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/sections_firstprivate_codeg...

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

1723 lines
116 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 -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK1
[OPENMP] -fopenmp enables OpenMP support (fix for http://llvm.org/PR23492) -fopenmp turns on OpenMP support and links libiomp5 as OpenMP library. Also there is -fopenmp={libiomp5|libgomp} option that allows to override effect of -fopenmp and link libgomp library (if -fopenmp=libgomp is specified). Differential Revision: http://reviews.llvm.org/D9736 llvm-svn: 237769
2015-05-20 12:24:19 +08:00
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-apple-darwin10 -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 -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK2
// RUN: %clang_cc1 -verify -fopenmp -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK3
// RUN: %clang_cc1 -verify -fopenmp -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK4
[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 -x c++ -triple x86_64-apple-darwin10 -emit-llvm %s -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 -x c++ -std=c++11 -triple x86_64-apple-darwin10 -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 -x c++ -triple x86_64-apple-darwin10 -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -std=c++11 -DLAMBDA -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -fblocks -DBLOCKS -triple x86_64-apple-darwin10 -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
[OPENMP] Codegen for 'firstprivate' clause in 'sections' directive. If there are 2 or more sections in a 'section' directive the following code is generated: <init for firstprivates> @__kmpc_cancel_barrier();// To avoid data race in firstprivate init @__kmpc_for_static_init_4(); <BODY for sections directive> @__kmpc_for_static_fini() If there is only one section, the following code is generated: if (@__kmpc_single()) { <init for firstprivates> @__kmpc_end_single(); } @__kmpc_cancel_barrier(); // To avoid data race in firstprivate init Differential Revision: http://reviews.llvm.org/D9214 llvm-svn: 235691
2015-04-24 11:37:03 +08:00
// expected-no-diagnostics
#ifndef HEADER
#define HEADER
struct St {
int a, b;
St() : a(0), b(0) {}
St(const St &st) : a(st.a + st.b), b(0) {}
~St() {}
};
volatile int g = 1212;
template <class T>
struct S {
T f;
S(T a) : f(a + g) {}
S() : f(g) {}
S(const S &s, St t = St()) : f(s.f + t.a) {}
operator T() { return T(); }
~S() {}
};
template <typename T>
T tmain() {
S<T> test;
T t_var = T();
T vec[] = {1, 2};
S<T> s_arr[] = {1, 2};
S<T> var(3);
#pragma omp parallel
#pragma omp sections firstprivate(t_var, vec, s_arr, var)
{
vec[0] = t_var;
#pragma omp section
s_arr[0] = var;
}
return T();
}
S<float> test;
int t_var = 333;
int vec[] = {1, 2};
S<float> s_arr[] = {1, 2};
S<float> var(3);
int main() {
Allow static local variables specified on data-sharing attribute clauses. http://reviews.llvm.org/D11619 llvm-svn: 247715
2015-09-16 02:56:58 +08:00
static int sivar;
[OPENMP] Codegen for 'firstprivate' clause in 'sections' directive. If there are 2 or more sections in a 'section' directive the following code is generated: <init for firstprivates> @__kmpc_cancel_barrier();// To avoid data race in firstprivate init @__kmpc_for_static_init_4(); <BODY for sections directive> @__kmpc_for_static_fini() If there is only one section, the following code is generated: if (@__kmpc_single()) { <init for firstprivates> @__kmpc_end_single(); } @__kmpc_cancel_barrier(); // To avoid data race in firstprivate init Differential Revision: http://reviews.llvm.org/D9214 llvm-svn: 235691
2015-04-24 11:37:03 +08:00
#ifdef LAMBDA
[&]() {
#pragma omp parallel
Allow static local variables specified on data-sharing attribute clauses. http://reviews.llvm.org/D11619 llvm-svn: 247715
2015-09-16 02:56:58 +08:00
#pragma omp sections firstprivate(g, sivar)
[OPENMP] Codegen for 'firstprivate' clause in 'sections' directive. If there are 2 or more sections in a 'section' directive the following code is generated: <init for firstprivates> @__kmpc_cancel_barrier();// To avoid data race in firstprivate init @__kmpc_for_static_init_4(); <BODY for sections directive> @__kmpc_for_static_fini() If there is only one section, the following code is generated: if (@__kmpc_single()) { <init for firstprivates> @__kmpc_end_single(); } @__kmpc_cancel_barrier(); // To avoid data race in firstprivate init Differential Revision: http://reviews.llvm.org/D9214 llvm-svn: 235691
2015-04-24 11:37:03 +08:00
{
// Skip temp vars for loop
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
Allow static local variables specified on data-sharing attribute clauses. http://reviews.llvm.org/D11619 llvm-svn: 247715
2015-09-16 02:56:58 +08:00
{
g = 1;
sivar = 10;
}
[OPENMP] Codegen for 'firstprivate' clause in 'sections' directive. If there are 2 or more sections in a 'section' directive the following code is generated: <init for firstprivates> @__kmpc_cancel_barrier();// To avoid data race in firstprivate init @__kmpc_for_static_init_4(); <BODY for sections directive> @__kmpc_for_static_fini() If there is only one section, the following code is generated: if (@__kmpc_single()) { <init for firstprivates> @__kmpc_end_single(); } @__kmpc_cancel_barrier(); // To avoid data race in firstprivate init Differential Revision: http://reviews.llvm.org/D9214 llvm-svn: 235691
2015-04-24 11:37:03 +08:00
#pragma omp section
[&]() {
g = 2;
Allow static local variables specified on data-sharing attribute clauses. http://reviews.llvm.org/D11619 llvm-svn: 247715
2015-09-16 02:56:58 +08:00
sivar = 20;
[OPENMP] Codegen for 'firstprivate' clause in 'sections' directive. If there are 2 or more sections in a 'section' directive the following code is generated: <init for firstprivates> @__kmpc_cancel_barrier();// To avoid data race in firstprivate init @__kmpc_for_static_init_4(); <BODY for sections directive> @__kmpc_for_static_fini() If there is only one section, the following code is generated: if (@__kmpc_single()) { <init for firstprivates> @__kmpc_end_single(); } @__kmpc_cancel_barrier(); // To avoid data race in firstprivate init Differential Revision: http://reviews.llvm.org/D9214 llvm-svn: 235691
2015-04-24 11:37:03 +08:00
}();
}
}();
return 0;
#elif defined(BLOCKS)
^{
#pragma omp parallel
Allow static local variables specified on data-sharing attribute clauses. http://reviews.llvm.org/D11619 llvm-svn: 247715
2015-09-16 02:56:58 +08:00
#pragma omp sections firstprivate(g, sivar)
[OPENMP] Codegen for 'firstprivate' clause in 'sections' directive. If there are 2 or more sections in a 'section' directive the following code is generated: <init for firstprivates> @__kmpc_cancel_barrier();// To avoid data race in firstprivate init @__kmpc_for_static_init_4(); <BODY for sections directive> @__kmpc_for_static_fini() If there is only one section, the following code is generated: if (@__kmpc_single()) { <init for firstprivates> @__kmpc_end_single(); } @__kmpc_cancel_barrier(); // To avoid data race in firstprivate init Differential Revision: http://reviews.llvm.org/D9214 llvm-svn: 235691
2015-04-24 11:37:03 +08:00
{
// Skip temp vars for loop
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-06 06:13:14 +08:00
Allow static local variables specified on data-sharing attribute clauses. http://reviews.llvm.org/D11619 llvm-svn: 247715
2015-09-16 02:56:58 +08:00
{
g = 1;
sivar = 10;
}
[OPENMP] Codegen for 'firstprivate' clause in 'sections' directive. If there are 2 or more sections in a 'section' directive the following code is generated: <init for firstprivates> @__kmpc_cancel_barrier();// To avoid data race in firstprivate init @__kmpc_for_static_init_4(); <BODY for sections directive> @__kmpc_for_static_fini() If there is only one section, the following code is generated: if (@__kmpc_single()) { <init for firstprivates> @__kmpc_end_single(); } @__kmpc_cancel_barrier(); // To avoid data race in firstprivate init Differential Revision: http://reviews.llvm.org/D9214 llvm-svn: 235691
2015-04-24 11:37:03 +08:00
#pragma omp section
^{
g = 2;
Allow static local variables specified on data-sharing attribute clauses. http://reviews.llvm.org/D11619 llvm-svn: 247715
2015-09-16 02:56:58 +08:00
sivar = 20;
[OPENMP] Codegen for 'firstprivate' clause in 'sections' directive. If there are 2 or more sections in a 'section' directive the following code is generated: <init for firstprivates> @__kmpc_cancel_barrier();// To avoid data race in firstprivate init @__kmpc_for_static_init_4(); <BODY for sections directive> @__kmpc_for_static_fini() If there is only one section, the following code is generated: if (@__kmpc_single()) { <init for firstprivates> @__kmpc_end_single(); } @__kmpc_cancel_barrier(); // To avoid data race in firstprivate init Differential Revision: http://reviews.llvm.org/D9214 llvm-svn: 235691
2015-04-24 11:37:03 +08:00
}();
}
}();
return 0;
#else
Allow static local variables specified on data-sharing attribute clauses. http://reviews.llvm.org/D11619 llvm-svn: 247715
2015-09-16 02:56:58 +08:00
#pragma omp sections firstprivate(t_var, vec, s_arr, var, sivar) nowait
[OPENMP] Codegen for 'firstprivate' clause in 'sections' directive. If there are 2 or more sections in a 'section' directive the following code is generated: <init for firstprivates> @__kmpc_cancel_barrier();// To avoid data race in firstprivate init @__kmpc_for_static_init_4(); <BODY for sections directive> @__kmpc_for_static_fini() If there is only one section, the following code is generated: if (@__kmpc_single()) { <init for firstprivates> @__kmpc_end_single(); } @__kmpc_cancel_barrier(); // To avoid data race in firstprivate init Differential Revision: http://reviews.llvm.org/D9214 llvm-svn: 235691
2015-04-24 11:37:03 +08:00
{
{
vec[0] = t_var;
s_arr[0] = var;
Allow static local variables specified on data-sharing attribute clauses. http://reviews.llvm.org/D11619 llvm-svn: 247715
2015-09-16 02:56:58 +08:00
sivar = 31;
[OPENMP] Codegen for 'firstprivate' clause in 'sections' directive. If there are 2 or more sections in a 'section' directive the following code is generated: <init for firstprivates> @__kmpc_cancel_barrier();// To avoid data race in firstprivate init @__kmpc_for_static_init_4(); <BODY for sections directive> @__kmpc_for_static_fini() If there is only one section, the following code is generated: if (@__kmpc_single()) { <init for firstprivates> @__kmpc_end_single(); } @__kmpc_cancel_barrier(); // To avoid data race in firstprivate init Differential Revision: http://reviews.llvm.org/D9214 llvm-svn: 235691
2015-04-24 11:37:03 +08:00
}
}
return tmain<int>();
#endif
}
// firstprivate t_var(t_var)
[OPENMP] Fixed codegen for firstprivate vars in standalone worksharing directives. If firstprivate variable is is captured by value in outlined region and then used as firstprivate variable in inner worksharing directive, the copy for this firstprivate variable was not created. Fixed this bug. llvm-svn: 270536
2016-05-24 15:40:12 +08:00
[OPENMP] Codegen for 'firstprivate' clause in 'sections' directive. If there are 2 or more sections in a 'section' directive the following code is generated: <init for firstprivates> @__kmpc_cancel_barrier();// To avoid data race in firstprivate init @__kmpc_for_static_init_4(); <BODY for sections directive> @__kmpc_for_static_fini() If there is only one section, the following code is generated: if (@__kmpc_single()) { <init for firstprivates> @__kmpc_end_single(); } @__kmpc_cancel_barrier(); // To avoid data race in firstprivate init Differential Revision: http://reviews.llvm.org/D9214 llvm-svn: 235691
2015-04-24 11:37:03 +08:00
// firstprivate vec(vec)
// firstprivate s_arr(s_arr)
// firstprivate var(var)
Allow static local variables specified on data-sharing attribute clauses. http://reviews.llvm.org/D11619 llvm-svn: 247715
2015-09-16 02:56:58 +08:00
// firstprivate isvar
[OPENMP] Generalize codegen for 'sections'-based directive. If 'sections' directive has only one sub-section, the code for 'single'-based directive was emitted. Removed this codegen, because it causes crashes in different cases. llvm-svn: 258495
2016-01-22 16:56:50 +08:00
Allow static local variables specified on data-sharing attribute clauses. http://reviews.llvm.org/D11619 llvm-svn: 247715
2015-09-16 02:56:58 +08:00
[OPENMP] Codegen for 'firstprivate' clause in 'sections' directive. If there are 2 or more sections in a 'section' directive the following code is generated: <init for firstprivates> @__kmpc_cancel_barrier();// To avoid data race in firstprivate init @__kmpc_for_static_init_4(); <BODY for sections directive> @__kmpc_for_static_fini() If there is only one section, the following code is generated: if (@__kmpc_single()) { <init for firstprivates> @__kmpc_end_single(); } @__kmpc_cancel_barrier(); // To avoid data race in firstprivate init Differential Revision: http://reviews.llvm.org/D9214 llvm-svn: 235691
2015-04-24 11:37:03 +08:00
// ~(firstprivate var), ~(firstprivate s_arr)
// Skip temp vars for loop
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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] Outlined function for parallel and other regions with list of captured variables. Currently all variables used in OpenMP regions are captured into a record and passed to outlined functions in this record. It may result in some poor performance because of too complex analysis later in optimization passes. Patch makes to emit outlined functions for parallel-based regions with a list of captured variables. It reduces code for 2*n GEPs, stores and loads at least. Codegen for task-based regions remains unchanged because runtime requires that all captured variables are passed in captured record. llvm-svn: 247251
2015-09-10 16:12:02 +08:00
[OPENMP] Codegen for 'firstprivate' clause in 'sections' directive. If there are 2 or more sections in a 'section' directive the following code is generated: <init for firstprivates> @__kmpc_cancel_barrier();// To avoid data race in firstprivate init @__kmpc_for_static_init_4(); <BODY for sections directive> @__kmpc_for_static_fini() If there is only one section, the following code is generated: if (@__kmpc_single()) { <init for firstprivates> @__kmpc_end_single(); } @__kmpc_cancel_barrier(); // To avoid data race in firstprivate init Differential Revision: http://reviews.llvm.org/D9214 llvm-svn: 235691
2015-04-24 11:37:03 +08:00
// firstprivate t_var(t_var)
// firstprivate vec(vec)
// firstprivate s_arr(s_arr)
// firstprivate var(var)
[OPENMP] Remove extra sync barriers for 'firstprivate' clause. Sync barrier will be emitted after generation of firstprivate variables only if one of the firstprivate vars is used in lastprivate clause. llvm-svn: 260877
2016-02-15 16:07:17 +08:00
// No synchronization for initialization.
