mirrors/llvm-project
0
0
Fork 0
mirror of https://github.com/llvm/llvm-project.git synced 2025-04-27 00:06:05 +00:00
Code Issues Projects Releases Wiki Activity
llvm-project/clang/test/OpenMP/target_parallel_codegen.cpp

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

3281 lines
231 KiB
C++
Raw Normal View History

[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --function-signature --include-generated-funcs --replace-value-regex "__omp_offloading_[0-9a-z]+_[0-9a-z]+" "reduction_size[.].+[.]" "pl_cond[.].+[.|,]" --prefix-filecheck-ir-name _
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
// Test host codegen.
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK1
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK3
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK3
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
[OPENMP] Support for -fopenmp-simd option with compilation of simd loops only. Added support for -fopenmp-simd option that allows compilation of simd-based constructs without emission of OpenMP runtime calls. llvm-svn: 321560
2017-12-29 18:07:07 +00:00
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
// Test target codegen - host bc file has to be created first.
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix=CHECK9
[OpenMP][OMPIRBuilder] Rename IsEmbedded and IsTargetCodegen flags This patch renames the `OpenMPIRBuilderConfig` flags to reduce confusion over their meaning. `IsTargetCodegen` becomes `IsGPU`, whereas `IsEmbedded` becomes `IsTargetDevice`. The `-fopenmp-is-device` compiler option is also renamed to `-fopenmp-is-target-device` and the `omp.is_device` MLIR attribute is renamed to `omp.is_target_device`. Getters and setters of all these renamed properties are also updated accordingly. Many unit tests have been updated to use the new names, but an alias for the `-fopenmp-is-device` option is created so that external programs do not stop working after the name change. `IsGPU` is set when the target triple is AMDGCN or NVIDIA PTX, and it is only valid if `IsTargetDevice` is specified as well. `IsTargetDevice` is set by the `-fopenmp-is-target-device` compiler frontend option, which is only added to the OpenMP device invocation for offloading-enabled programs. Differential Revision: https://reviews.llvm.org/D154591
2023-07-06 10:28:24 +01:00
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK9
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix=CHECK11
[OpenMP][OMPIRBuilder] Rename IsEmbedded and IsTargetCodegen flags This patch renames the `OpenMPIRBuilderConfig` flags to reduce confusion over their meaning. `IsTargetCodegen` becomes `IsGPU`, whereas `IsEmbedded` becomes `IsTargetDevice`. The `-fopenmp-is-device` compiler option is also renamed to `-fopenmp-is-target-device` and the `omp.is_device` MLIR attribute is renamed to `omp.is_target_device`. Getters and setters of all these renamed properties are also updated accordingly. Many unit tests have been updated to use the new names, but an alias for the `-fopenmp-is-device` option is created so that external programs do not stop working after the name change. `IsGPU` is set when the target triple is AMDGCN or NVIDIA PTX, and it is only valid if `IsTargetDevice` is specified as well. `IsTargetDevice` is set by the `-fopenmp-is-target-device` compiler frontend option, which is only added to the OpenMP device invocation for offloading-enabled programs. Differential Revision: https://reviews.llvm.org/D154591
2023-07-06 10:28:24 +01:00
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK11
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
[OpenMP][OMPIRBuilder] Rename IsEmbedded and IsTargetCodegen flags This patch renames the `OpenMPIRBuilderConfig` flags to reduce confusion over their meaning. `IsTargetCodegen` becomes `IsGPU`, whereas `IsEmbedded` becomes `IsTargetDevice`. The `-fopenmp-is-device` compiler option is also renamed to `-fopenmp-is-target-device` and the `omp.is_device` MLIR attribute is renamed to `omp.is_target_device`. Getters and setters of all these renamed properties are also updated accordingly. Many unit tests have been updated to use the new names, but an alias for the `-fopenmp-is-device` option is created so that external programs do not stop working after the name change. `IsGPU` is set when the target triple is AMDGCN or NVIDIA PTX, and it is only valid if `IsTargetDevice` is specified as well. `IsTargetDevice` is set by the `-fopenmp-is-target-device` compiler frontend option, which is only added to the OpenMP device invocation for offloading-enabled programs. Differential Revision: https://reviews.llvm.org/D154591
2023-07-06 10:28:24 +01:00
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
[OpenMP][OMPIRBuilder] Rename IsEmbedded and IsTargetCodegen flags This patch renames the `OpenMPIRBuilderConfig` flags to reduce confusion over their meaning. `IsTargetCodegen` becomes `IsGPU`, whereas `IsEmbedded` becomes `IsTargetDevice`. The `-fopenmp-is-device` compiler option is also renamed to `-fopenmp-is-target-device` and the `omp.is_device` MLIR attribute is renamed to `omp.is_target_device`. Getters and setters of all these renamed properties are also updated accordingly. Many unit tests have been updated to use the new names, but an alias for the `-fopenmp-is-device` option is created so that external programs do not stop working after the name change. `IsGPU` is set when the target triple is AMDGCN or NVIDIA PTX, and it is only valid if `IsTargetDevice` is specified as well. `IsTargetDevice` is set by the `-fopenmp-is-target-device` compiler frontend option, which is only added to the OpenMP device invocation for offloading-enabled programs. Differential Revision: https://reviews.llvm.org/D154591
2023-07-06 10:28:24 +01:00
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp-simd -fopenmp-version=45 -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
[OpenMP] Ensure testing for versions 4.5 and default - Part 1 Many OpenMP Clang tests do not RUN for version 4.5 and the default version. This first patch in the series only handles test cases which do not require any modifications in the CHECK lines after adding RUN lines for default version. Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D84844
2020-08-13 07:27:15 +00:00
// Test host codegen.
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK1
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --check-prefix=CHECK3
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK3
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
[OpenMP] Ensure testing for versions 4.5 and default - Part 1 Many OpenMP Clang tests do not RUN for version 4.5 and the default version. This first patch in the series only handles test cases which do not require any modifications in the CHECK lines after adding RUN lines for default version. Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D84844
2020-08-13 07:27:15 +00:00
// Test target codegen - host bc file has to be created first.
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --check-prefix=CHECK9
[OpenMP][OMPIRBuilder] Rename IsEmbedded and IsTargetCodegen flags This patch renames the `OpenMPIRBuilderConfig` flags to reduce confusion over their meaning. `IsTargetCodegen` becomes `IsGPU`, whereas `IsEmbedded` becomes `IsTargetDevice`. The `-fopenmp-is-device` compiler option is also renamed to `-fopenmp-is-target-device` and the `omp.is_device` MLIR attribute is renamed to `omp.is_target_device`. Getters and setters of all these renamed properties are also updated accordingly. Many unit tests have been updated to use the new names, but an alias for the `-fopenmp-is-device` option is created so that external programs do not stop working after the name change. `IsGPU` is set when the target triple is AMDGCN or NVIDIA PTX, and it is only valid if `IsTargetDevice` is specified as well. `IsTargetDevice` is set by the `-fopenmp-is-target-device` compiler frontend option, which is only added to the OpenMP device invocation for offloading-enabled programs. Differential Revision: https://reviews.llvm.org/D154591
2023-07-06 10:28:24 +01:00
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK9
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --check-prefix=CHECK11
[OpenMP][OMPIRBuilder] Rename IsEmbedded and IsTargetCodegen flags This patch renames the `OpenMPIRBuilderConfig` flags to reduce confusion over their meaning. `IsTargetCodegen` becomes `IsGPU`, whereas `IsEmbedded` becomes `IsTargetDevice`. The `-fopenmp-is-device` compiler option is also renamed to `-fopenmp-is-target-device` and the `omp.is_device` MLIR attribute is renamed to `omp.is_target_device`. Getters and setters of all these renamed properties are also updated accordingly. Many unit tests have been updated to use the new names, but an alias for the `-fopenmp-is-device` option is created so that external programs do not stop working after the name change. `IsGPU` is set when the target triple is AMDGCN or NVIDIA PTX, and it is only valid if `IsTargetDevice` is specified as well. `IsTargetDevice` is set by the `-fopenmp-is-target-device` compiler frontend option, which is only added to the OpenMP device invocation for offloading-enabled programs. Differential Revision: https://reviews.llvm.org/D154591
2023-07-06 10:28:24 +01:00
// RUN: %clang_cc1 -fopenmp -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --check-prefix=CHECK11
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm-bc %s -o %t-ppc-host.bc
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
[OpenMP][OMPIRBuilder] Rename IsEmbedded and IsTargetCodegen flags This patch renames the `OpenMPIRBuilderConfig` flags to reduce confusion over their meaning. `IsTargetCodegen` becomes `IsGPU`, whereas `IsEmbedded` becomes `IsTargetDevice`. The `-fopenmp-is-device` compiler option is also renamed to `-fopenmp-is-target-device` and the `omp.is_device` MLIR attribute is renamed to `omp.is_target_device`. Getters and setters of all these renamed properties are also updated accordingly. Many unit tests have been updated to use the new names, but an alias for the `-fopenmp-is-device` option is created so that external programs do not stop working after the name change. `IsGPU` is set when the target triple is AMDGCN or NVIDIA PTX, and it is only valid if `IsTargetDevice` is specified as well. `IsTargetDevice` is set by the `-fopenmp-is-target-device` compiler frontend option, which is only added to the OpenMP device invocation for offloading-enabled programs. Differential Revision: https://reviews.llvm.org/D154591
2023-07-06 10:28:24 +01:00
// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple powerpc64le-unknown-unknown -fopenmp-targets=powerpc64le-ibm-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-ppc-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm-bc %s -o %t-x86-host.bc
// RUN: %clang_cc1 -verify -Wno-vla -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-llvm %s -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
[OpenMP][OMPIRBuilder] Rename IsEmbedded and IsTargetCodegen flags This patch renames the `OpenMPIRBuilderConfig` flags to reduce confusion over their meaning. `IsTargetCodegen` becomes `IsGPU`, whereas `IsEmbedded` becomes `IsTargetDevice`. The `-fopenmp-is-device` compiler option is also renamed to `-fopenmp-is-target-device` and the `omp.is_device` MLIR attribute is renamed to `omp.is_target_device`. Getters and setters of all these renamed properties are also updated accordingly. Many unit tests have been updated to use the new names, but an alias for the `-fopenmp-is-device` option is created so that external programs do not stop working after the name change. `IsGPU` is set when the target triple is AMDGCN or NVIDIA PTX, and it is only valid if `IsTargetDevice` is specified as well. `IsTargetDevice` is set by the `-fopenmp-is-target-device` compiler frontend option, which is only added to the OpenMP device invocation for offloading-enabled programs. Differential Revision: https://reviews.llvm.org/D154591
2023-07-06 10:28:24 +01:00
// RUN: %clang_cc1 -fopenmp-simd -x c++ -std=c++11 -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -emit-pch -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -o %t %s
Diagnose use of VLAs in C++ by default Reapplication of 7339c0f782d5c70e0928f8991b0c05338a90c84c with a fix for a crash involving arrays without a size expression. Clang supports VLAs in C++ as an extension, but we currently only warn on their use when you pass -Wvla, -Wvla-extension, or -pedantic. However, VLAs as they're expressed in C have been considered by WG21 and rejected, are easy to use accidentally to the surprise of users (e.g., https://ddanilov.me/default-non-standard-features/), and they have potential security implications beyond constant-size arrays (https://wiki.sei.cmu.edu/confluence/display/c/ARR32-C.+Ensure+size+arguments+for+variable+length+arrays+are+in+a+valid+range). C++ users should strongly consider using other functionality such as std::vector instead. This seems like sufficiently compelling evidence to warn users about VLA use by default in C++ modes. This patch enables the -Wvla-extension diagnostic group in C++ language modes by default, and adds the warning group to -Wall in GNU++ language modes. The warning is still opt-in in C language modes, where support for VLAs is somewhat less surprising to users. RFC: https://discourse.llvm.org/t/rfc-diagnosing-use-of-vlas-in-c/73109 Fixes https://github.com/llvm/llvm-project/issues/62836 Differential Revision: https://reviews.llvm.org/D156565
2023-10-20 13:08:50 -04:00
// RUN: %clang_cc1 -fopenmp-simd -x c++ -triple i386-unknown-unknown -fopenmp-targets=i386-pc-linux-gnu -std=c++11 -fopenmp-is-target-device -fopenmp-host-ir-file-path %t-x86-host.bc -include-pch %t -verify -Wno-vla %s -emit-llvm -o - | FileCheck %s --implicit-check-not="{{__kmpc|__tgt}}"
[OPENMP] Support for -fopenmp-simd option with compilation of simd loops only. Added support for -fopenmp-simd option that allows compilation of simd-based constructs without emission of OpenMP runtime calls. llvm-svn: 321560
2017-12-29 18:07:07 +00:00
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
// expected-no-diagnostics
#ifndef HEADER
#define HEADER
The patch adds support for the new map interface between clang and libomptarget. The changes in the interface are the following: device IDs are now 64-bit integers (as opposed to 32-bit) map flags are 64-bit long (used to be 32-bit) mappings for partially mapped structs are now calculated at compile time and members of partially mapped structs are flagged using the MEMBER_OF field Support for is_device_ptr on struct members was dropped - this functionality is not supported by the OpenMP standard and its implementation is technically infeasible (however, use_device_ptr on struct members works as a non-standard extension of the compiler) llvm-svn: 337468
2018-07-19 16:34:13 +00:00
// We have 8 target regions, but only 6 that actually will generate offloading
// code and have mapped arguments, and only 4 have all-constant map sizes.
