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/target_firstprivate_codegen...

612 lines
40 KiB
C++
Raw Normal View History

[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// Test host codegen.
[OpenMP] Use fopenmp prefix for all options introduced by the offloading implementation. Summary: This patch changes the options used by offloading to start with -fopenmp instead of -fomp. This makes the option naming more consistent and materializes a suggestion by Richard Smith in http://reviews.llvm.org/D9888. Reviewers: hfinkel, carlo.bertolli, arpith-jacob, ABataev Subscribers: kkwli0, cfe-commits, caomhin Differential Revision: http://reviews.llvm.org/D21841 llvm-svn: 274283
2016-07-01 05:22:08 +08:00
// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-64
// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix CHECK --check-prefix CHECK-32
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
[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 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY0 %s
// SIMD-ONLY0-NOT: {{__kmpc|__tgt}}
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// Test target codegen - host bc file has to be created first.
[OpenMP] Use fopenmp prefix for all options introduced by the offloading implementation. Summary: This patch changes the options used by offloading to start with -fopenmp instead of -fomp. This makes the option naming more consistent and materializes a suggestion by Richard Smith in http://reviews.llvm.org/D9888. Reviewers: hfinkel, carlo.bertolli, arpith-jacob, ABataev Subscribers: kkwli0, cfe-commits, caomhin Differential Revision: http://reviews.llvm.org/D21841 llvm-svn: 274283
2016-07-01 05:22:08 +08:00
// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-64
// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck %s --check-prefix TCHECK --check-prefix TCHECK-32
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
[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 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck --check-prefix SIMD-ONLY1 %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify %s -emit-llvm -o - | FileCheck --check-prefix SIMD-ONLY1 %s
// SIMD-ONLY1-NOT: {{__kmpc|__tgt}}
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// expected-no-diagnostics
#ifndef HEADER
#define HEADER
template<typename tx, typename ty>
struct TT{
tx X;
ty Y;
};
// CHECK: [[TT:%.+]] = type { i64, i8 }
// CHECK: [[S1:%.+]] = type { double }
// TCHECK: [[TT:%.+]] = type { i64, i8 }
// TCHECK: [[S1:%.+]] = type { double }
// CHECK-DAG: [[SIZET:@.+]] = private unnamed_addr constant [1 x i{{32|64}}] [i[[SZ:32|64]] 4]
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: [[MAPT:@.+]] = private unnamed_addr constant [1 x i64] [i64 800]
// CHECK-DAG: [[MAPT2:@.+]] = private unnamed_addr constant [9 x i64] [i64 800, i64 673, i64 288, i64 673, i64 673, i64 288, i64 288, i64 673, i64 673]
[OpenMP] Adjust map type bits according to latest spec and use zero size array sections for pointers. Summary: This patch changes the bits used to specify the map types according to the latest version of the libomptarget document and add the support for zero size array section when pointers are being implicitly mapped. This completes the missing new 4.5 map semantics. Reviewers: hfinkel, carlo.bertolli, arpith-jacob, kkwli0, ABataev Subscribers: caomhin, cfe-commits Differential Revision: http://reviews.llvm.org/D20111 llvm-svn: 270868
2016-05-27 00:48:10 +08:00
// CHECK-DAG: [[SIZET3:@.+]] = private unnamed_addr constant [1 x i{{32|64}}] zeroinitializer
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK-DAG: [[MAPT3:@.+]] = private unnamed_addr constant [1 x i64] [i64 544]
// CHECK-DAG: [[MAPT4:@.+]] = private unnamed_addr constant [6 x i64] [i64 32, i64 281474976711171, i64 800, i64 288, i64 288, i64 673]
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK-DAG: [[SIZET5:@.+]] = private unnamed_addr constant [3 x i{{32|64}}] [i[[SZ]] 4, i[[SZ]] 1, i[[SZ]] 40]
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK-DAG: [[MAPT5:@.+]] = private unnamed_addr constant [3 x i64] [i64 800, i64 800, i64 673]
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK-DAG: [[SIZET6:@.+]] = private unnamed_addr constant [2 x i{{32|64}}] [i[[SZ]] 4, i[[SZ]] 40]
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK-DAG: [[MAPT6:@.+]] = private unnamed_addr constant [2 x i64] [i64 800, i64 673]
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: define {{.*}}[[FOO:@.+]](
int foo(int n, double *ptr) {
int a = 0;
short aa = 0;
float b[10];
float bn[n];
double c[5][10];
double cn[5][n];
TT<long long, char> d;
#pragma omp target firstprivate(a)
{
}
// a is passed by value to tgt_target
// CHECK: [[N_ADDR:%.+]] = alloca i{{[0-9]+}},
// CHECK: [[PTR_ADDR:%.+]] = alloca double*,
// CHECK: [[A:%.+]] = alloca i{{[0-9]+}},
// CHECK: [[A2:%.+]] = alloca i{{[0-9]+}},
// CHECK: [[B:%.+]] = alloca [10 x float],
// CHECK: [[SSTACK:%.+]] = alloca i8*,
// CHECK: [[C:%.+]] = alloca [5 x [10 x double]],
// CHECK: [[D:%.+]] = alloca [[TT]],
// CHECK: [[ACAST:%.+]] = alloca i{{[0-9]+}},
// CHECK: [[BASE_PTR_ARR:%.+]] = alloca [1 x i8*],
// CHECK: [[PTR_ARR:%.+]] = alloca [1 x i8*],
// CHECK: [[A2CAST:%.+]] = alloca i{{[0-9]+}},
// CHECK: [[BASE_PTR_ARR2:%.+]] = alloca [9 x i8*],
// CHECK: [[PTR_ARR2:%.+]] = alloca [9 x i8*],
// CHECK: [[SIZET2:%.+]] = alloca [9 x i{{[0-9]+}}],
// CHECK: [[BASE_PTR_ARR3:%.+]] = alloca [1 x i8*],
// CHECK: [[PTR_ARR3:%.+]] = alloca [1 x i8*],
// CHECK: [[N_ADDR_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[N_ADDR]],
// CHECK-64: [[N_EXT:%.+]] = zext i{{[0-9]+}} [[N_ADDR_VAL]] to i{{[0-9]+}}
// CHECK: [[SSAVE_RET:%.+]] = call i8* @llvm.stacksave()
// CHECK: store i8* [[SSAVE_RET]], i8** [[SSTACK]],
// CHECK-64: [[BN_VLA:%.+]] = alloca float, i{{[0-9]+}} [[N_EXT]],
// CHECK-32: [[BN_VLA:%.+]] = alloca float, i{{[0-9]+}} [[N_ADDR_VAL]],
// CHECK: [[N_ADDR_VAL2:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[N_ADDR]],
// CHECK-64: [[N_EXT2:%.+]] = zext i{{[0-9]+}} [[N_ADDR_VAL2]] to i{{[0-9]+}}
// CHECK-64: [[CN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_EXT2]]
// CHECK-32: [[CN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_ADDR_VAL2]]
// CHECK: [[CN_VLA:%.+]] = alloca double, i{{[0-9]+}} [[CN_SIZE]],
// CHECK: [[AVAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A]],
// CHECK-64: [[CONV:%.+]] = bitcast i{{[0-9]+}}* [[ACAST]] to i{{[0-9]+}}*
// CHECK-64: store i{{[0-9]+}} [[AVAL]], i{{[0-9]+}}* [[CONV]],
// CHECK-32: store i{{[0-9]+}} [[AVAL]], i{{[0-9]+}}* [[ACAST]],
// CHECK: [[ACAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[ACAST]],