[OPENMP] Codegen for 'firstprivate' clause in 'sections' directive. If there are 2 or more sections in a 'section' directive the following code is generated: <init for firstprivates> @__kmpc_cancel_barrier();// To avoid data race in firstprivate init @__kmpc_for_static_init_4(); <BODY for sections directive> @__kmpc_for_static_fini() If there is only one section, the following code is generated: if (@__kmpc_single()) { <init for firstprivates> @__kmpc_end_single(); } @__kmpc_cancel_barrier(); // To avoid data race in firstprivate init Differential Revision: http://reviews.llvm.org/D9214 llvm-svn: 235691
2015-04-24 11:37:03 +08:00
// ~(firstprivate var), ~(firstprivate s_arr)
#endif
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 {{[^@]+}}@__cxx_global_var_init
// CHECK1-SAME: () #[[ATTR0:[0-9]+]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK1-NEXT: entry:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIfEC1Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) @test)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S*)* @_ZN1SIfED1Ev to void (i8*)*), i8* bitcast (%struct.S* @test to i8*), i8* @__dso_handle) #[[ATTR2:[0-9]+]]
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] 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.S*, align 8
// CHECK1-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIfEC2Ev(%struct.S* noundef 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
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfED1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.S*, align 8
// CHECK1-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIfED2Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 {{[^@]+}}@_ZN1SIfEC2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
[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: entry:
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK1-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[TMP0:%.*]] = load volatile i32, i32* @g, align 4
// CHECK1-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP0]] to float
// CHECK1-NEXT: store float [[CONV]], float* [[F]], align 4
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfED2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
[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: entry:
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK1-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// 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 {{[^@]+}}@__cxx_global_var_init.1
// CHECK1-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK1-NEXT: entry:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIfEC1Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 0, i64 0), float noundef 1.000000e+00)
// CHECK1-NEXT: call void @_ZN1SIfEC1Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 0, i64 1), float noundef 2.000000e+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
// CHECK1-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR2]]
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ef
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], float noundef [[A:%.*]]) unnamed_addr #[[ATTR1]] 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.S*, align 8
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca float, align 4
// CHECK1-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK1-NEXT: store float [[A]], float* [[A_ADDR]], align 4
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load float, float* [[A_ADDR]], align 4
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIfEC2Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS1]], float noundef [[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 {{[^@]+}}@__cxx_global_array_dtor
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 17:53:11 +08:00
// CHECK1-SAME: (i8* noundef [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK1-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
[ConstantFold] Remove unnecessary bounded index restriction The fold for merging a GEP of GEP into a single GEP currently bails if doing so would result in notional overindexing. The justification given in the comment above this check is dangerously incorrect: GEPs with notional overindexing are perfectly fine, and if some code treats them incorrectly, then that code is broken, not the GEP. Such a GEP might legally appear in source IR, so only preventing its creation cannot be sufficient. (The constant folder also ends up canonicalizing the GEP to remove the notional overindexing, but that's neither here nor there.) This check dates back to https://github.com/llvm/llvm-project/commit/bd4fef4a8939db18f39b108e19097b25e2c7c47a, and as far as I can tell the original issue this was trying to patch around has since been resolved. Differential Revision: https://reviews.llvm.org/D116587
2022-01-04 18:53:28 +08:00
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIfED1Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
// 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 {{[^@]+}}@_ZN1SIfEC2Ef
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], float noundef [[A:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
[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: entry:
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca float, align 4
// CHECK1-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK1-NEXT: store float [[A]], float* [[A_ADDR]], align 4
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[TMP0:%.*]] = load float, float* [[A_ADDR]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = load volatile i32, i32* @g, align 4
// CHECK1-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP1]] to float
// CHECK1-NEXT: [[ADD:%.*]] = fadd float [[TMP0]], [[CONV]]
// CHECK1-NEXT: store float [[ADD]], float* [[F]], align 4
// 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 {{[^@]+}}@__cxx_global_var_init.2
// CHECK1-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK1-NEXT: entry:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIfEC1Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) @var, float noundef 3.000000e+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
// CHECK1-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S*)* @_ZN1SIfED1Ev to void (i8*)*), i8* bitcast (%struct.S* @var to i8*), i8* @__dso_handle) #[[ATTR2]]
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@main
// CHECK1-SAME: () #[[ATTR3:[0-9]+]] {
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[RETVAL:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTOMP_SECTIONS_LB_:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTOMP_SECTIONS_UB_:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTOMP_SECTIONS_ST_:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTOMP_SECTIONS_IL_:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTOMP_SECTIONS_IV_:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[T_VAR:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[VEC:%.*]] = alloca [2 x i32], align 4
// CHECK1-NEXT: [[S_ARR:%.*]] = alloca [2 x %struct.S], align 4
// CHECK1-NEXT: [[AGG_TMP:%.*]] = alloca [[STRUCT_ST:%.*]], align 4
// CHECK1-NEXT: [[VAR:%.*]] = alloca [[STRUCT_S:%.*]], align 4
// CHECK1-NEXT: [[AGG_TMP2:%.*]] = alloca [[STRUCT_ST]], align 4
// CHECK1-NEXT: [[SIVAR:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2:[0-9]+]])
// CHECK1-NEXT: store i32 0, i32* [[RETVAL]], align 4
// CHECK1-NEXT: store i32 0, i32* [[DOTOMP_SECTIONS_LB_]], align 4
// CHECK1-NEXT: store i32 0, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK1-NEXT: store i32 1, i32* [[DOTOMP_SECTIONS_ST_]], align 4
// CHECK1-NEXT: store i32 0, i32* [[DOTOMP_SECTIONS_IL_]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = load i32, i32* @t_var, align 4
// CHECK1-NEXT: store i32 [[TMP1]], i32* [[T_VAR]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = bitcast [2 x i32]* [[VEC]] to i8*
// CHECK1-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP2]], i8* align 4 bitcast ([2 x i32]* @vec to i8*), i64 8, i1 false)
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i32 0, i32 0
// CHECK1-NEXT: [[TMP3:%.*]] = getelementptr [[STRUCT_S]], %struct.S* [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq %struct.S* [[ARRAY_BEGIN]], [[TMP3]]
// CHECK1-NEXT: br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE1:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
// CHECK1: omp.arraycpy.body:
// CHECK1-NEXT: [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
// CHECK1-NEXT: [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %struct.S* [ [[ARRAY_BEGIN]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN2StC1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP]])
// CHECK1-NEXT: call void @_ZN1SIfEC1ERKS0_2St(%struct.S* noundef nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %struct.S* noundef nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]], %struct.St* noundef [[AGG_TMP]])
// CHECK1-NEXT: call void @_ZN2StD1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 [[STRUCT_S]], %struct.S* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
// CHECK1-NEXT: [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[STRUCT_S]], %struct.S* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
// CHECK1-NEXT: [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP3]]
// CHECK1-NEXT: br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE1]], label [[OMP_ARRAYCPY_BODY]]
// CHECK1: omp.arraycpy.done1:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN2StC1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP2]])
// CHECK1-NEXT: call void @_ZN1SIfEC1ERKS0_2St(%struct.S* noundef nonnull align 4 dereferenceable(4) [[VAR]], %struct.S* noundef nonnull align 4 dereferenceable(4) @var, %struct.St* noundef [[AGG_TMP2]])
// CHECK1-NEXT: call void @_ZN2StD1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP2]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: [[TMP4:%.*]] = load i32, i32* @_ZZ4mainE5sivar, align 4
// CHECK1-NEXT: store i32 [[TMP4]], i32* [[SIVAR]], align 4
// CHECK1-NEXT: call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP0]], i32 34, i32* [[DOTOMP_SECTIONS_IL_]], i32* [[DOTOMP_SECTIONS_LB_]], i32* [[DOTOMP_SECTIONS_UB_]], i32* [[DOTOMP_SECTIONS_ST_]], i32 1, i32 1)
// CHECK1-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = icmp slt i32 [[TMP5]], 0
// CHECK1-NEXT: [[TMP7:%.*]] = select i1 [[TMP6]], i32 [[TMP5]], i32 0
// CHECK1-NEXT: store i32 [[TMP7]], i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_LB_]], align 4
// CHECK1-NEXT: store i32 [[TMP8]], i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK1-NEXT: br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1: omp.inner.for.cond:
// CHECK1-NEXT: [[TMP9:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK1-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK1-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]]
// CHECK1-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1: omp.inner.for.body:
// CHECK1-NEXT: [[TMP11:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK1-NEXT: switch i32 [[TMP11]], label [[DOTOMP_SECTIONS_EXIT:%.*]] [
// CHECK1-NEXT: i32 0, label [[DOTOMP_SECTIONS_CASE:%.*]]
// CHECK1-NEXT: ]
// CHECK1: .omp.sections.case:
// CHECK1-NEXT: [[TMP12:%.*]] = load i32, i32* [[T_VAR]], align 4
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 0
// CHECK1-NEXT: store i32 [[TMP12]], i32* [[ARRAYIDX]], align 4
// CHECK1-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i64 0, i64 0
// CHECK1-NEXT: [[TMP13:%.*]] = bitcast %struct.S* [[ARRAYIDX3]] to i8*
// CHECK1-NEXT: [[TMP14:%.*]] = bitcast %struct.S* [[VAR]] to i8*
// CHECK1-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i64 4, i1 false)
// CHECK1-NEXT: store i32 31, i32* [[SIVAR]], align 4
// CHECK1-NEXT: br label [[DOTOMP_SECTIONS_EXIT]]
// CHECK1: .omp.sections.exit:
// CHECK1-NEXT: br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1: omp.inner.for.inc:
// CHECK1-NEXT: [[TMP15:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK1-NEXT: [[INC:%.*]] = add nsw i32 [[TMP15]], 1
// CHECK1-NEXT: store i32 [[INC]], i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK1-NEXT: br label [[OMP_INNER_FOR_COND]]
// CHECK1: omp.inner.for.end:
// CHECK1-NEXT: call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIfED1Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) [[VAR]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: [[ARRAY_BEGIN4:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i32 0, i32 0
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAY_BEGIN4]], i64 2
// CHECK1-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK1: arraydestroy.body:
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ [[TMP16]], [[OMP_INNER_FOR_END]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIfED1Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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_DONE:%.*]] = icmp eq %struct.S* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN4]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE5:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done5:
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 17:53:11 +08:00
// CHECK1-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 i32 [[CALL]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN2StC1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.St* noundef nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.St*, align 8
// CHECK1-NEXT: store %struct.St* [[THIS]], %struct.St** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.St*, %struct.St** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN2StC2Ev(%struct.St* noundef nonnull align 4 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 {{[^@]+}}@_ZN1SIfEC1ERKS0_2St
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], %struct.S* noundef nonnull align 4 dereferenceable(4) [[S:%.*]], %struct.St* noundef [[T:%.*]]) unnamed_addr #[[ATTR1]] 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.S*, align 8
// CHECK1-NEXT: [[S_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK1-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK1-NEXT: store %struct.S* [[S]], %struct.S** [[S_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load %struct.S*, %struct.S** [[S_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIfEC2ERKS0_2St(%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS1]], %struct.S* noundef nonnull align 4 dereferenceable(4) [[TMP0]], %struct.St* noundef [[T]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 {{[^@]+}}@_ZN2StD1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.St* noundef nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.St*, align 8
// CHECK1-NEXT: store %struct.St* [[THIS]], %struct.St** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.St*, %struct.St** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN2StD2Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[THIS1]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 {{[^@]+}}@_Z5tmainIiET_v
// CHECK1-SAME: () #[[ATTR5:[0-9]+]] {
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[RETVAL:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[TEST:%.*]] = alloca [[STRUCT_S_0:%.*]], align 4