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
// Check target registration is registered as a Ctor.
template<typename tx, typename ty>
struct TT{
tx X;
ty Y;
};
int foo(int n) {
int a = 0;
short aa = 0;
float b[10];
float bn[n];
double c[5][10];
double cn[5][n];
TT<long long, char> d;
[OPENMP] Add codegen for `nowait` clause in target directives. Added basic codegen for `nowait` clauses in target-based directives. llvm-svn: 320613
2017-12-13 21:04:20 +00:00
#pragma omp target parallel nowait
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
{
}
#pragma omp target parallel if(target: 0)
{
a += 1;
}
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
#pragma omp target parallel if(target: 1)
{
aa += 1;
[OPENMP]Allow cancellation constructs in target parallel regions. Summary: omp cancellation point parallel and omp cancel parallel directives are allowed in target paralle regions. Reviewers: jdoerfert Subscribers: yaxunl, guansong, caomhin, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D78941
2020-04-27 11:37:35 -04:00
#pragma omp cancel parallel
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
}
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
[OPENMP] Simplify codegen for non-offloading code. Simplified and generalized codegen for non-offloading part that works if offloading is failed or condition of the `if` clause is `false`. llvm-svn: 314670
2017-10-02 14:20:58 +00:00
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
#pragma omp target parallel if(target: n>10)
{
a += 1;
aa += 1;
}
// We capture 3 VLA sizes in this target region
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
// The names below are not necessarily consistent with the names used for the
// addresses above as some are repeated.
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
#pragma omp target parallel if(target: n>20)
{
a += 1;
b[2] += 1.0;
bn[3] += 1.0;
c[1][2] += 1.0;
cn[1][3] += 1.0;
d.X += 1;
d.Y += 1;
}
return a;
}
// Check that the offloading functions are emitted and that the arguments are
// correct and loaded correctly for the target regions in foo().
// Create stack storage and store argument in there.
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
// Create stack storage and store argument in there.
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
// Create stack storage and store argument in there.
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
// Create local storage for each capture.
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
// To reduce complexity, we're only going as far as validating the signature of the outlined parallel function.
template<typename tx>
tx ftemplate(int n) {
tx a = 0;
short aa = 0;
tx b[10];
#pragma omp target parallel if(target: n>40)
{
a += 1;
aa += 1;
b[2] += 1;
}
return a;
}
static
int fstatic(int n) {
int a = 0;
short aa = 0;
char aaa = 0;
int b[10];
#pragma omp target parallel if(target: n>50)
{
a += 1;
aa += 1;
aaa += 1;
b[2] += 1;
}
return a;
}
struct S1 {
double a;
int r1(int n){
int b = n+1;
short int c[2][n];
#pragma omp target parallel if(target: n>60)
{
this->a = (double)b + 1.5;
c[1][1] = ++a;
}
return c[1][1] + (int)b;
}
};
int bar(int n){
int a = 0;
a += foo(n);
S1 S;
a += S.r1(n);
a += fstatic(n);
a += ftemplate<int>(n);
return a;
}
// We capture 2 VLA sizes in this target region
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
// The names below are not necessarily consistent with the names used for the
// addresses above as some are repeated.
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
// Check that the offloading functions are emitted and that the arguments are
// correct and loaded correctly for the target regions of the callees of bar().
// Create local storage for each capture.
// Store captures in the context.
// To reduce complexity, we're only going as far as validating the signature of the outlined parallel function.
// Create local storage for each capture.
// Store captures in the context.
// To reduce complexity, we're only going as far as validating the signature of the outlined parallel function.
// Create local storage for each capture.
// Store captures in the context.
// To reduce complexity, we're only going as far as validating the signature of the outlined parallel function.
[OpenMP] Add support for registering requires directives with the runtime Summary: This patch adds support for the registration of the requires directives with the runtime. Each requires directive clause will enable a particular flag to be set. The set of flags is passed to the runtime to be checked for compatibility with other such flags coming from other object files. The registration function is called whenever OpenMP is present even if a requires directive is not present. This helps detect cases in which requires directives are used inconsistently. Reviewers: ABataev, AlexEichenberger, caomhin Reviewed By: ABataev, AlexEichenberger Subscribers: jholewinski, guansong, jfb, jdoerfert, cfe-commits Tags: #clang Differential Revision: https://reviews.llvm.org/D60568 llvm-svn: 361298
2019-05-21 19:42:01 +00:00
[OpenMP] Codegen support for 'target parallel' on the host. This patch adds support for codegen of 'target parallel' on the host. It is also the first combined directive that requires two or more captured statements. Support for this functionality is included in the patch. A combined directive such as 'target parallel' has two captured statements, one for the 'target' and the other for the 'parallel' region. Two captured statements are required because each has different implicit parameters (see SemaOpenMP.cpp). For example, the 'parallel' has 'global_tid' and 'bound_tid' while the 'target' does not. The patch adds support for handling multiple captured statements based on the combined directive. When codegen'ing the 'target parallel' directive, the 'target' outlined function is created using the outer captured statement and the 'parallel' outlined function is created using the inner captured statement. Reviewers: ABataev Differential Revision: https://reviews.llvm.org/D28753 llvm-svn: 292419
2017-01-18 18:18:53 +00:00
#endif
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-LABEL: define {{[^@]+}}@_Z3fooi
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK1-SAME: (i32 noundef signext [[N:%.*]]) #[[ATTR0:[0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[A:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[AA:%.*]] = alloca i16, align 2
// CHECK1-NEXT: [[B:%.*]] = alloca [10 x float], align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[SAVED_STACK:%.*]] = alloca ptr, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[__VLA_EXPR0:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[C:%.*]] = alloca [5 x [10 x double]], align 8
// CHECK1-NEXT: [[__VLA_EXPR1:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[D:%.*]] = alloca [[STRUCT_TT:%.*]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]], align 1
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AA_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 8
// CHECK1-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 8
// CHECK1-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 8
[Clang][OpenMP] Insert alloca for kernel args at function entry block instead of the launch point. If an inlined kernel is called in a loop, the launch point alloca would lead to increasing stack usage every time the kernel is invoked. This could make the application run out of stack space and crash. This problem is fixed by using the alloca insertion point while creating the alloca instruction. Fixes https://github.com/llvm/llvm-project/issues/60602 Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D145820
2023-03-17 12:16:17 -04:00
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[A_CASTED2:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AA_CASTED3:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[DOTOFFLOAD_BASEPTRS4:%.*]] = alloca [2 x ptr], align 8
// CHECK1-NEXT: [[DOTOFFLOAD_PTRS5:%.*]] = alloca [2 x ptr], align 8
// CHECK1-NEXT: [[DOTOFFLOAD_MAPPERS6:%.*]] = alloca [2 x ptr], align 8
[Clang][OpenMP] Insert alloca for kernel args at function entry block instead of the launch point. If an inlined kernel is called in a loop, the launch point alloca would lead to increasing stack usage every time the kernel is invoked. This could make the application run out of stack space and crash. This problem is fixed by using the alloca insertion point while creating the alloca instruction. Fixes https://github.com/llvm/llvm-project/issues/60602 Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D145820
2023-03-17 12:16:17 -04:00
// CHECK1-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[A_CASTED10:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[DOTOFFLOAD_BASEPTRS13:%.*]] = alloca [9 x ptr], align 8
// CHECK1-NEXT: [[DOTOFFLOAD_PTRS14:%.*]] = alloca [9 x ptr], align 8
// CHECK1-NEXT: [[DOTOFFLOAD_MAPPERS15:%.*]] = alloca [9 x ptr], align 8
Revert "[OPENMP]Fix PR50347: Mapping of global scope deep object fails." This reverts commit 638938117aeae5518d6cacd066ffd9830ef4fc9a. Need to fix reported fail https://lab.llvm.org/buildbot/#/builders/193/builds/7496
2022-02-24 12:04:39 -08:00
// CHECK1-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [9 x i64], align 8
[Clang][OpenMP] Insert alloca for kernel args at function entry block instead of the launch point. If an inlined kernel is called in a loop, the launch point alloca would lead to increasing stack usage every time the kernel is invoked. This could make the application run out of stack space and crash. This problem is fixed by using the alloca insertion point while creating the alloca instruction. Fixes https://github.com/llvm/llvm-project/issues/60602 Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D145820
2023-03-17 12:16:17 -04:00
// CHECK1-NEXT: [[KERNEL_ARGS16:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
// CHECK1-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK1-NEXT: store i32 0, ptr [[A]], align 4
// CHECK1-NEXT: store i16 0, ptr [[AA]], align 2
// CHECK1-NEXT: [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[TMP2:%.*]] = zext i32 [[TMP1]] to i64
Intrinsics: Add type overload to stacksave and stackstore This allows use with non-0 address space stacks. llvm_ptr_ty should never be used. This could use some more percolation up through mlir, but this is enough to fix existing tests. https://reviews.llvm.org/D156666
2023-07-29 09:15:31 -04:00
// CHECK1-NEXT: [[TMP3:%.*]] = call ptr @llvm.stacksave.p0()
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[TMP3]], ptr [[SAVED_STACK]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[VLA:%.*]] = alloca float, i64 [[TMP2]], align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i64 [[TMP2]], ptr [[__VLA_EXPR0]], align 8
// CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[N_ADDR]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[TMP5:%.*]] = zext i32 [[TMP4]] to i64
// CHECK1-NEXT: [[TMP6:%.*]] = mul nuw i64 5, [[TMP5]]
// CHECK1-NEXT: [[VLA1:%.*]] = alloca double, i64 [[TMP6]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i64 [[TMP5]], ptr [[__VLA_EXPR1]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = call ptr @__kmpc_omp_target_task_alloc(ptr @[[GLOB1]], i32 [[TMP0]], i32 1, i64 40, i64 1, ptr @.omp_task_entry., i64 -1)
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP7]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[TMP9:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP7]])
// CHECK1-NEXT: [[TMP10:%.*]] = load i32, ptr [[A]], align 4
// CHECK1-NEXT: store i32 [[TMP10]], ptr [[A_CASTED]], align 4
// CHECK1-NEXT: [[TMP11:%.*]] = load i64, ptr [[A_CASTED]], align 8
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l104(i64 [[TMP11]]) #[[ATTR3:[0-9]+]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[TMP12:%.*]] = load i16, ptr [[AA]], align 2
// CHECK1-NEXT: store i16 [[TMP12]], ptr [[AA_CASTED]], align 2
// CHECK1-NEXT: [[TMP13:%.*]] = load i64, ptr [[AA_CASTED]], align 8
// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP13]], ptr [[TMP15]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
// CHECK1-NEXT: store ptr null, ptr [[TMP16]], align 8
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
[OpenMP] Increment kernel args version, used by runtime for detecting dyn_ptr. (#85363) A kernel implicit parameter (dyn_ptr) was introduced some time back. This patch increments the kernel args version for a compiler supporting dyn_ptr. The version will be used by the runtime to determine whether the implicit parameter is generated by the compiler. The versioning is required to support use cases where code generated by an older compiler is linked with a newer runtime. If approved, this patch should be backported to release 18.