// CHECK: [[BASE_PTR_GEP:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BASE_PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[ACAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} [[ACAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[PTR_GEP:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[ACAST_TOPTR2:%.+]] = bitcast i8** [[PTR_GEP]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} [[ACAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR2]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[BASE_PTR_GEP_ARG:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BASE_PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
// CHECK: [[PTR_GEP_ARG:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PTR_ARR]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
[Clang][OpenMP] New clang/libomptarget map interface: new function signatures, clang-side This clang patch changes the __tgt_* API function signatures in preparation for the new map interface. Changes are: Device IDs 32bits --> 64bits, Flags 32bits --> 64bits Differential revision: https://reviews.llvm.org/D40281 llvm-svn: 318789
2017-11-22 02:25:12 +08:00
// CHECK: {{.+}} = call i32 @__tgt_target(i64 -1, {{.+}}, i32 1, i8** [[BASE_PTR_GEP_ARG]], i8** [[PTR_GEP_ARG]], i[[SZ]]* getelementptr inbounds ([1 x i[[SZ]]], [1 x i[[SZ]]]* [[SIZET]], i32 0, i32 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* [[MAPT]], i32 0, i32 0))
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
[OPENMP, NVPTX] Fix linkage of the global entries. The linkage of the global entries must be weak to enable support of redefinition of the same target regions in multiple compilation units. llvm-svn: 331768
2018-05-08 22:16:57 +08:00
// TCHECK: define weak void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]])
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}},
[OPENMP] Pass scalar firstprivate vars by value. For better performance and to unify code with offloading part we pass scalar firstprivate values by value, instead of by reference. It will remove some extra copying operations. llvm-svn: 269751
2016-05-17 16:55:33 +08:00
// TCHECK-NOT: alloca i{{[0-9]+}},
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]],
[OPENMP] Pass scalar firstprivate vars by value. For better performance and to unify code with offloading part we pass scalar firstprivate values by value, instead of by reference. It will remove some extra copying operations. llvm-svn: 269751
2016-05-17 16:55:33 +08:00
// TCHECK-NOT: store i{{[0-9]+}} %
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: ret void
#pragma omp target firstprivate(aa,b,bn,c,cn,d)
{
aa += 1;
b[2] = 1.0;
bn[3] = 1.0;
c[1][2] = 1.0;
cn[1][3] = 1.0;
d.X = 1;
d.Y = 1;
}
// CHECK: [[A2VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A2]],
// CHECK: [[A2CASTCONV:%.+]] = bitcast i{{[0-9]+}}* [[A2CAST]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} [[A2VAL]], i{{[0-9]+}}* [[A2CASTCONV]],
// CHECK: [[A2CAST_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[A2CAST]],
// CHECK-64: [[BN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} [[N_EXT]], 4
// CHECK-32: [[BN_SIZE:%.+]] = mul{{.+}} i{{[0-9]+}} [[N_ADDR_VAL]], 4
// CHECK-64: [[CN_SIZE_1:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_EXT2]]
// CHECK-32: [[CN_SIZE_1:%.+]] = mul{{.+}} i{{[0-9]+}} 5, [[N_ADDR_VAL2]]
// CHECK: [[CN_SIZE_2:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SIZE_1]], 8
// firstprivate(aa) --> base_ptr = aa, ptr = aa, size = 2 (short)
// CHECK: [[BASE_PTR_GEP2_0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[ACAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_0]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} [[A2CAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[PTR_GEP2_0:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[ACAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_0]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} [[A2CAST_VAL]], i{{[0-9]+}}* [[ACAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[SIZE_GEPA2:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
// CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[SIZE_GEPA2]],
// firstprivate(b): base_ptr = &b[0], ptr = &b[0], size = 40 (sizeof(float)*10)
// CHECK: [[BASE_PTR_GEP2_1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_1]] to [10 x float]**
// CHECK: store [10 x float]* [[B]], [10 x float]** [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[PTR_GEP2_1:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_1]] to [10 x float]**
// CHECK: store [10 x float]* [[B]], [10 x float]** [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[SIZE_GEPB:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
// CHECK: store i{{[0-9]+}} 40, i{{[0-9]+}}* [[SIZE_GEPB]],
// firstprivate(bn), 2 entries, n and bn: (1) base_ptr = n, ptr = n, size = 8 ; (2) base_ptr = &c[0], ptr = &c[0], size = n*sizeof(float)
// CHECK: [[BASE_PTR_GEP2_2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_2]] to i{{[0-9]+}}*
// CHECK-64: store i{{[0-9]+}} [[N_EXT]], i{{[0-9]+}}* [[BCAST_TOPTR]],
// CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL]], i{{[0-9]+}}* [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[PTR_GEP2_2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_2]] to i{{[0-9]+}}*
// CHECK-64: store i{{[0-9]+}} [[N_EXT]], i{{[0-9]+}}* [[BCAST_TOPTR]],
// CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL]], i{{[0-9]+}}* [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[SIZE_GEPBN_1:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
// CHECK: store i{{[0-9]+}} {{[0-9]}}, i{{[0-9]+}}* [[SIZE_GEPBN_1]],
// CHECK: [[BASE_PTR_GEP2_3:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 3
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_3]] to float**
// CHECK: store float* [[BN_VLA]], float** [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[SIZE_GEPBN_3:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 3
// CHECK: store i{{[0-9]+}} [[BN_SIZE]], i{{[0-9]+}}* [[SIZE_GEPBN_3]]
// firstprivate(c): base_ptr = &c[0], ptr = &c[0], size = 400 (5*10*sizeof(double))
// CHECK: [[BASE_PTR_GEP2_4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 4
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_4]] to [5 x [10 x double]]**
// CHECK: store [5 x [10 x double]]* [[C]], [5 x [10 x double]]** [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[PTR_GEP2_4:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 4
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_4]] to [5 x [10 x double]]**
// CHECK: store [5 x [10 x double]]* [[C]], [5 x [10 x double]]** [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[SIZE_GEPC_4:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 4
// CHECK: store i{{[0-9]+}} 400, i{{[0-9]+}}* [[SIZE_GEPC_4]],
// firstprivate(cn), 3 entries, 5, n, cn: (1) base_ptr = 5, ptr = 5, size = 8; (2) (1) base_ptr = n, ptr = n, size = 8; (3) base_ptr = &cn[0], ptr = &cn[0], size = 5*n*sizeof(double)