// CHECK1-NEXT: [[T_VAR:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[VEC:%.*]] = alloca [2 x i32], align 4
// CHECK1-NEXT: [[S_ARR:%.*]] = alloca [2 x %struct.S.0], align 4
// CHECK1-NEXT: [[VAR:%.*]] = alloca [[STRUCT_S_0]], align 4
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIiEC1Ev(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[TEST]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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* [[T_VAR]], align 4
// CHECK1-NEXT: [[TMP0:%.*]] = bitcast [2 x i32]* [[VEC]] to i8*
// CHECK1-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP0]], i8* align 4 bitcast ([2 x i32]* @__const._Z5tmainIiET_v.vec to i8*), i64 8, i1 false)
// CHECK1-NEXT: [[ARRAYINIT_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i64 0, i64 0
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIiEC1Ei(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[ARRAYINIT_BEGIN]], i32 noundef 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: [[ARRAYINIT_ELEMENT:%.*]] = getelementptr inbounds [[STRUCT_S_0]], %struct.S.0* [[ARRAYINIT_BEGIN]], i64 1
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIiEC1Ei(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[ARRAYINIT_ELEMENT]], i32 noundef 2)
// CHECK1-NEXT: call void @_ZN1SIiEC1Ei(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[VAR]], i32 noundef 3)
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB2]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, [2 x i32]*, [2 x %struct.S.0]*, %struct.S.0*)* @.omp_outlined. to void (i32*, i32*, ...)*), i32* [[T_VAR]], [2 x i32]* [[VEC]], [2 x %struct.S.0]* [[S_ARR]], %struct.S.0* [[VAR]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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* [[RETVAL]], align 4
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[VAR]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i32 0, i32 0
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_S_0]], %struct.S.0* [[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:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S.0* [ [[TMP1]], [[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 [[STRUCT_S_0]], %struct.S.0* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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_DONE:%.*]] = icmp eq %struct.S.0* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done1:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[TMP2:%.*]] = load i32, i32* [[RETVAL]], align 4
// CHECK1-NEXT: ret i32 [[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-LABEL: define {{[^@]+}}@_ZN2StC2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.St* noundef nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.St*, align 8
// CHECK1-NEXT: store %struct.St* [[THIS]], %struct.St** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.St*, %struct.St** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.St* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: store i32 0, i32* [[A]], align 4
// CHECK1-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.St* [[THIS1]], i32 0, i32 1
// CHECK1-NEXT: store i32 0, i32* [[B]], align 4
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIfEC2ERKS0_2St
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], %struct.S* noundef nonnull align 4 dereferenceable(4) [[S:%.*]], %struct.St* noundef [[T:%.*]]) unnamed_addr #[[ATTR1]] 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.S*, align 8
// CHECK1-NEXT: [[S_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK1-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK1-NEXT: store %struct.S* [[S]], %struct.S** [[S_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[TMP0:%.*]] = load %struct.S*, %struct.S** [[S_ADDR]], align 8
// CHECK1-NEXT: [[F2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[TMP0]], i32 0, i32 0
// CHECK1-NEXT: [[TMP1:%.*]] = load float, float* [[F2]], align 4
// CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.St* [[T]], i32 0, i32 0
// CHECK1-NEXT: [[TMP2:%.*]] = load i32, i32* [[A]], align 4
// CHECK1-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP2]] to float
// CHECK1-NEXT: [[ADD:%.*]] = fadd float [[TMP1]], [[CONV]]
// CHECK1-NEXT: store float [[ADD]], float* [[F]], align 4
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN2StD2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.St* noundef nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.St*, align 8
// CHECK1-NEXT: store %struct.St* [[THIS]], %struct.St** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.St*, %struct.St** [[THIS_ADDR]], align 8
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK1-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIiEC2Ev(%struct.S.0* noundef 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
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ei
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 noundef [[A:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: store i32 [[A]], i32* [[A_ADDR]], align 4
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIiEC2Ei(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS1]], i32 noundef [[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 {{[^@]+}}@.omp_outlined.
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 17:53:11 +08:00
// CHECK1-SAME: (i32* noalias noundef [[DOTGLOBAL_TID_:%.*]], i32* noalias noundef [[DOTBOUND_TID_:%.*]], i32* noundef nonnull align 4 dereferenceable(4) [[T_VAR:%.*]], [2 x i32]* noundef nonnull align 4 dereferenceable(8) [[VEC:%.*]], [2 x %struct.S.0]* noundef nonnull align 4 dereferenceable(8) [[S_ARR:%.*]], %struct.S.0* noundef nonnull align 4 dereferenceable(4) [[VAR:%.*]]) #[[ATTR6:[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:
// 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 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[T_VAR_ADDR:%.*]] = alloca i32*, align 8
// CHECK1-NEXT: [[VEC_ADDR:%.*]] = alloca [2 x i32]*, align 8
// CHECK1-NEXT: [[S_ARR_ADDR:%.*]] = alloca [2 x %struct.S.0]*, align 8
// CHECK1-NEXT: [[VAR_ADDR:%.*]] = alloca %struct.S.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
// CHECK1-NEXT: [[DOTOMP_SECTIONS_LB_:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTOMP_SECTIONS_UB_:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTOMP_SECTIONS_ST_:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTOMP_SECTIONS_IL_:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[DOTOMP_SECTIONS_IV_:%.*]] = alloca i32, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[T_VAR1:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[VEC2:%.*]] = alloca [2 x i32], align 4
// CHECK1-NEXT: [[S_ARR3:%.*]] = alloca [2 x %struct.S.0], align 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: [[AGG_TMP:%.*]] = alloca [[STRUCT_ST:%.*]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[VAR5:%.*]] = alloca [[STRUCT_S_0:%.*]], align 4
// CHECK1-NEXT: [[AGG_TMP6:%.*]] = alloca [[STRUCT_ST]], align 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: 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 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: store i32* [[T_VAR]], i32** [[T_VAR_ADDR]], align 8
// CHECK1-NEXT: store [2 x i32]* [[VEC]], [2 x i32]** [[VEC_ADDR]], align 8
// CHECK1-NEXT: store [2 x %struct.S.0]* [[S_ARR]], [2 x %struct.S.0]** [[S_ARR_ADDR]], align 8
// CHECK1-NEXT: store %struct.S.0* [[VAR]], %struct.S.0** [[VAR_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load i32*, i32** [[T_VAR_ADDR]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load [2 x i32]*, [2 x i32]** [[VEC_ADDR]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load [2 x %struct.S.0]*, [2 x %struct.S.0]** [[S_ARR_ADDR]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = load %struct.S.0*, %struct.S.0** [[VAR_ADDR]], 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_SECTIONS_LB_]], align 4
// CHECK1-NEXT: store i32 1, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK1-NEXT: store i32 1, i32* [[DOTOMP_SECTIONS_ST_]], align 4
// CHECK1-NEXT: store i32 0, i32* [[DOTOMP_SECTIONS_IL_]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK1-NEXT: store i32 [[TMP4]], i32* [[T_VAR1]], align 4
// CHECK1-NEXT: [[TMP5:%.*]] = bitcast [2 x i32]* [[VEC2]] to i8*
// CHECK1-NEXT: [[TMP6:%.*]] = bitcast [2 x i32]* [[TMP1]] to i8*
// CHECK1-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP5]], i8* align 4 [[TMP6]], i64 8, i1 false)
// CHECK1-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR3]], i32 0, i32 0
// CHECK1-NEXT: [[TMP7:%.*]] = bitcast [2 x %struct.S.0]* [[TMP2]] to %struct.S.0*
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr [[STRUCT_S_0]], %struct.S.0* [[ARRAY_BEGIN]], i64 2
// CHECK1-NEXT: [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq %struct.S.0* [[ARRAY_BEGIN]], [[TMP8]]
// CHECK1-NEXT: br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_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: omp.arraycpy.body:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %struct.S.0* [ [[TMP7]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_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: [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %struct.S.0* [ [[ARRAY_BEGIN]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN2StC1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP]])
// CHECK1-NEXT: call void @_ZN1SIiEC1ERKS0_2St(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %struct.S.0* noundef nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]], %struct.St* noundef [[AGG_TMP]])
// CHECK1-NEXT: call void @_ZN2StD1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 [[STRUCT_S_0]], %struct.S.0* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
// CHECK1-NEXT: [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[STRUCT_S_0]], %struct.S.0* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S.0* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP8]]
// CHECK1-NEXT: br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE4]], label [[OMP_ARRAYCPY_BODY]]
// CHECK1: omp.arraycpy.done4:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN2StC1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP6]])
// CHECK1-NEXT: call void @_ZN1SIiEC1ERKS0_2St(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[VAR5]], %struct.S.0* noundef nonnull align 4 dereferenceable(4) [[TMP3]], %struct.St* noundef [[AGG_TMP6]])
// CHECK1-NEXT: call void @_ZN2StD1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP6]]) #[[ATTR2]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[TMP9:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: [[TMP10:%.*]] = load i32, i32* [[TMP9]], align 4
// CHECK1-NEXT: call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP10]], i32 34, i32* [[DOTOMP_SECTIONS_IL_]], i32* [[DOTOMP_SECTIONS_LB_]], i32* [[DOTOMP_SECTIONS_UB_]], i32* [[DOTOMP_SECTIONS_ST_]], i32 1, i32 1)
// CHECK1-NEXT: [[TMP11:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK1-NEXT: [[TMP12:%.*]] = icmp slt i32 [[TMP11]], 1
// CHECK1-NEXT: [[TMP13:%.*]] = select i1 [[TMP12]], i32 [[TMP11]], i32 1
// CHECK1-NEXT: store i32 [[TMP13]], i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK1-NEXT: [[TMP14:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_LB_]], align 4
// CHECK1-NEXT: store i32 [[TMP14]], i32* [[DOTOMP_SECTIONS_IV_]], align 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: br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK1: omp.inner.for.cond:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[TMP15:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK1-NEXT: [[TMP16:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK1-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP15]], [[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: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK1: omp.inner.for.body:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[TMP17:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK1-NEXT: switch i32 [[TMP17]], label [[DOTOMP_SECTIONS_EXIT:%.*]] [
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: i32 0, label [[DOTOMP_SECTIONS_CASE:%.*]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: i32 1, label [[DOTOMP_SECTIONS_CASE7:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: ]
// CHECK1: .omp.sections.case:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[TMP18:%.*]] = load i32, i32* [[T_VAR1]], align 4
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC2]], i64 0, i64 0
// CHECK1-NEXT: store i32 [[TMP18]], i32* [[ARRAYIDX]], align 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: br label [[DOTOMP_SECTIONS_EXIT]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1: .omp.sections.case7:
// CHECK1-NEXT: [[ARRAYIDX8:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR3]], i64 0, i64 0
// CHECK1-NEXT: [[TMP19:%.*]] = bitcast %struct.S.0* [[ARRAYIDX8]] to i8*
// CHECK1-NEXT: [[TMP20:%.*]] = bitcast %struct.S.0* [[VAR5]] to i8*
// CHECK1-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP19]], i8* align 4 [[TMP20]], i64 4, i1 false)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 [[DOTOMP_SECTIONS_EXIT]]
// CHECK1: .omp.sections.exit:
// CHECK1-NEXT: br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK1: omp.inner.for.inc:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[TMP21:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK1-NEXT: [[INC:%.*]] = add nsw i32 [[TMP21]], 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: store i32 [[INC]], i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK1-NEXT: br label [[OMP_INNER_FOR_COND]]
// CHECK1: omp.inner.for.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[TMP22:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: [[TMP23:%.*]] = load i32, i32* [[TMP22]], align 4
// CHECK1-NEXT: call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP23]])
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[VAR5]]) #[[ATTR2]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[ARRAY_BEGIN9:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR3]], i32 0, i32 0
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT_S_0]], %struct.S.0* [[ARRAY_BEGIN9]], 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:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S.0* [ [[TMP24]], [[OMP_INNER_FOR_END]] ], [ [[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 [[STRUCT_S_0]], %struct.S.0* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIiED1Ev(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK1-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S.0* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN9]]
// CHECK1-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE10:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK1: arraydestroy.done10:
// CHECK1-NEXT: [[TMP25:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: [[TMP26:%.*]] = load i32, i32* [[TMP25]], align 4
// CHECK1-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB3:[0-9]+]], i32 [[TMP26]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 {{[^@]+}}@_ZN1SIiEC1ERKS0_2St
