-NEXT: store i8* [[TMP16]], i8** [[TMP15]], align 8, !dbg [[DBG168]]
// CHECK5-NEXT: [[TMP17:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 2, !dbg [[DBG168]]
// CHECK5-NEXT: [[TMP18:%.*]] = load i32*, i32** [[_TMP5]], align 8, !dbg [[DBG172:![0-9]+]]
// CHECK5-NEXT: [[TMP19:%.*]] = bitcast i32* [[TMP18]] to i8*, !dbg [[DBG168]]
// CHECK5-NEXT: store i8* [[TMP19]], i8** [[TMP17]], align 8, !dbg [[DBG168]]
// CHECK5-NEXT: [[TMP20:%.*]] = bitcast [3 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*, !dbg [[DBG168]]
// CHECK5-NEXT: [[TMP21:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4, !dbg [[DBG168]]
// CHECK5-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB28]], i32 [[TMP4]], i64 24, i8* [[TMP20]], void (i8*, i8*)* @.omp.copyprivate.copy_func.11, i32 [[TMP21]]), !dbg [[DBG168]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: ret void, !dbg [[DBG173:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func.11
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR9]] !dbg [[DBG174:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK5-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8, !dbg [[DBG175:![0-9]+]]
// CHECK5-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [3 x i8*]*, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [3 x i8*]*, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 0, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to i32*, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 0, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to i32*, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP12:%.*]] = load i32, i32* [[TMP11]], align 4, !dbg [[DBG176:![0-9]+]]
// CHECK5-NEXT: store i32 [[TMP12]], i32* [[TMP8]], align 4, !dbg [[DBG176]]
// CHECK5-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 1, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to i32*, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP16:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 1, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to i32*, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP19:%.*]] = load i32, i32* [[TMP18]], align 4, !dbg [[DBG177:![0-9]+]]
// CHECK5-NEXT: store i32 [[TMP19]], i32* [[TMP15]], align 4, !dbg [[DBG177]]
// CHECK5-NEXT: [[TMP20:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP3]], i64 0, i64 2, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP21:%.*]] = load i8*, i8** [[TMP20]], align 8, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP22:%.*]] = bitcast i8* [[TMP21]] to i32*, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP23:%.*]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[TMP5]], i64 0, i64 2, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP24:%.*]] = load i8*, i8** [[TMP23]], align 8, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP25:%.*]] = bitcast i8* [[TMP24]] to i32*, !dbg [[DBG175]]
// CHECK5-NEXT: [[TMP26:%.*]] = load i32, i32* [[TMP25]], align 4, !dbg [[DBG178:![0-9]+]]
// CHECK5-NEXT: store i32 [[TMP26]], i32* [[TMP22]], align 4, !dbg [[DBG178]]
// CHECK5-NEXT: ret void, !dbg [[DBG178]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@_Z15parallel_singlev
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-SAME: () #[[ATTR10]] !dbg [[DBG179:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB35:[0-9]+]], i32 0, void (i32*, i32*, ...)* bitcast (void (i32*, i32*)* @.omp_outlined..12 to void (i32*, i32*, ...)*)), !dbg [[DBG180:![0-9]+]]
Return "[CGCall] Annotate `this` argument with alignment" The original change was reverted because it was discovered that clang mishandles thunks, and they receive wrong attributes for their this/return types - the ones for the function they will call, not the ones they have. While i have tried to fix this in https://reviews.llvm.org/D100388 that patch has been up and stuck for a month now, with little signs of progress. So while it will be good to solve this for real, for now we can simply avoid introducing the bug, by not annotating this/return for thunks. This reverts commit 6270b3a1eafaba4279e021418c5a2c5a35abc002, relanding 0aa0458f1429372038ca6a4edc7e94c96cd9a753.