// CHECK: [[BASE_PTR_GEP2_5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 5
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_5]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} 5, i{{[0-9]+}}* [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[PTR_GEP2_5:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 5
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_5]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} 5, i{{[0-9]+}}* [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[SIZE_GEPCN_5:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 5
// CHECK: store i{{[0-9]+}} {{[0-9]}}, i{{[0-9]+}}* [[SIZE_GEPCN_5]],
// CHECK: [[BASE_PTR_GEP2_6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 6
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_6]] to i{{[0-9]+}}*
// CHECK-64: store i{{[0-9]+}} [[N_EXT2]], i{{[0-9]+}}* [[BCAST_TOPTR]],
// CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL2]], i{{[0-9]+}}* [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[PTR_GEP2_6:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 6
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_6]] to i{{[0-9]+}}*
// CHECK-64: store i{{[0-9]+}} [[N_EXT2]], i{{[0-9]+}}* [[BCAST_TOPTR]],
// CHECK-32: store i{{[0-9]+}} [[N_ADDR_VAL2]], i{{[0-9]+}}* [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[SIZE_GEPCN_6:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 6
// CHECK: store i{{[0-9]+}} {{[0-9]}}, i{{[0-9]+}}* [[SIZE_GEPCN_6]],
// CHECK: [[BASE_PTR_GEP2_7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 7
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_7]] to double**
// CHECK: store double* [[CN_VLA]], double** [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[PTR_GEP2_7:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 7
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_7]] to double**
// CHECK: store double* [[CN_VLA]], double** [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[SIZE_GEPCN_7:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 7
// CHECK: store i{{[0-9]+}} [[CN_SIZE_2]], i{{[0-9]+}}* [[SIZE_GEPCN_7]],
// firstprivate(d): base_ptr = &d, ptr = &d, size = 16
// CHECK: [[BASE_PTR_GEP2_8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 8
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP2_8]] to [[TT]]**
// CHECK: store [[TT]]* [[D]], [[TT]]** [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[PTR_GEP2_8:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 8
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP2_8]] to [[TT]]**
// CHECK: store [[TT]]* [[D]], [[TT]]** [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[SIZE_GEPCN_8:%.+]] = getelementptr inbounds [9 x i{{[0-9]+}}], [9 x i{{[0-9]+}}]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 8
// CHECK: store i{{[0-9]+}} {{[0-9]+}}, i{{[0-9]+}}* [[SIZE_GEPCN_8]],
// CHECK: [[BASE_PTR_GEP_ARG2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[BASE_PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
// CHECK: [[PTR_GEP_ARG2:%.+]] = getelementptr inbounds [9 x i8*], [9 x i8*]* [[PTR_ARR2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
// CHECK: [[SIZES_ARG2:%.+]] = getelementptr inbounds [9 x i[[SZ]]], [9 x i[[SZ]]]* [[SIZET2]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
[Clang][OpenMP] New clang/libomptarget map interface: new function signatures, clang-side This clang patch changes the __tgt_* API function signatures in preparation for the new map interface. Changes are: Device IDs 32bits --> 64bits, Flags 32bits --> 64bits Differential revision: https://reviews.llvm.org/D40281 llvm-svn: 318789
2017-11-22 02:25:12 +08:00
// CHECK: {{.+}} = call i32 @__tgt_target(i64 -1, {{.+}}, i32 9, i8** [[BASE_PTR_GEP_ARG2]], i8** [[PTR_GEP_ARG2]], i[[SZ]]* [[SIZES_ARG2]], i64* getelementptr inbounds ([9 x i64], [9 x i64]* [[MAPT2]], i32 0, i32 0))
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// make sure that firstprivate variables are generated in all cases and that we use those instances for operations inside the
// target region
[OPENMP] Fix PR35013: Fix passing VLAs captures to outlined functions. Fixed passing of VLAs and variably-modified types to outlined functions. Synchronized passing with the types codegen. llvm-svn: 316488
2017-10-25 03:52:31 +08:00
// TCHECK: define {{.*}}void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A2_IN:%.+]], [10 x float]* {{.+}} [[B_IN:%.+]], i{{[0-9]+}} [[BN_SZ:%.+]], float* {{.+}} [[BN_IN:%.+]], [5 x [10 x double]]* {{.+}} [[C_IN:%.+]], i{{[0-9]+}} [[CN_SZ1:%.+]], i{{[0-9]+}} [[CN_SZ2:%.+]], double* {{.+}} [[CN_IN:%.+]], [[TT]]* {{.+}} [[D_IN:%.+]])
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: [[A2_ADDR:%.+]] = alloca i{{[0-9]+}},
// TCHECK: [[B_ADDR:%.+]] = alloca [10 x float]*,
// TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}},
// TCHECK: [[BN_ADDR:%.+]] = alloca float*,
// TCHECK: [[C_ADDR:%.+]] = alloca [5 x [10 x double]]*,
// TCHECK: [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}},
// TCHECK: [[VLA_ADDR4:%.+]] = alloca i{{[0-9]+}},
// TCHECK: [[CN_ADDR:%.+]] = alloca double*,
// TCHECK: [[D_ADDR:%.+]] = alloca [[TT]]*,
[OPENMP] Pass scalar firstprivate vars by value. For better performance and to unify code with offloading part we pass scalar firstprivate values by value, instead of by reference. It will remove some extra copying operations. llvm-svn: 269751
2016-05-17 16:55:33 +08:00
// TCHECK-NOT: alloca i{{[0-9]+}},
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: [[B_PRIV:%.+]] = alloca [10 x float],
// TCHECK: [[SSTACK:%.+]] = alloca i8*,
// TCHECK: [[C_PRIV:%.+]] = alloca [5 x [10 x double]],
// TCHECK: [[D_PRIV:%.+]] = alloca [[TT]],
// TCHECK: store i{{[0-9]+}} [[A2_IN]], i{{[0-9]+}}* [[A2_ADDR]],
// TCHECK: store [10 x float]* [[B_IN]], [10 x float]** [[B_ADDR]],
// TCHECK: store i{{[0-9]+}} [[BN_SZ]], i{{[0-9]+}}* [[VLA_ADDR]],
// TCHECK: store float* [[BN_IN]], float** [[BN_ADDR]],
// TCHECK: store [5 x [10 x double]]* [[C_IN]], [5 x [10 x double]]** [[C_ADDR]],
// TCHECK: store i{{[0-9]+}} [[CN_SZ1]], i{{[0-9]+}}* [[VLA_ADDR2]],
// TCHECK: store i{{[0-9]+}} [[CN_SZ2]], i{{[0-9]+}}* [[VLA_ADDR4]],
// TCHECK: store double* [[CN_IN]], double** [[CN_ADDR]],
// TCHECK: store [[TT]]* [[D_IN]], [[TT]]** [[D_ADDR]],
// TCHECK: [[CONV_A2ADDR:%.+]] = bitcast i{{[0-9]+}}* [[A2_ADDR]] to i{{[0-9]+}}*
// TCHECK: [[B_ADDR_REF:%.+]] = load [10 x float]*, [10 x float]** [[B_ADDR]],
// TCHECK: [[BN_SZ_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR]],
// TCHECK: [[BN_ADDR_REF:%.+]] = load float*, float** [[BN_ADDR]],
// TCHECK: [[C_ADDR_REF:%.+]] = load [5 x [10 x double]]*, [5 x [10 x double]]** [[C_ADDR]],