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], %struct.S.0* noundef nonnull align 4 dereferenceable(4) [[S:%.*]], %struct.St* noundef [[T:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK1-NEXT: [[S_ADDR:%.*]] = alloca %struct.S.0*, align 8
// CHECK1-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: store %struct.S.0* [[S]], %struct.S.0** [[S_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load %struct.S.0*, %struct.S.0** [[S_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIiEC2ERKS0_2St(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS1]], %struct.S.0* noundef nonnull align 4 dereferenceable(4) [[TMP0]], %struct.St* noundef [[T]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 {{[^@]+}}@_ZN1SIiED1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK1-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-NEXT: call void @_ZN1SIiED2Ev(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 {{[^@]+}}@_ZN1SIiEC2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK1-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], %struct.S.0* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[TMP0:%.*]] = load volatile i32, i32* @g, align 4
// CHECK1-NEXT: store i32 [[TMP0]], i32* [[F]], align 4
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIiEC2Ei
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 noundef [[A:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: store i32 [[A]], i32* [[A_ADDR]], align 4
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], %struct.S.0* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = load volatile i32, i32* @g, align 4
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP0]], [[TMP1]]
// CHECK1-NEXT: store i32 [[ADD]], i32* [[F]], align 4
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIiEC2ERKS0_2St
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], %struct.S.0* noundef nonnull align 4 dereferenceable(4) [[S:%.*]], %struct.St* noundef [[T:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK1-NEXT: [[S_ADDR:%.*]] = alloca %struct.S.0*, align 8
// CHECK1-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: store %struct.S.0* [[S]], %struct.S.0** [[S_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], %struct.S.0* [[THIS1]], i32 0, i32 0
// CHECK1-NEXT: [[TMP0:%.*]] = load %struct.S.0*, %struct.S.0** [[S_ADDR]], align 8
// CHECK1-NEXT: [[F2:%.*]] = getelementptr inbounds [[STRUCT_S_0]], %struct.S.0* [[TMP0]], i32 0, i32 0
// CHECK1-NEXT: [[TMP1:%.*]] = load i32, i32* [[F2]], align 4
// CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.St* [[T]], i32 0, i32 0
// CHECK1-NEXT: [[TMP2:%.*]] = load i32, i32* [[A]], align 4
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], [[TMP2]]
// CHECK1-NEXT: store i32 [[ADD]], i32* [[F]], align 4
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN1SIiED2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK1-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK1-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_sections_firstprivate_codegen.cpp
// CHECK1-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK1-NEXT: entry:
// CHECK1-NEXT: call void @__cxx_global_var_init()
// CHECK1-NEXT: call void @__cxx_global_var_init.1()
// CHECK1-NEXT: call void @__cxx_global_var_init.2()
// CHECK1-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@__cxx_global_var_init
// CHECK2-SAME: () #[[ATTR0:[0-9]+]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK2-NEXT: entry:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIfEC1Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) @test)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 (%struct.S*)* @_ZN1SIfED1Ev to void (i8*)*), i8* bitcast (%struct.S* @test to i8*), i8* @__dso_handle) #[[ATTR2:[0-9]+]]
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] 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.S*, align 8
// CHECK2-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIfEC2Ev(%struct.S* noundef 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 {{[^@]+}}@_ZN1SIfED1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.S*, align 8
// CHECK2-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIfED2Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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
//
//
[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
// CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
[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
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK2-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT: [[TMP0:%.*]] = load volatile i32, i32* @g, align 4
// CHECK2-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP0]] to float
// CHECK2-NEXT: store float [[CONV]], float* [[F]], align 4
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfED2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
[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
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK2-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK2-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
// CHECK2-LABEL: define {{[^@]+}}@__cxx_global_var_init.1
// CHECK2-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK2-NEXT: entry:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIfEC1Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 0, i64 0), float noundef 1.000000e+00)
// CHECK2-NEXT: call void @_ZN1SIfEC1Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 0, i64 1), float noundef 2.000000e+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
// CHECK2-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR2]]
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ef
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], float noundef [[A:%.*]]) unnamed_addr #[[ATTR1]] 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.S*, align 8
// CHECK2-NEXT: [[A_ADDR:%.*]] = alloca float, align 4
// CHECK2-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK2-NEXT: store float [[A]], float* [[A_ADDR]], align 4
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[TMP0:%.*]] = load float, float* [[A_ADDR]], align 4
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIfEC2Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS1]], float noundef [[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 {{[^@]+}}@__cxx_global_array_dtor
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 17:53:11 +08:00
// CHECK2-SAME: (i8* noundef [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK2-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK2-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK2: arraydestroy.body:
[ConstantFold] Remove unnecessary bounded index restriction The fold for merging a GEP of GEP into a single GEP currently bails if doing so would result in notional overindexing. The justification given in the comment above this check is dangerously incorrect: GEPs with notional overindexing are perfectly fine, and if some code treats them incorrectly, then that code is broken, not the GEP. Such a GEP might legally appear in source IR, so only preventing its creation cannot be sufficient. (The constant folder also ends up canonicalizing the GEP to remove the notional overindexing, but that's neither here nor there.) This check dates back to https://github.com/llvm/llvm-project/commit/bd4fef4a8939db18f39b108e19097b25e2c7c47a, and as far as I can tell the original issue this was trying to patch around has since been resolved. Differential Revision: https://reviews.llvm.org/D116587
2022-01-04 18:53:28 +08:00
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIfED1Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
// CHECK2-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK2: arraydestroy.done1:
// CHECK2-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
// CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], float noundef [[A:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
[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
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK2-NEXT: [[A_ADDR:%.*]] = alloca float, align 4
// CHECK2-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK2-NEXT: store float [[A]], float* [[A_ADDR]], align 4
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT: [[TMP0:%.*]] = load float, float* [[A_ADDR]], align 4
// CHECK2-NEXT: [[TMP1:%.*]] = load volatile i32, i32* @g, align 4
// CHECK2-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP1]] to float
// CHECK2-NEXT: [[ADD:%.*]] = fadd float [[TMP0]], [[CONV]]
// CHECK2-NEXT: store float [[ADD]], float* [[F]], align 4
// CHECK2-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
// CHECK2-LABEL: define {{[^@]+}}@__cxx_global_var_init.2
// CHECK2-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK2-NEXT: entry:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIfEC1Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) @var, float noundef 3.000000e+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
// CHECK2-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S*)* @_ZN1SIfED1Ev to void (i8*)*), i8* bitcast (%struct.S* @var to i8*), i8* @__dso_handle) #[[ATTR2]]
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@main
// CHECK2-SAME: () #[[ATTR3:[0-9]+]] {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[RETVAL:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTOMP_SECTIONS_LB_:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTOMP_SECTIONS_UB_:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTOMP_SECTIONS_ST_:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTOMP_SECTIONS_IL_:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTOMP_SECTIONS_IV_:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[T_VAR:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[VEC:%.*]] = alloca [2 x i32], align 4
// CHECK2-NEXT: [[S_ARR:%.*]] = alloca [2 x %struct.S], align 4
// CHECK2-NEXT: [[AGG_TMP:%.*]] = alloca [[STRUCT_ST:%.*]], align 4
// CHECK2-NEXT: [[VAR:%.*]] = alloca [[STRUCT_S:%.*]], align 4
// CHECK2-NEXT: [[AGG_TMP2:%.*]] = alloca [[STRUCT_ST]], align 4
// CHECK2-NEXT: [[SIVAR:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(%struct.ident_t* @[[GLOB2:[0-9]+]])
// CHECK2-NEXT: store i32 0, i32* [[RETVAL]], align 4
// CHECK2-NEXT: store i32 0, i32* [[DOTOMP_SECTIONS_LB_]], align 4
// CHECK2-NEXT: store i32 0, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK2-NEXT: store i32 1, i32* [[DOTOMP_SECTIONS_ST_]], align 4
// CHECK2-NEXT: store i32 0, i32* [[DOTOMP_SECTIONS_IL_]], align 4
// CHECK2-NEXT: [[TMP1:%.*]] = load i32, i32* @t_var, align 4
// CHECK2-NEXT: store i32 [[TMP1]], i32* [[T_VAR]], align 4
// CHECK2-NEXT: [[TMP2:%.*]] = bitcast [2 x i32]* [[VEC]] to i8*
// CHECK2-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP2]], i8* align 4 bitcast ([2 x i32]* @vec to i8*), i64 8, i1 false)
// CHECK2-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i32 0, i32 0
// CHECK2-NEXT: [[TMP3:%.*]] = getelementptr [[STRUCT_S]], %struct.S* [[ARRAY_BEGIN]], i64 2
// CHECK2-NEXT: [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq %struct.S* [[ARRAY_BEGIN]], [[TMP3]]
// CHECK2-NEXT: br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE1:%.*]], label [[OMP_ARRAYCPY_BODY:%.*]]
// CHECK2: omp.arraycpy.body:
// CHECK2-NEXT: [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0), [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
// CHECK2-NEXT: [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %struct.S* [ [[ARRAY_BEGIN]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN2StC1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP]])
// CHECK2-NEXT: call void @_ZN1SIfEC1ERKS0_2St(%struct.S* noundef nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %struct.S* noundef nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]], %struct.St* noundef [[AGG_TMP]])
// CHECK2-NEXT: call void @_ZN2StD1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 [[STRUCT_S]], %struct.S* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
// CHECK2-NEXT: [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[STRUCT_S]], %struct.S* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
// CHECK2-NEXT: [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP3]]
// CHECK2-NEXT: br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE1]], label [[OMP_ARRAYCPY_BODY]]
// CHECK2: omp.arraycpy.done1:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN2StC1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP2]])
// CHECK2-NEXT: call void @_ZN1SIfEC1ERKS0_2St(%struct.S* noundef nonnull align 4 dereferenceable(4) [[VAR]], %struct.S* noundef nonnull align 4 dereferenceable(4) @var, %struct.St* noundef [[AGG_TMP2]])
// CHECK2-NEXT: call void @_ZN2StD1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP2]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: [[TMP4:%.*]] = load i32, i32* @_ZZ4mainE5sivar, align 4
// CHECK2-NEXT: store i32 [[TMP4]], i32* [[SIVAR]], align 4
// CHECK2-NEXT: call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP0]], i32 34, i32* [[DOTOMP_SECTIONS_IL_]], i32* [[DOTOMP_SECTIONS_LB_]], i32* [[DOTOMP_SECTIONS_UB_]], i32* [[DOTOMP_SECTIONS_ST_]], i32 1, i32 1)
// CHECK2-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK2-NEXT: [[TMP6:%.*]] = icmp slt i32 [[TMP5]], 0
// CHECK2-NEXT: [[TMP7:%.*]] = select i1 [[TMP6]], i32 [[TMP5]], i32 0
// CHECK2-NEXT: store i32 [[TMP7]], i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK2-NEXT: [[TMP8:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_LB_]], align 4
// CHECK2-NEXT: store i32 [[TMP8]], i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK2-NEXT: br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2: omp.inner.for.cond:
// CHECK2-NEXT: [[TMP9:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK2-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK2-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]]
// CHECK2-NEXT: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2: omp.inner.for.body:
// CHECK2-NEXT: [[TMP11:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK2-NEXT: switch i32 [[TMP11]], label [[DOTOMP_SECTIONS_EXIT:%.*]] [
// CHECK2-NEXT: i32 0, label [[DOTOMP_SECTIONS_CASE:%.*]]
// CHECK2-NEXT: ]
// CHECK2: .omp.sections.case:
// CHECK2-NEXT: [[TMP12:%.*]] = load i32, i32* [[T_VAR]], align 4
// CHECK2-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC]], i64 0, i64 0
// CHECK2-NEXT: store i32 [[TMP12]], i32* [[ARRAYIDX]], align 4
// CHECK2-NEXT: [[ARRAYIDX3:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i64 0, i64 0
// CHECK2-NEXT: [[TMP13:%.*]] = bitcast %struct.S* [[ARRAYIDX3]] to i8*
// CHECK2-NEXT: [[TMP14:%.*]] = bitcast %struct.S* [[VAR]] to i8*
// CHECK2-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP13]], i8* align 4 [[TMP14]], i64 4, i1 false)
// CHECK2-NEXT: store i32 31, i32* [[SIVAR]], align 4
// CHECK2-NEXT: br label [[DOTOMP_SECTIONS_EXIT]]
// CHECK2: .omp.sections.exit:
// CHECK2-NEXT: br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2: omp.inner.for.inc:
// CHECK2-NEXT: [[TMP15:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK2-NEXT: [[INC:%.*]] = add nsw i32 [[TMP15]], 1
