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/simd_codegen.cpp

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

888 lines
38 KiB
C++
Raw Normal View History

[OPENMP50]Codegen for lastprivate conditional list items. Added codegen support for lastprivate conditional. According to the standard, if when the conditional modifier appears on the clause, if an assignment to a list item is encountered in the construct then the original list item is assigned the value that is assigned to the new list item in the sequentially last iteration or lexically last section in which such an assignment is encountered. We look for the assignment operations and check if the left side references lastprivate conditional variable. Then the next code is emitted: if (last_iv_a <= iv) { last_iv_a = iv; last_a = lp_a; } At the end the implicit barrier is generated to wait for the end of all threads and then in the check for the last iteration the private copy is assigned the last value. if (last_iter) { lp_a = last_a; // <--- new code a = lp_a; // <--- store of private value to the original variable. }
2019-12-27 22:44:43 +08:00
// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=45 | FileCheck %s
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=45
// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=45 | FileCheck %s
// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - -fopenmp-version=45 | FileCheck %s --check-prefix=TERM_DEBUG
// RUN: %clang_cc1 -verify -fopenmp -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP50 --check-prefix=OMP50RT
[OPENMP50]Add if clause in simd directive. According to OpenMP 5.0, if clause can be used in simd directive. If condition in the if clause if false, the non-vectorized version of the loop must be executed.
2019-11-20 01:07:54 +08:00
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=50 -DOMP5
[OPENMP50]Codegen for lastprivate conditional list items. Added codegen support for lastprivate conditional. According to the standard, if when the conditional modifier appears on the clause, if an assignment to a list item is encountered in the construct then the original list item is assigned the value that is assigned to the new list item in the sequentially last iteration or lexically last section in which such an assignment is encountered. We look for the assignment operations and check if the left side references lastprivate conditional variable. Then the next code is emitted: if (last_iv_a <= iv) { last_iv_a = iv; last_a = lp_a; } At the end the implicit barrier is generated to wait for the end of all threads and then in the check for the last iteration the private copy is assigned the last value. if (last_iter) { lp_a = last_a; // <--- new code a = lp_a; // <--- store of private value to the original variable. }
2019-12-27 22:44:43 +08:00
// RUN: %clang_cc1 -fopenmp -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP50 --check-prefix=OMP50RT
[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
[OPENMP50]Codegen for lastprivate conditional list items. Added codegen support for lastprivate conditional. According to the standard, if when the conditional modifier appears on the clause, if an assignment to a list item is encountered in the construct then the original list item is assigned the value that is assigned to the new list item in the sequentially last iteration or lexically last section in which such an assignment is encountered. We look for the assignment operations and check if the left side references lastprivate conditional variable. Then the next code is emitted: if (last_iv_a <= iv) { last_iv_a = iv; last_a = lp_a; } At the end the implicit barrier is generated to wait for the end of all threads and then in the check for the last iteration the private copy is assigned the last value. if (last_iter) { lp_a = last_a; // <--- new code a = lp_a; // <--- store of private value to the original variable. }
2019-12-27 22:44:43 +08:00
// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=45 | FileCheck %s
// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=45
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=45 | FileCheck %s
// RUN: %clang_cc1 -verify -triple x86_64-apple-darwin10 -fopenmp-simd -fexceptions -fcxx-exceptions -debug-info-kind=line-tables-only -x c++ -emit-llvm %s -o - -fopenmp-version=45 | FileCheck --check-prefix=TERM_DEBUG %s
[OPENMP50]Add if clause in simd directive. According to OpenMP 5.0, if clause can be used in simd directive. If condition in the if clause if false, the non-vectorized version of the loop must be executed.
2019-11-20 01:07:54 +08:00
// RUN: %clang_cc1 -verify -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -emit-llvm %s -fexceptions -fcxx-exceptions -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP50
// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -emit-pch -o %t %s -fopenmp-version=50 -DOMP5
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple x86_64-unknown-unknown -fexceptions -fcxx-exceptions -debug-info-kind=limited -std=c++11 -include-pch %t -verify %s -emit-llvm -o - -fopenmp-version=50 -DOMP5 | FileCheck %s --check-prefix=CHECK --check-prefix=OMP50
[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
// expected-no-diagnostics
#ifndef HEADER
#define HEADER
[OPENMP50]Codegen for lastprivate conditional list items. Added codegen support for lastprivate conditional. According to the standard, if when the conditional modifier appears on the clause, if an assignment to a list item is encountered in the construct then the original list item is assigned the value that is assigned to the new list item in the sequentially last iteration or lexically last section in which such an assignment is encountered. We look for the assignment operations and check if the left side references lastprivate conditional variable. Then the next code is emitted: if (last_iv_a <= iv) { last_iv_a = iv; last_a = lp_a; } At the end the implicit barrier is generated to wait for the end of all threads and then in the check for the last iteration the private copy is assigned the last value. if (last_iter) { lp_a = last_a; // <--- new code a = lp_a; // <--- store of private value to the original variable. }
2019-12-27 22:44:43 +08:00
#ifdef OMP5
#define CONDITIONAL conditional :
#else
#define CONDITIONAL
#endif //OMP5
[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
// CHECK: [[SS_TY:%.+]] = type { i32 }
[OPENMP50]Codegen for lastprivate conditional list items. Added codegen support for lastprivate conditional. According to the standard, if when the conditional modifier appears on the clause, if an assignment to a list item is encountered in the construct then the original list item is assigned the value that is assigned to the new list item in the sequentially last iteration or lexically last section in which such an assignment is encountered. We look for the assignment operations and check if the left side references lastprivate conditional variable. Then the next code is emitted: if (last_iv_a <= iv) { last_iv_a = iv; last_a = lp_a; } At the end the implicit barrier is generated to wait for the end of all threads and then in the check for the last iteration the private copy is assigned the last value. if (last_iter) { lp_a = last_a; // <--- new code a = lp_a; // <--- store of private value to the original variable. }
2019-12-27 22:44:43 +08:00
// OMP50-DAG: [[LAST_IV:@.+]] = {{.*}}common global i64 0
// OMP50-DAG: [[LAST_A:@.+]] = {{.*}}common global i32 0
[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
long long get_val() { return 0; }
double *g_ptr;
[OPENMP50]Codegen for nontemporal clause. Summary: Basic codegen for the declarations marked as nontemporal. Also, if the base declaration in the member expression is marked as nontemporal, lvalue for member decl access inherits nonteporal flag from the base lvalue. Reviewers: rjmccall, hfinkel, jdoerfert Subscribers: guansong, arphaman, caomhin, kkwli0, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D71708
2019-12-19 23:01:10 +08:00
struct S {
int a, b;
};
[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
// CHECK-LABEL: define {{.*void}} @{{.*}}simple{{.*}}(float* {{.+}}, float* {{.+}}, float* {{.+}}, float* {{.+}})
void simple(float *a, float *b, float *c, float *d) {
[OPENMP50]Codegen for nontemporal clause. Summary: Basic codegen for the declarations marked as nontemporal. Also, if the base declaration in the member expression is marked as nontemporal, lvalue for member decl access inherits nonteporal flag from the base lvalue. Reviewers: rjmccall, hfinkel, jdoerfert Subscribers: guansong, arphaman, caomhin, kkwli0, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D71708
2019-12-19 23:01:10 +08:00
S s, *p;
[OPENMP50]Add if clause in simd directive. According to OpenMP 5.0, if clause can be used in simd directive. If condition in the if clause if false, the non-vectorized version of the loop must be executed.