// TCHECK: [[CN_SZ1_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR2]],
// TCHECK: [[CN_SZ2_VAL:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR4]],
// TCHECK: [[CN_ADDR_REF:%.+]] = load double*, double** [[CN_ADDR]],
// TCHECK: [[D_ADDR_REF:%.+]] = load [[TT]]*, [[TT]]** [[D_ADDR]],
// firstprivate(aa): a_priv = a_in
[OPENMP] Pass scalar firstprivate vars by value. For better performance and to unify code with offloading part we pass scalar firstprivate values by value, instead of by reference. It will remove some extra copying operations. llvm-svn: 269751
2016-05-17 16:55:33 +08:00
// TCHECK-NOT: store i{{[0-9]+}} %
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// firstprivate(b): memcpy(b_priv,b_in)
// TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x float]* [[B_PRIV]] to i8*
// TCHECK: [[B_ADDR_REF_BCAST:%.+]] = bitcast [10 x float]* [[B_ADDR_REF]] to i8*
Change memcpy/memove/memset to have dest and source alignment attributes (Step 1). Summary: Upstream LLVM is changing the the prototypes of the @llvm.memcpy/memmove/memset intrinsics. This change updates the Clang tests for this change. The @llvm.memcpy/memmove/memset intrinsics currently have an explicit argument which is required to be a constant integer. It represents the alignment of the dest (and source), and so must be the minimum of the actual alignment of the two. This change removes the alignment argument in favour of placing the alignment attribute on the source and destination pointers of the memory intrinsic call. For example, code which used to read: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 100, i32 4, i1 false) will now read call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 %dest, i8* align 4 %src, i32 100, i1 false) At this time the source and destination alignments must be the same (Step 1). Step 2 of the change, to be landed shortly, will relax that contraint and allow the source and destination to have different alignments. llvm-svn: 322964
2018-01-20 01:12:54 +08:00
// TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_ADDR_REF_BCAST]], {{.+}})
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: [[RET_STACK:%.+]] = call i8* @llvm.stacksave()
// TCHECK: store i8* [[RET_STACK]], i8** [[SSTACK]],
// firstprivate(bn)
// TCHECK: [[BN_PRIV:%.+]] = alloca float, i{{[0-9]+}} [[BN_SZ_VAL]],
// TCHECK: [[BN_COPY_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[BN_SZ_VAL]], 4
// TCHECK: [[BN_PRIV__BCAST:%.+]] = bitcast float* [[BN_PRIV]] to i8*
// TCHECK: [[BN_REF_IN_BCAST:%.+]] = bitcast float* [[BN_ADDR_REF]] to i8*
Change memcpy/memove/memset to have dest and source alignment attributes (Step 1). Summary: Upstream LLVM is changing the the prototypes of the @llvm.memcpy/memmove/memset intrinsics. This change updates the Clang tests for this change. The @llvm.memcpy/memmove/memset intrinsics currently have an explicit argument which is required to be a constant integer. It represents the alignment of the dest (and source), and so must be the minimum of the actual alignment of the two. This change removes the alignment argument in favour of placing the alignment attribute on the source and destination pointers of the memory intrinsic call. For example, code which used to read: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 100, i32 4, i1 false) will now read call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 %dest, i8* align 4 %src, i32 100, i1 false) At this time the source and destination alignments must be the same (Step 1). Step 2 of the change, to be landed shortly, will relax that contraint and allow the source and destination to have different alignments. llvm-svn: 322964
2018-01-20 01:12:54 +08:00
// TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[BN_PRIV__BCAST]], i8* align {{[0-9]+}} [[BN_REF_IN_BCAST]], i{{[0-9]+}} [[BN_COPY_SZ]],{{.+}})
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// firstprivate(c)
// TCHECK: [[C_PRIV_BCAST:%.+]] = bitcast [5 x [10 x double]]* [[C_PRIV]] to i8*
// TCHECK: [[C_IN_BCAST:%.+]] = bitcast [5 x [10 x double]]* [[C_ADDR_REF]] to i8*
Change memcpy/memove/memset to have dest and source alignment attributes (Step 1). Summary: Upstream LLVM is changing the the prototypes of the @llvm.memcpy/memmove/memset intrinsics. This change updates the Clang tests for this change. The @llvm.memcpy/memmove/memset intrinsics currently have an explicit argument which is required to be a constant integer. It represents the alignment of the dest (and source), and so must be the minimum of the actual alignment of the two. This change removes the alignment argument in favour of placing the alignment attribute on the source and destination pointers of the memory intrinsic call. For example, code which used to read: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 100, i32 4, i1 false) will now read call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 %dest, i8* align 4 %src, i32 100, i1 false) At this time the source and destination alignments must be the same (Step 1). Step 2 of the change, to be landed shortly, will relax that contraint and allow the source and destination to have different alignments. llvm-svn: 322964
2018-01-20 01:12:54 +08:00
// TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[C_PRIV_BCAST]], i8* align {{[0-9]+}} [[C_IN_BCAST]],{{.+}})
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// firstprivate(cn)
// TCHECK: [[CN_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ1_VAL]], [[CN_SZ2_VAL]]
// TCHECK: [[CN_PRIV:%.+]] = alloca double, i{{[0-9]+}} [[CN_SZ]],
// TCHECK: [[CN_SZ2:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ1_VAL]], [[CN_SZ2_VAL]]
// TCHECK: [[CN_SZ2_CPY:%.+]] = mul{{.+}} i{{[0-9]+}} [[CN_SZ2]], 8
// TCHECK: [[CN_PRIV_BCAST:%.+]] = bitcast double* [[CN_PRIV]] to i8*
// TCHECK: [[CN_IN_BCAST:%.+]] = bitcast double* [[CN_ADDR_REF]] to i8*
Change memcpy/memove/memset to have dest and source alignment attributes (Step 1). Summary: Upstream LLVM is changing the the prototypes of the @llvm.memcpy/memmove/memset intrinsics. This change updates the Clang tests for this change. The @llvm.memcpy/memmove/memset intrinsics currently have an explicit argument which is required to be a constant integer. It represents the alignment of the dest (and source), and so must be the minimum of the actual alignment of the two. This change removes the alignment argument in favour of placing the alignment attribute on the source and destination pointers of the memory intrinsic call. For example, code which used to read: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 100, i32 4, i1 false) will now read call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 %dest, i8* align 4 %src, i32 100, i1 false) At this time the source and destination alignments must be the same (Step 1). Step 2 of the change, to be landed shortly, will relax that contraint and allow the source and destination to have different alignments. llvm-svn: 322964