// CHECK2-NEXT: store i32 [[INC]], i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK2-NEXT: br label [[OMP_INNER_FOR_COND]]
// CHECK2: omp.inner.for.end:
// CHECK2-NEXT: call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP0]])
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIfED1Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) [[VAR]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: [[ARRAY_BEGIN4:%.*]] = getelementptr inbounds [2 x %struct.S], [2 x %struct.S]* [[S_ARR]], i32 0, i32 0
// CHECK2-NEXT: [[TMP16:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAY_BEGIN4]], i64 2
// CHECK2-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK2: arraydestroy.body:
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ [[TMP16]], [[OMP_INNER_FOR_END]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIfED1Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 %struct.S* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN4]]
// CHECK2-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE5:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK2: arraydestroy.done5:
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 17:53:11 +08:00
// CHECK2-NEXT: [[CALL:%.*]] = call noundef i32 @_Z5tmainIiET_v()
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 i32 [[CALL]]
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN2StC1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.St* noundef nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.St*, align 8
// CHECK2-NEXT: store %struct.St* [[THIS]], %struct.St** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.St*, %struct.St** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN2StC2Ev(%struct.St* noundef nonnull align 4 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 {{[^@]+}}@_ZN1SIfEC1ERKS0_2St
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], %struct.S* noundef nonnull align 4 dereferenceable(4) [[S:%.*]], %struct.St* noundef [[T:%.*]]) unnamed_addr #[[ATTR1]] 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.S*, align 8
// CHECK2-NEXT: [[S_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK2-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK2-NEXT: store %struct.S* [[S]], %struct.S** [[S_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[TMP0:%.*]] = load %struct.S*, %struct.S** [[S_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIfEC2ERKS0_2St(%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS1]], %struct.S* noundef nonnull align 4 dereferenceable(4) [[TMP0]], %struct.St* noundef [[T]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 {{[^@]+}}@_ZN2StD1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.St* noundef nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.St*, align 8
// CHECK2-NEXT: store %struct.St* [[THIS]], %struct.St** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.St*, %struct.St** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN2StD2Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[THIS1]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 {{[^@]+}}@_Z5tmainIiET_v
// CHECK2-SAME: () #[[ATTR5:[0-9]+]] {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: [[RETVAL:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[TEST:%.*]] = alloca [[STRUCT_S_0:%.*]], align 4
// CHECK2-NEXT: [[T_VAR:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[VEC:%.*]] = alloca [2 x i32], align 4
// CHECK2-NEXT: [[S_ARR:%.*]] = alloca [2 x %struct.S.0], align 4
// CHECK2-NEXT: [[VAR:%.*]] = alloca [[STRUCT_S_0]], align 4
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIiEC1Ev(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[TEST]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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* [[T_VAR]], align 4
// CHECK2-NEXT: [[TMP0:%.*]] = bitcast [2 x i32]* [[VEC]] to i8*
// CHECK2-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP0]], i8* align 4 bitcast ([2 x i32]* @__const._Z5tmainIiET_v.vec to i8*), i64 8, i1 false)
// CHECK2-NEXT: [[ARRAYINIT_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i64 0, i64 0
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIiEC1Ei(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[ARRAYINIT_BEGIN]], i32 noundef 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: [[ARRAYINIT_ELEMENT:%.*]] = getelementptr inbounds [[STRUCT_S_0]], %struct.S.0* [[ARRAYINIT_BEGIN]], i64 1
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIiEC1Ei(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[ARRAYINIT_ELEMENT]], i32 noundef 2)
// CHECK2-NEXT: call void @_ZN1SIiEC1Ei(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[VAR]], i32 noundef 3)
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB2]], i32 4, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*, [2 x i32]*, [2 x %struct.S.0]*, %struct.S.0*)* @.omp_outlined. to void (i32*, i32*, ...)*), i32* [[T_VAR]], [2 x i32]* [[VEC]], [2 x %struct.S.0]* [[S_ARR]], %struct.S.0* [[VAR]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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* [[RETVAL]], align 4
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIiED1Ev(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[VAR]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR]], i32 0, i32 0
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[TMP1:%.*]] = getelementptr inbounds [[STRUCT_S_0]], %struct.S.0* [[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
// CHECK2-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK2: arraydestroy.body:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S.0* [ [[TMP1]], [[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
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], %struct.S.0* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIiED1Ev(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 %struct.S.0* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN]]
// CHECK2-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK2: arraydestroy.done1:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIiED1Ev(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[TEST]]) #[[ATTR2]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[TMP2:%.*]] = load i32, i32* [[RETVAL]], align 4
// CHECK2-NEXT: ret i32 [[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-LABEL: define {{[^@]+}}@_ZN2StC2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.St* noundef nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.St*, align 8
// CHECK2-NEXT: store %struct.St* [[THIS]], %struct.St** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.St*, %struct.St** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.St* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT: store i32 0, i32* [[A]], align 4
// CHECK2-NEXT: [[B:%.*]] = getelementptr inbounds [[STRUCT_ST]], %struct.St* [[THIS1]], i32 0, i32 1
// CHECK2-NEXT: store i32 0, i32* [[B]], align 4
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN1SIfEC2ERKS0_2St
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], %struct.S* noundef nonnull align 4 dereferenceable(4) [[S:%.*]], %struct.St* noundef [[T:%.*]]) unnamed_addr #[[ATTR1]] 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.S*, align 8
// CHECK2-NEXT: [[S_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK2-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK2-NEXT: store %struct.S* [[S]], %struct.S** [[S_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT: [[TMP0:%.*]] = load %struct.S*, %struct.S** [[S_ADDR]], align 8
// CHECK2-NEXT: [[F2:%.*]] = getelementptr inbounds [[STRUCT_S]], %struct.S* [[TMP0]], i32 0, i32 0
// CHECK2-NEXT: [[TMP1:%.*]] = load float, float* [[F2]], align 4
// CHECK2-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.St* [[T]], i32 0, i32 0
// CHECK2-NEXT: [[TMP2:%.*]] = load i32, i32* [[A]], align 4
// CHECK2-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP2]] to float
// CHECK2-NEXT: [[ADD:%.*]] = fadd float [[TMP1]], [[CONV]]
// CHECK2-NEXT: store float [[ADD]], float* [[F]], align 4
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN2StD2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.St* noundef nonnull align 4 dereferenceable(8) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.St*, align 8
// CHECK2-NEXT: store %struct.St* [[THIS]], %struct.St** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.St*, %struct.St** [[THIS_ADDR]], align 8
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN1SIiEC1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK2-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIiEC2Ev(%struct.S.0* noundef 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 {{[^@]+}}@_ZN1SIiEC1Ei
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 noundef [[A:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK2-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: store i32 [[A]], i32* [[A_ADDR]], align 4
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIiEC2Ei(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS1]], i32 noundef [[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 {{[^@]+}}@.omp_outlined.
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 17:53:11 +08:00
// CHECK2-SAME: (i32* noalias noundef [[DOTGLOBAL_TID_:%.*]], i32* noalias noundef [[DOTBOUND_TID_:%.*]], i32* noundef nonnull align 4 dereferenceable(4) [[T_VAR:%.*]], [2 x i32]* noundef nonnull align 4 dereferenceable(8) [[VEC:%.*]], [2 x %struct.S.0]* noundef nonnull align 4 dereferenceable(8) [[S_ARR:%.*]], %struct.S.0* noundef nonnull align 4 dereferenceable(4) [[VAR:%.*]]) #[[ATTR6:[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
// 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 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[T_VAR_ADDR:%.*]] = alloca i32*, align 8
// CHECK2-NEXT: [[VEC_ADDR:%.*]] = alloca [2 x i32]*, align 8
// CHECK2-NEXT: [[S_ARR_ADDR:%.*]] = alloca [2 x %struct.S.0]*, align 8
// CHECK2-NEXT: [[VAR_ADDR:%.*]] = alloca %struct.S.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: [[DOTOMP_SECTIONS_LB_:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTOMP_SECTIONS_UB_:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTOMP_SECTIONS_ST_:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTOMP_SECTIONS_IL_:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[DOTOMP_SECTIONS_IV_:%.*]] = alloca i32, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[T_VAR1:%.*]] = alloca i32, align 4
// CHECK2-NEXT: [[VEC2:%.*]] = alloca [2 x i32], align 4
// CHECK2-NEXT: [[S_ARR3:%.*]] = alloca [2 x %struct.S.0], align 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
// CHECK2-NEXT: [[AGG_TMP:%.*]] = alloca [[STRUCT_ST:%.*]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[VAR5:%.*]] = alloca [[STRUCT_S_0:%.*]], align 4
// CHECK2-NEXT: [[AGG_TMP6:%.*]] = alloca [[STRUCT_ST]], align 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
// 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 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: store i32* [[T_VAR]], i32** [[T_VAR_ADDR]], align 8
// CHECK2-NEXT: store [2 x i32]* [[VEC]], [2 x i32]** [[VEC_ADDR]], align 8
// CHECK2-NEXT: store [2 x %struct.S.0]* [[S_ARR]], [2 x %struct.S.0]** [[S_ARR_ADDR]], align 8
// CHECK2-NEXT: store %struct.S.0* [[VAR]], %struct.S.0** [[VAR_ADDR]], align 8
// CHECK2-NEXT: [[TMP0:%.*]] = load i32*, i32** [[T_VAR_ADDR]], align 8
// CHECK2-NEXT: [[TMP1:%.*]] = load [2 x i32]*, [2 x i32]** [[VEC_ADDR]], align 8
// CHECK2-NEXT: [[TMP2:%.*]] = load [2 x %struct.S.0]*, [2 x %struct.S.0]** [[S_ARR_ADDR]], align 8
// CHECK2-NEXT: [[TMP3:%.*]] = load %struct.S.0*, %struct.S.0** [[VAR_ADDR]], 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_SECTIONS_LB_]], align 4
// CHECK2-NEXT: store i32 1, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK2-NEXT: store i32 1, i32* [[DOTOMP_SECTIONS_ST_]], align 4
// CHECK2-NEXT: store i32 0, i32* [[DOTOMP_SECTIONS_IL_]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK2-NEXT: store i32 [[TMP4]], i32* [[T_VAR1]], align 4
// CHECK2-NEXT: [[TMP5:%.*]] = bitcast [2 x i32]* [[VEC2]] to i8*
// CHECK2-NEXT: [[TMP6:%.*]] = bitcast [2 x i32]* [[TMP1]] to i8*
// CHECK2-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP5]], i8* align 4 [[TMP6]], i64 8, i1 false)
// CHECK2-NEXT: [[ARRAY_BEGIN:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR3]], i32 0, i32 0
// CHECK2-NEXT: [[TMP7:%.*]] = bitcast [2 x %struct.S.0]* [[TMP2]] to %struct.S.0*
// CHECK2-NEXT: [[TMP8:%.*]] = getelementptr [[STRUCT_S_0]], %struct.S.0* [[ARRAY_BEGIN]], i64 2
// CHECK2-NEXT: [[OMP_ARRAYCPY_ISEMPTY:%.*]] = icmp eq %struct.S.0* [[ARRAY_BEGIN]], [[TMP8]]
// CHECK2-NEXT: br i1 [[OMP_ARRAYCPY_ISEMPTY]], label [[OMP_ARRAYCPY_DONE4:%.*]], label [[OMP_ARRAYCPY_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
// CHECK2: omp.arraycpy.body:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[OMP_ARRAYCPY_SRCELEMENTPAST:%.*]] = phi %struct.S.0* [ [[TMP7]], [[ENTRY:%.*]] ], [ [[OMP_ARRAYCPY_SRC_ELEMENT:%.*]], [[OMP_ARRAYCPY_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
// CHECK2-NEXT: [[OMP_ARRAYCPY_DESTELEMENTPAST:%.*]] = phi %struct.S.0* [ [[ARRAY_BEGIN]], [[ENTRY]] ], [ [[OMP_ARRAYCPY_DEST_ELEMENT:%.*]], [[OMP_ARRAYCPY_BODY]] ]
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN2StC1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP]])
// CHECK2-NEXT: call void @_ZN1SIiEC1ERKS0_2St(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_DESTELEMENTPAST]], %struct.S.0* noundef nonnull align 4 dereferenceable(4) [[OMP_ARRAYCPY_SRCELEMENTPAST]], %struct.St* noundef [[AGG_TMP]])
// CHECK2-NEXT: call void @_ZN2StD1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 [[STRUCT_S_0]], %struct.S.0* [[OMP_ARRAYCPY_DESTELEMENTPAST]], i32 1
// CHECK2-NEXT: [[OMP_ARRAYCPY_SRC_ELEMENT]] = getelementptr [[STRUCT_S_0]], %struct.S.0* [[OMP_ARRAYCPY_SRCELEMENTPAST]], i32 1
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[OMP_ARRAYCPY_DONE:%.*]] = icmp eq %struct.S.0* [[OMP_ARRAYCPY_DEST_ELEMENT]], [[TMP8]]
// CHECK2-NEXT: br i1 [[OMP_ARRAYCPY_DONE]], label [[OMP_ARRAYCPY_DONE4]], label [[OMP_ARRAYCPY_BODY]]
// CHECK2: omp.arraycpy.done4:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN2StC1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP6]])
// CHECK2-NEXT: call void @_ZN1SIiEC1ERKS0_2St(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[VAR5]], %struct.S.0* noundef nonnull align 4 dereferenceable(4) [[TMP3]], %struct.St* noundef [[AGG_TMP6]])
// CHECK2-NEXT: call void @_ZN2StD1Ev(%struct.St* noundef nonnull align 4 dereferenceable(8) [[AGG_TMP6]]) #[[ATTR2]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[TMP9:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT: [[TMP10:%.*]] = load i32, i32* [[TMP9]], align 4
// CHECK2-NEXT: call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1]], i32 [[TMP10]], i32 34, i32* [[DOTOMP_SECTIONS_IL_]], i32* [[DOTOMP_SECTIONS_LB_]], i32* [[DOTOMP_SECTIONS_UB_]], i32* [[DOTOMP_SECTIONS_ST_]], i32 1, i32 1)
// CHECK2-NEXT: [[TMP11:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK2-NEXT: [[TMP12:%.*]] = icmp slt i32 [[TMP11]], 1