2019-11-20 01:07:54 +08:00
#ifdef OMP5
[OPENMP50]Codegen for nontemporal clause. Summary: Basic codegen for the declarations marked as nontemporal. Also, if the base declaration in the member expression is marked as nontemporal, lvalue for member decl access inherits nonteporal flag from the base lvalue. Reviewers: rjmccall, hfinkel, jdoerfert Subscribers: guansong, arphaman, caomhin, kkwli0, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D71708
2019-12-19 23:01:10 +08:00
#pragma omp simd if (simd: true) nontemporal(a, b, c, d, s)
[OPENMP50]Add if clause in simd directive. According to OpenMP 5.0, if clause can be used in simd directive. If condition in the if clause if false, the non-vectorized version of the loop must be executed.
2019-11-20 01:07:54 +08:00
#else
[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
#pragma omp simd
[OPENMP50]Add if clause in simd directive. According to OpenMP 5.0, if clause can be used in simd directive. If condition in the if clause if false, the non-vectorized version of the loop must be executed.
2019-11-20 01:07:54 +08:00
#endif
// CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]]
[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
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[CMP:%.+]] = icmp slt i32 [[IV]], 6
// CHECK-NEXT: br i1 [[CMP]], label %[[SIMPLE_LOOP1_BODY:.+]], label %[[SIMPLE_LOOP1_END:[^,]+]]
for (int i = 3; i < 32; i += 5) {
// CHECK: [[SIMPLE_LOOP1_BODY]]:
// Start of body: calculate i from IV:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV1_1:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK: [[CALC_I_1:%.+]] = mul nsw i32 [[IV1_1]], 5
// CHECK-NEXT: [[CALC_I_2:%.+]] = add nsw i32 3, [[CALC_I_1]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]]{{.*}}!llvm.access.group
[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
// ... loop body ...
// End of body: store into a[i]:
[OPENMP50]Codegen for nontemporal clause. Summary: Basic codegen for the declarations marked as nontemporal. Also, if the base declaration in the member expression is marked as nontemporal, lvalue for member decl access inherits nonteporal flag from the base lvalue. Reviewers: rjmccall, hfinkel, jdoerfert Subscribers: guansong, arphaman, caomhin, kkwli0, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D71708
2019-12-19 23:01:10 +08:00
// OMP45-NOT: load float*,{{.*}}!nontemporal
// CHECK-NOT: load float,{{.*}}!nontemporal
// OMP50: load float*,{{.*}}!nontemporal
// OMP50: load float*,{{.*}}!nontemporal
// OMP50: load float*,{{.*}}!nontemporal
// OMP50: load i32,{{.*}}!nontemporal
// OMP50-NOT: load i32,{{.*}}!nontemporal
// OMP50: load float*,{{.*}}!nontemporal
// CHECK-NOT: load float,{{.*}}!nontemporal
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: store float [[RESULT:%.+]], float* {{%.+}}{{.*}}!llvm.access.group
[OPENMP50]Codegen for nontemporal clause. Summary: Basic codegen for the declarations marked as nontemporal. Also, if the base declaration in the member expression is marked as nontemporal, lvalue for member decl access inherits nonteporal flag from the base lvalue. Reviewers: rjmccall, hfinkel, jdoerfert Subscribers: guansong, arphaman, caomhin, kkwli0, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D71708
2019-12-19 23:01:10 +08:00
a[i] = b[i] * c[i] * d[i] + s.a + p->a;
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[ADD1_2:%.+]] = add nsw i32 [[IV1_2]], 1
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i32 [[ADD1_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// br label %{{.+}}, !llvm.loop !{{.+}}
}
// CHECK: [[SIMPLE_LOOP1_END]]:
long long k = get_val();
#pragma omp simd linear(k : 3)
// CHECK: [[K0:%.+]] = call {{.*}}i64 @{{.*}}get_val
// CHECK-NEXT: store i64 [[K0]], i64* [[K_VAR:%[^,]+]]
// CHECK: store i32 0, i32* [[OMP_IV2:%[^,]+]]
// CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_VAR]]
// CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV2]], 9
// CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP2_BODY:.+]], label %[[SIMPLE_LOOP2_END:[^,]+]]
for (int i = 10; i > 1; i--) {
// CHECK: [[SIMPLE_LOOP2_BODY]]:
// Start of body: calculate i from IV:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV2_0:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
[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
// FIXME: It is interesting, why the following "mul 1" was not constant folded?
// CHECK-NEXT: [[IV2_1:%.+]] = mul nsw i32 [[IV2_0]], 1
// CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV2_1]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
[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
//
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
// CHECK-NEXT: [[IV2_2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV2_2]], 3
// CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
// CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
// Update of the privatized version of linear variable!
// CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
a[k]++;
k = k + 3;
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV2_2:%.+]] = load i32, i32* [[OMP_IV2]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV2_2]], 1
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV2]]{{.*}}!llvm.access.group
[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
// br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP2_ID]]
}
// CHECK: [[SIMPLE_LOOP2_END]]:
//
// Update linear vars after loop, as the loop was operating on a private version.
// CHECK: [[LIN0_2:%.+]] = load i64, i64* [[LIN0]]
// CHECK-NEXT: [[LIN_ADD2:%.+]] = add nsw i64 [[LIN0_2]], 27
// CHECK-NEXT: store i64 [[LIN_ADD2]], i64* [[K_VAR]]
//
int lin = 12;
#pragma omp simd linear(lin : get_val()), linear(g_ptr)
// Init linear private var.
// CHECK: store i32 12, i32* [[LIN_VAR:%[^,]+]]
// CHECK: store i64 0, i64* [[OMP_IV3:%[^,]+]]
// CHECK: [[LIN_LOAD:%.+]] = load i32, i32* [[LIN_VAR]]
// CHECK-NEXT: store i32 [[LIN_LOAD]], i32* [[LIN_START:%[^,]+]]
// Remember linear step.
// CHECK: [[CALL_VAL:%.+]] = invoke
// CHECK: store i64 [[CALL_VAL]], i64* [[LIN_STEP:%[^,]+]]
// CHECK: [[GLIN_LOAD:%.+]] = load double*, double** [[GLIN_VAR:@[^,]+]]
// CHECK-NEXT: store double* [[GLIN_LOAD]], double** [[GLIN_START:%[^,]+]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV3:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[CMP3:%.+]] = icmp ult i64 [[IV3]], 4
// CHECK-NEXT: br i1 [[CMP3]], label %[[SIMPLE_LOOP3_BODY:.+]], label %[[SIMPLE_LOOP3_END:[^,]+]]
for (unsigned long long it = 2000; it >= 600; it-=400) {
// CHECK: [[SIMPLE_LOOP3_BODY]]:
// Start of body: calculate it from IV:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV3_0:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[LC_IT_1:%.+]] = mul i64 [[IV3_0]], 400
// CHECK-NEXT: [[LC_IT_2:%.+]] = sub i64 2000, [[LC_IT_1]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i64 [[LC_IT_2]], i64* {{.+}}, !llvm.access.group
[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
//
// Linear start and step are used to calculate current value of the linear variable.
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[LINSTART:.+]] = load i32, i32* [[LIN_START]]{{.*}}!llvm.access.group
// CHECK: [[LINSTEP:.+]] = load i64, i64* [[LIN_STEP]]{{.*}}!llvm.access.group
// CHECK-NOT: store i32 {{.+}}, i32* [[LIN_VAR]],{{.*}}!llvm.access.group
// CHECK: [[GLINSTART:.+]] = load double*, double** [[GLIN_START]]{{.*}}!llvm.access.group
// CHECK-NEXT: [[IV3_1:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[MUL:%.+]] = mul i64 [[IV3_1]], 1
// CHECK: [[GEP:%.+]] = getelementptr{{.*}}[[GLINSTART]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store double* [[GEP]], double** [[G_PTR_CUR:%[^,]+]]{{.*}}!llvm.access.group
[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
*g_ptr++ = 0.0;
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[GEP_VAL:%.+]] = load double{{.*}}[[G_PTR_CUR]]{{.*}}!llvm.access.group
// CHECK: store double{{.*}}[[GEP_VAL]]{{.*}}!llvm.access.group
[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
a[it + lin]++;
// CHECK: [[FLT_INC:%.+]] = fadd float
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store float [[FLT_INC]],{{.*}}!llvm.access.group
// CHECK: [[IV3_2:%.+]] = load i64, i64* [[OMP_IV3]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[ADD3_2:%.+]] = add i64 [[IV3_2]], 1
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i64 [[ADD3_2]], i64* [[OMP_IV3]]{{.*}}!llvm.access.group
[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
}
// CHECK: [[SIMPLE_LOOP3_END]]:
//
// Linear start and step are used to calculate final value of the linear variables.