2024-03-19 16:40:22 -07:00
// CHECK1-NEXT: store i32 3, ptr [[TMP19]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 1, ptr [[TMP20]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[TMP17]], ptr [[TMP21]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[TMP18]], ptr [[TMP22]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr @.offload_sizes, ptr [[TMP23]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr @.offload_maptypes, ptr [[TMP24]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr null, ptr [[TMP26]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i64 0, ptr [[TMP27]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i64 0, ptr [[TMP28]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP29]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP30]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i32 0, ptr [[TMP31]], align 4
// CHECK1-NEXT: [[TMP32:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110.region_id, ptr [[KERNEL_ARGS]])
// CHECK1-NEXT: [[TMP33:%.*]] = icmp ne i32 [[TMP32]], 0
// CHECK1-NEXT: br i1 [[TMP33]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1: omp_offload.failed:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110(i64 [[TMP13]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: [[TMP34:%.*]] = load i32, ptr [[A]], align 4
// CHECK1-NEXT: store i32 [[TMP34]], ptr [[A_CASTED2]], align 4
// CHECK1-NEXT: [[TMP35:%.*]] = load i64, ptr [[A_CASTED2]], align 8
// CHECK1-NEXT: [[TMP36:%.*]] = load i16, ptr [[AA]], align 2
// CHECK1-NEXT: store i16 [[TMP36]], ptr [[AA_CASTED3]], align 2
// CHECK1-NEXT: [[TMP37:%.*]] = load i64, ptr [[AA_CASTED3]], align 8
// CHECK1-NEXT: [[TMP38:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP38]], 10
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK1: omp_if.then:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP35]], ptr [[TMP39]], align 8
// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP35]], ptr [[TMP40]], align 8
// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 0
// CHECK1-NEXT: store ptr null, ptr [[TMP41]], align 8
// CHECK1-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
// CHECK1-NEXT: store i64 [[TMP37]], ptr [[TMP42]], align 8
// CHECK1-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
// CHECK1-NEXT: store i64 [[TMP37]], ptr [[TMP43]], align 8
// CHECK1-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i64 0, i64 1
// CHECK1-NEXT: store ptr null, ptr [[TMP44]], align 8
// CHECK1-NEXT: [[TMP45:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
// CHECK1-NEXT: [[TMP46:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP47:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
[OpenMP] Increment kernel args version, used by runtime for detecting dyn_ptr. (#85363) A kernel implicit parameter (dyn_ptr) was introduced some time back. This patch increments the kernel args version for a compiler supporting dyn_ptr. The version will be used by the runtime to determine whether the implicit parameter is generated by the compiler. The versioning is required to support use cases where code generated by an older compiler is linked with a newer runtime. If approved, this patch should be backported to release 18.
2024-03-19 16:40:22 -07:00
// CHECK1-NEXT: store i32 3, ptr [[TMP47]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP48:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i32 2, ptr [[TMP48]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP49:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store ptr [[TMP45]], ptr [[TMP49]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP50:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store ptr [[TMP46]], ptr [[TMP50]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP51:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: store ptr @.offload_sizes.1, ptr [[TMP51]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP52:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: store ptr @.offload_maptypes.2, ptr [[TMP52]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP53:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store ptr null, ptr [[TMP53]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP54:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store ptr null, ptr [[TMP54]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP55:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i64 0, ptr [[TMP55]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP56:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i64 0, ptr [[TMP56]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP57:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP57]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP58:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP58]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP59:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i32 0, ptr [[TMP59]], align 4
// CHECK1-NEXT: [[TMP60:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119.region_id, ptr [[KERNEL_ARGS7]])
// CHECK1-NEXT: [[TMP61:%.*]] = icmp ne i32 [[TMP60]], 0
// CHECK1-NEXT: br i1 [[TMP61]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1: omp_offload.failed8:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119(i64 [[TMP35]], i64 [[TMP37]]) #[[ATTR3]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK1: omp_offload.cont9:
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK1: omp_if.else:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119(i64 [[TMP35]], i64 [[TMP37]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: [[TMP62:%.*]] = load i32, ptr [[A]], align 4
// CHECK1-NEXT: store i32 [[TMP62]], ptr [[A_CASTED10]], align 4
// CHECK1-NEXT: [[TMP63:%.*]] = load i64, ptr [[A_CASTED10]], align 8
// CHECK1-NEXT: [[TMP64:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK1-NEXT: [[CMP11:%.*]] = icmp sgt i32 [[TMP64]], 20
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: br i1 [[CMP11]], label [[OMP_IF_THEN12:%.*]], label [[OMP_IF_ELSE19:%.*]]
// CHECK1: omp_if.then12:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: [[TMP65:%.*]] = mul nuw i64 [[TMP2]], 4
// CHECK1-NEXT: [[TMP66:%.*]] = mul nuw i64 5, [[TMP5]]
// CHECK1-NEXT: [[TMP67:%.*]] = mul nuw i64 [[TMP66]], 8
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes.3, i64 72, i1 false)
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: [[TMP68:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP63]], ptr [[TMP68]], align 8
// CHECK1-NEXT: [[TMP69:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP63]], ptr [[TMP69]], align 8
// CHECK1-NEXT: [[TMP70:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 0
// CHECK1-NEXT: store ptr null, ptr [[TMP70]], align 8
// CHECK1-NEXT: [[TMP71:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 1
// CHECK1-NEXT: store ptr [[B]], ptr [[TMP71]], align 8
// CHECK1-NEXT: [[TMP72:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 1
// CHECK1-NEXT: store ptr [[B]], ptr [[TMP72]], align 8
// CHECK1-NEXT: [[TMP73:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 1
// CHECK1-NEXT: store ptr null, ptr [[TMP73]], align 8
// CHECK1-NEXT: [[TMP74:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 2
// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP74]], align 8
// CHECK1-NEXT: [[TMP75:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 2
// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP75]], align 8
// CHECK1-NEXT: [[TMP76:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 2
// CHECK1-NEXT: store ptr null, ptr [[TMP76]], align 8
// CHECK1-NEXT: [[TMP77:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 3
// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP77]], align 8
// CHECK1-NEXT: [[TMP78:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 3
// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP78]], align 8
// CHECK1-NEXT: [[TMP79:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
// CHECK1-NEXT: store i64 [[TMP65]], ptr [[TMP79]], align 8
// CHECK1-NEXT: [[TMP80:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 3
// CHECK1-NEXT: store ptr null, ptr [[TMP80]], align 8
// CHECK1-NEXT: [[TMP81:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 4
// CHECK1-NEXT: store ptr [[C]], ptr [[TMP81]], align 8
// CHECK1-NEXT: [[TMP82:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 4
// CHECK1-NEXT: store ptr [[C]], ptr [[TMP82]], align 8
// CHECK1-NEXT: [[TMP83:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 4
// CHECK1-NEXT: store ptr null, ptr [[TMP83]], align 8
// CHECK1-NEXT: [[TMP84:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 5
// CHECK1-NEXT: store i64 5, ptr [[TMP84]], align 8
// CHECK1-NEXT: [[TMP85:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 5
// CHECK1-NEXT: store i64 5, ptr [[TMP85]], align 8
// CHECK1-NEXT: [[TMP86:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 5
// CHECK1-NEXT: store ptr null, ptr [[TMP86]], align 8
// CHECK1-NEXT: [[TMP87:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 6
// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP87]], align 8
// CHECK1-NEXT: [[TMP88:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 6
// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP88]], align 8
// CHECK1-NEXT: [[TMP89:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 6
// CHECK1-NEXT: store ptr null, ptr [[TMP89]], align 8
// CHECK1-NEXT: [[TMP90:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 7
// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP90]], align 8
// CHECK1-NEXT: [[TMP91:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 7
// CHECK1-NEXT: store ptr [[VLA1]], ptr [[TMP91]], align 8
// CHECK1-NEXT: [[TMP92:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
// CHECK1-NEXT: store i64 [[TMP67]], ptr [[TMP92]], align 8
// CHECK1-NEXT: [[TMP93:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 7
// CHECK1-NEXT: store ptr null, ptr [[TMP93]], align 8
// CHECK1-NEXT: [[TMP94:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 8
// CHECK1-NEXT: store ptr [[D]], ptr [[TMP94]], align 8
// CHECK1-NEXT: [[TMP95:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 8
// CHECK1-NEXT: store ptr [[D]], ptr [[TMP95]], align 8
// CHECK1-NEXT: [[TMP96:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i64 0, i64 8
// CHECK1-NEXT: store ptr null, ptr [[TMP96]], align 8
// CHECK1-NEXT: [[TMP97:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
// CHECK1-NEXT: [[TMP98:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
// CHECK1-NEXT: [[TMP99:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP100:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 0
[OpenMP] Increment kernel args version, used by runtime for detecting dyn_ptr. (#85363) A kernel implicit parameter (dyn_ptr) was introduced some time back. This patch increments the kernel args version for a compiler supporting dyn_ptr. The version will be used by the runtime to determine whether the implicit parameter is generated by the compiler. The versioning is required to support use cases where code generated by an older compiler is linked with a newer runtime. If approved, this patch should be backported to release 18.
2024-03-19 16:40:22 -07:00
// CHECK1-NEXT: store i32 3, ptr [[TMP100]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP101:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 1
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i32 9, ptr [[TMP101]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP102:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 2
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store ptr [[TMP97]], ptr [[TMP102]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP103:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 3
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store ptr [[TMP98]], ptr [[TMP103]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP104:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 4
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store ptr [[TMP99]], ptr [[TMP104]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP105:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 5
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: store ptr @.offload_maptypes.4, ptr [[TMP105]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP106:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 6
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store ptr null, ptr [[TMP106]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP107:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 7
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store ptr null, ptr [[TMP107]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP108:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 8
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i64 0, ptr [[TMP108]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP109:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 9
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i64 0, ptr [[TMP109]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP110:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 10
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP110]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP111:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 11
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP112:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 12
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i32 0, ptr [[TMP112]], align 4
// CHECK1-NEXT: [[TMP113:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144.region_id, ptr [[KERNEL_ARGS16]])
// CHECK1-NEXT: [[TMP114:%.*]] = icmp ne i32 [[TMP113]], 0
// CHECK1-NEXT: br i1 [[TMP114]], label [[OMP_OFFLOAD_FAILED17:%.*]], label [[OMP_OFFLOAD_CONT18:%.*]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1: omp_offload.failed17:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144(i64 [[TMP63]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR3]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT18]]
// CHECK1: omp_offload.cont18:
// CHECK1-NEXT: br label [[OMP_IF_END20:%.*]]
// CHECK1: omp_if.else19:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144(i64 [[TMP63]], ptr [[B]], i64 [[TMP2]], ptr [[VLA]], ptr [[C]], i64 5, i64 [[TMP5]], ptr [[VLA1]], ptr [[D]]) #[[ATTR3]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: br label [[OMP_IF_END20]]
// CHECK1: omp_if.end20:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: [[TMP115:%.*]] = load i32, ptr [[A]], align 4
// CHECK1-NEXT: [[TMP116:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
Intrinsics: Add type overload to stacksave and stackstore This allows use with non-0 address space stacks. llvm_ptr_ty should never be used. This could use some more percolation up through mlir, but this is enough to fix existing tests. https://reviews.llvm.org/D156666
2023-07-29 09:15:31 -04:00
// CHECK1-NEXT: call void @llvm.stackrestore.p0(ptr [[TMP116]])
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: ret i32 [[TMP115]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100
// CHECK1-SAME: () #[[ATTR2:[0-9]+]] {
// CHECK1-NEXT: entry:
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 0, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100.omp_outlined)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@.omp_task_entry.