// CHECK2-NEXT: [[TMP13:%.*]] = select i1 [[TMP12]], i32 [[TMP11]], i32 1
// CHECK2-NEXT: store i32 [[TMP13]], i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK2-NEXT: [[TMP14:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_LB_]], align 4
// CHECK2-NEXT: store i32 [[TMP14]], i32* [[DOTOMP_SECTIONS_IV_]], align 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
// CHECK2-NEXT: br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK2: omp.inner.for.cond:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[TMP15:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK2-NEXT: [[TMP16:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK2-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP15]], [[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: br i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK2: omp.inner.for.body:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[TMP17:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK2-NEXT: switch i32 [[TMP17]], label [[DOTOMP_SECTIONS_EXIT:%.*]] [
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: i32 0, label [[DOTOMP_SECTIONS_CASE:%.*]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: i32 1, label [[DOTOMP_SECTIONS_CASE7:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: ]
// CHECK2: .omp.sections.case:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[TMP18:%.*]] = load i32, i32* [[T_VAR1]], align 4
// CHECK2-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [2 x i32], [2 x i32]* [[VEC2]], i64 0, i64 0
// CHECK2-NEXT: store i32 [[TMP18]], i32* [[ARRAYIDX]], align 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
// CHECK2-NEXT: br label [[DOTOMP_SECTIONS_EXIT]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2: .omp.sections.case7:
// CHECK2-NEXT: [[ARRAYIDX8:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR3]], i64 0, i64 0
// CHECK2-NEXT: [[TMP19:%.*]] = bitcast %struct.S.0* [[ARRAYIDX8]] to i8*
// CHECK2-NEXT: [[TMP20:%.*]] = bitcast %struct.S.0* [[VAR5]] to i8*
// CHECK2-NEXT: call void @llvm.memcpy.p0i8.p0i8.i64(i8* align 4 [[TMP19]], i8* align 4 [[TMP20]], i64 4, i1 false)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 [[DOTOMP_SECTIONS_EXIT]]
// CHECK2: .omp.sections.exit:
// CHECK2-NEXT: br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK2: omp.inner.for.inc:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[TMP21:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK2-NEXT: [[INC:%.*]] = add nsw i32 [[TMP21]], 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: store i32 [[INC]], i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK2-NEXT: br label [[OMP_INNER_FOR_COND]]
// CHECK2: omp.inner.for.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[TMP22:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT: [[TMP23:%.*]] = load i32, i32* [[TMP22]], align 4
// CHECK2-NEXT: call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP23]])
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIiED1Ev(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[VAR5]]) #[[ATTR2]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[ARRAY_BEGIN9:%.*]] = getelementptr inbounds [2 x %struct.S.0], [2 x %struct.S.0]* [[S_ARR3]], i32 0, i32 0
// CHECK2-NEXT: [[TMP24:%.*]] = getelementptr inbounds [[STRUCT_S_0]], %struct.S.0* [[ARRAY_BEGIN9]], 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
// CHECK2-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK2: arraydestroy.body:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S.0* [ [[TMP24]], [[OMP_INNER_FOR_END]] ], [ [[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
// CHECK2-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S_0]], %struct.S.0* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIiED1Ev(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK2-NEXT: [[ARRAYDESTROY_DONE:%.*]] = icmp eq %struct.S.0* [[ARRAYDESTROY_ELEMENT]], [[ARRAY_BEGIN9]]
// CHECK2-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE10:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK2: arraydestroy.done10:
// CHECK2-NEXT: [[TMP25:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK2-NEXT: [[TMP26:%.*]] = load i32, i32* [[TMP25]], align 4
// CHECK2-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB3:[0-9]+]], i32 [[TMP26]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 {{[^@]+}}@_ZN1SIiEC1ERKS0_2St
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], %struct.S.0* noundef nonnull align 4 dereferenceable(4) [[S:%.*]], %struct.St* noundef [[T:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK2-NEXT: [[S_ADDR:%.*]] = alloca %struct.S.0*, align 8
// CHECK2-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: store %struct.S.0* [[S]], %struct.S.0** [[S_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[TMP0:%.*]] = load %struct.S.0*, %struct.S.0** [[S_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIiEC2ERKS0_2St(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS1]], %struct.S.0* noundef nonnull align 4 dereferenceable(4) [[TMP0]], %struct.St* noundef [[T]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 {{[^@]+}}@_ZN1SIiED1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK2-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-NEXT: call void @_ZN1SIiED2Ev(%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 {{[^@]+}}@_ZN1SIiEC2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK2-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], %struct.S.0* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT: [[TMP0:%.*]] = load volatile i32, i32* @g, align 4
// CHECK2-NEXT: store i32 [[TMP0]], i32* [[F]], align 4
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN1SIiEC2Ei
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], i32 noundef [[A:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK2-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK2-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: store i32 [[A]], i32* [[A_ADDR]], align 4
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], %struct.S.0* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT: [[TMP0:%.*]] = load i32, i32* [[A_ADDR]], align 4
// CHECK2-NEXT: [[TMP1:%.*]] = load volatile i32, i32* @g, align 4
// CHECK2-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP0]], [[TMP1]]
// CHECK2-NEXT: store i32 [[ADD]], i32* [[F]], align 4
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN1SIiEC2ERKS0_2St
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], %struct.S.0* noundef nonnull align 4 dereferenceable(4) [[S:%.*]], %struct.St* noundef [[T:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK2-NEXT: [[S_ADDR:%.*]] = alloca %struct.S.0*, align 8
// CHECK2-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: store %struct.S.0* [[S]], %struct.S.0** [[S_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S_0:%.*]], %struct.S.0* [[THIS1]], i32 0, i32 0
// CHECK2-NEXT: [[TMP0:%.*]] = load %struct.S.0*, %struct.S.0** [[S_ADDR]], align 8
// CHECK2-NEXT: [[F2:%.*]] = getelementptr inbounds [[STRUCT_S_0]], %struct.S.0* [[TMP0]], i32 0, i32 0
// CHECK2-NEXT: [[TMP1:%.*]] = load i32, i32* [[F2]], align 4
// CHECK2-NEXT: [[A:%.*]] = getelementptr inbounds [[STRUCT_ST:%.*]], %struct.St* [[T]], i32 0, i32 0
// CHECK2-NEXT: [[TMP2:%.*]] = load i32, i32* [[A]], align 4
// CHECK2-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], [[TMP2]]
// CHECK2-NEXT: store i32 [[ADD]], i32* [[F]], align 4
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_ZN1SIiED2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK2-SAME: (%struct.S.0* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.S.0*, align 8
// CHECK2-NEXT: store %struct.S.0* [[THIS]], %struct.S.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: [[THIS1:%.*]] = load %struct.S.0*, %struct.S.0** [[THIS_ADDR]], align 8
// CHECK2-NEXT: ret void
//
//
// CHECK2-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_sections_firstprivate_codegen.cpp
// CHECK2-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK2-NEXT: entry:
// CHECK2-NEXT: call void @__cxx_global_var_init()
// CHECK2-NEXT: call void @__cxx_global_var_init.1()
// CHECK2-NEXT: call void @__cxx_global_var_init.2()
// CHECK2-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@__cxx_global_var_init
// CHECK3-SAME: () #[[ATTR0:[0-9]+]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK3-NEXT: entry:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK3-NEXT: call void @_ZN1SIfEC1Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) @test)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S*)* @_ZN1SIfED1Ev to void (i8*)*), i8* bitcast (%struct.S* @test to i8*), i8* @__dso_handle) #[[ATTR2:[0-9]+]]
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK3-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] 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.S*, align 8
// CHECK3-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK3-NEXT: call void @_ZN1SIfEC2Ev(%struct.S* noundef 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 {{[^@]+}}@_ZN1SIfED1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK3-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.S*, align 8
// CHECK3-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK3-NEXT: call void @_ZN1SIfED2Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 {{[^@]+}}@_ZN1SIfEC2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK3-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
[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: entry:
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK3-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[TMP0:%.*]] = load volatile i32, i32* @g, align 4
// CHECK3-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP0]] to float
// CHECK3-NEXT: store float [[CONV]], float* [[F]], align 4
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfED2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK3-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
[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: entry:
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK3-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// 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 {{[^@]+}}@__cxx_global_var_init.1
// CHECK3-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK3-NEXT: entry:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK3-NEXT: call void @_ZN1SIfEC1Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 0, i64 0), float noundef 1.000000e+00)
// CHECK3-NEXT: call void @_ZN1SIfEC1Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 0, i64 1), float noundef 2.000000e+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
// CHECK3-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR2]]
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ef
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK3-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], float noundef [[A:%.*]]) unnamed_addr #[[ATTR1]] 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.S*, align 8
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca float, align 4
// CHECK3-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK3-NEXT: store float [[A]], float* [[A_ADDR]], align 4
// CHECK3-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = load float, float* [[A_ADDR]], align 4
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK3-NEXT: call void @_ZN1SIfEC2Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS1]], float noundef [[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 {{[^@]+}}@__cxx_global_array_dtor
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 17:53:11 +08:00
// CHECK3-SAME: (i8* noundef [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK3: arraydestroy.body:
[ConstantFold] Remove unnecessary bounded index restriction The fold for merging a GEP of GEP into a single GEP currently bails if doing so would result in notional overindexing. The justification given in the comment above this check is dangerously incorrect: GEPs with notional overindexing are perfectly fine, and if some code treats them incorrectly, then that code is broken, not the GEP. Such a GEP might legally appear in source IR, so only preventing its creation cannot be sufficient. (The constant folder also ends up canonicalizing the GEP to remove the notional overindexing, but that's neither here nor there.) This check dates back to https://github.com/llvm/llvm-project/commit/bd4fef4a8939db18f39b108e19097b25e2c7c47a, and as far as I can tell the original issue this was trying to patch around has since been resolved. Differential Revision: https://reviews.llvm.org/D116587
2022-01-04 18:53:28 +08:00
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK3-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK3-NEXT: call void @_ZN1SIfED1Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
// 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 {{[^@]+}}@_ZN1SIfEC2Ef
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK3-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], float noundef [[A:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
[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: entry:
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca float, align 4
// CHECK3-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK3-NEXT: store float [[A]], float* [[A_ADDR]], align 4
// CHECK3-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK3-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[THIS1]], i32 0, i32 0
// CHECK3-NEXT: [[TMP0:%.*]] = load float, float* [[A_ADDR]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = load volatile i32, i32* @g, align 4
// CHECK3-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP1]] to float
// CHECK3-NEXT: [[ADD:%.*]] = fadd float [[TMP0]], [[CONV]]
// CHECK3-NEXT: store float [[ADD]], float* [[F]], align 4
// 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 {{[^@]+}}@__cxx_global_var_init.2
// CHECK3-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK3-NEXT: entry:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK3-NEXT: call void @_ZN1SIfEC1Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) @var, float noundef 3.000000e+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
// CHECK3-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S*)* @_ZN1SIfED1Ev to void (i8*)*), i8* bitcast (%struct.S* @var to i8*), i8* @__dso_handle) #[[ATTR2]]
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@main
// CHECK3-SAME: () #[[ATTR3:[0-9]+]] {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[RETVAL:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON:%.*]], align 8
// CHECK3-NEXT: store i32 0, i32* [[RETVAL]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = getelementptr inbounds [[CLASS_ANON]], %class.anon* [[REF_TMP]], i32 0, i32 0
// CHECK3-NEXT: store i32* @_ZZ4mainE5sivar, i32** [[TMP0]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK3-NEXT: call void @"_ZZ4mainENK3$_0clEv"(%class.anon* noundef nonnull align 8 dereferenceable(8) [[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 i32 0
//
//
// CHECK3-LABEL: define {{[^@]+}}@.omp_outlined.