// CHECK: [[LINSTART:.+]] = load i32, i32* [[LIN_START]]
// CHECK: [[LINSTEP:.+]] = load i64, i64* [[LIN_STEP]]
// CHECK: store i32 {{.+}}, i32* [[LIN_VAR]],
// CHECK: [[GLINSTART:.+]] = load double*, double** [[GLIN_START]]
// CHECK: store double* {{.*}}[[GLIN_VAR]]
#pragma omp simd
// CHECK: store i32 0, i32* [[OMP_IV4:%[^,]+]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV4:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[CMP4:%.+]] = icmp slt i32 [[IV4]], 4
// CHECK-NEXT: br i1 [[CMP4]], label %[[SIMPLE_LOOP4_BODY:.+]], label %[[SIMPLE_LOOP4_END:[^,]+]]
for (short it = 6; it <= 20; it-=-4) {
// CHECK: [[SIMPLE_LOOP4_BODY]]:
// Start of body: calculate it from IV:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV4_0:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i32 [[IV4_0]], 4
// CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i32 6, [[LC_IT_1]]
// CHECK-NEXT: [[LC_IT_3:%.+]] = trunc i32 [[LC_IT_2]] to i16
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i16 [[LC_IT_3]], i16* {{.+}}, !llvm.access.group
[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
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV4_2:%.+]] = load i32, i32* [[OMP_IV4]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[ADD4_2:%.+]] = add nsw i32 [[IV4_2]], 1
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i32 [[ADD4_2]], i32* [[OMP_IV4]]{{.*}}!llvm.access.group
[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
}
// CHECK: [[SIMPLE_LOOP4_END]]:
#pragma omp simd
// CHECK: store i32 0, i32* [[OMP_IV5:%[^,]+]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV5:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[CMP5:%.+]] = icmp slt i32 [[IV5]], 26
// CHECK-NEXT: br i1 [[CMP5]], label %[[SIMPLE_LOOP5_BODY:.+]], label %[[SIMPLE_LOOP5_END:[^,]+]]
for (unsigned char it = 'z'; it >= 'a'; it+=-1) {
// CHECK: [[SIMPLE_LOOP5_BODY]]:
// Start of body: calculate it from IV:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV5_0:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[IV5_1:%.+]] = mul nsw i32 [[IV5_0]], 1
// CHECK-NEXT: [[LC_IT_1:%.+]] = sub nsw i32 122, [[IV5_1]]
// CHECK-NEXT: [[LC_IT_2:%.+]] = trunc i32 [[LC_IT_1]] to i8
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i8 [[LC_IT_2]], i8* {{.+}}, !llvm.access.group
[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
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV5_2:%.+]] = load i32, i32* [[OMP_IV5]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[ADD5_2:%.+]] = add nsw i32 [[IV5_2]], 1
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i32 [[ADD5_2]], i32* [[OMP_IV5]]{{.*}}!llvm.access.group
[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
}
// CHECK: [[SIMPLE_LOOP5_END]]:
// CHECK-NOT: mul i32 %{{.+}}, 10
#pragma omp simd
for (unsigned i=100; i<10; i+=10) {
}
int A;
// CHECK: store i32 -1, i32* [[A:%.+]],
A = -1;
[OPENMP50]Codegen for lastprivate conditional list items. Added codegen support for lastprivate conditional. According to the standard, if when the conditional modifier appears on the clause, if an assignment to a list item is encountered in the construct then the original list item is assigned the value that is assigned to the new list item in the sequentially last iteration or lexically last section in which such an assignment is encountered. We look for the assignment operations and check if the left side references lastprivate conditional variable. Then the next code is emitted: if (last_iv_a <= iv) { last_iv_a = iv; last_a = lp_a; } At the end the implicit barrier is generated to wait for the end of all threads and then in the check for the last iteration the private copy is assigned the last value. if (last_iter) { lp_a = last_a; // <--- new code a = lp_a; // <--- store of private value to the original variable. }
2019-12-27 22:44:43 +08:00
#pragma omp simd lastprivate(CONDITIONAL A)
[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
// CHECK: store i64 0, i64* [[OMP_IV7:%[^,]+]]
// CHECK: br label %[[SIMD_LOOP7_COND:[^,]+]]
// CHECK: [[SIMD_LOOP7_COND]]:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: [[IV7:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[CMP7:%.+]] = icmp slt i64 [[IV7]], 7
// CHECK-NEXT: br i1 [[CMP7]], label %[[SIMPLE_LOOP7_BODY:.+]], label %[[SIMPLE_LOOP7_END:[^,]+]]
for (long long i = -10; i < 10; i += 3) {
// CHECK: [[SIMPLE_LOOP7_BODY]]:
// Start of body: calculate i from IV:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV7_0:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV7_0]], 3
// CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{.+}}!llvm.access.group
// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]{{.+}}!llvm.access.group
[OPENMP50]Codegen for lastprivate conditional list items. Added codegen support for lastprivate conditional. According to the standard, if when the conditional modifier appears on the clause, if an assignment to a list item is encountered in the construct then the original list item is assigned the value that is assigned to the new list item in the sequentially last iteration or lexically last section in which such an assignment is encountered. We look for the assignment operations and check if the left side references lastprivate conditional variable. Then the next code is emitted: if (last_iv_a <= iv) { last_iv_a = iv; last_a = lp_a; } At the end the implicit barrier is generated to wait for the end of all threads and then in the check for the last iteration the private copy is assigned the last value. if (last_iter) { lp_a = last_a; // <--- new code a = lp_a; // <--- store of private value to the original variable. }
2019-12-27 22:44:43 +08:00
// CHECK-NEXT: [[A_VAL:%.+]] = load i32, i32* [[A_PRIV:%[^,]+]],{{.+}}!llvm.access.group
// CHECK-NEXT: [[CAST:%.+]] = sext i32 [[A_VAL]] to i64
// CHECK-NEXT: [[ADD:%.+]] = add nsw i64 [[CAST]], [[LC_VAL]]
// CHECK-NEXT: [[CONV:%.+]] = trunc i64 [[ADD]] to i32
// CHECK-NEXT: store i32 [[CONV]], i32* [[A_PRIV]],{{.+}}!llvm.access.group
// OMP50-NEXT: [[IV:%.+]] = load i64, i64* [[OMP_IV7]],{{.+}}!llvm.access.group
// OMP50RT: call void @__kmpc_critical(%struct.ident_t* {{.+}}, i32 [[GTID:%.+]], [8 x i32]* [[A_REGION:@.+]]),{{.+}}!llvm.access.group
// OMP50-NEXT: [[LAST_IV_VAL:%.+]] = load i64, i64* [[LAST_IV]],{{.+}}!llvm.access.group
// OMP50-NEXT: [[CMP:%.+]] = icmp sle i64 [[LAST_IV_VAL]], [[IV]]
// OMP50-NEXT: br i1 [[CMP]], label %[[LP_THEN:.+]], label %[[LP_DONE:[^,]+]]
// OMP50: [[LP_THEN]]:
// OMP50-NEXT: store i64 [[IV]], i64* [[LAST_IV]],{{.+}}!llvm.access.group
// OMP50-NEXT: [[A_VAL:%.+]] = load i32, i32* [[A_PRIV]],{{.+}}!llvm.access.group
// OMP50-NEXT: store i32 [[A_VAL]], i32* [[LAST_A]],{{.+}}!llvm.access.group
// OMP50-NEXT: br label %[[LP_DONE]]
// OMP50: [[LP_DONE]]:
// OMP50RT-NEXT: call void @__kmpc_end_critical(%struct.ident_t* {{.+}}, i32 [[GTID]], [8 x i32]* [[A_REGION]]),{{.+}}!llvm.access.group
A += i;
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV7_2:%.+]] = load i64, i64* [[OMP_IV7]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[ADD7_2:%.+]] = add nsw i64 [[IV7_2]], 1
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i64 [[ADD7_2]], i64* [[OMP_IV7]]{{.*}}!llvm.access.group
[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
}
// CHECK: [[SIMPLE_LOOP7_END]]:
// CHECK-NEXT: store i64 11, i64*
[OPENMP50]Codegen for lastprivate conditional list items. Added codegen support for lastprivate conditional. According to the standard, if when the conditional modifier appears on the clause, if an assignment to a list item is encountered in the construct then the original list item is assigned the value that is assigned to the new list item in the sequentially last iteration or lexically last section in which such an assignment is encountered. We look for the assignment operations and check if the left side references lastprivate conditional variable. Then the next code is emitted: if (last_iv_a <= iv) { last_iv_a = iv; last_a = lp_a; } At the end the implicit barrier is generated to wait for the end of all threads and then in the check for the last iteration the private copy is assigned the last value. if (last_iter) { lp_a = last_a; // <--- new code a = lp_a; // <--- store of private value to the original variable. }
2019-12-27 22:44:43 +08:00
// OMP50-NEXT: [[LAST_A_VAL:%.+]] = load i32, i32* [[LAST_A]],
// OMP50-NEXT: store i32 [[LAST_A_VAL]], i32* [[A_PRIV]],
[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
// CHECK-NEXT: [[A_PRIV_VAL:%.+]] = load i32, i32* [[A_PRIV]],
// CHECK-NEXT: store i32 [[A_PRIV_VAL]], i32* [[A]],
int R;
// CHECK: store i32 -1, i32* [[R:%[^,]+]],
R = -1;
// CHECK: store i64 0, i64* [[OMP_IV8:%[^,]+]],
// CHECK: store i32 1, i32* [[R_PRIV:%[^,]+]],
[OPENMP50]Add if clause in simd directive. According to OpenMP 5.0, if clause can be used in simd directive. If condition in the if clause if false, the non-vectorized version of the loop must be executed.
2019-11-20 01:07:54 +08:00
#ifdef OMP5
#pragma omp simd reduction(*:R) if(A)
#else
[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
#pragma omp simd reduction(*:R)
[OPENMP50]Add if clause in simd directive. According to OpenMP 5.0, if clause can be used in simd directive. If condition in the if clause if false, the non-vectorized version of the loop must be executed.
2019-11-20 01:07:54 +08:00
#endif
// OMP50: [[A_VAL:%.+]] = load i32, i32* [[A]],
// OMP50-NEXT: [[COND:%.+]] = icmp ne i32 [[A_VAL]], 0
// OMP50-NEXT: br i1 [[COND]], label {{%?}}[[THEN:[^,]+]], label {{%?}}[[ELSE:[^,]+]]
// OMP50: [[THEN]]:
[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
// CHECK: br label %[[SIMD_LOOP8_COND:[^,]+]]
// CHECK: [[SIMD_LOOP8_COND]]:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: [[IV8:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[CMP8:%.+]] = icmp slt i64 [[IV8]], 7
// CHECK-NEXT: br i1 [[CMP8]], label %[[SIMPLE_LOOP8_BODY:.+]], label %[[SIMPLE_LOOP8_END:[^,]+]]
for (long long i = -10; i < 10; i += 3) {
// CHECK: [[SIMPLE_LOOP8_BODY]]:
// Start of body: calculate i from IV:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV8_0:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV8_0]], 3
// CHECK-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{.+}}!llvm.access.group
// CHECK-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]]{{.+}}!llvm.access.group
// CHECK: store i32 %{{.+}}, i32* [[R_PRIV]],{{.+}}!llvm.access.group
[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
R *= i;
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV8_2:%.+]] = load i64, i64* [[OMP_IV8]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[ADD8_2:%.+]] = add nsw i64 [[IV8_2]], 1
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i64 [[ADD8_2]], i64* [[OMP_IV8]]{{.*}}!llvm.access.group
[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
}
// CHECK: [[SIMPLE_LOOP8_END]]:
[OPENMP50]Add if clause in simd directive. According to OpenMP 5.0, if clause can be used in simd directive. If condition in the if clause if false, the non-vectorized version of the loop must be executed.
2019-11-20 01:07:54 +08:00
// OMP50: br label {{%?}}[[EXIT:[^,]+]]
// OMP50: br label %[[SIMD_LOOP8_COND:[^,]+]]
// OMP50: [[SIMD_LOOP8_COND]]:
// OMP50-NEXT: [[IV8:%.+]] = load i64, i64* [[OMP_IV8]],{{[^!]*}}
// OMP50-NEXT: [[CMP8:%.+]] = icmp slt i64 [[IV8]], 7
// OMP50-NEXT: br i1 [[CMP8]], label %[[SIMPLE_LOOP8_BODY:.+]], label %[[SIMPLE_LOOP8_END:[^,]+]]
// OMP50: [[SIMPLE_LOOP8_BODY]]:
// Start of body: calculate i from IV:
// OMP50: [[IV8_0:%.+]] = load i64, i64* [[OMP_IV8]],{{[^!]*}}
// OMP50-NEXT: [[LC_IT_1:%.+]] = mul nsw i64 [[IV8_0]], 3
// OMP50-NEXT: [[LC_IT_2:%.+]] = add nsw i64 -10, [[LC_IT_1]]
// OMP50-NEXT: store i64 [[LC_IT_2]], i64* [[LC:%[^,]+]],{{[^!]*}}
// OMP50-NEXT: [[LC_VAL:%.+]] = load i64, i64* [[LC]],{{[^!]*}}
// OMP50: store i32 %{{.+}}, i32* [[R_PRIV]],{{[^!]*}}
// OMP50: [[IV8_2:%.+]] = load i64, i64* [[OMP_IV8]],{{[^!]*}}
// OMP50-NEXT: [[ADD8_2:%.+]] = add nsw i64 [[IV8_2]], 1
// OMP50-NEXT: store i64 [[ADD8_2]], i64* [[OMP_IV8]],{{[^!]*}}
// OMP50: br label {{%?}}[[SIMD_LOOP8_COND]], {{.*}}!llvm.loop ![[DISABLE_VECT:.+]]
// OMP50: [[SIMPLE_LOOP8_END]]:
// OMP50: br label {{%?}}[[EXIT]]
// OMP50: [[EXIT]]:
[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
// CHECK-DAG: [[R_VAL:%.+]] = load i32, i32* [[R]],
// CHECK-DAG: [[R_PRIV_VAL:%.+]] = load i32, i32* [[R_PRIV]],
// CHECK: [[RED:%.+]] = mul nsw i32 [[R_VAL]], [[R_PRIV_VAL]]
// CHECK-NEXT: store i32 [[RED]], i32* [[R]],
// CHECK-NEXT: ret void
}
template <class T, unsigned K> T tfoo(T a) { return a + K; }
template <typename T, unsigned N>
int templ1(T a, T *z) {
#pragma omp simd collapse(N)
for (int i = 0; i < N * 2; i++) {
for (long long j = 0; j < (N + N + N + N); j += 2) {
z[i + j] = a + tfoo<T, N>(i + j);
}
}
return 0;
}
// Instatiation templ1<float,2>
// CHECK-LABEL: define {{.*i32}} @{{.*}}templ1{{.*}}(float {{.+}}, float* {{.+}})
// CHECK: store i64 0, i64* [[T1_OMP_IV:[^,]+]]
// ...