2018-01-20 01:12:54 +08:00
// TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[CN_PRIV_BCAST]], i8* align {{[0-9]+}} [[CN_IN_BCAST]], i{{[0-9]+}} [[CN_SZ2_CPY]],{{.+}})
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// firstprivate(d)
// TCHECK: [[D_PRIV_BCAST:%.+]] = bitcast [[TT]]* [[D_PRIV]] to i8*
// TCHECK: [[D_IN_BCAST:%.+]] = bitcast [[TT]]* [[D_ADDR_REF]] to i8*
Change memcpy/memove/memset to have dest and source alignment attributes (Step 1). Summary: Upstream LLVM is changing the the prototypes of the @llvm.memcpy/memmove/memset intrinsics. This change updates the Clang tests for this change. The @llvm.memcpy/memmove/memset intrinsics currently have an explicit argument which is required to be a constant integer. It represents the alignment of the dest (and source), and so must be the minimum of the actual alignment of the two. This change removes the alignment argument in favour of placing the alignment attribute on the source and destination pointers of the memory intrinsic call. For example, code which used to read: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 100, i32 4, i1 false) will now read call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 %dest, i8* align 4 %src, i32 100, i1 false) At this time the source and destination alignments must be the same (Step 1). Step 2 of the change, to be landed shortly, will relax that contraint and allow the source and destination to have different alignments. llvm-svn: 322964
2018-01-20 01:12:54 +08:00
// TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[D_PRIV_BCAST]], i8* align {{[0-9]+}} [[D_IN_BCAST]],{{.+}})
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
#pragma omp target firstprivate(ptr)
{
ptr[0]++;
}
// CHECK: [[PTR_ADDR_REF:%.+]] = load double*, double** [[PTR_ADDR]],
// CHECK: [[BASE_PTR_GEP3_0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BASE_PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTR_GEP3_0]] to double**
// CHECK: store double* [[PTR_ADDR_REF]], double** [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[PTR_GEP3_0:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTR_GEP3_0]] to double**
// CHECK: store double* [[PTR_ADDR_REF]], double** [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[BASE_PTR_GEP_ARG3:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[BASE_PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
// CHECK: [[PTR_GEP_ARG3:%.+]] = getelementptr inbounds [1 x i8*], [1 x i8*]* [[PTR_ARR3]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
[Clang][OpenMP] New clang/libomptarget map interface: new function signatures, clang-side This clang patch changes the __tgt_* API function signatures in preparation for the new map interface. Changes are: Device IDs 32bits --> 64bits, Flags 32bits --> 64bits Differential revision: https://reviews.llvm.org/D40281 llvm-svn: 318789
2017-11-22 02:25:12 +08:00
// CHECK: {{.+}} = call i32 @__tgt_target(i64 -1, {{.+}}, i32 1, i8** [[BASE_PTR_GEP_ARG3]], i8** [[PTR_GEP_ARG3]], i[[SZ]]* getelementptr inbounds ([1 x i[[SZ]]], [1 x i[[SZ]]]* [[SIZET3]], i32 0, i32 0), i64* getelementptr inbounds ([1 x i64], [1 x i64]* [[MAPT3]], i32 0, i32 0))
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
[OPENMP, NVPTX] Fix linkage of the global entries. The linkage of the global entries must be weak to enable support of redefinition of the same target regions in multiple compilation units. llvm-svn: 331768
2018-05-08 22:16:57 +08:00
// TCHECK: define weak void @__omp_offloading_{{.+}}(double* [[PTR_IN:%.+]])
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: [[PTR_ADDR:%.+]] = alloca double*,
[OPENMP] Pass scalar firstprivate vars by value. For better performance and to unify code with offloading part we pass scalar firstprivate values by value, instead of by reference. It will remove some extra copying operations. llvm-svn: 269751
2016-05-17 16:55:33 +08:00
// TCHECK-NOT: alloca double*,
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: store double* [[PTR_IN]], double** [[PTR_ADDR]],
[OPENMP] Pass scalar firstprivate vars by value. For better performance and to unify code with offloading part we pass scalar firstprivate values by value, instead of by reference. It will remove some extra copying operations. llvm-svn: 269751
2016-05-17 16:55:33 +08:00
// TCHECK-NOT: store double* %
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
return a;
}
template<typename tx>
tx ftemplate(int n) {
tx a = 0;
tx b[10];
#pragma omp target firstprivate(a,b)
{
a += 1;
b[2] += 1;
}
return a;
}
static
int fstatic(int n) {
int a = 0;
char aaa = 0;
int b[10];
#pragma omp target firstprivate(a,aaa,b)
{
a += 1;
aaa += 1;
b[2] += 1;
}
return a;
}
[OPENMP, NVPTX] Fix linkage of the global entries. The linkage of the global entries must be weak to enable support of redefinition of the same target regions in multiple compilation units. llvm-svn: 331768
2018-05-08 22:16:57 +08:00
// TCHECK: define weak void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]], i{{[0-9]+}} [[A3_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]])
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}},
// TCHECK: [[A3_ADDR:%.+]] = alloca i{{[0-9]+}},
// TCHECK: [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*,
[OPENMP] Pass scalar firstprivate vars by value. For better performance and to unify code with offloading part we pass scalar firstprivate values by value, instead of by reference. It will remove some extra copying operations. llvm-svn: 269751
2016-05-17 16:55:33 +08:00
// TCHECK-NOT: alloca i{{[0-9]+}},
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: [[B_PRIV:%.+]] = alloca [10 x i{{[0-9]+}}],
// TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]],
// TCHECK: store i{{[0-9]+}} [[A3_IN]], i{{[0-9]+}}* [[A3_ADDR]],
// TCHECK: store [10 x i{{[0-9]+}}]* [[B_IN]], [10 x i{{[0-9]+}}]** [[B_ADDR]],
// TCHECK-64: [[A_CONV:%.+]] = bitcast i{{[0-9]+}}* [[A_ADDR]] to i{{[0-9]+}}*
// TCHECK: [[A3_CONV:%.+]] = bitcast i{{[0-9]+}}* [[A3_ADDR]] to i8*
// TCHECK: [[B_ADDR_REF:%.+]] = load [10 x i{{[0-9]+}}]*, [10 x i{{[0-9]+}}]** [[B_ADDR]],
// firstprivate(a): a_priv = a_in
// firstprivate(aaa)
[OPENMP] Pass scalar firstprivate vars by value. For better performance and to unify code with offloading part we pass scalar firstprivate values by value, instead of by reference. It will remove some extra copying operations. llvm-svn: 269751
2016-05-17 16:55:33 +08:00
// TCHECK-NOT: store i{{[0-9]+}} %
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// firstprivate(b)
// TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8*