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 17:53:11 +08:00
// CHECK3-SAME: (i32* noalias noundef [[DOTGLOBAL_TID_:%.*]], i32* noalias noundef [[DOTBOUND_TID_:%.*]], i32* noundef nonnull align 4 dereferenceable(4) [[SIVAR:%.*]]) #[[ATTR5:[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: [[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 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK3-NEXT: [[SIVAR_ADDR:%.*]] = 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_SECTIONS_LB_:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[DOTOMP_SECTIONS_UB_:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[DOTOMP_SECTIONS_ST_:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[DOTOMP_SECTIONS_IL_:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[DOTOMP_SECTIONS_IV_:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[G:%.*]] = alloca i32, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK3-NEXT: [[SIVAR1:%.*]] = alloca i32, align 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
// CHECK3-NEXT: [[REF_TMP:%.*]] = alloca [[CLASS_ANON_0:%.*]], 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 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK3-NEXT: store i32* [[SIVAR]], i32** [[SIVAR_ADDR]], align 8
// CHECK3-NEXT: [[TMP0:%.*]] = load i32*, i32** [[SIVAR_ADDR]], 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_SECTIONS_LB_]], align 4
// CHECK3-NEXT: store i32 1, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK3-NEXT: store i32 1, i32* [[DOTOMP_SECTIONS_ST_]], align 4
// CHECK3-NEXT: store i32 0, i32* [[DOTOMP_SECTIONS_IL_]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK3-NEXT: [[TMP1:%.*]] = load volatile i32, i32* @g, align 4
// CHECK3-NEXT: store i32 [[TMP1]], i32* [[G]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK3-NEXT: store i32 [[TMP2]], i32* [[SIVAR1]], align 4
// CHECK3-NEXT: [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK3-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK3-NEXT: call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP4]], i32 34, i32* [[DOTOMP_SECTIONS_IL_]], i32* [[DOTOMP_SECTIONS_LB_]], i32* [[DOTOMP_SECTIONS_UB_]], i32* [[DOTOMP_SECTIONS_ST_]], i32 1, i32 1)
// CHECK3-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = icmp slt i32 [[TMP5]], 1
// CHECK3-NEXT: [[TMP7:%.*]] = select i1 [[TMP6]], i32 [[TMP5]], i32 1
// CHECK3-NEXT: store i32 [[TMP7]], i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_LB_]], align 4
// CHECK3-NEXT: store i32 [[TMP8]], i32* [[DOTOMP_SECTIONS_IV_]], align 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
// CHECK3-NEXT: br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK3: omp.inner.for.cond:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK3-NEXT: [[TMP9:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK3-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK3-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK3: omp.inner.for.body:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK3-NEXT: [[TMP11:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK3-NEXT: switch i32 [[TMP11]], label [[DOTOMP_SECTIONS_EXIT:%.*]] [
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: i32 0, label [[DOTOMP_SECTIONS_CASE:%.*]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK3-NEXT: i32 1, label [[DOTOMP_SECTIONS_CASE2:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: ]
// CHECK3: .omp.sections.case:
// CHECK3-NEXT: store i32 1, i32* [[G]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK3-NEXT: store i32 10, i32* [[SIVAR1]], align 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
// CHECK3-NEXT: br label [[DOTOMP_SECTIONS_EXIT]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK3: .omp.sections.case2:
// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 0
// CHECK3-NEXT: store i32* [[G]], i32** [[TMP12]], align 8
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [[CLASS_ANON_0]], %class.anon.0* [[REF_TMP]], i32 0, i32 1
// CHECK3-NEXT: store i32* [[SIVAR1]], i32** [[TMP13]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK3-NEXT: call void @"_ZZZ4mainENK3$_0clEvENKUlvE_clEv"(%class.anon.0* noundef 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: br label [[DOTOMP_SECTIONS_EXIT]]
// CHECK3: .omp.sections.exit:
// CHECK3-NEXT: br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK3: omp.inner.for.inc:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK3-NEXT: [[TMP14:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK3-NEXT: [[INC:%.*]] = add nsw i32 [[TMP14]], 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: store i32 [[INC]], i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK3-NEXT: br label [[OMP_INNER_FOR_COND]]
// CHECK3: omp.inner.for.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK3-NEXT: call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK3-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], 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: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_sections_firstprivate_codegen.cpp
// CHECK3-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK3-NEXT: entry:
// CHECK3-NEXT: call void @__cxx_global_var_init()
// CHECK3-NEXT: call void @__cxx_global_var_init.1()
// CHECK3-NEXT: call void @__cxx_global_var_init.2()
// CHECK3-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@__cxx_global_var_init
// CHECK4-SAME: () #[[ATTR0:[0-9]+]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK4-NEXT: entry:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK4-NEXT: call void @_ZN1SIfEC1Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) @test)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S*)* @_ZN1SIfED1Ev to void (i8*)*), i8* bitcast (%struct.S* @test to i8*), i8* @__dso_handle) #[[ATTR2:[0-9]+]]
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK4-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1:[0-9]+]] 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.S*, align 8
// CHECK4-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK4-NEXT: call void @_ZN1SIfEC2Ev(%struct.S* noundef 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 {{[^@]+}}@_ZN1SIfED1Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK4-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] 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.S*, align 8
// CHECK4-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK4-NEXT: call void @_ZN1SIfED2Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS1]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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
//
//
[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
// CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK4-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
[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
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK4-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[THIS1]], i32 0, i32 0
// CHECK4-NEXT: [[TMP0:%.*]] = load volatile i32, i32* @g, align 4
// CHECK4-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP0]] to float
// CHECK4-NEXT: store float [[CONV]], float* [[F]], align 4
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfED2Ev
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK4-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
[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
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK4-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK4-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
// CHECK4-LABEL: define {{[^@]+}}@__cxx_global_var_init.1
// CHECK4-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK4-NEXT: entry:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK4-NEXT: call void @_ZN1SIfEC1Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 0, i64 0), float noundef 1.000000e+00)
// CHECK4-NEXT: call void @_ZN1SIfEC1Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 0, i64 1), float noundef 2.000000e+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
// CHECK4-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* @__cxx_global_array_dtor, i8* null, i8* @__dso_handle) #[[ATTR2]]
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC1Ef
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK4-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], float noundef [[A:%.*]]) unnamed_addr #[[ATTR1]] 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.S*, align 8
// CHECK4-NEXT: [[A_ADDR:%.*]] = alloca float, align 4
// CHECK4-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK4-NEXT: store float [[A]], float* [[A_ADDR]], align 4
// CHECK4-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[TMP0:%.*]] = load float, float* [[A_ADDR]], align 4
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK4-NEXT: call void @_ZN1SIfEC2Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS1]], float noundef [[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 {{[^@]+}}@__cxx_global_array_dtor
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 17:53:11 +08:00
// CHECK4-SAME: (i8* noundef [[TMP0:%.*]]) #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: [[DOTADDR:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: store i8* [[TMP0]], i8** [[DOTADDR]], align 8
// CHECK4-NEXT: br label [[ARRAYDESTROY_BODY:%.*]]
// CHECK4: arraydestroy.body:
[ConstantFold] Remove unnecessary bounded index restriction The fold for merging a GEP of GEP into a single GEP currently bails if doing so would result in notional overindexing. The justification given in the comment above this check is dangerously incorrect: GEPs with notional overindexing are perfectly fine, and if some code treats them incorrectly, then that code is broken, not the GEP. Such a GEP might legally appear in source IR, so only preventing its creation cannot be sufficient. (The constant folder also ends up canonicalizing the GEP to remove the notional overindexing, but that's neither here nor there.) This check dates back to https://github.com/llvm/llvm-project/commit/bd4fef4a8939db18f39b108e19097b25e2c7c47a, and as far as I can tell the original issue this was trying to patch around has since been resolved. Differential Revision: https://reviews.llvm.org/D116587
2022-01-04 18:53:28 +08:00
// CHECK4-NEXT: [[ARRAYDESTROY_ELEMENTPAST:%.*]] = phi %struct.S* [ getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i64 1, i64 0), [[ENTRY:%.*]] ], [ [[ARRAYDESTROY_ELEMENT:%.*]], [[ARRAYDESTROY_BODY]] ]
// CHECK4-NEXT: [[ARRAYDESTROY_ELEMENT]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[ARRAYDESTROY_ELEMENTPAST]], i64 -1
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK4-NEXT: call void @_ZN1SIfED1Ev(%struct.S* noundef nonnull align 4 dereferenceable(4) [[ARRAYDESTROY_ELEMENT]]) #[[ATTR2]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 %struct.S* [[ARRAYDESTROY_ELEMENT]], getelementptr inbounds ([2 x %struct.S], [2 x %struct.S]* @s_arr, i32 0, i32 0)
// CHECK4-NEXT: br i1 [[ARRAYDESTROY_DONE]], label [[ARRAYDESTROY_DONE1:%.*]], label [[ARRAYDESTROY_BODY]]
// CHECK4: arraydestroy.done1:
// 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
// CHECK4-LABEL: define {{[^@]+}}@_ZN1SIfEC2Ef
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK4-SAME: (%struct.S* noundef nonnull align 4 dereferenceable(4) [[THIS:%.*]], float noundef [[A:%.*]]) unnamed_addr #[[ATTR1]] align 2 {
[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
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[THIS_ADDR:%.*]] = alloca %struct.S*, align 8
// CHECK4-NEXT: [[A_ADDR:%.*]] = alloca float, align 4
// CHECK4-NEXT: store %struct.S* [[THIS]], %struct.S** [[THIS_ADDR]], align 8
// CHECK4-NEXT: store float [[A]], float* [[A_ADDR]], align 4
// CHECK4-NEXT: [[THIS1:%.*]] = load %struct.S*, %struct.S** [[THIS_ADDR]], align 8
// CHECK4-NEXT: [[F:%.*]] = getelementptr inbounds [[STRUCT_S:%.*]], %struct.S* [[THIS1]], i32 0, i32 0
// CHECK4-NEXT: [[TMP0:%.*]] = load float, float* [[A_ADDR]], align 4
// CHECK4-NEXT: [[TMP1:%.*]] = load volatile i32, i32* @g, align 4
// CHECK4-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP1]] to float
// CHECK4-NEXT: [[ADD:%.*]] = fadd float [[TMP0]], [[CONV]]
// CHECK4-NEXT: store float [[ADD]], float* [[F]], align 4
// CHECK4-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
// CHECK4-LABEL: define {{[^@]+}}@__cxx_global_var_init.2
// CHECK4-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK4-NEXT: entry:
[IRGen] Do not overwrite existing attributes in CGCall. When adding new attributes, existing attributes are dropped. While this appears to be a longstanding issue, this was highlighted by D105169 which dropped a lot of attributes due to adding the new noundef attribute. Ahmed Bougacha (@ab) tracked down the issue and provided the fix in CGCall.cpp. I bundled it up and updated the tests.