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[CMP1:%.+]] = icmp slt i64 [[IV]], 16
// CHECK-NEXT: br i1 [[CMP1]], label %[[T1_BODY:.+]], label %[[T1_END:[^,]+]]
// CHECK: [[T1_BODY]]:
// Loop counters i and j updates:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV1:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[I_1:%.+]] = sdiv i64 [[IV1]], 4
// CHECK-NEXT: [[I_1_MUL1:%.+]] = mul nsw i64 [[I_1]], 1
// CHECK-NEXT: [[I_1_ADD0:%.+]] = add nsw i64 0, [[I_1_MUL1]]
// CHECK-NEXT: [[I_2:%.+]] = trunc i64 [[I_1_ADD0]] to i32
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i32 [[I_2]], i32* {{%.+}}{{.*}}!llvm.access.group
// CHECK: [[IV2:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
[OpenMP] Avoid remainder operations for loop index values on a collapsed loop nest. Summary: Change the strategy for computing loop index variables after collapsing a loop nest via the collapse clause by replacing the expensive remainder operation with multiplications and additions. Reviewers: ABataev, caomhin Reviewed By: ABataev Subscribers: guansong, arphaman, cfe-commits Differential Revision: https://reviews.llvm.org/D56413 llvm-svn: 350759
2019-01-10 04:45:26 +08:00
// CHECK: [[IV2_1:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
// CHECK-NEXT: [[J_1_DIV1:%.+]] = sdiv i64 [[IV2_1]], 4
// CHECK-NEXT: [[J_1_MUL1:%.+]] = mul nsw i64 [[J_1_DIV1]], 4
// CHECK-NEXT: [[J_1_SUB0:%.+]] = sub nsw i64 [[IV2]], [[J_1_MUL1]]
// CHECK-NEXT: [[J_2:%.+]] = mul nsw i64 [[J_1_SUB0]], 2
[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
// CHECK-NEXT: [[J_2_ADD0:%.+]] = add nsw i64 0, [[J_2]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i64 [[J_2_ADD0]], i64* {{%.+}}{{.*}}!llvm.access.group
[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
// simd.for.inc:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV3:%.+]] = load i64, i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[INC:%.+]] = add nsw i64 [[IV3]], 1
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i64 [[INC]], i64* [[T1_OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: br label {{%.+}}
// CHECK: [[T1_END]]:
// CHECK: ret i32 0
//
void inst_templ1() {
float a;
float z[100];
templ1<float,2> (a, z);
}
typedef int MyIdx;
class IterDouble {
double *Ptr;
public:
IterDouble operator++ () const {
IterDouble n;
n.Ptr = Ptr + 1;
return n;
}
bool operator < (const IterDouble &that) const {
return Ptr < that.Ptr;
}
double & operator *() const {
return *Ptr;
}
MyIdx operator - (const IterDouble &that) const {
return (MyIdx) (Ptr - that.Ptr);
}
IterDouble operator + (int Delta) {
IterDouble re;
re.Ptr = Ptr + Delta;
return re;
}
///~IterDouble() {}
};
// CHECK-LABEL: define {{.*void}} @{{.*}}iter_simple{{.*}}
void iter_simple(IterDouble ia, IterDouble ib, IterDouble ic) {
//
// Calculate number of iterations before the loop body.
// CHECK: [[DIFF1:%.+]] = invoke {{.*}}i32 @{{.*}}IterDouble{{.*}}
// CHECK: [[DIFF2:%.+]] = sub nsw i32 [[DIFF1]], 1
// CHECK-NEXT: [[DIFF3:%.+]] = add nsw i32 [[DIFF2]], 1
// CHECK-NEXT: [[DIFF4:%.+]] = sdiv i32 [[DIFF3]], 1
// CHECK-NEXT: [[DIFF5:%.+]] = sub nsw i32 [[DIFF4]], 1
// CHECK-NEXT: store i32 [[DIFF5]], i32* [[OMP_LAST_IT:%[^,]+]]{{.+}}
// CHECK: store i32 0, i32* [[IT_OMP_IV:%[^,]+]]
#pragma omp simd
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}} !llvm.access.group
// CHECK-NEXT: [[LAST_IT:%.+]] = load i32, i32* [[OMP_LAST_IT]]{{.+}}!llvm.access.group
[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
// CHECK-NEXT: [[NUM_IT:%.+]] = add nsw i32 [[LAST_IT]], 1
// CHECK-NEXT: [[CMP:%.+]] = icmp slt i32 [[IV]], [[NUM_IT]]
// CHECK-NEXT: br i1 [[CMP]], label %[[IT_BODY:[^,]+]], label %[[IT_END:[^,]+]]
for (IterDouble i = ia; i < ib; ++i) {
// CHECK: [[IT_BODY]]:
// Start of body: calculate i from index:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV1:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}}!llvm.access.group
[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
// Call of operator+ (i, IV).
// CHECK: {{%.+}} = invoke {{.+}} @{{.*}}IterDouble{{.*}}
// ... loop body ...
*i = *ic * 0.5;
// Float multiply and save result.
// CHECK: [[MULR:%.+]] = fmul double {{%.+}}, 5.000000e-01
// CHECK-NEXT: invoke {{.+}} @{{.*}}IterDouble{{.*}}
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: store double [[MULR:%.+]], double* [[RESULT_ADDR:%.+]], !llvm.access.group
[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
++ic;
//
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV2:%.+]] = load i32, i32* [[IT_OMP_IV]]{{.+}}!llvm.access.group
[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
// CHECK-NEXT: [[ADD2:%.+]] = add nsw i32 [[IV2]], 1
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i32 [[ADD2]], i32* [[IT_OMP_IV]]{{.+}}!llvm.access.group
[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
// br label %{{.*}}, !llvm.loop ![[ITER_LOOP_ID]]
}
// CHECK: [[IT_END]]:
// CHECK: ret void
}
// CHECK-LABEL: define {{.*void}} @{{.*}}collapsed{{.*}}
void collapsed(float *a, float *b, float *c, float *d) {
int i; // outer loop counter
unsigned j; // middle loop couter, leads to unsigned icmp in loop header.
// k declared in the loop init below
short l; // inner loop counter
// CHECK: store i32 0, i32* [[OMP_IV:[^,]+]]
//
#pragma omp simd collapse(4)
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
[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
// CHECK-NEXT: [[CMP:%.+]] = icmp ult i32 [[IV]], 120
// CHECK-NEXT: br i1 [[CMP]], label %[[COLL1_BODY:[^,]+]], label %[[COLL1_END:[^,]+]]
for (i = 1; i < 3; i++) // 2 iterations
for (j = 2u; j < 5u; j++) //3 iterations
for (int k = 3; k <= 6; k++) // 4 iterations
for (l = 4; l < 9; ++l) // 5 iterations
{
// CHECK: [[COLL1_BODY]]:
// Start of body: calculate i from index:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
[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
// Calculation of the loop counters values.