// TCHECK: [[B_IN_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_ADDR_REF]] to i8*
Change memcpy/memove/memset to have dest and source alignment attributes (Step 1). Summary: Upstream LLVM is changing the the prototypes of the @llvm.memcpy/memmove/memset intrinsics. This change updates the Clang tests for this change. The @llvm.memcpy/memmove/memset intrinsics currently have an explicit argument which is required to be a constant integer. It represents the alignment of the dest (and source), and so must be the minimum of the actual alignment of the two. This change removes the alignment argument in favour of placing the alignment attribute on the source and destination pointers of the memory intrinsic call. For example, code which used to read: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 100, i32 4, i1 false) will now read call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 %dest, i8* align 4 %src, i32 100, i1 false) At this time the source and destination alignments must be the same (Step 1). Step 2 of the change, to be landed shortly, will relax that contraint and allow the source and destination to have different alignments. llvm-svn: 322964
2018-01-20 01:12:54 +08:00
// TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_IN_BCAST]],{{.+}})
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: ret void
struct S1 {
double a;
int r1(int n){
int b = n+1;
short int c[2][n];
#pragma omp target firstprivate(b,c)
{
this->a = (double)b + 1.5;
c[1][1] = ++a;
}
return c[1][1] + (int)b;
}
// on the host side, we first generate r1, then the static function and the template above
// CHECK: define{{.+}} i32 {{.+}}([[S1]]* {{.+}}, i{{[0-9]+}} {{.+}})
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: [[BASE_PTRS4:%.+]] = alloca [6 x i8*],
// CHECK: [[PTRS4:%.+]] = alloca [6 x i8*],
// CHECK: [[SIZET4:%.+]] = alloca [6 x i{{[0-9]+}}],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// map(this: this ptr is implicitly captured (not firstprivate matter)
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: [[BP0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
// CHECK: [[CBP0:%.+]] = bitcast i8** [[BP0]] to %struct.S1**
// CHECK: store %struct.S1* [[THIS:%.+]], %struct.S1** [[CBP0]],
// CHECK: [[P0:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
// CHECK: [[CP0:%.+]] = bitcast i8** [[P0]] to double**
// CHECK: store double* [[A:%.+]], double** [[CP0]],
// CHECK: [[SZ0:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
// CHECK: store i{{[0-9]+}} %{{.+}}, i{{[0-9]+}}* [[SZ0]],
// CHECK: [[BP1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
// CHECK: [[CBP1:%.+]] = bitcast i8** [[BP1]] to %struct.S1**
// CHECK: store %struct.S1* [[THIS]], %struct.S1** [[CBP1]],
// CHECK: [[P1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
// CHECK: [[CP1:%.+]] = bitcast i8** [[P1]] to double**
// CHECK: store double* [[A]], double** [[CP1]],
// CHECK: [[SZ1:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
// CHECK: store i{{[0-9]+}} 8, i{{[0-9]+}}* [[SZ1]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// firstprivate(b): base_ptr = b, ptr = b, size = 4 (pass by-value)
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: [[BASE_PTRS_GEP4_1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_1]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} [[B_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]],
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: [[PTRS_GEP4_1:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_1]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} [[B_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]],
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: [[SIZES_GEP4_1:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIZES_GEP4_1]],
// firstprivate(c), 3 entries: 2, n, c
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: [[BASE_PTRS_GEP4_2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_2]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[BCAST_TOPTR]],
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: [[PTRS_GEP4_2:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_2]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} 2, i{{[0-9]+}}* [[BCAST_TOPTR]],
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: [[SIZES_GEP4_2:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 3
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK-64: store i{{[0-9]+}} 8, i{{[0-9]+}}* [[SIZES_GEP4_2]],
// CHECK-32: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIZES_GEP4_2]],
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: [[BASE_PTRS_GEP4_3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_3]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} [[N:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]],
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: [[PTRS_GEP4_3:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_3]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} [[N]], i{{[0-9]+}}* [[BCAST_TOPTR]],
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: [[SIZES_GEP4_3:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 4
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK-64: store i{{[0-9]+}} 8, i{{[0-9]+}}* [[SIZES_GEP4_3]],
// CHECK-32: store i{{[0-9]+}} 4, i{{[0-9]+}}* [[SIZES_GEP4_3]],
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: [[BASE_PTRS_GEP4_4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[BASE_PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 5
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP4_4]] to i{{[0-9]+}}**
// CHECK: store i{{[0-9]+}}* [[B:%.+]], i{{[0-9]+}}** [[BCAST_TOPTR]],
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: [[PTRS_GEP4_4:%.+]] = getelementptr inbounds [6 x i8*], [6 x i8*]* [[PTRS4]], i{{[0-9]+}} 0, i{{[0-9]+}} 5
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP4_4]] to i{{[0-9]+}}**
// CHECK: store i{{[0-9]+}}* [[B]], i{{[0-9]+}}** [[BCAST_TOPTR]],
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: [[SIZES_GEP4_4:%.+]] = getelementptr inbounds [6 x i{{[0-9]+}}], [6 x i{{[0-9]+}}]* [[SIZET4]], i{{[0-9]+}} 0, i{{[0-9]+}} 5
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: store i{{[0-9]+}} [[B_SIZE:%.+]], i{{[0-9]+}}* [[SIZES_GEP4_4]],
// only check that we use the map types stored in the global variable
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-20 00:34:13 +08:00
// CHECK: call i32 @__tgt_target(i64 -1, {{.+}}, i32 6, i8** {{.+}}, i8** {{.+}}, i{{[0-9]+}}* {{.+}}, i64* getelementptr inbounds ([6 x i64], [6 x i64]* [[MAPT4]], i32 0, i32 0))
[OPENMP, NVPTX] Fix linkage of the global entries. The linkage of the global entries must be weak to enable support of redefinition of the same target regions in multiple compilation units. llvm-svn: 331768