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-SAME: (i32 noundef signext [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR4:[0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR_I:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[DOTPART_ID__ADDR_I:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[DOTPRIVATES__ADDR_I:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[DOTCOPY_FN__ADDR_I:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[DOTTASK_T__ADDR_I:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[__CONTEXT_ADDR_I:%.*]] = alloca ptr, align 8
[OpenMP] Change OpenMP code generation for target region entries This patch changes the code we generate to enter a target region on the device. This is in-line with the new definition in the runtime that was added previously. Additionally we implement this in the OpenMPIRBuilder so that this code can be shared with Flang in the future. Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D128550
2022-06-24 09:13:53 -04:00
// CHECK1-NEXT: [[KERNEL_ARGS_I:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[DOTADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[DOTADDR1:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: store i32 [[TMP0]], ptr [[DOTADDR]], align 4
// CHECK1-NEXT: store ptr [[TMP1]], ptr [[DOTADDR1]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTADDR]], align 4
// CHECK1-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP4:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP3]], i32 0, i32 0
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP4]], i32 0, i32 2
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T]], ptr [[TMP4]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 8
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-NEXT: call void @llvm.experimental.noalias.scope.decl(metadata [[META12:![0-9]+]])
// CHECK1-NEXT: call void @llvm.experimental.noalias.scope.decl(metadata [[META15:![0-9]+]])
// CHECK1-NEXT: call void @llvm.experimental.noalias.scope.decl(metadata [[META17:![0-9]+]])
// CHECK1-NEXT: call void @llvm.experimental.noalias.scope.decl(metadata [[META19:![0-9]+]])
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK1-NEXT: store i32 [[TMP2]], ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias [[META21:![0-9]+]]
// CHECK1-NEXT: store ptr [[TMP5]], ptr [[DOTPART_ID__ADDR_I]], align 8, !noalias [[META21]]
// CHECK1-NEXT: store ptr null, ptr [[DOTPRIVATES__ADDR_I]], align 8, !noalias [[META21]]
// CHECK1-NEXT: store ptr null, ptr [[DOTCOPY_FN__ADDR_I]], align 8, !noalias [[META21]]
// CHECK1-NEXT: store ptr [[TMP3]], ptr [[DOTTASK_T__ADDR_I]], align 8, !noalias [[META21]]
// CHECK1-NEXT: store ptr [[TMP7]], ptr [[__CONTEXT_ADDR_I]], align 8, !noalias [[META21]]
// CHECK1-NEXT: [[TMP8:%.*]] = load ptr, ptr [[__CONTEXT_ADDR_I]], align 8, !noalias [[META21]]
[OpenMP] Increment kernel args version, used by runtime for detecting dyn_ptr. (#85363) A kernel implicit parameter (dyn_ptr) was introduced some time back. This patch increments the kernel args version for a compiler supporting dyn_ptr. The version will be used by the runtime to determine whether the implicit parameter is generated by the compiler. The versioning is required to support use cases where code generated by an older compiler is linked with a newer runtime. If approved, this patch should be backported to release 18.
2024-03-19 16:40:22 -07:00
// CHECK1-NEXT: store i32 3, ptr [[KERNEL_ARGS_I]], align 4, !noalias [[META21]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK1-NEXT: store i32 0, ptr [[TMP9]], align 4, !noalias [[META21]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK1-NEXT: store ptr null, ptr [[TMP10]], align 8, !noalias [[META21]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK1-NEXT: store ptr null, ptr [[TMP11]], align 8, !noalias [[META21]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8, !noalias [[META21]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8, !noalias [[META21]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK1-NEXT: store ptr null, ptr [[TMP14]], align 8, !noalias [[META21]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8, !noalias [[META21]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK1-NEXT: store i64 0, ptr [[TMP16]], align 8, !noalias [[META21]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK1-NEXT: store i64 1, ptr [[TMP17]], align 8, !noalias [[META21]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4, !noalias [[META21]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4, !noalias [[META21]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK1-NEXT: store i32 0, ptr [[TMP20]], align 4, !noalias [[META21]]
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100.region_id, ptr [[KERNEL_ARGS_I]])
// CHECK1-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__EXIT:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1: omp_offload.failed.i:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100() #[[ATTR3]]
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: br label [[DOTOMP_OUTLINED__EXIT]]
// CHECK1: .omp_outlined..exit:
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret i32 0
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l104
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-SAME: (i64 noundef [[A:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK1-NEXT: store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l104.omp_outlined, i64 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l104.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP0]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK1-SAME: (i64 noundef [[AA:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[AA_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK1-NEXT: store i16 [[TMP0]], ptr [[AA_CASTED]], align 2
// CHECK1-NEXT: [[TMP1:%.*]] = load i64, ptr [[AA_CASTED]], align 8
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110.omp_outlined, i64 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[AA:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP0]] to i32
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[CONV]], 1
// CHECK1-NEXT: [[CONV1:%.*]] = trunc i32 [[ADD]] to i16
// CHECK1-NEXT: store i16 [[CONV1]], ptr [[AA_ADDR]], align 2
// CHECK1-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK1-NEXT: [[TMP3:%.*]] = call i32 @__kmpc_cancel(ptr @[[GLOB1]], i32 [[TMP2]], i32 1)
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[TMP4:%.*]] = icmp ne i32 [[TMP3]], 0
// CHECK1-NEXT: br i1 [[TMP4]], label [[DOTCANCEL_EXIT:%.*]], label [[DOTCANCEL_CONTINUE:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1: .cancel.exit:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_cancel_barrier(ptr @[[GLOB2:[0-9]+]], i32 [[TMP2]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: br label [[DOTCANCEL_CONTINUE]]
// CHECK1: .cancel.continue:
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK1-SAME: (i64 noundef [[A:%.*]], i64 noundef [[AA:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AA_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK1-NEXT: store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK1-NEXT: store i16 [[TMP2]], ptr [[AA_CASTED]], align 2
// CHECK1-NEXT: [[TMP3:%.*]] = load i64, ptr [[AA_CASTED]], align 8
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 2, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119.omp_outlined, i64 [[TMP1]], i64 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]], i64 noundef [[AA:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP0]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP1]] to i32
// CHECK1-NEXT: [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
// CHECK1-NEXT: [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
// CHECK1-NEXT: store i16 [[CONV2]], ptr [[AA_ADDR]], align 2
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-SAME: (i64 noundef [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[BN:%.*]], ptr noundef nonnull align 8 dereferenceable(400) [[C:%.*]], i64 noundef [[VLA1:%.*]], i64 noundef [[VLA3:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[CN:%.*]], ptr noundef nonnull align 8 dereferenceable(16) [[D:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[BN_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[VLA_ADDR2:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[VLA_ADDR4:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[CN_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[D_ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8
// CHECK1-NEXT: store ptr [[BN]], ptr [[BN_ADDR]], align 8
// CHECK1-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT: store i64 [[VLA1]], ptr [[VLA_ADDR2]], align 8
// CHECK1-NEXT: store i64 [[VLA3]], ptr [[VLA_ADDR4]], align 8
// CHECK1-NEXT: store ptr [[CN]], ptr [[CN_ADDR]], align 8
// CHECK1-NEXT: store ptr [[D]], ptr [[D_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load i64, ptr [[VLA_ADDR]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load ptr, ptr [[BN_ADDR]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
// CHECK1-NEXT: [[TMP4:%.*]] = load i64, ptr [[VLA_ADDR2]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = load i64, ptr [[VLA_ADDR4]], align 8
// CHECK1-NEXT: [[TMP6:%.*]] = load ptr, ptr [[CN_ADDR]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = load ptr, ptr [[D_ADDR]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK1-NEXT: store i32 [[TMP8]], ptr [[A_CASTED]], align 4
// CHECK1-NEXT: [[TMP9:%.*]] = load i64, ptr [[A_CASTED]], align 8
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 9, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144.omp_outlined, i64 [[TMP9]], ptr [[TMP0]], i64 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], i64 [[TMP4]], i64 [[TMP5]], ptr [[TMP6]], ptr [[TMP7]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[BN:%.*]], ptr noundef nonnull align 8 dereferenceable(400) [[C:%.*]], i64 noundef [[VLA1:%.*]], i64 noundef [[VLA3:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[CN:%.*]], ptr noundef nonnull align 8 dereferenceable(16) [[D:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[BN_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[VLA_ADDR2:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[VLA_ADDR4:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[CN_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[D_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8
// CHECK1-NEXT: store ptr [[BN]], ptr [[BN_ADDR]], align 8
// CHECK1-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT: store i64 [[VLA1]], ptr [[VLA_ADDR2]], align 8
// CHECK1-NEXT: store i64 [[VLA3]], ptr [[VLA_ADDR4]], align 8
// CHECK1-NEXT: store ptr [[CN]], ptr [[CN_ADDR]], align 8
// CHECK1-NEXT: store ptr [[D]], ptr [[D_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load i64, ptr [[VLA_ADDR]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load ptr, ptr [[BN_ADDR]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
// CHECK1-NEXT: [[TMP4:%.*]] = load i64, ptr [[VLA_ADDR2]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = load i64, ptr [[VLA_ADDR4]], align 8
// CHECK1-NEXT: [[TMP6:%.*]] = load ptr, ptr [[CN_ADDR]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = load ptr, ptr [[D_ADDR]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP8]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], ptr [[TMP0]], i64 0, i64 2
// CHECK1-NEXT: [[TMP9:%.*]] = load float, ptr [[ARRAYIDX]], align 4
// CHECK1-NEXT: [[CONV:%.*]] = fpext float [[TMP9]] to double
// CHECK1-NEXT: [[ADD5:%.*]] = fadd double [[CONV]], 1.000000e+00
// CHECK1-NEXT: [[CONV6:%.*]] = fptrunc double [[ADD5]] to float
// CHECK1-NEXT: store float [[CONV6]], ptr [[ARRAYIDX]], align 4
// CHECK1-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds float, ptr [[TMP2]], i64 3
// CHECK1-NEXT: [[TMP10:%.*]] = load float, ptr [[ARRAYIDX7]], align 4
// CHECK1-NEXT: [[CONV8:%.*]] = fpext float [[TMP10]] to double
// CHECK1-NEXT: [[ADD9:%.*]] = fadd double [[CONV8]], 1.000000e+00
// CHECK1-NEXT: [[CONV10:%.*]] = fptrunc double [[ADD9]] to float
// CHECK1-NEXT: store float [[CONV10]], ptr [[ARRAYIDX7]], align 4
// CHECK1-NEXT: [[ARRAYIDX11:%.*]] = getelementptr inbounds [5 x [10 x double]], ptr [[TMP3]], i64 0, i64 1
// CHECK1-NEXT: [[ARRAYIDX12:%.*]] = getelementptr inbounds [10 x double], ptr [[ARRAYIDX11]], i64 0, i64 2
// CHECK1-NEXT: [[TMP11:%.*]] = load double, ptr [[ARRAYIDX12]], align 8
// CHECK1-NEXT: [[ADD13:%.*]] = fadd double [[TMP11]], 1.000000e+00
// CHECK1-NEXT: store double [[ADD13]], ptr [[ARRAYIDX12]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[TMP12:%.*]] = mul nsw i64 1, [[TMP5]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[ARRAYIDX14:%.*]] = getelementptr inbounds double, ptr [[TMP6]], i64 [[TMP12]]
// CHECK1-NEXT: [[ARRAYIDX15:%.*]] = getelementptr inbounds double, ptr [[ARRAYIDX14]], i64 3
// CHECK1-NEXT: [[TMP13:%.*]] = load double, ptr [[ARRAYIDX15]], align 8
// CHECK1-NEXT: [[ADD16:%.*]] = fadd double [[TMP13]], 1.000000e+00
// CHECK1-NEXT: store double [[ADD16]], ptr [[ARRAYIDX15]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[X:%.*]] = getelementptr inbounds nuw [[STRUCT_TT:%.*]], ptr [[TMP7]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[TMP14:%.*]] = load i64, ptr [[X]], align 8
// CHECK1-NEXT: [[ADD17:%.*]] = add nsw i64 [[TMP14]], 1
// CHECK1-NEXT: store i64 [[ADD17]], ptr [[X]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[Y:%.*]] = getelementptr inbounds nuw [[STRUCT_TT]], ptr [[TMP7]], i32 0, i32 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[TMP15:%.*]] = load i8, ptr [[Y]], align 8
// CHECK1-NEXT: [[CONV18:%.*]] = sext i8 [[TMP15]] to i32
// CHECK1-NEXT: [[ADD19:%.*]] = add nsw i32 [[CONV18]], 1
// CHECK1-NEXT: [[CONV20:%.*]] = trunc i32 [[ADD19]] to i8
// CHECK1-NEXT: store i8 [[CONV20]], ptr [[Y]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@_Z3bari
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK1-SAME: (i32 noundef signext [[N:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[A:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[S:%.*]] = alloca [[STRUCT_S1:%.*]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK1-NEXT: store i32 0, ptr [[A]], align 4
// CHECK1-NEXT: [[TMP0:%.*]] = load i32, ptr [[N_ADDR]], align 4
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK1-NEXT: [[CALL:%.*]] = call noundef signext i32 @_Z3fooi(i32 noundef signext [[TMP0]])
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[TMP1:%.*]] = load i32, ptr [[A]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], [[CALL]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 [[ADD]], ptr [[A]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK1-NEXT: [[CALL1:%.*]] = call noundef signext i32 @_ZN2S12r1Ei(ptr noundef nonnull align 8 dereferenceable(8) [[S]], i32 noundef signext [[TMP2]])