// CHECK: [[CALC_I_1:%.+]] = udiv i32 [[IV1]], 60
// CHECK-NEXT: [[CALC_I_1_MUL1:%.+]] = mul i32 [[CALC_I_1]], 1
// CHECK-NEXT: [[CALC_I_2:%.+]] = add i32 1, [[CALC_I_1_MUL1]]
// CHECK-NEXT: store i32 [[CALC_I_2]], i32* [[LC_I:.+]]
[OpenMP] Avoid remainder operations for loop index values on a collapsed loop nest. Summary: Change the strategy for computing loop index variables after collapsing a loop nest via the collapse clause by replacing the expensive remainder operation with multiplications and additions. Reviewers: ABataev, caomhin Reviewed By: ABataev Subscribers: guansong, arphaman, cfe-commits Differential Revision: https://reviews.llvm.org/D56413 llvm-svn: 350759
2019-01-10 04:45:26 +08:00
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV1_2:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
[OpenMP] Avoid remainder operations for loop index values on a collapsed loop nest. Summary: Change the strategy for computing loop index variables after collapsing a loop nest via the collapse clause by replacing the expensive remainder operation with multiplications and additions. Reviewers: ABataev, caomhin Reviewed By: ABataev Subscribers: guansong, arphaman, cfe-commits Differential Revision: https://reviews.llvm.org/D56413 llvm-svn: 350759
2019-01-10 04:45:26 +08:00
// CHECK: [[IV1_2_1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
// CHECK-NEXT: [[CALC_J_1:%.+]] = udiv i32 [[IV1_2_1]], 60
// CHECK-NEXT: [[MUL_1:%.+]] = mul i32 [[CALC_J_1]], 60
// CHECK-NEXT: [[SUB_3:%.+]] = sub i32 [[IV1_2]], [[MUL_1]]
// CHECK-NEXT: [[CALC_J_2:%.+]] = udiv i32 [[SUB_3]], 20
[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
// CHECK-NEXT: [[CALC_J_2_MUL1:%.+]] = mul i32 [[CALC_J_2]], 1
// CHECK-NEXT: [[CALC_J_3:%.+]] = add i32 2, [[CALC_J_2_MUL1]]
// CHECK-NEXT: store i32 [[CALC_J_3]], i32* [[LC_J:.+]]
[OpenMP] Avoid remainder operations for loop index values on a collapsed loop nest. Summary: Change the strategy for computing loop index variables after collapsing a loop nest via the collapse clause by replacing the expensive remainder operation with multiplications and additions. Reviewers: ABataev, caomhin Reviewed By: ABataev Subscribers: guansong, arphaman, cfe-commits Differential Revision: https://reviews.llvm.org/D56413 llvm-svn: 350759
2019-01-10 04:45:26 +08:00
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV1_3:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
[OpenMP] Avoid remainder operations for loop index values on a collapsed loop nest. Summary: Change the strategy for computing loop index variables after collapsing a loop nest via the collapse clause by replacing the expensive remainder operation with multiplications and additions. Reviewers: ABataev, caomhin Reviewed By: ABataev Subscribers: guansong, arphaman, cfe-commits Differential Revision: https://reviews.llvm.org/D56413 llvm-svn: 350759
2019-01-10 04:45:26 +08:00
// CHECK: [[IV1_3_1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
// CHECK-NEXT: [[DIV_1:%.+]] = udiv i32 [[IV1_3_1]], 60
// CHECK-NEXT: [[MUL_2:%.+]] = mul i32 [[DIV_1]], 60
// CHECK-NEXT: [[ADD_3:%.+]] = sub i32 [[IV1_3]], [[MUL_2]]
// CHECK: [[IV1_4:%.+]] = load i32, i32* [[OMP_IV]]
// CHECK: [[IV1_4_1:%.+]] = load i32, i32* [[OMP_IV]]
// CHECK-NEXT: [[DIV_2:%.+]] = udiv i32 [[IV1_4_1]], 60
// CHECK-NEXT: [[MUL_3:%.+]] = mul i32 [[DIV_2]], 60
// CHECK-NEXT: [[SUB_6:%.+]] = sub i32 [[IV1_4]], [[MUL_3]]
// CHECK-NEXT: [[DIV_3:%.+]] = udiv i32 [[SUB_6]], 20
// CHECK-NEXT: [[MUL_4:%.+]] = mul i32 [[DIV_3]], 20
// CHECK-NEXT: [[ADD_5:%.+]] = sub i32 [[ADD_3]], [[MUL_4]]
// CHECK-NEXT: [[DIV_4:%.+]] = udiv i32 [[ADD_5]], 5
// CHECK-NEXT: [[MUL_5:%.+]] = mul i32 [[DIV_4]], 1
// CHECK-NEXT: [[ADD_6:%.+]] = add i32 3, [[MUL_5]]
// CHECK-NEXT: store i32 [[ADD_6]], i32* [[LC_K:.+]]
// CHECK: [[IV1_5:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
// CHECK: [[IV1_5_1:%.+]] = load i32, i32* [[OMP_IV]]{{.+}}!llvm.access.group
// CHECK-NEXT: [[DIV_5:%.+]] = udiv i32 [[IV1_5_1]], 60
// CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[DIV_5]], 60
// CHECK-NEXT: [[SUB_7:%.+]] = sub i32 [[IV1_5]], [[MUL_6]]
// CHECK: [[IV1_6:%.+]] = load i32, i32* [[OMP_IV]]
// CHECK: [[IV1_6_1:%.+]] = load i32, i32* [[OMP_IV]]
// CHECK-NEXT: [[DIV_6:%.+]] = udiv i32 [[IV1_6_1]], 60
// CHECK-NEXT: [[MUL_7:%.+]] = mul i32 [[DIV_6]], 60
// CHECK-NEXT: [[SUB_10:%.+]] = sub i32 [[IV1_6]], [[MUL_7]]
// CHECK-NEXT: [[DIV_7:%.+]] = udiv i32 [[SUB_10]], 20
// CHECK-NEXT: [[MUL_8:%.+]] = mul i32 [[DIV_7]], 20
// CHECK-NEXT: [[SUB_11:%.+]] = sub i32 [[SUB_7]], [[MUL_8]]
// CHECK: [[IV1_7:%.+]] = load i32, i32* [[OMP_IV]]
// CHECK: [[IV1_7_1:%.+]] = load i32, i32* [[OMP_IV]]
// CHECK-NEXT: [[DIV_8:%.+]] = udiv i32 [[IV1_7_1]], 60
// CHECK-NEXT: [[MUL_9:%.+]] = mul i32 [[DIV_8]], 60
// CHECK-NEXT: [[SUB_12:%.+]] = sub i32 [[IV1_7]], [[MUL_9]]
// CHECK: [[IV1_8:%.+]] = load i32, i32* [[OMP_IV]]
// CHECK: [[IV1_8_1:%.+]] = load i32, i32* [[OMP_IV]]
// CHECK-NEXT: [[DIV_3:%.+]] = udiv i32 [[IV1_8_1]], 60
// CHECK-NEXT: [[MUL_4:%.+]] = mul i32 [[DIV_3]], 60
// CHECK-NEXT: [[SUB_7:%.+]] = sub i32 [[IV1_8]], [[MUL_4]]
// CHECK-NEXT: [[DIV_4:%.+]] = udiv i32 [[SUB_7]], 20
// CHECK-NEXT: [[MUL_5:%.+]] = mul i32 [[DIV_4]], 20
// CHECK-NEXT: [[SUB_8:%.+]] = sub i32 [[SUB_12]], [[MUL_5]]
// CHECK-NEXT: [[DIV_5:%.+]] = udiv i32 [[SUB_8]], 5
// CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[DIV_5]], 5
// CHECK-NEXT: [[SUB_9:%.+]] = sub i32 [[SUB_11]], [[MUL_6]]
// CHECK-NEXT: [[MUL_6:%.+]] = mul i32 [[SUB_9]], 1
// CHECK-NEXT: [[CALC_L_2:%.+]] = add i32 4, [[MUL_6]]
[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
// CHECK-NEXT: [[CALC_L_3:%.+]] = trunc i32 [[CALC_L_2]] to i16
// CHECK-NEXT: store i16 [[CALC_L_3]], i16* [[LC_L:.+]]
// ... loop body ...