2018-05-08 22:16:57 +08:00
// TCHECK: define weak void @__omp_offloading_{{.+}}([[S1]]* [[TH:%.+]], i{{[0-9]+}} [[B_IN:%.+]], i{{[0-9]+}} [[VLA:%.+]], i{{[0-9]+}} [[VLA1:%.+]], i{{[0-9]+}}{{.+}} [[C_IN:%.+]])
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: [[TH_ADDR:%.+]] = alloca [[S1]]*,
// TCHECK: [[B_ADDR:%.+]] = alloca i{{[0-9]+}},
// TCHECK: [[VLA_ADDR:%.+]] = alloca i{{[0-9]+}},
// TCHECK: [[VLA_ADDR2:%.+]] = alloca i{{[0-9]+}},
// TCHECK: [[C_ADDR:%.+]] = alloca i{{[0-9]+}}*,
[OPENMP] Pass scalar firstprivate vars by value. For better performance and to unify code with offloading part we pass scalar firstprivate values by value, instead of by reference. It will remove some extra copying operations. llvm-svn: 269751
2016-05-17 16:55:33 +08:00
// TCHECK-NOT: alloca i{{[0-9]+}},
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: [[SSTACK:%.+]] = alloca i8*,
// TCHECK: store [[S1]]* [[TH]], [[S1]]** [[TH_ADDR]],
// TCHECK: store i{{[0-9]+}} [[B_IN]], i{{[0-9]+}}* [[B_ADDR]],
// TCHECK: store i{{[0-9]+}} [[VLA]], i{{[0-9]+}}* [[VLA_ADDR]],
// TCHECK: store i{{[0-9]+}} [[VLA1]], i{{[0-9]+}}* [[VLA_ADDR2]],
// TCHECK: store i{{[0-9]+}}* [[C_IN]], i{{[0-9]+}}** [[C_ADDR]],
// TCHECK: [[TH_ADDR_REF:%.+]] = load [[S1]]*, [[S1]]** [[TH_ADDR]],
// TCHECK-64: [[B_ADDR_CONV:%.+]] = bitcast i{{[0-9]+}}* [[B_ADDR]] to i{{[0-9]+}}*
// TCHECK: [[VLA_ADDR_REF:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR]],
// TCHECK: [[VLA_ADDR_REF2:%.+]] = load i{{[0-9]+}}, i{{[0-9]+}}* [[VLA_ADDR2]],
// TCHECK: [[C_ADDR_REF:%.+]] = load i{{[0-9]+}}*, i{{[0-9]+}}** [[C_ADDR]],
// firstprivate(b)
[OPENMP] Pass scalar firstprivate vars by value. For better performance and to unify code with offloading part we pass scalar firstprivate values by value, instead of by reference. It will remove some extra copying operations. llvm-svn: 269751
2016-05-17 16:55:33 +08:00
// TCHECK-NOT: store i{{[0-9]+}} %
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: [[RET_STACK:%.+]] = call i8* @llvm.stacksave()
// TCHECK: store i8* [[RET_STACK:%.+]], i8** [[SSTACK]],
// firstprivate(c)
// TCHECK: [[C_SZ:%.+]] = mul{{.+}} i{{[0-9]+}} [[VLA_ADDR_REF]], [[VLA_ADDR_REF2]]
// TCHECK: [[C_PRIV:%.+]] = alloca i{{[0-9]+}}, i{{[0-9]+}} [[C_SZ]],
// TCHECK: [[C_SZ2:%.+]] = mul{{.+}} i{{[0-9]+}} [[VLA_ADDR_REF]], [[VLA_ADDR_REF2]]
// TCHECK: [[C_SZ_CPY:%.+]] = mul{{.+}} i{{[0-9]+}} [[C_SZ2]], 2
// TCHECK: [[C_PRIV_BCAST:%.+]] = bitcast i{{[0-9]+}}* [[C_PRIV]] to i8*
// TCHECK: [[C_IN_BCAST:%.+]] = bitcast i{{[0-9]+}}* [[C_ADDR_REF]] to i8*
Change memcpy/memove/memset to have dest and source alignment attributes (Step 1). Summary: Upstream LLVM is changing the the prototypes of the @llvm.memcpy/memmove/memset intrinsics. This change updates the Clang tests for this change. The @llvm.memcpy/memmove/memset intrinsics currently have an explicit argument which is required to be a constant integer. It represents the alignment of the dest (and source), and so must be the minimum of the actual alignment of the two. This change removes the alignment argument in favour of placing the alignment attribute on the source and destination pointers of the memory intrinsic call. For example, code which used to read: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 100, i32 4, i1 false) will now read call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 %dest, i8* align 4 %src, i32 100, i1 false) At this time the source and destination alignments must be the same (Step 1). Step 2 of the change, to be landed shortly, will relax that contraint and allow the source and destination to have different alignments. llvm-svn: 322964
2018-01-20 01:12:54 +08:00
// TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[C_PRIV_BCAST]], i8* align {{[0-9]+}} [[C_IN_BCAST]],{{.+}})
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// finish
// TCHECK: [[RELOAD_SSTACK:%.+]] = load i8*, i8** [[SSTACK]],
// TCHECK: call void @llvm.stackrestore(i8* [[RELOAD_SSTACK]])
// TCHECK: ret void
// static host function
// CHECK: define{{.+}} i32 {{.+}}(i{{[0-9]+}} {{.+}})
// CHECK: [[BASE_PTRS5:%.+]] = alloca [3 x i8*],
// CHECK: [[PTRS5:%.+]] = alloca [3 x i8*],
// firstprivate(a): by value
// CHECK: [[BASE_PTRS_GEP5_0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP5_0]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} [[A_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[PTRS_GEP5_0:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP5_0]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} [[A_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// firstprivate(aaa): by value
// CHECK: [[BASE_PTRS_GEP5_1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP5_1]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} [[A3_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[PTRS_GEP5_1:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP5_1]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} [[A3_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// firstprivate(b): base_ptr = &b[0], ptr= &b[0]
// CHECK: [[BASE_PTRS_GEP5_2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[BASE_PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP5_2]] to [10 x i{{[0-9]+}}]**
// CHECK: store [10 x i{{[0-9]+}}]* [[B:%.+]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[PTRS_GEP5_2:%.+]] = getelementptr inbounds [3 x i8*], [3 x i8*]* [[PTRS5]], i{{[0-9]+}} 0, i{{[0-9]+}} 2
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP5_2]] to [10 x i{{[0-9]+}}]**
// CHECK: store [10 x i{{[0-9]+}}]* [[B]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// only check that the right sizes and map types are used
[Clang][OpenMP] New clang/libomptarget map interface: new function signatures, clang-side This clang patch changes the __tgt_* API function signatures in preparation for the new map interface. Changes are: Device IDs 32bits --> 64bits, Flags 32bits --> 64bits Differential revision: https://reviews.llvm.org/D40281 llvm-svn: 318789
2017-11-22 02:25:12 +08:00
// CHECK: call i32 @__tgt_target(i64 -1, {{.+}}, i32 3, i8** {{.+}}, i8** {{.+}}, i[[SZ]]* getelementptr inbounds ([3 x i[[SZ]]], [3 x i[[SZ]]]* [[SIZET5]], i32 0, i32 0), i64* getelementptr inbounds ([3 x i64], [3 x i64]* [[MAPT5]], i32 0, i32 0))
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
};
int bar(int n, double *ptr){
int a = 0;
a += foo(n, ptr);
S1 S;
a += S.r1(n);
a += fstatic(n);
a += ftemplate<int>(n);
return a;
}
// template host and device
// CHECK: define{{.+}} i32 {{.+}}(i{{[0-9]+}} {{.+}})
// CHECK: [[BASE_PTRS6:%.+]] = alloca [2 x i8*],
// CHECK: [[PTRS6:%.+]] = alloca [2 x i8*],
// firstprivate(a): by value