// CHECK1-NEXT: [[TMP3:%.*]] = load i32, ptr [[A]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP3]], [[CALL1]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 [[ADD2]], ptr [[A]], align 4
// CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[N_ADDR]], align 4
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK1-NEXT: [[CALL3:%.*]] = call noundef signext i32 @_ZL7fstatici(i32 noundef signext [[TMP4]])
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[TMP5:%.*]] = load i32, ptr [[A]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[ADD4:%.*]] = add nsw i32 [[TMP5]], [[CALL3]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 [[ADD4]], ptr [[A]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = load i32, ptr [[N_ADDR]], align 4
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK1-NEXT: [[CALL5:%.*]] = call noundef signext i32 @_Z9ftemplateIiET_i(i32 noundef signext [[TMP6]])
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[TMP7:%.*]] = load i32, ptr [[A]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP7]], [[CALL5]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 [[ADD6]], ptr [[A]], align 4
// CHECK1-NEXT: [[TMP8:%.*]] = load i32, ptr [[A]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret i32 [[TMP8]]
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZN2S12r1Ei
[clang][CodeGenCXX] Improve handling of itanium ABI member function alignment requirements The itanium ABI for certain platforms requires a minimum alignments for member function pointers to reserve certain bits for distinguishing virtual and non-virtual functions. Our implementation of this however depends on the alignment of the function involved, which may however not reflect the true alignment of function pointers on certain targets for which the alignment is independent of the function (e.g. AIX). Worse, the 2-byte alignment we use may be less than the ABI minimum for the target, and in the case we are using explicit sections will result in invalid codegen. This patch attempts to correct this situation by considering the target alignment of function pointers as part of making the decision about whether we need to adjust the function alignment to conform to the ABI. Targets which do not provide the function ptr alignment information will return a value of 1 when queried and will conservatively retain the old alignment. Differential Revision: https://reviews.llvm.org/D147184
2023-03-29 09:25:37 -04:00
// CHECK1-SAME: (ptr noundef nonnull align 8 dereferenceable(8) [[THIS:%.*]], i32 noundef signext [[N:%.*]]) #[[ATTR0]] comdat {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[B:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[SAVED_STACK:%.*]] = alloca ptr, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[__VLA_EXPR0:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[B_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [5 x ptr], align 8
// CHECK1-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [5 x ptr], align 8
// CHECK1-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [5 x ptr], align 8
Revert "[OPENMP]Fix PR50347: Mapping of global scope deep object fails." This reverts commit 638938117aeae5518d6cacd066ffd9830ef4fc9a. Need to fix reported fail https://lab.llvm.org/buildbot/#/builders/193/builds/7496
2022-02-24 12:04:39 -08:00
// CHECK1-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [5 x i64], align 8
[Clang][OpenMP] Insert alloca for kernel args at function entry block instead of the launch point. If an inlined kernel is called in a loop, the launch point alloca would lead to increasing stack usage every time the kernel is invoked. This could make the application run out of stack space and crash. This problem is fixed by using the alloca insertion point while creating the alloca instruction. Fixes https://github.com/llvm/llvm-project/issues/60602 Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D145820
2023-03-17 12:16:17 -04:00
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK1-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK1-NEXT: [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load i32, ptr [[N_ADDR]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP0]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 [[ADD]], ptr [[B]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[TMP2:%.*]] = zext i32 [[TMP1]] to i64
Intrinsics: Add type overload to stacksave and stackstore This allows use with non-0 address space stacks. llvm_ptr_ty should never be used. This could use some more percolation up through mlir, but this is enough to fix existing tests. https://reviews.llvm.org/D156666
2023-07-29 09:15:31 -04:00
// CHECK1-NEXT: [[TMP3:%.*]] = call ptr @llvm.stacksave.p0()
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[TMP3]], ptr [[SAVED_STACK]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[TMP4:%.*]] = mul nuw i64 2, [[TMP2]]
// CHECK1-NEXT: [[VLA:%.*]] = alloca i16, i64 [[TMP4]], align 2
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i64 [[TMP2]], ptr [[__VLA_EXPR0]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = load i32, ptr [[B]], align 4
// CHECK1-NEXT: store i32 [[TMP5]], ptr [[B_CASTED]], align 4
// CHECK1-NEXT: [[TMP6:%.*]] = load i64, ptr [[B_CASTED]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = load i32, ptr [[N_ADDR]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP7]], 60
// CHECK1-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK1: omp_if.then:
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds nuw [[STRUCT_S1:%.*]], ptr [[THIS1]], i32 0, i32 0
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[TMP8:%.*]] = mul nuw i64 2, [[TMP2]]
// CHECK1-NEXT: [[TMP9:%.*]] = mul nuw i64 [[TMP8]], 2
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: call void @llvm.memcpy.p0.p0.i64(ptr align 8 [[DOTOFFLOAD_SIZES]], ptr align 8 @.offload_sizes.5, i64 40, i1 false)
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: store ptr [[THIS1]], ptr [[TMP10]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: store ptr [[A]], ptr [[TMP11]], align 8
// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
// CHECK1-NEXT: store i64 [[TMP6]], ptr [[TMP13]], align 8
// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
// CHECK1-NEXT: store i64 [[TMP6]], ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
// CHECK1-NEXT: store i64 2, ptr [[TMP16]], align 8
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
// CHECK1-NEXT: store i64 2, ptr [[TMP17]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 3
// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP19]], align 8
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 3
// CHECK1-NEXT: store i64 [[TMP2]], ptr [[TMP20]], align 8
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 3
// CHECK1-NEXT: store ptr null, ptr [[TMP21]], align 8
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 4
// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP22]], align 8
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 4
// CHECK1-NEXT: store ptr [[VLA]], ptr [[TMP23]], align 8
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 4
// CHECK1-NEXT: store i64 [[TMP9]], ptr [[TMP24]], align 8
// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 4
// CHECK1-NEXT: store ptr null, ptr [[TMP25]], align 8
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
[OpenMP] Increment kernel args version, used by runtime for detecting dyn_ptr. (#85363) A kernel implicit parameter (dyn_ptr) was introduced some time back. This patch increments the kernel args version for a compiler supporting dyn_ptr. The version will be used by the runtime to determine whether the implicit parameter is generated by the compiler. The versioning is required to support use cases where code generated by an older compiler is linked with a newer runtime. If approved, this patch should be backported to release 18.
2024-03-19 16:40:22 -07:00
// CHECK1-NEXT: store i32 3, ptr [[TMP29]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 5, ptr [[TMP30]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[TMP26]], ptr [[TMP31]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[TMP27]], ptr [[TMP32]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[TMP28]], ptr [[TMP33]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP34:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP34]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP35:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr null, ptr [[TMP35]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP36:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr null, ptr [[TMP36]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP37:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i64 0, ptr [[TMP37]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP38:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i64 0, ptr [[TMP38]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP39:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP39]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP40:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP41:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i32 0, ptr [[TMP41]], align 4
// CHECK1-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.region_id, ptr [[KERNEL_ARGS]])
// CHECK1-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
// CHECK1-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1: omp_offload.failed:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK1: omp_if.else:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216(ptr [[THIS1]], i64 [[TMP6]], i64 2, i64 [[TMP2]], ptr [[VLA]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: [[TMP44:%.*]] = mul nsw i64 1, [[TMP2]]
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i64 [[TMP44]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i64 1
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
// CHECK1-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
// CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
// CHECK1-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 8
Intrinsics: Add type overload to stacksave and stackstore This allows use with non-0 address space stacks. llvm_ptr_ty should never be used. This could use some more percolation up through mlir, but this is enough to fix existing tests. https://reviews.llvm.org/D156666
2023-07-29 09:15:31 -04:00
// CHECK1-NEXT: call void @llvm.stackrestore.p0(ptr [[TMP47]])
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: ret i32 [[ADD3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
//
//
// CHECK1-LABEL: define {{[^@]+}}@_ZL7fstatici
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK1-SAME: (i32 noundef signext [[N:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[A:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[AA:%.*]] = alloca i16, align 2
// CHECK1-NEXT: [[AAA:%.*]] = alloca i8, align 1
// CHECK1-NEXT: [[B:%.*]] = alloca [10 x i32], align 4
// CHECK1-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AA_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AAA_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [4 x ptr], align 8
// CHECK1-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [4 x ptr], align 8
// CHECK1-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [4 x ptr], align 8
[Clang][OpenMP] Insert alloca for kernel args at function entry block instead of the launch point. If an inlined kernel is called in a loop, the launch point alloca would lead to increasing stack usage every time the kernel is invoked. This could make the application run out of stack space and crash. This problem is fixed by using the alloca insertion point while creating the alloca instruction. Fixes https://github.com/llvm/llvm-project/issues/60602 Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D145820
2023-03-17 12:16:17 -04:00
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK1-NEXT: store i32 0, ptr [[A]], align 4
// CHECK1-NEXT: store i16 0, ptr [[AA]], align 2
// CHECK1-NEXT: store i8 0, ptr [[AAA]], align 1
// CHECK1-NEXT: [[TMP0:%.*]] = load i32, ptr [[A]], align 4
// CHECK1-NEXT: store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA]], align 2
// CHECK1-NEXT: store i16 [[TMP2]], ptr [[AA_CASTED]], align 2
// CHECK1-NEXT: [[TMP3:%.*]] = load i64, ptr [[AA_CASTED]], align 8
// CHECK1-NEXT: [[TMP4:%.*]] = load i8, ptr [[AAA]], align 1
// CHECK1-NEXT: store i8 [[TMP4]], ptr [[AAA_CASTED]], align 1
// CHECK1-NEXT: [[TMP5:%.*]] = load i64, ptr [[AAA_CASTED]], align 8
// CHECK1-NEXT: [[TMP6:%.*]] = load i32, ptr [[N_ADDR]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP6]], 50
// CHECK1-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK1: omp_if.then:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP1]], ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP1]], ptr [[TMP8]], align 8
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
// CHECK1-NEXT: store ptr null, ptr [[TMP9]], align 8
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
// CHECK1-NEXT: store i64 [[TMP3]], ptr [[TMP10]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
// CHECK1-NEXT: store i64 [[TMP3]], ptr [[TMP11]], align 8
// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
// CHECK1-NEXT: store ptr null, ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP13]], align 8
// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
// CHECK1-NEXT: store i64 [[TMP5]], ptr [[TMP14]], align 8
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
// CHECK1-NEXT: store ptr null, ptr [[TMP15]], align 8
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 3
// CHECK1-NEXT: store ptr [[B]], ptr [[TMP16]], align 8
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 3
// CHECK1-NEXT: store ptr [[B]], ptr [[TMP17]], align 8
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 3
// CHECK1-NEXT: store ptr null, ptr [[TMP18]], align 8
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
[OpenMP] Increment kernel args version, used by runtime for detecting dyn_ptr. (#85363) A kernel implicit parameter (dyn_ptr) was introduced some time back. This patch increments the kernel args version for a compiler supporting dyn_ptr. The version will be used by the runtime to determine whether the implicit parameter is generated by the compiler. The versioning is required to support use cases where code generated by an older compiler is linked with a newer runtime. If approved, this patch should be backported to release 18.