// End of body: store into a[i]:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: store float [[RESULT:%.+]], float* [[RESULT_ADDR:%.+]]{{.+}}!llvm.access.group
[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
float res = b[j] * c[k];
a[i] = res * d[l];
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[ADD2:%.+]] = add i32 [[IV2]], 1
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i32 [[ADD2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// br label %{{[^,]+}}, !llvm.loop ![[COLL1_LOOP_ID]]
// CHECK: [[COLL1_END]]:
}
// i,j,l are updated; k is not updated.
// CHECK: store i32 3, i32*
// CHECK-NEXT: store i32 5, i32*
// CHECK-NEXT: store i32 7, i32*
// CHECK-NEXT: store i16 9, i16*
// CHECK: ret void
}
extern char foo();
extern double globalfloat;
// CHECK-LABEL: define {{.*void}} @{{.*}}widened{{.*}}
void widened(float *a, float *b, float *c, float *d) {
int i; // outer loop counter
short j; // inner loop counter
globalfloat = 1.0;
int localint = 1;
// CHECK: store double {{.+}}, double* [[GLOBALFLOAT:@.+]]
// Counter is widened to 64 bits.
// CHECK: store i64 0, i64* [[OMP_IV:[^,]+]]
//
#pragma omp simd collapse(2) private(globalfloat, localint)
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.access.group
// CHECK-NEXT: [[LI:%.+]] = load i64, i64* [[OMP_LI:%[^,]+]]{{.+}}!llvm.access.group
[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
// CHECK-NEXT: [[NUMIT:%.+]] = add nsw i64 [[LI]], 1
// CHECK-NEXT: [[CMP:%.+]] = icmp slt i64 [[IV]], [[NUMIT]]
// CHECK-NEXT: br i1 [[CMP]], label %[[WIDE1_BODY:[^,]+]], label %[[WIDE1_END:[^,]+]]
for (i = 1; i < 3; i++) // 2 iterations
for (j = 0; j < foo(); j++) // foo() iterations
{
// CHECK: [[WIDE1_BODY]]:
// Start of body: calculate i from index:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV1:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.access.group
[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
// Calculation of the loop counters values...
// CHECK: store i32 {{[^,]+}}, i32* [[LC_I:.+]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV1_2:%.+]] = load i64, i64* [[OMP_IV]]{{.+}}!llvm.access.group
[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
// CHECK: store i16 {{[^,]+}}, i16* [[LC_J:.+]]
// ... loop body ...
//
// Here we expect store into private double var, not global
// CHECK-NOT: store double {{.+}}, double* [[GLOBALFLOAT]]
globalfloat = (float)j/i;
float res = b[j] * c[j];
// Store into a[i]:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: store float [[RESULT:%.+]], float* [[RESULT_ADDR:%.+]]{{.+}}!llvm.access.group
[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
a[i] = res * d[i];
// Then there's a store into private var localint:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: store i32 {{.+}}, i32* [[LOCALINT:%[^,]+]]{{.+}}!llvm.access.group
[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
localint = (int)j;
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV2:%.+]] = load i64, i64* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[ADD2:%.+]] = add nsw i64 [[IV2]], 1
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i64 [[ADD2]], i64* [[OMP_IV]]{{.*}}!llvm.access.group
[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
//
// br label %{{[^,]+}}, !llvm.loop ![[WIDE1_LOOP_ID]]
// CHECK: [[WIDE1_END]]:
}
// i,j are updated.
// CHECK: store i32 3, i32* [[I:%[^,]+]]
// CHECK: store i16
//
// Here we expect store into original localint, not its privatized version.
// CHECK-NOT: store i32 {{.+}}, i32* [[LOCALINT]]
localint = (int)j;
// CHECK: ret void
}
// CHECK-LABEL: define {{.*void}} @{{.*}}linear{{.*}}(float* {{.+}})
void linear(float *a) {
// CHECK: [[VAL_ADDR:%.+]] = alloca i64,
// CHECK: [[K_ADDR:%.+]] = alloca i64*,
long long val = 0;
long long &k = val;
#pragma omp simd linear(k : 3)
// CHECK: store i64* [[VAL_ADDR]], i64** [[K_ADDR]],
// CHECK: [[VAL_REF:%.+]] = load i64*, i64** [[K_ADDR]],
// CHECK: store i64* [[VAL_REF]], i64** [[K_ADDR_REF:%.+]],
// CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]]
// CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]],
// CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_REF]]
// CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9
// CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]]
for (int i = 10; i > 1; i--) {
// CHECK: [[SIMPLE_LOOP_BODY]]:
// Start of body: calculate i from IV:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// FIXME: It is interesting, why the following "mul 1" was not constant folded?
// CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1
// CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
[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
//
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
// CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3
// CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
// CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
// Update of the privatized version of linear variable!
// CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
a[k]++;
k = k + 3;
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]]
}
// CHECK: [[SIMPLE_LOOP_END]]:
//
// Update linear vars after loop, as the loop was operating on a private version.
// CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]],
// CHECK: store i64* [[K_REF]], i64** [[K_PRIV_REF:%.+]],
// CHECK: [[LIN0_2:%.+]] = load i64, i64* [[LIN0]]
// CHECK-NEXT: [[LIN_ADD2:%.+]] = add nsw i64 [[LIN0_2]], 27
// CHECK-NEXT: [[K_REF:%.+]] = load i64*, i64** [[K_PRIV_REF]],
// CHECK-NEXT: store i64 [[LIN_ADD2]], i64* [[K_REF]]
//
#pragma omp simd linear(val(k) : 3)
// CHECK: [[VAL_REF:%.+]] = load i64*, i64** [[K_ADDR]],
// CHECK: store i64* [[VAL_REF]], i64** [[K_ADDR_REF:%.+]],
// CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]]
// CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]],
// CHECK: [[K0LOAD:%.+]] = load i64, i64* [[K_REF]]
// CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9
// CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]]
for (int i = 10; i > 1; i--) {
// CHECK: [[SIMPLE_LOOP_BODY]]:
// Start of body: calculate i from IV:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// FIXME: It is interesting, why the following "mul 1" was not constant folded?
// CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1
// CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
[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
//
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
// CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3
// CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
// CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
// Update of the privatized version of linear variable!
// CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
a[k]++;
k = k + 3;
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]]
}
// CHECK: [[SIMPLE_LOOP_END]]:
//
// Update linear vars after loop, as the loop was operating on a private version.
// CHECK: [[K_REF:%.+]] = load i64*, i64** [[K_ADDR_REF]],
// CHECK: store i64* [[K_REF]], i64** [[K_PRIV_REF:%.+]],
// CHECK: [[LIN0_2:%.+]] = load i64, i64* [[LIN0]]
// CHECK-NEXT: [[LIN_ADD2:%.+]] = add nsw i64 [[LIN0_2]], 27
// CHECK-NEXT: [[K_REF:%.+]] = load i64*, i64** [[K_PRIV_REF]],
// CHECK-NEXT: store i64 [[LIN_ADD2]], i64* [[K_REF]]
//
#pragma omp simd linear(uval(k) : 3)
// CHECK: store i32 0, i32* [[OMP_IV:%[^,]+]]
// CHECK: [[K0LOAD:%.+]] = load i64, i64* [[VAL_ADDR]]
// CHECK-NEXT: store i64 [[K0LOAD]], i64* [[LIN0:%[^,]+]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[CMP2:%.+]] = icmp slt i32 [[IV]], 9
// CHECK-NEXT: br i1 [[CMP2]], label %[[SIMPLE_LOOP_BODY:.+]], label %[[SIMPLE_LOOP_END:[^,]+]]
for (int i = 10; i > 1; i--) {
// CHECK: [[SIMPLE_LOOP_BODY]]:
// Start of body: calculate i from IV:
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV_0:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// FIXME: It is interesting, why the following "mul 1" was not constant folded?