// CHECK: [[BASE_PTRS_GEP6_0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP6_0]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} [[AT_CAST:%.+]], i{{[0-9]+}}* [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[PTRS_GEP6_0:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 0
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP6_0]] to i{{[0-9]+}}*
// CHECK: store i{{[0-9]+}} [[AT_CAST]], i{{[0-9]+}}* [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// firstprivate(b): pointer
// CHECK: [[BASE_PTRS_GEP6_1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[BASE_PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[BASE_PTRS_GEP6_1]] to [10 x i{{[0-9]+}}]**
// CHECK: store [10 x i{{[0-9]+}}]* [[B:%.+]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// CHECK: [[PTRS_GEP6_1:%.+]] = getelementptr inbounds [2 x i8*], [2 x i8*]* [[PTRS6]], i{{[0-9]+}} 0, i{{[0-9]+}} 1
[OPENMP][DEBUG] Generate second function with correct arg types. Currently, if the some of the parameters are captured by value, this argument is converted to uintptr_t type and thus we loosing the debug info about real type of the argument (captured variable): ``` void @.outlined_function.(uintptr %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.outlined_function.(uintptr %a.casted.val) ... ``` To resolve this problem, in debug mode a speciall external wrapper function is generated, that calls the outlined function with the correct parameters types: ``` void @.wrapper.(uintptr %par) { %a = alloca i32 %cast = bitcast i32* %a to uintptr* store uintptr %par, uintptr *%cast %a.val = load i32, i32* %a call void @.outlined_function.(i32 %a) ret void } void @.outlined_function.(i32 %par); ... %a = alloca i32 %a.casted = alloca uintptr %cast = bitcast uintptr* %a.casted to i32* %a.val = load i32, i32 *%a store i32 %a.val, i32 *%cast %a.casted.val = load uintptr, uintptr* %a.casted call void @.wrapper.(uintptr %a.casted.val) ... ``` llvm-svn: 306697
2017-06-30 00:43:05 +08:00
// CHECK: [[BCAST_TOPTR:%.+]] = bitcast i8** [[PTRS_GEP6_1]] to [10 x i{{[0-9]+}}]**
// CHECK: store [10 x i{{[0-9]+}}]* [[B]], [10 x i{{[0-9]+}}]** [[BCAST_TOPTR]],
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
[Clang][OpenMP] New clang/libomptarget map interface: new function signatures, clang-side This clang patch changes the __tgt_* API function signatures in preparation for the new map interface. Changes are: Device IDs 32bits --> 64bits, Flags 32bits --> 64bits Differential revision: https://reviews.llvm.org/D40281 llvm-svn: 318789
2017-11-22 02:25:12 +08:00
// CHECK: call i32 @__tgt_target(i64 -1, {{.+}}, i32 2, i8** {{.+}}, i8** {{.+}}, i[[SZ]]* getelementptr inbounds ([2 x i[[SZ]]], [2 x i[[SZ]]]* [[SIZET6]], i32 0, i32 0), i64* getelementptr inbounds ([2 x i64], [2 x i64]* [[MAPT6]], i32 0, i32 0))
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
[OPENMP, NVPTX] Fix linkage of the global entries. The linkage of the global entries must be weak to enable support of redefinition of the same target regions in multiple compilation units. llvm-svn: 331768
2018-05-08 22:16:57 +08:00
// TCHECK: define weak void @__omp_offloading_{{.+}}(i{{[0-9]+}} [[A_IN:%.+]], [10 x i{{[0-9]+}}]*{{.+}} [[B_IN:%.+]])
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: [[A_ADDR:%.+]] = alloca i{{[0-9]+}},
// TCHECK: [[B_ADDR:%.+]] = alloca [10 x i{{[0-9]+}}]*,
[OPENMP] Pass scalar firstprivate vars by value. For better performance and to unify code with offloading part we pass scalar firstprivate values by value, instead of by reference. It will remove some extra copying operations. llvm-svn: 269751
2016-05-17 16:55:33 +08:00
// TCHECK-NOT: alloca i{{[0-9]+}},
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: [[B_PRIV:%.+]] = alloca [10 x i{{[0-9]+}}],
// TCHECK: store i{{[0-9]+}} [[A_IN]], i{{[0-9]+}}* [[A_ADDR]],
// TCHECK: store [10 x i{{[0-9]+}}]* [[B_IN]], [10 x i{{[0-9]+}}]** [[B_ADDR]],
// TCHECK-64: [[A_ADDR_CONV:%.+]] = bitcast i{{[0-9]+}}* [[A_ADDR]] to i{{[0-9]+}}*
// TCHECK: [[B_ADDR_REF:%.+]] = load [10 x i{{[0-9]+}}]*, [10 x i{{[0-9]+}}]** [[B_ADDR]],
// firstprivate(a)
[OPENMP] Pass scalar firstprivate vars by value. For better performance and to unify code with offloading part we pass scalar firstprivate values by value, instead of by reference. It will remove some extra copying operations. llvm-svn: 269751
2016-05-17 16:55:33 +08:00
// TCHECK-NOT: store i{{[0-9]+}} %
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// firstprivate(b)
// TCHECK: [[B_PRIV_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_PRIV]] to i8*
// TCHECK: [[B_IN_BCAST:%.+]] = bitcast [10 x i{{[0-9]+}}]* [[B_ADDR_REF]] to i8*
Change memcpy/memove/memset to have dest and source alignment attributes (Step 1). Summary: Upstream LLVM is changing the the prototypes of the @llvm.memcpy/memmove/memset intrinsics. This change updates the Clang tests for this change. The @llvm.memcpy/memmove/memset intrinsics currently have an explicit argument which is required to be a constant integer. It represents the alignment of the dest (and source), and so must be the minimum of the actual alignment of the two. This change removes the alignment argument in favour of placing the alignment attribute on the source and destination pointers of the memory intrinsic call. For example, code which used to read: call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dest, i8* %src, i32 100, i32 4, i1 false) will now read call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 4 %dest, i8* align 4 %src, i32 100, i1 false) At this time the source and destination alignments must be the same (Step 1). Step 2 of the change, to be landed shortly, will relax that contraint and allow the source and destination to have different alignments. llvm-svn: 322964
2018-01-20 01:12:54 +08:00
// TCHECK: call void @llvm.memcpy.{{.+}}(i8* align {{[0-9]+}} [[B_PRIV_BCAST]], i8* align {{[0-9]+}} [[B_IN_BCAST]],{{.+}})
[OPENMP] Implementation of codegen for firstprivate clause of target directive This patch implements the following aspects: It extends sema to check that a variable is not reference in both a map clause and firstprivate or private. This is needed to ensure correct functioning at codegen level, apart from being useful for the user. It implements firstprivate for target in codegen. The implementation applies to both host and nvptx devices. It adds regression tests for codegen of firstprivate, host and device side when using the host as device, and nvptx side. Please note that the regression test for nvptx codegen is missing VLAs. This is because VLAs currently require saving and restoring the stack which appears not to be a supported operation by nvptx backend. It adds a check in sema regression tests for target map, firstprivate, and private clauses. http://reviews.llvm.org/D18203 llvm-svn: 263837
2016-03-19 05:43:32 +08:00
// TCHECK: ret void
#endif