2024-03-19 16:40:22 -07:00
// CHECK1-NEXT: store i32 3, ptr [[TMP21]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 4, ptr [[TMP22]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: store ptr @.offload_sizes.7, ptr [[TMP25]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP26]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr null, ptr [[TMP27]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr null, ptr [[TMP28]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP29:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i64 0, ptr [[TMP29]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP30:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i64 0, ptr [[TMP30]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP31:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP32:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP33:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i32 0, ptr [[TMP33]], align 4
// CHECK1-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198.region_id, ptr [[KERNEL_ARGS]])
// CHECK1-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
// CHECK1-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1: omp_offload.failed:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], ptr [[B]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK1: omp_if.else:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198(i64 [[TMP1]], i64 [[TMP3]], i64 [[TMP5]], ptr [[B]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: [[TMP36:%.*]] = load i32, ptr [[A]], align 4
// CHECK1-NEXT: ret i32 [[TMP36]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
//
//
// CHECK1-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK1-SAME: (i32 noundef signext [[N:%.*]]) #[[ATTR0]] comdat {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[A:%.*]] = alloca i32, align 4
// CHECK1-NEXT: [[AA:%.*]] = alloca i16, align 2
// CHECK1-NEXT: [[B:%.*]] = alloca [10 x i32], align 4
// CHECK1-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AA_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [3 x ptr], align 8
// CHECK1-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [3 x ptr], align 8
// CHECK1-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [3 x ptr], align 8
[Clang][OpenMP] Insert alloca for kernel args at function entry block instead of the launch point. If an inlined kernel is called in a loop, the launch point alloca would lead to increasing stack usage every time the kernel is invoked. This could make the application run out of stack space and crash. This problem is fixed by using the alloca insertion point while creating the alloca instruction. Fixes https://github.com/llvm/llvm-project/issues/60602 Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D145820
2023-03-17 12:16:17 -04:00
// CHECK1-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK1-NEXT: store i32 0, ptr [[A]], align 4
// CHECK1-NEXT: store i16 0, ptr [[AA]], align 2
// CHECK1-NEXT: [[TMP0:%.*]] = load i32, ptr [[A]], align 4
// CHECK1-NEXT: store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK1-NEXT: [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA]], align 2
// CHECK1-NEXT: store i16 [[TMP2]], ptr [[AA_CASTED]], align 2
// CHECK1-NEXT: [[TMP3:%.*]] = load i64, ptr [[AA_CASTED]], align 8
// CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[N_ADDR]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP4]], 40
// CHECK1-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK1: omp_if.then:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[TMP5:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP1]], ptr [[TMP5]], align 8
// CHECK1-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK1-NEXT: store i64 [[TMP1]], ptr [[TMP6]], align 8
// CHECK1-NEXT: [[TMP7:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 0
// CHECK1-NEXT: store ptr null, ptr [[TMP7]], align 8
// CHECK1-NEXT: [[TMP8:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
// CHECK1-NEXT: store i64 [[TMP3]], ptr [[TMP8]], align 8
// CHECK1-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
// CHECK1-NEXT: store i64 [[TMP3]], ptr [[TMP9]], align 8
// CHECK1-NEXT: [[TMP10:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 1
// CHECK1-NEXT: store ptr null, ptr [[TMP10]], align 8
// CHECK1-NEXT: [[TMP11:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
// CHECK1-NEXT: store ptr [[B]], ptr [[TMP11]], align 8
// CHECK1-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
// CHECK1-NEXT: store ptr [[B]], ptr [[TMP12]], align 8
// CHECK1-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i64 0, i64 2
// CHECK1-NEXT: store ptr null, ptr [[TMP13]], align 8
// CHECK1-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK1-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
[OpenMP] Increment kernel args version, used by runtime for detecting dyn_ptr. (#85363) A kernel implicit parameter (dyn_ptr) was introduced some time back. This patch increments the kernel args version for a compiler supporting dyn_ptr. The version will be used by the runtime to determine whether the implicit parameter is generated by the compiler. The versioning is required to support use cases where code generated by an older compiler is linked with a newer runtime. If approved, this patch should be backported to release 18.
2024-03-19 16:40:22 -07:00
// CHECK1-NEXT: store i32 3, ptr [[TMP16]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 3, ptr [[TMP17]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[TMP14]], ptr [[TMP18]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[TMP15]], ptr [[TMP19]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: store ptr @.offload_sizes.9, ptr [[TMP20]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: store ptr @.offload_maptypes.10, ptr [[TMP21]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP22:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr null, ptr [[TMP22]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP23:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr null, ptr [[TMP23]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP24:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i64 0, ptr [[TMP24]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP25:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i64 0, ptr [[TMP25]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP26:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP27:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[TMP28:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: store i32 0, ptr [[TMP28]], align 4
// CHECK1-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181.region_id, ptr [[KERNEL_ARGS]])
// CHECK1-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
// CHECK1-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1: omp_offload.failed:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK1: omp_offload.cont:
// CHECK1-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK1: omp_if.else:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181(i64 [[TMP1]], i64 [[TMP3]], ptr [[B]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: br label [[OMP_IF_END]]
// CHECK1: omp_if.end:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK1-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
// CHECK1-NEXT: ret i32 [[TMP31]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-SAME: (ptr noundef [[THIS:%.*]], i64 noundef [[B:%.*]], i64 noundef [[VLA:%.*]], i64 noundef [[VLA1:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[C:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[VLA_ADDR2:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[B_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK1-NEXT: store i64 [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8
// CHECK1-NEXT: store i64 [[VLA1]], ptr [[VLA_ADDR2]], align 8
// CHECK1-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load i64, ptr [[VLA_ADDR]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load i64, ptr [[VLA_ADDR2]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
// CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[B_ADDR]], align 4
// CHECK1-NEXT: store i32 [[TMP4]], ptr [[B_CASTED]], align 4
// CHECK1-NEXT: [[TMP5:%.*]] = load i64, ptr [[B_CASTED]], align 8
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.omp_outlined, ptr [[TMP0]], i64 [[TMP5]], i64 [[TMP1]], i64 [[TMP2]], ptr [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]], i64 noundef [[B:%.*]], i64 noundef [[VLA:%.*]], i64 noundef [[VLA1:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[C:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[VLA_ADDR2:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK1-NEXT: store i64 [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8
// CHECK1-NEXT: store i64 [[VLA1]], ptr [[VLA_ADDR2]], align 8
// CHECK1-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load i64, ptr [[VLA_ADDR]], align 8
// CHECK1-NEXT: [[TMP2:%.*]] = load i64, ptr [[VLA_ADDR2]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
// CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[B_ADDR]], align 4
// CHECK1-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP4]] to double
// CHECK1-NEXT: [[ADD:%.*]] = fadd double [[CONV]], 1.500000e+00
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[A:%.*]] = getelementptr inbounds nuw [[STRUCT_S1:%.*]], ptr [[TMP0]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store double [[ADD]], ptr [[A]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK1-NEXT: [[A3:%.*]] = getelementptr inbounds nuw [[STRUCT_S1]], ptr [[TMP0]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[TMP5:%.*]] = load double, ptr [[A3]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[INC:%.*]] = fadd double [[TMP5]], 1.000000e+00
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store double [[INC]], ptr [[A3]], align 8
// CHECK1-NEXT: [[CONV4:%.*]] = fptosi double [[INC]] to i16
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[TMP6:%.*]] = mul nsw i64 1, [[TMP2]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[TMP3]], i64 [[TMP6]]
// CHECK1-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i64 1
// CHECK1-NEXT: store i16 [[CONV4]], ptr [[ARRAYIDX5]], align 2
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-SAME: (i64 noundef [[A:%.*]], i64 noundef [[AA:%.*]], i64 noundef [[AAA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AAA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AA_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AAA_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK1-NEXT: store i64 [[AAA]], ptr [[AAA_ADDR]], align 8
// CHECK1-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK1-NEXT: store i32 [[TMP1]], ptr [[A_CASTED]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK1-NEXT: store i16 [[TMP3]], ptr [[AA_CASTED]], align 2
// CHECK1-NEXT: [[TMP4:%.*]] = load i64, ptr [[AA_CASTED]], align 8
// CHECK1-NEXT: [[TMP5:%.*]] = load i8, ptr [[AAA_ADDR]], align 1
// CHECK1-NEXT: store i8 [[TMP5]], ptr [[AAA_CASTED]], align 1
// CHECK1-NEXT: [[TMP6:%.*]] = load i64, ptr [[AAA_CASTED]], align 8
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198.omp_outlined, i64 [[TMP2]], i64 [[TMP4]], i64 [[TMP6]], ptr [[TMP0]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]], i64 noundef [[AA:%.*]], i64 noundef [[AAA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AAA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK1-NEXT: store i64 [[AAA]], ptr [[AAA_ADDR]], align 8
// CHECK1-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP2]] to i32
// CHECK1-NEXT: [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
// CHECK1-NEXT: [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
// CHECK1-NEXT: store i16 [[CONV2]], ptr [[AA_ADDR]], align 2
// CHECK1-NEXT: [[TMP3:%.*]] = load i8, ptr [[AAA_ADDR]], align 1
// CHECK1-NEXT: [[CONV3:%.*]] = sext i8 [[TMP3]] to i32
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: [[ADD4:%.*]] = add nsw i32 [[CONV3]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[CONV5:%.*]] = trunc i32 [[ADD4]] to i8
// CHECK1-NEXT: store i8 [[CONV5]], ptr [[AAA_ADDR]], align 1
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 2
// CHECK1-NEXT: [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
// CHECK1-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP4]], 1
// CHECK1-NEXT: store i32 [[ADD6]], ptr [[ARRAYIDX]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret void
//
//
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-SAME: (i64 noundef [[A:%.*]], i64 noundef [[AA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AA_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK1-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK1-NEXT: store i32 [[TMP1]], ptr [[A_CASTED]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK1-NEXT: [[TMP3:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK1-NEXT: store i16 [[TMP3]], ptr [[AA_CASTED]], align 2
// CHECK1-NEXT: [[TMP4:%.*]] = load i64, ptr [[AA_CASTED]], align 8
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181.omp_outlined, i64 [[TMP2]], i64 [[TMP4]], ptr [[TMP0]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK1-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK1-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]], i64 noundef [[AA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK1-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK1-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK1-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK1-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK1-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK1-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK1-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK1-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK1-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK1-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK1-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK1-NEXT: [[CONV:%.*]] = sext i16 [[TMP2]] to i32
// CHECK1-NEXT: [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
// CHECK1-NEXT: [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
// CHECK1-NEXT: store i16 [[CONV2]], ptr [[AA_ADDR]], align 2
// CHECK1-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 2
// CHECK1-NEXT: [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
// CHECK1-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP3]], 1
// CHECK1-NEXT: store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK1-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_Z3fooi
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK3-SAME: (i32 noundef [[N:%.*]]) #[[ATTR0:[0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[A:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA:%.*]] = alloca i16, align 2
// CHECK3-NEXT: [[B:%.*]] = alloca [10 x float], align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[SAVED_STACK:%.*]] = alloca ptr, align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[__VLA_EXPR0:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[C:%.*]] = alloca [5 x [10 x double]], align 8
// CHECK3-NEXT: [[__VLA_EXPR1:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[D:%.*]] = alloca [[STRUCT_TT:%.*]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[AGG_CAPTURED:%.*]] = alloca [[STRUCT_ANON:%.*]], align 1
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[A_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [1 x ptr], align 4
// CHECK3-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [1 x ptr], align 4
// CHECK3-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [1 x ptr], align 4
[Clang][OpenMP] Insert alloca for kernel args at function entry block instead of the launch point. If an inlined kernel is called in a loop, the launch point alloca would lead to increasing stack usage every time the kernel is invoked. This could make the application run out of stack space and crash. This problem is fixed by using the alloca insertion point while creating the alloca instruction. Fixes https://github.com/llvm/llvm-project/issues/60602 Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D145820
2023-03-17 12:16:17 -04:00
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[A_CASTED2:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA_CASTED3:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[DOTOFFLOAD_BASEPTRS4:%.*]] = alloca [2 x ptr], align 4
// CHECK3-NEXT: [[DOTOFFLOAD_PTRS5:%.*]] = alloca [2 x ptr], align 4
// CHECK3-NEXT: [[DOTOFFLOAD_MAPPERS6:%.*]] = alloca [2 x ptr], align 4
[Clang][OpenMP] Insert alloca for kernel args at function entry block instead of the launch point. If an inlined kernel is called in a loop, the launch point alloca would lead to increasing stack usage every time the kernel is invoked. This could make the application run out of stack space and crash. This problem is fixed by using the alloca insertion point while creating the alloca instruction. Fixes https://github.com/llvm/llvm-project/issues/60602 Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D145820