2022-01-20 21:45:19 +08:00
// CHECK4-NEXT: call void @_ZN1SIfEC1Ef(%struct.S* noundef nonnull align 4 dereferenceable(4) @var, float noundef 3.000000e+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
// CHECK4-NEXT: [[TMP0:%.*]] = call i32 @__cxa_atexit(void (i8*)* bitcast (void (%struct.S*)* @_ZN1SIfED1Ev to void (i8*)*), i8* bitcast (%struct.S* @var to i8*), i8* @__dso_handle) #[[ATTR2]]
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@main
// CHECK4-SAME: () #[[ATTR3:[0-9]+]] {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: [[RETVAL:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[BLOCK:%.*]] = alloca <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>, align 8
// CHECK4-NEXT: store i32 0, i32* [[RETVAL]], align 4
// CHECK4-NEXT: [[BLOCK_ISA:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>* [[BLOCK]], i32 0, i32 0
// CHECK4-NEXT: store i8* bitcast (i8** @_NSConcreteStackBlock to i8*), i8** [[BLOCK_ISA]], align 8
// CHECK4-NEXT: [[BLOCK_FLAGS:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>* [[BLOCK]], i32 0, i32 1
// CHECK4-NEXT: store i32 1073741824, i32* [[BLOCK_FLAGS]], align 8
// CHECK4-NEXT: [[BLOCK_RESERVED:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>* [[BLOCK]], i32 0, i32 2
// CHECK4-NEXT: store i32 0, i32* [[BLOCK_RESERVED]], align 4
// CHECK4-NEXT: [[BLOCK_INVOKE:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>* [[BLOCK]], i32 0, i32 3
// CHECK4-NEXT: store i8* bitcast (void (i8*)* @__main_block_invoke to i8*), i8** [[BLOCK_INVOKE]], align 8
// CHECK4-NEXT: [[BLOCK_DESCRIPTOR:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>* [[BLOCK]], i32 0, i32 4
// CHECK4-NEXT: store %struct.__block_descriptor* bitcast ({ i64, i64, i8*, i8* }* @__block_descriptor_tmp.3 to %struct.__block_descriptor*), %struct.__block_descriptor** [[BLOCK_DESCRIPTOR]], align 8
// CHECK4-NEXT: [[BLOCK_CAPTURED:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>* [[BLOCK]], i32 0, i32 5
// CHECK4-NEXT: [[TMP0:%.*]] = load i32, i32* @_ZZ4mainE5sivar, align 4
// CHECK4-NEXT: store i32 [[TMP0]], i32* [[BLOCK_CAPTURED]], align 8
// CHECK4-NEXT: [[TMP1:%.*]] = bitcast <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>* [[BLOCK]] to void ()*
// CHECK4-NEXT: [[BLOCK_LITERAL:%.*]] = bitcast void ()* [[TMP1]] to %struct.__block_literal_generic*
// CHECK4-NEXT: [[TMP2:%.*]] = getelementptr inbounds [[STRUCT___BLOCK_LITERAL_GENERIC:%.*]], %struct.__block_literal_generic* [[BLOCK_LITERAL]], i32 0, i32 3
// CHECK4-NEXT: [[TMP3:%.*]] = bitcast %struct.__block_literal_generic* [[BLOCK_LITERAL]] to i8*
// CHECK4-NEXT: [[TMP4:%.*]] = load i8*, i8** [[TMP2]], align 8
// CHECK4-NEXT: [[TMP5:%.*]] = bitcast i8* [[TMP4]] to void (i8*)*
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 17:53:11 +08:00
// CHECK4-NEXT: call void [[TMP5]](i8* noundef [[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: ret i32 0
//
//
// CHECK4-LABEL: define {{[^@]+}}@__main_block_invoke
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 17:53:11 +08:00
// CHECK4-SAME: (i8* noundef [[DOTBLOCK_DESCRIPTOR:%.*]]) #[[ATTR1]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: [[DOTBLOCK_DESCRIPTOR_ADDR:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: [[BLOCK_ADDR:%.*]] = alloca <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>*, align 8
// CHECK4-NEXT: store i8* [[DOTBLOCK_DESCRIPTOR]], i8** [[DOTBLOCK_DESCRIPTOR_ADDR]], align 8
// CHECK4-NEXT: [[BLOCK:%.*]] = bitcast i8* [[DOTBLOCK_DESCRIPTOR]] to <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>*
// CHECK4-NEXT: store <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>* [[BLOCK]], <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32 }>** [[BLOCK_ADDR]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK4-NEXT: call void (%struct.ident_t*, i32, void (i32*, i32*, ...)*, ...) @__kmpc_fork_call(%struct.ident_t* @[[GLOB3:[0-9]+]], i32 1, void (i32*, i32*, ...)* bitcast (void (i32*, i32*, i32*)* @.omp_outlined. to void (i32*, i32*, ...)*), i32* @_ZZ4mainE5sivar)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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.
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 17:53:11 +08:00
// CHECK4-SAME: (i32* noalias noundef [[DOTGLOBAL_TID_:%.*]], i32* noalias noundef [[DOTBOUND_TID_:%.*]], i32* noundef nonnull align 4 dereferenceable(4) [[SIVAR:%.*]]) #[[ATTR4:[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
// 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 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK4-NEXT: [[SIVAR_ADDR:%.*]] = 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_SECTIONS_LB_:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[DOTOMP_SECTIONS_UB_:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[DOTOMP_SECTIONS_ST_:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[DOTOMP_SECTIONS_IL_:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[DOTOMP_SECTIONS_IV_:%.*]] = alloca i32, align 4
// CHECK4-NEXT: [[G:%.*]] = alloca i32, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK4-NEXT: [[SIVAR1:%.*]] = alloca i32, align 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
// CHECK4-NEXT: [[BLOCK:%.*]] = alloca <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>, 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 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK4-NEXT: store i32* [[SIVAR]], i32** [[SIVAR_ADDR]], align 8
// CHECK4-NEXT: [[TMP0:%.*]] = load i32*, i32** [[SIVAR_ADDR]], 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_SECTIONS_LB_]], align 4
// CHECK4-NEXT: store i32 1, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK4-NEXT: store i32 1, i32* [[DOTOMP_SECTIONS_ST_]], align 4
// CHECK4-NEXT: store i32 0, i32* [[DOTOMP_SECTIONS_IL_]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK4-NEXT: [[TMP1:%.*]] = load volatile i32, i32* @g, align 4
// CHECK4-NEXT: store i32 [[TMP1]], i32* [[G]], align 4
// CHECK4-NEXT: [[TMP2:%.*]] = load i32, i32* [[TMP0]], align 4
// CHECK4-NEXT: store i32 [[TMP2]], i32* [[SIVAR1]], align 4
// CHECK4-NEXT: [[TMP3:%.*]] = load i32*, i32** [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK4-NEXT: [[TMP4:%.*]] = load i32, i32* [[TMP3]], align 4
// CHECK4-NEXT: call void @__kmpc_for_static_init_4(%struct.ident_t* @[[GLOB1:[0-9]+]], i32 [[TMP4]], i32 34, i32* [[DOTOMP_SECTIONS_IL_]], i32* [[DOTOMP_SECTIONS_LB_]], i32* [[DOTOMP_SECTIONS_UB_]], i32* [[DOTOMP_SECTIONS_ST_]], i32 1, i32 1)
// CHECK4-NEXT: [[TMP5:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK4-NEXT: [[TMP6:%.*]] = icmp slt i32 [[TMP5]], 1
// CHECK4-NEXT: [[TMP7:%.*]] = select i1 [[TMP6]], i32 [[TMP5]], i32 1
// CHECK4-NEXT: store i32 [[TMP7]], i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK4-NEXT: [[TMP8:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_LB_]], align 4
// CHECK4-NEXT: store i32 [[TMP8]], i32* [[DOTOMP_SECTIONS_IV_]], align 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
// CHECK4-NEXT: br label [[OMP_INNER_FOR_COND:%.*]]
// CHECK4: omp.inner.for.cond:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK4-NEXT: [[TMP9:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK4-NEXT: [[TMP10:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_UB_]], align 4
// CHECK4-NEXT: [[CMP:%.*]] = icmp sle i32 [[TMP9]], [[TMP10]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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 i1 [[CMP]], label [[OMP_INNER_FOR_BODY:%.*]], label [[OMP_INNER_FOR_END:%.*]]
// CHECK4: omp.inner.for.body:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK4-NEXT: [[TMP11:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK4-NEXT: switch i32 [[TMP11]], label [[DOTOMP_SECTIONS_EXIT:%.*]] [
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: i32 0, label [[DOTOMP_SECTIONS_CASE:%.*]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK4-NEXT: i32 1, label [[DOTOMP_SECTIONS_CASE2:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: ]
// CHECK4: .omp.sections.case:
// CHECK4-NEXT: store i32 1, i32* [[G]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK4-NEXT: store i32 10, i32* [[SIVAR1]], align 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
// CHECK4-NEXT: br label [[DOTOMP_SECTIONS_EXIT]]
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK4: .omp.sections.case2:
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: [[BLOCK_ISA:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>* [[BLOCK]], i32 0, i32 0
// CHECK4-NEXT: store i8* bitcast (i8** @_NSConcreteStackBlock to i8*), i8** [[BLOCK_ISA]], align 8
// CHECK4-NEXT: [[BLOCK_FLAGS:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>* [[BLOCK]], i32 0, i32 1
// CHECK4-NEXT: store i32 1073741824, i32* [[BLOCK_FLAGS]], align 8
// CHECK4-NEXT: [[BLOCK_RESERVED:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>* [[BLOCK]], i32 0, i32 2
// CHECK4-NEXT: store i32 0, i32* [[BLOCK_RESERVED]], align 4
// CHECK4-NEXT: [[BLOCK_INVOKE:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>* [[BLOCK]], i32 0, i32 3
// CHECK4-NEXT: store i8* bitcast (void (i8*)* @var_block_invoke to i8*), i8** [[BLOCK_INVOKE]], align 8
// CHECK4-NEXT: [[BLOCK_DESCRIPTOR:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>* [[BLOCK]], i32 0, i32 4
// CHECK4-NEXT: store %struct.__block_descriptor* bitcast ({ i64, i64, i8*, i8* }* @__block_descriptor_tmp to %struct.__block_descriptor*), %struct.__block_descriptor** [[BLOCK_DESCRIPTOR]], align 8
// CHECK4-NEXT: [[BLOCK_CAPTURED:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>* [[BLOCK]], i32 0, i32 5
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK4-NEXT: [[TMP12:%.*]] = load volatile i32, i32* [[G]], align 4
// CHECK4-NEXT: store volatile i32 [[TMP12]], i32* [[BLOCK_CAPTURED]], align 8
// CHECK4-NEXT: [[BLOCK_CAPTURED3:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>* [[BLOCK]], i32 0, i32 6
// CHECK4-NEXT: [[TMP13:%.*]] = load i32, i32* [[SIVAR1]], align 4
// CHECK4-NEXT: store i32 [[TMP13]], i32* [[BLOCK_CAPTURED3]], align 4
// CHECK4-NEXT: [[TMP14:%.*]] = bitcast <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>* [[BLOCK]] to void ()*
// CHECK4-NEXT: [[BLOCK_LITERAL:%.*]] = bitcast void ()* [[TMP14]] to %struct.__block_literal_generic*
// CHECK4-NEXT: [[TMP15:%.*]] = getelementptr inbounds [[STRUCT___BLOCK_LITERAL_GENERIC:%.*]], %struct.__block_literal_generic* [[BLOCK_LITERAL]], i32 0, i32 3
// CHECK4-NEXT: [[TMP16:%.*]] = bitcast %struct.__block_literal_generic* [[BLOCK_LITERAL]] to i8*
// CHECK4-NEXT: [[TMP17:%.*]] = load i8*, i8** [[TMP15]], align 8
// CHECK4-NEXT: [[TMP18:%.*]] = bitcast i8* [[TMP17]] to void (i8*)*
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 17:53:11 +08:00
// CHECK4-NEXT: call void [[TMP18]](i8* noundef [[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: br label [[DOTOMP_SECTIONS_EXIT]]
// CHECK4: .omp.sections.exit:
// CHECK4-NEXT: br label [[OMP_INNER_FOR_INC:%.*]]
// CHECK4: omp.inner.for.inc:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK4-NEXT: [[TMP19:%.*]] = load i32, i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK4-NEXT: [[INC:%.*]] = add nsw i32 [[TMP19]], 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: store i32 [[INC]], i32* [[DOTOMP_SECTIONS_IV_]], align 4
// CHECK4-NEXT: br label [[OMP_INNER_FOR_COND]]
// CHECK4: omp.inner.for.end:
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-22 04:20:39 +08:00
// CHECK4-NEXT: call void @__kmpc_for_static_fini(%struct.ident_t* @[[GLOB1]], i32 [[TMP4]])
// CHECK4-NEXT: call void @__kmpc_barrier(%struct.ident_t* @[[GLOB2:[0-9]+]], 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: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@var_block_invoke
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 17:53:11 +08:00
// CHECK4-SAME: (i8* noundef [[DOTBLOCK_DESCRIPTOR:%.*]]) #[[ATTR1]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code 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: [[DOTBLOCK_DESCRIPTOR_ADDR:%.*]] = alloca i8*, align 8
// CHECK4-NEXT: [[BLOCK_ADDR:%.*]] = alloca <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>*, align 8
// CHECK4-NEXT: store i8* [[DOTBLOCK_DESCRIPTOR]], i8** [[DOTBLOCK_DESCRIPTOR_ADDR]], align 8
// CHECK4-NEXT: [[BLOCK:%.*]] = bitcast i8* [[DOTBLOCK_DESCRIPTOR]] to <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>*
// CHECK4-NEXT: store <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>* [[BLOCK]], <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>** [[BLOCK_ADDR]], align 8
// CHECK4-NEXT: [[BLOCK_CAPTURE_ADDR:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>* [[BLOCK]], i32 0, i32 5
// CHECK4-NEXT: store i32 2, i32* [[BLOCK_CAPTURE_ADDR]], align 8
// CHECK4-NEXT: [[BLOCK_CAPTURE_ADDR1:%.*]] = getelementptr inbounds <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>, <{ i8*, i32, i32, i8*, %struct.__block_descriptor*, i32, i32 }>* [[BLOCK]], i32 0, i32 6
// CHECK4-NEXT: store i32 20, i32* [[BLOCK_CAPTURE_ADDR1]], align 4
// CHECK4-NEXT: ret void
//
//
// CHECK4-LABEL: define {{[^@]+}}@_GLOBAL__sub_I_sections_firstprivate_codegen.cpp
// CHECK4-SAME: () #[[ATTR0]] section "__TEXT,__StaticInit,regular,pure_instructions" {
// CHECK4-NEXT: entry:
// CHECK4-NEXT: call void @__cxx_global_var_init()
// CHECK4-NEXT: call void @__cxx_global_var_init.1()
// CHECK4-NEXT: call void @__cxx_global_var_init.2()
// CHECK4-NEXT: ret void
//