2021-05-13 23:20:37 +08:00
// CHECK5-NEXT: ret void, !dbg [[DBG181:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@.omp_outlined..12
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: (i32* noalias [[DOTGLOBAL_TID_:%.*]], i32* noalias [[DOTBOUND_TID_:%.*]]) #[[ATTR12]] personality i8* bitcast (i32 (...)* @__gxx_personality_v0 to i8*) !dbg [[DBG182:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
// CHECK5-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK5-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca i32*, align 8
// CHECK5-NEXT: [[EXN_SLOT:%.*]] = alloca i8*, align 8
// CHECK5-NEXT: [[EHSELECTOR_SLOT:%.*]] = alloca i32, align 4
// CHECK5-NEXT: store i32* [[DOTGLOBAL_TID_]], i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK5-NEXT: store i32* [[DOTBOUND_TID_]], i32** [[DOTBOUND_TID__ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP0:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8, !dbg [[DBG183:![0-9]+]]
// CHECK5-NEXT: [[TMP1:%.*]] = load i32, i32* [[TMP0]], align 4, !dbg [[DBG183]]
// CHECK5-NEXT: [[TMP2:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB32:[0-9]+]], i32 [[TMP1]]), !dbg [[DBG183]]
// CHECK5-NEXT: [[TMP3:%.*]] = icmp ne i32 [[TMP2]], 0, !dbg [[DBG183]]
// CHECK5-NEXT: br i1 [[TMP3]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]], !dbg [[DBG183]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.then:
// CHECK5-NEXT: invoke void @_Z3foov()
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: to label [[INVOKE_CONT:%.*]] unwind label [[LPAD:%.*]], !dbg [[DBG184:![0-9]+]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: invoke.cont:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB32]], i32 [[TMP1]]), !dbg [[DBG184]]
// CHECK5-NEXT: br label [[OMP_IF_END]], !dbg [[DBG184]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: lpad:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[TMP4:%.*]] = landingpad { i8*, i32 }
// CHECK5-NEXT: catch i8* null, !dbg [[DBG185:![0-9]+]]
// CHECK5-NEXT: [[TMP5:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 0, !dbg [[DBG185]]
// CHECK5-NEXT: store i8* [[TMP5]], i8** [[EXN_SLOT]], align 8, !dbg [[DBG185]]
// CHECK5-NEXT: [[TMP6:%.*]] = extractvalue { i8*, i32 } [[TMP4]], 1, !dbg [[DBG185]]
// CHECK5-NEXT: store i32 [[TMP6]], i32* [[EHSELECTOR_SLOT]], align 4, !dbg [[DBG185]]
// CHECK5-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB32]], i32 [[TMP1]]), !dbg [[DBG184]]
// CHECK5-NEXT: br label [[TERMINATE_HANDLER:%.*]], !dbg [[DBG184]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: omp_if.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB33:[0-9]+]], i32 [[TMP1]]), !dbg [[DBG186:![0-9]+]]
// CHECK5-NEXT: ret void, !dbg [[DBG186]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5: terminate.handler:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: [[EXN:%.*]] = load i8*, i8** [[EXN_SLOT]], align 8, !dbg [[DBG184]]
// CHECK5-NEXT: call void @__clang_call_terminate(i8* [[EXN]]) #[[ATTR13]], !dbg [[DBG184]]
// CHECK5-NEXT: unreachable, !dbg [[DBG184]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
//
//
// CHECK5-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_single_codegen.cpp
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" !dbg [[DBG187:![0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: entry:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
// CHECK5-NEXT: call void @__cxx_global_var_init(), !dbg [[DBG188:![0-9]+]]
// CHECK5-NEXT: call void @__cxx_global_var_init.4(), !dbg [[DBG188]]
// CHECK5-NEXT: call void @.__omp_threadprivate_init_.(), !dbg [[DBG188]]
// CHECK5-NEXT: call void @.__omp_threadprivate_init_..3(), !dbg [[DBG188]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
// CHECK5-NEXT: ret void
//
//
// CHECK6-LABEL: define {{[^@]+}}@_Z10array_funciPiP2St
// CHECK6-SAME: (i32 [[N:%.*]], i32* [[A:%.*]], %struct.St* [[S:%.*]]) #[[ATTR0:[0-9]+]] {
// CHECK6-NEXT: entry:
// CHECK6-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK6-NEXT: [[A_ADDR:%.*]] = alloca i32*, align 8