2023-03-17 12:16:17 -04:00
// CHECK3-NEXT: [[KERNEL_ARGS7:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[A_CASTED10:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[DOTOFFLOAD_BASEPTRS13:%.*]] = alloca [9 x ptr], align 4
// CHECK3-NEXT: [[DOTOFFLOAD_PTRS14:%.*]] = alloca [9 x ptr], align 4
// CHECK3-NEXT: [[DOTOFFLOAD_MAPPERS15:%.*]] = alloca [9 x ptr], align 4
Revert "[OPENMP]Fix PR50347: Mapping of global scope deep object fails." This reverts commit 638938117aeae5518d6cacd066ffd9830ef4fc9a. Need to fix reported fail https://lab.llvm.org/buildbot/#/builders/193/builds/7496
2022-02-24 12:04:39 -08:00
// CHECK3-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [9 x i64], align 4
[Clang][OpenMP] Insert alloca for kernel args at function entry block instead of the launch point. If an inlined kernel is called in a loop, the launch point alloca would lead to increasing stack usage every time the kernel is invoked. This could make the application run out of stack space and crash. This problem is fixed by using the alloca insertion point while creating the alloca instruction. Fixes https://github.com/llvm/llvm-project/issues/60602 Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D145820
2023-03-17 12:16:17 -04:00
// CHECK3-NEXT: [[KERNEL_ARGS16:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[TMP0:%.*]] = call i32 @__kmpc_global_thread_num(ptr @[[GLOB1:[0-9]+]])
// CHECK3-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK3-NEXT: store i32 0, ptr [[A]], align 4
// CHECK3-NEXT: store i16 0, ptr [[AA]], align 2
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
Intrinsics: Add type overload to stacksave and stackstore This allows use with non-0 address space stacks. llvm_ptr_ty should never be used. This could use some more percolation up through mlir, but this is enough to fix existing tests. https://reviews.llvm.org/D156666
2023-07-29 09:15:31 -04:00
// CHECK3-NEXT: [[TMP2:%.*]] = call ptr @llvm.stacksave.p0()
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[TMP2]], ptr [[SAVED_STACK]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[VLA:%.*]] = alloca float, i32 [[TMP1]], align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[TMP1]], ptr [[__VLA_EXPR0]], align 4
// CHECK3-NEXT: [[TMP3:%.*]] = load i32, ptr [[N_ADDR]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[TMP4:%.*]] = mul nuw i32 5, [[TMP3]]
// CHECK3-NEXT: [[VLA1:%.*]] = alloca double, i32 [[TMP4]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[TMP3]], ptr [[__VLA_EXPR1]], align 4
// CHECK3-NEXT: [[TMP5:%.*]] = call ptr @__kmpc_omp_target_task_alloc(ptr @[[GLOB1]], i32 [[TMP0]], i32 1, i32 20, i32 1, ptr @.omp_task_entry., i64 -1)
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP5]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[TMP7:%.*]] = call i32 @__kmpc_omp_task(ptr @[[GLOB1]], i32 [[TMP0]], ptr [[TMP5]])
// CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[A]], align 4
// CHECK3-NEXT: store i32 [[TMP8]], ptr [[A_CASTED]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = load i32, ptr [[A_CASTED]], align 4
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l104(i32 [[TMP9]]) #[[ATTR3:[0-9]+]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[TMP10:%.*]] = load i16, ptr [[AA]], align 2
// CHECK3-NEXT: store i16 [[TMP10]], ptr [[AA_CASTED]], align 2
// CHECK3-NEXT: [[TMP11:%.*]] = load i32, ptr [[AA_CASTED]], align 4
// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK3-NEXT: store i32 [[TMP11]], ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: store i32 [[TMP11]], ptr [[TMP13]], align 4
// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [1 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
[OpenMP] Increment kernel args version, used by runtime for detecting dyn_ptr. (#85363) A kernel implicit parameter (dyn_ptr) was introduced some time back. This patch increments the kernel args version for a compiler supporting dyn_ptr. The version will be used by the runtime to determine whether the implicit parameter is generated by the compiler. The versioning is required to support use cases where code generated by an older compiler is linked with a newer runtime. If approved, this patch should be backported to release 18.
2024-03-19 16:40:22 -07:00
// CHECK3-NEXT: store i32 3, ptr [[TMP17]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 1, ptr [[TMP18]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP19]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[TMP16]], ptr [[TMP20]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr @.offload_sizes, ptr [[TMP21]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr @.offload_maptypes, ptr [[TMP22]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr null, ptr [[TMP23]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr null, ptr [[TMP24]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i64 0, ptr [[TMP25]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i64 0, ptr [[TMP26]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP27]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP28]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i32 0, ptr [[TMP29]], align 4
// CHECK3-NEXT: [[TMP30:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110.region_id, ptr [[KERNEL_ARGS]])
// CHECK3-NEXT: [[TMP31:%.*]] = icmp ne i32 [[TMP30]], 0
// CHECK3-NEXT: br i1 [[TMP31]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3: omp_offload.failed:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110(i32 [[TMP11]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: [[TMP32:%.*]] = load i32, ptr [[A]], align 4
// CHECK3-NEXT: store i32 [[TMP32]], ptr [[A_CASTED2]], align 4
// CHECK3-NEXT: [[TMP33:%.*]] = load i32, ptr [[A_CASTED2]], align 4
// CHECK3-NEXT: [[TMP34:%.*]] = load i16, ptr [[AA]], align 2
// CHECK3-NEXT: store i16 [[TMP34]], ptr [[AA_CASTED3]], align 2
// CHECK3-NEXT: [[TMP35:%.*]] = load i32, ptr [[AA_CASTED3]], align 4
// CHECK3-NEXT: [[TMP36:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP36]], 10
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK3: omp_if.then:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
// CHECK3-NEXT: store i32 [[TMP33]], ptr [[TMP37]], align 4
// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
// CHECK3-NEXT: store i32 [[TMP33]], ptr [[TMP38]], align 4
// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 0
// CHECK3-NEXT: store ptr null, ptr [[TMP39]], align 4
// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 1
// CHECK3-NEXT: store i32 [[TMP35]], ptr [[TMP40]], align 4
// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 1
// CHECK3-NEXT: store i32 [[TMP35]], ptr [[TMP41]], align 4
// CHECK3-NEXT: [[TMP42:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_MAPPERS6]], i32 0, i32 1
// CHECK3-NEXT: store ptr null, ptr [[TMP42]], align 4
// CHECK3-NEXT: [[TMP43:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_BASEPTRS4]], i32 0, i32 0
// CHECK3-NEXT: [[TMP44:%.*]] = getelementptr inbounds [2 x ptr], ptr [[DOTOFFLOAD_PTRS5]], i32 0, i32 0
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP45:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 0
[OpenMP] Increment kernel args version, used by runtime for detecting dyn_ptr. (#85363) A kernel implicit parameter (dyn_ptr) was introduced some time back. This patch increments the kernel args version for a compiler supporting dyn_ptr. The version will be used by the runtime to determine whether the implicit parameter is generated by the compiler. The versioning is required to support use cases where code generated by an older compiler is linked with a newer runtime. If approved, this patch should be backported to release 18.
2024-03-19 16:40:22 -07:00
// CHECK3-NEXT: store i32 3, ptr [[TMP45]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP46:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 1
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i32 2, ptr [[TMP46]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP47:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 2
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store ptr [[TMP43]], ptr [[TMP47]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP48:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 3
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store ptr [[TMP44]], ptr [[TMP48]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP49:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: store ptr @.offload_sizes.1, ptr [[TMP49]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP50:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 5
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: store ptr @.offload_maptypes.2, ptr [[TMP50]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP51:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 6
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store ptr null, ptr [[TMP51]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP52:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 7
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store ptr null, ptr [[TMP52]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP53:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 8
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i64 0, ptr [[TMP53]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP54:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 9
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i64 0, ptr [[TMP54]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP55:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 10
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP55]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP56:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 11
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP56]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP57:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS7]], i32 0, i32 12
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i32 0, ptr [[TMP57]], align 4
// CHECK3-NEXT: [[TMP58:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119.region_id, ptr [[KERNEL_ARGS7]])
// CHECK3-NEXT: [[TMP59:%.*]] = icmp ne i32 [[TMP58]], 0
// CHECK3-NEXT: br i1 [[TMP59]], label [[OMP_OFFLOAD_FAILED8:%.*]], label [[OMP_OFFLOAD_CONT9:%.*]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3: omp_offload.failed8:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119(i32 [[TMP33]], i32 [[TMP35]]) #[[ATTR3]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT9]]
// CHECK3: omp_offload.cont9:
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK3: omp_if.else:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119(i32 [[TMP33]], i32 [[TMP35]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: [[TMP60:%.*]] = load i32, ptr [[A]], align 4
// CHECK3-NEXT: store i32 [[TMP60]], ptr [[A_CASTED10]], align 4
// CHECK3-NEXT: [[TMP61:%.*]] = load i32, ptr [[A_CASTED10]], align 4
// CHECK3-NEXT: [[TMP62:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK3-NEXT: [[CMP11:%.*]] = icmp sgt i32 [[TMP62]], 20
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: br i1 [[CMP11]], label [[OMP_IF_THEN12:%.*]], label [[OMP_IF_ELSE19:%.*]]
// CHECK3: omp_if.then12:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: [[TMP63:%.*]] = mul nuw i32 [[TMP1]], 4
// CHECK3-NEXT: [[TMP64:%.*]] = sext i32 [[TMP63]] to i64
// CHECK3-NEXT: [[TMP65:%.*]] = mul nuw i32 5, [[TMP3]]
// CHECK3-NEXT: [[TMP66:%.*]] = mul nuw i32 [[TMP65]], 8
// CHECK3-NEXT: [[TMP67:%.*]] = sext i32 [[TMP66]] to i64
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES]], ptr align 4 @.offload_sizes.3, i32 72, i1 false)
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: [[TMP68:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
// CHECK3-NEXT: store i32 [[TMP61]], ptr [[TMP68]], align 4
// CHECK3-NEXT: [[TMP69:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
// CHECK3-NEXT: store i32 [[TMP61]], ptr [[TMP69]], align 4
// CHECK3-NEXT: [[TMP70:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 0
// CHECK3-NEXT: store ptr null, ptr [[TMP70]], align 4
// CHECK3-NEXT: [[TMP71:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 1
// CHECK3-NEXT: store ptr [[B]], ptr [[TMP71]], align 4
// CHECK3-NEXT: [[TMP72:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 1
// CHECK3-NEXT: store ptr [[B]], ptr [[TMP72]], align 4
// CHECK3-NEXT: [[TMP73:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 1
// CHECK3-NEXT: store ptr null, ptr [[TMP73]], align 4
// CHECK3-NEXT: [[TMP74:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 2
// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP74]], align 4
// CHECK3-NEXT: [[TMP75:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 2
// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP75]], align 4
// CHECK3-NEXT: [[TMP76:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 2
// CHECK3-NEXT: store ptr null, ptr [[TMP76]], align 4
// CHECK3-NEXT: [[TMP77:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 3
// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP77]], align 4
// CHECK3-NEXT: [[TMP78:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 3
// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP78]], align 4
// CHECK3-NEXT: [[TMP79:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 3
// CHECK3-NEXT: store i64 [[TMP64]], ptr [[TMP79]], align 4
// CHECK3-NEXT: [[TMP80:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 3
// CHECK3-NEXT: store ptr null, ptr [[TMP80]], align 4
// CHECK3-NEXT: [[TMP81:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 4
// CHECK3-NEXT: store ptr [[C]], ptr [[TMP81]], align 4
// CHECK3-NEXT: [[TMP82:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 4
// CHECK3-NEXT: store ptr [[C]], ptr [[TMP82]], align 4
// CHECK3-NEXT: [[TMP83:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 4
// CHECK3-NEXT: store ptr null, ptr [[TMP83]], align 4
// CHECK3-NEXT: [[TMP84:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 5
// CHECK3-NEXT: store i32 5, ptr [[TMP84]], align 4
// CHECK3-NEXT: [[TMP85:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 5
// CHECK3-NEXT: store i32 5, ptr [[TMP85]], align 4
// CHECK3-NEXT: [[TMP86:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 5
// CHECK3-NEXT: store ptr null, ptr [[TMP86]], align 4
// CHECK3-NEXT: [[TMP87:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 6
// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP87]], align 4
// CHECK3-NEXT: [[TMP88:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 6
// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP88]], align 4
// CHECK3-NEXT: [[TMP89:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 6
// CHECK3-NEXT: store ptr null, ptr [[TMP89]], align 4
// CHECK3-NEXT: [[TMP90:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 7
// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP90]], align 4
// CHECK3-NEXT: [[TMP91:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 7
// CHECK3-NEXT: store ptr [[VLA1]], ptr [[TMP91]], align 4
// CHECK3-NEXT: [[TMP92:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 7
// CHECK3-NEXT: store i64 [[TMP67]], ptr [[TMP92]], align 4
// CHECK3-NEXT: [[TMP93:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 7
// CHECK3-NEXT: store ptr null, ptr [[TMP93]], align 4
// CHECK3-NEXT: [[TMP94:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 8
// CHECK3-NEXT: store ptr [[D]], ptr [[TMP94]], align 4
// CHECK3-NEXT: [[TMP95:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 8
// CHECK3-NEXT: store ptr [[D]], ptr [[TMP95]], align 4
// CHECK3-NEXT: [[TMP96:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_MAPPERS15]], i32 0, i32 8
// CHECK3-NEXT: store ptr null, ptr [[TMP96]], align 4
// CHECK3-NEXT: [[TMP97:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_BASEPTRS13]], i32 0, i32 0
// CHECK3-NEXT: [[TMP98:%.*]] = getelementptr inbounds [9 x ptr], ptr [[DOTOFFLOAD_PTRS14]], i32 0, i32 0
// CHECK3-NEXT: [[TMP99:%.*]] = getelementptr inbounds [9 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP100:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 0
[OpenMP] Increment kernel args version, used by runtime for detecting dyn_ptr. (#85363) A kernel implicit parameter (dyn_ptr) was introduced some time back. This patch increments the kernel args version for a compiler supporting dyn_ptr. The version will be used by the runtime to determine whether the implicit parameter is generated by the compiler. The versioning is required to support use cases where code generated by an older compiler is linked with a newer runtime. If approved, this patch should be backported to release 18.