// CHECK-NEXT: [[IV_1:%.+]] = mul nsw i32 [[IV_0]], 1
// CHECK-NEXT: [[LC_I_1:%.+]] = sub nsw i32 10, [[IV_1]]
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i32 [[LC_I_1]], i32* {{.+}}, !llvm.access.group
[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
//
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: [[LIN0_1:%.+]] = load i64, i64* [[LIN0]]{{.*}}!llvm.access.group
// CHECK-NEXT: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[LIN_MUL1:%.+]] = mul nsw i32 [[IV_2]], 3
// CHECK-NEXT: [[LIN_EXT1:%.+]] = sext i32 [[LIN_MUL1]] to i64
// CHECK-NEXT: [[LIN_ADD1:%.+]] = add nsw i64 [[LIN0_1]], [[LIN_EXT1]]
// Update of the privatized version of linear variable!
// CHECK-NEXT: store i64 [[LIN_ADD1]], i64* [[K_PRIVATIZED:%[^,]+]]
a[k]++;
k = k + 3;
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK: [[IV_2:%.+]] = load i32, i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// CHECK-NEXT: [[ADD2_2:%.+]] = add nsw i32 [[IV_2]], 1
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// CHECK-NEXT: store i32 [[ADD2_2]], i32* [[OMP_IV]]{{.*}}!llvm.access.group
[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
// br label {{.+}}, !llvm.loop ![[SIMPLE_LOOP_ID]]
}
// CHECK: [[SIMPLE_LOOP_END]]:
//
// Update linear vars after loop, as the loop was operating on a private version.
// CHECK: [[LIN0_2:%.+]] = load i64, i64* [[LIN0]]
// CHECK-NEXT: [[LIN_ADD2:%.+]] = add nsw i64 [[LIN0_2]], 27
// CHECK-NEXT: store i64 [[LIN_ADD2]], i64* [[VAL_ADDR]]
//
}
[OPENMP50]Codegen for nontemporal clause. Summary: Basic codegen for the declarations marked as nontemporal. Also, if the base declaration in the member expression is marked as nontemporal, lvalue for member decl access inherits nonteporal flag from the base lvalue. Reviewers: rjmccall, hfinkel, jdoerfert Subscribers: guansong, arphaman, caomhin, kkwli0, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D71708
2019-12-19 23:01:10 +08:00
#ifdef OMP5
// OMP50-LABEL: inner_simd
void inner_simd() {
double a, b;
#pragma omp simd nontemporal(a)
for (int i = 0; i < 10; ++i) {
#pragma omp simd nontemporal(b)
for (int k = 0; k < 10; ++k) {
// OMP50: load double,{{.*}}!nontemporal
// OMP50: store double{{.*}}!nontemporal
a = b;
}
// OMP50-NOT: load double,{{.*}}!nontemporal
// OMP50: load double,
// OMP50: store double{{.*}}!nontemporal
a = b;
}
}
extern struct T t;
struct Base {
float a;
};
struct T : public Base {
void foo() {
#pragma omp simd nontemporal(Base::a)
for (int i = 0; i < 10; ++i) {
// OMP50: store float{{.*}}!nontemporal
// OMP50-NOT: nontemporal
// OMP50-NEXT: store float
Base::a = 0;
t.a = 0;
}
}
} t;
void bartfoo() {
t.foo();
}
#endif // OMP5
[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
// TERM_DEBUG-LABEL: bar
int bar() {return 0;};
// TERM_DEBUG-LABEL: parallel_simd
void parallel_simd(float *a) {
#pragma omp parallel
#pragma omp simd
// TERM_DEBUG-NOT: __kmpc_global_thread_num
// TERM_DEBUG: invoke i32 {{.*}}bar{{.*}}()
[CodeGen] Generate llvm.loop.parallel_accesses instead of llvm.mem.parallel_loop_access metadata. Instead of generating llvm.mem.parallel_loop_access metadata, generate llvm.access.group on instructions and llvm.loop.parallel_accesses on loops. There is one access group per generated loop. This is clang part of D52116/r349725. Differential Revision: https://reviews.llvm.org/D52117 llvm-svn: 349823
2018-12-21 05:24:54 +08:00
// TERM_DEBUG: unwind label %[[TERM_LPAD:[^,]+]],
[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
// TERM_DEBUG-NOT: __kmpc_global_thread_num
// TERM_DEBUG: [[TERM_LPAD]]
// TERM_DEBUG: call void @__clang_call_terminate
// TERM_DEBUG: unreachable
for (unsigned i = 131071; i <= 2147483647; i += 127)
a[i] += bar();
}
// TERM_DEBUG: !{{[0-9]+}} = !DILocation(line: [[@LINE-11]],
// CHECK-LABEL: S8
// CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64
// CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64
// CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64
// CHECK-DAG: ptrtoint [[SS_TY]]* %{{.+}} to i64
// CHECK-DAG: and i64 %{{.+}}, 15
// CHECK-DAG: icmp eq i64 %{{.+}}, 0
// CHECK-DAG: call void @llvm.assume(i1
// CHECK-DAG: and i64 %{{.+}}, 7
// CHECK-DAG: icmp eq i64 %{{.+}}, 0
// CHECK-DAG: call void @llvm.assume(i1
// CHECK-DAG: and i64 %{{.+}}, 15
// CHECK-DAG: icmp eq i64 %{{.+}}, 0
// CHECK-DAG: call void @llvm.assume(i1
// CHECK-DAG: and i64 %{{.+}}, 3
// CHECK-DAG: icmp eq i64 %{{.+}}, 0
// CHECK-DAG: call void @llvm.assume(i1
struct SS {
SS(): a(0) {}
SS(int v) : a(v) {}
int a;
typedef int type;
};
template <typename T>
class S7 : public T {
protected:
T *a;
T b[2];
S7() : a(0) {}
public:
S7(typename T::type &v) : a((T*)&v) {
#pragma omp simd aligned(a)
for (int k = 0; k < a->a; ++k)
++this->a->a;
#pragma omp simd aligned(this->b : 8)
for (int k = 0; k < a->a; ++k)
++a->a;
}
};
class S8 : private IterDouble, public S7<SS> {
S8() {}
public:
S8(int v) : S7<SS>(v){
#pragma omp parallel private(a)
#pragma omp simd aligned(S7<SS>::a)
for (int k = 0; k < a->a; ++k)
++this->a->a;
#pragma omp parallel shared(b)
#pragma omp simd aligned(this->b: 4)
for (int k = 0; k < a->a; ++k)
++a->a;
}
};
S8 s8(0);
[OPENMP] Remove more empty SourceLocations() from the code. Removed more empty SourceLocations() from the OpenMP code and replaced with the correct locations for better debug info emission. llvm-svn: 323232
2018-01-24 02:12:38 +08:00
// TERM_DEBUG-NOT: line: 0,
// TERM_DEBUG: distinct !DISubprogram(linkageName: "_GLOBAL__sub_I_simd_codegen.cpp",
[OPENMP50]Add if clause in simd directive. According to OpenMP 5.0, if clause can be used in simd directive. If condition in the if clause if false, the non-vectorized version of the loop must be executed.
2019-11-20 01:07:54 +08:00
// OMP50-DAG: ![[NOVECT:.+]] = !{!"llvm.loop.vectorize.enable", i1 false}
// OMP50-DAG: ![[DISABLE_VECT]] = distinct !{{.*}}![[NOVECT]]{{[,}]}}
[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
#endif // HEADER