// CHECK6-NEXT: [[S_ADDR:%.*]] = alloca %struct.St*, align 8
// CHECK6-NEXT: [[DOTOMP_COPYPRIVATE_DID_IT:%.*]] = alloca i32, align 4
// CHECK6-NEXT: [[DOTOMP_COPYPRIVATE_CPR_LIST:%.*]] = alloca [2 x i8*], align 8
// CHECK6-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB1:[0-9]+]])
// CHECK6-NEXT: store i32 [[N]], i32* [[N_ADDR]], align 4
// CHECK6-NEXT: store i32* [[A]], i32** [[A_ADDR]], align 8
// CHECK6-NEXT: store %struct.St* [[S]], %struct.St** [[S_ADDR]], align 8
// CHECK6-NEXT: [[TMP1:%.*]] = load i32, i32* [[N_ADDR]], align 4
// CHECK6-NEXT: [[TMP2:%.*]] = zext i32 [[TMP1]] to i64
// CHECK6-NEXT: store i32 0, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK6-NEXT: [[TMP3:%.*]] = call i32 @__kmpc_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK6-NEXT: [[TMP4:%.*]] = icmp ne i32 [[TMP3]], 0
// CHECK6-NEXT: br i1 [[TMP4]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_END:%.*]]
// CHECK6: omp_if.then:
// CHECK6-NEXT: call void @__kmpc_end_single(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
// CHECK6-NEXT: store i32 1, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK6-NEXT: br label [[OMP_IF_END]]
// CHECK6: omp_if.end:
// CHECK6-NEXT: [[TMP5:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 0
// CHECK6-NEXT: [[TMP6:%.*]] = bitcast i32** [[A_ADDR]] to i8*
// CHECK6-NEXT: store i8* [[TMP6]], i8** [[TMP5]], align 8
// CHECK6-NEXT: [[TMP7:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]], i64 0, i64 1
// CHECK6-NEXT: [[TMP8:%.*]] = bitcast %struct.St** [[S_ADDR]] to i8*
// CHECK6-NEXT: store i8* [[TMP8]], i8** [[TMP7]], align 8
// CHECK6-NEXT: [[TMP9:%.*]] = bitcast [2 x i8*]* [[DOTOMP_COPYPRIVATE_CPR_LIST]] to i8*
// CHECK6-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTOMP_COPYPRIVATE_DID_IT]], align 4
// CHECK6-NEXT: call void @__kmpc_copyprivate(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]], i64 16, i8* [[TMP9]], void (i8*, i8*)* @.omp.copyprivate.copy_func, i32 [[TMP10]])
// CHECK6-NEXT: ret void
//
//
// CHECK6-LABEL: define {{[^@]+}}@.omp.copyprivate.copy_func
// CHECK6-SAME: (i8* [[TMP0:%.*]], i8* [[TMP1:%.*]]) #[[ATTR3:[0-9]+]] {
// CHECK6-NEXT: entry:
// CHECK6-NEXT: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK6-NEXT: [[DOTADDR1:%.*]] = alloca i8*, align 8
// CHECK6-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK6-NEXT: store i8* [[TMP1]], i8** [[DOTADDR1]], align 8
// CHECK6-NEXT: [[TMP2:%.*]] = load i8*, i8** [[DOTADDR]], align 8
// CHECK6-NEXT: [[TMP3:%.*]] = bitcast i8* [[TMP2]] to [2 x i8*]*
// CHECK6-NEXT: [[TMP4:%.*]] = load i8*, i8** [[DOTADDR1]], align 8
// CHECK6-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to [2 x i8*]*
// CHECK6-NEXT: [[TMP6:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP3]], i64 0, i64 0
// CHECK6-NEXT: [[TMP7:%.*]] = load i8*, i8** [[TMP6]], align 8
// CHECK6-NEXT: [[TMP8:%.*]] = bitcast i8* [[TMP7]] to i32**
// CHECK6-NEXT: [[TMP9:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 0
// CHECK6-NEXT: [[TMP10:%.*]] = load i8*, i8** [[TMP9]], align 8
// CHECK6-NEXT: [[TMP11:%.*]] = bitcast i8* [[TMP10]] to i32**
// CHECK6-NEXT: [[TMP12:%.*]] = load i32*, i32** [[TMP11]], align 8
// CHECK6-NEXT: store i32* [[TMP12]], i32** [[TMP8]], align 8
// CHECK6-NEXT: [[TMP13:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP3]], i64 0, i64 1
// CHECK6-NEXT: [[TMP14:%.*]] = load i8*, i8** [[TMP13]], align 8
// CHECK6-NEXT: [[TMP15:%.*]] = bitcast i8* [[TMP14]] to %struct.St**
// CHECK6-NEXT: [[TMP16:%.*]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[TMP5]], i64 0, i64 1
// CHECK6-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP16]], align 8
// CHECK6-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to %struct.St**
// CHECK6-NEXT: [[TMP19:%.*]] = load %struct.St*, %struct.St** [[TMP18]], align 8
// CHECK6-NEXT: store %struct.St* [[TMP19]], %struct.St** [[TMP15]], align 8
// CHECK6-NEXT: ret void
//
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit e9c7291cb25f071f1a1dfa4049ed9f5a8a217b3e. Fix failing tests
2021-07-19 22:54:26 +08:00
//