2024-03-19 16:40:22 -07:00
// CHECK3-NEXT: store i32 3, ptr [[TMP100]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP101:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 1
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i32 9, ptr [[TMP101]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP102:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 2
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store ptr [[TMP97]], ptr [[TMP102]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP103:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 3
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store ptr [[TMP98]], ptr [[TMP103]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP104:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 4
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store ptr [[TMP99]], ptr [[TMP104]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP105:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 5
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: store ptr @.offload_maptypes.4, ptr [[TMP105]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP106:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 6
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store ptr null, ptr [[TMP106]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP107:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 7
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store ptr null, ptr [[TMP107]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP108:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 8
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i64 0, ptr [[TMP108]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP109:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 9
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i64 0, ptr [[TMP109]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP110:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 10
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP110]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP111:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 11
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP111]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP112:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS16]], i32 0, i32 12
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i32 0, ptr [[TMP112]], align 4
// CHECK3-NEXT: [[TMP113:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144.region_id, ptr [[KERNEL_ARGS16]])
// CHECK3-NEXT: [[TMP114:%.*]] = icmp ne i32 [[TMP113]], 0
// CHECK3-NEXT: br i1 [[TMP114]], label [[OMP_OFFLOAD_FAILED17:%.*]], label [[OMP_OFFLOAD_CONT18:%.*]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3: omp_offload.failed17:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144(i32 [[TMP61]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR3]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT18]]
// CHECK3: omp_offload.cont18:
// CHECK3-NEXT: br label [[OMP_IF_END20:%.*]]
// CHECK3: omp_if.else19:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144(i32 [[TMP61]], ptr [[B]], i32 [[TMP1]], ptr [[VLA]], ptr [[C]], i32 5, i32 [[TMP3]], ptr [[VLA1]], ptr [[D]]) #[[ATTR3]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: br label [[OMP_IF_END20]]
// CHECK3: omp_if.end20:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: [[TMP115:%.*]] = load i32, ptr [[A]], align 4
// CHECK3-NEXT: [[TMP116:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
Intrinsics: Add type overload to stacksave and stackstore This allows use with non-0 address space stacks. llvm_ptr_ty should never be used. This could use some more percolation up through mlir, but this is enough to fix existing tests. https://reviews.llvm.org/D156666
2023-07-29 09:15:31 -04:00
// CHECK3-NEXT: call void @llvm.stackrestore.p0(ptr [[TMP116]])
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: ret i32 [[TMP115]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100
// CHECK3-SAME: () #[[ATTR2:[0-9]+]] {
// CHECK3-NEXT: entry:
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 0, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100.omp_outlined)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@.omp_task_entry.
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-SAME: (i32 noundef [[TMP0:%.*]], ptr noalias noundef [[TMP1:%.*]]) #[[ATTR4:[0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR_I:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[DOTPART_ID__ADDR_I:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[DOTPRIVATES__ADDR_I:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[DOTCOPY_FN__ADDR_I:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[DOTTASK_T__ADDR_I:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[__CONTEXT_ADDR_I:%.*]] = alloca ptr, align 4
[OpenMP] Change OpenMP code generation for target region entries This patch changes the code we generate to enter a target region on the device. This is in-line with the new definition in the runtime that was added previously. Additionally we implement this in the OpenMPIRBuilder so that this code can be shared with Flang in the future. Reviewed By: ABataev Differential Revision: https://reviews.llvm.org/D128550
2022-06-24 09:13:53 -04:00
// CHECK3-NEXT: [[KERNEL_ARGS_I:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[DOTADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[DOTADDR1:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: store i32 [[TMP0]], ptr [[DOTADDR]], align 4
// CHECK3-NEXT: store ptr [[TMP1]], ptr [[DOTADDR1]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[DOTADDR]], align 4
// CHECK3-NEXT: [[TMP3:%.*]] = load ptr, ptr [[DOTADDR1]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP4:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T_WITH_PRIVATES:%.*]], ptr [[TMP3]], i32 0, i32 0
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T:%.*]], ptr [[TMP4]], i32 0, i32 2
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds nuw [[STRUCT_KMP_TASK_T]], ptr [[TMP4]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[TMP7:%.*]] = load ptr, ptr [[TMP6]], align 4
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-NEXT: call void @llvm.experimental.noalias.scope.decl(metadata [[META13:![0-9]+]])
// CHECK3-NEXT: call void @llvm.experimental.noalias.scope.decl(metadata [[META16:![0-9]+]])
// CHECK3-NEXT: call void @llvm.experimental.noalias.scope.decl(metadata [[META18:![0-9]+]])
// CHECK3-NEXT: call void @llvm.experimental.noalias.scope.decl(metadata [[META20:![0-9]+]])
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK3-NEXT: store i32 [[TMP2]], ptr [[DOTGLOBAL_TID__ADDR_I]], align 4, !noalias [[META22:![0-9]+]]
// CHECK3-NEXT: store ptr [[TMP5]], ptr [[DOTPART_ID__ADDR_I]], align 4, !noalias [[META22]]
// CHECK3-NEXT: store ptr null, ptr [[DOTPRIVATES__ADDR_I]], align 4, !noalias [[META22]]
// CHECK3-NEXT: store ptr null, ptr [[DOTCOPY_FN__ADDR_I]], align 4, !noalias [[META22]]
// CHECK3-NEXT: store ptr [[TMP3]], ptr [[DOTTASK_T__ADDR_I]], align 4, !noalias [[META22]]
// CHECK3-NEXT: store ptr [[TMP7]], ptr [[__CONTEXT_ADDR_I]], align 4, !noalias [[META22]]
// CHECK3-NEXT: [[TMP8:%.*]] = load ptr, ptr [[__CONTEXT_ADDR_I]], align 4, !noalias [[META22]]
[OpenMP] Increment kernel args version, used by runtime for detecting dyn_ptr. (#85363) A kernel implicit parameter (dyn_ptr) was introduced some time back. This patch increments the kernel args version for a compiler supporting dyn_ptr. The version will be used by the runtime to determine whether the implicit parameter is generated by the compiler. The versioning is required to support use cases where code generated by an older compiler is linked with a newer runtime. If approved, this patch should be backported to release 18.
2024-03-19 16:40:22 -07:00
// CHECK3-NEXT: store i32 3, ptr [[KERNEL_ARGS_I]], align 4, !noalias [[META22]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 1
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK3-NEXT: store i32 0, ptr [[TMP9]], align 4, !noalias [[META22]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 2
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK3-NEXT: store ptr null, ptr [[TMP10]], align 4, !noalias [[META22]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 3
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK3-NEXT: store ptr null, ptr [[TMP11]], align 4, !noalias [[META22]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 4
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4, !noalias [[META22]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 5
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK3-NEXT: store ptr null, ptr [[TMP13]], align 4, !noalias [[META22]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 6
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK3-NEXT: store ptr null, ptr [[TMP14]], align 4, !noalias [[META22]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 7
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 4, !noalias [[META22]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 8
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK3-NEXT: store i64 0, ptr [[TMP16]], align 8, !noalias [[META22]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 9
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK3-NEXT: store i64 1, ptr [[TMP17]], align 8, !noalias [[META22]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 10
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP18]], align 4, !noalias [[META22]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 11
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP19]], align 4, !noalias [[META22]]
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS_I]], i32 0, i32 12
[OpenMP] Remove `register_requires` global constructor (#80460) Summary: Currently, OpenMP handles the `omp requires` clause by emitting a global constructor into the runtime for every translation unit that requires it. However, this is not a great solution because it prevents us from having a defined order in which the runtime is accessed and used. This patch changes the approach to no longer use global constructors, but to instead group the flag with the other offloading entires that we already handle. This has the effect of still registering each flag per requires TU, but now we have a single constructor that handles everything. This function removes support for the old `__tgt_register_requires` and replaces it with a warning message. We just had a recent release, and the OpenMP policy for the past four releases since we switched to LLVM is that we do not provide strict backwards compatibility between major LLVM releases now that the library is versioned. This means that a user will need to recompile if they have an old binary that relied on `register_requires` having the old behavior. It is important that we actively deprecate this, as otherwise it would not solve the problem of having no defined init and shutdown order for `libomptarget`. The problem of `libomptarget` not having a define init and shutdown order cascades into a lot of other issues so I have a strong incentive to be rid of it. It is worth noting that the current `__tgt_offload_entry` only has space for a 32-bit integer here. I am planning to overhaul these at some point as well.
2024-02-21 11:33:32 -06:00
// CHECK3-NEXT: store i32 0, ptr [[TMP20]], align 4, !noalias [[META22]]
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: [[TMP21:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100.region_id, ptr [[KERNEL_ARGS_I]])
// CHECK3-NEXT: [[TMP22:%.*]] = icmp ne i32 [[TMP21]], 0
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: br i1 [[TMP22]], label [[OMP_OFFLOAD_FAILED_I:%.*]], label [[DOTOMP_OUTLINED__EXIT:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3: omp_offload.failed.i:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100() #[[ATTR3]]
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: br label [[DOTOMP_OUTLINED__EXIT]]
// CHECK3: .omp_outlined..exit:
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret i32 0
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l104
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-SAME: (i32 noundef [[A:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[A_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK3-NEXT: store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l104.omp_outlined, i32 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l104.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP0]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-21 13:20:39 -07:00
// CHECK3-NEXT: ret void
[OpenMP] Codegen aggregate for outlined function captures Parallel regions are outlined as functions with capture variables explicitly generated as distinct parameters in the function's argument list. That complicates the fork_call interface in the OpenMP runtime: (1) the fork_call is variadic since there is a variable number of arguments to forward to the outlined function, (2) wrapping/unwrapping arguments happens in the OpenMP runtime, which is sub-optimal, has been a source of ABI bugs, and has a hardcoded limit (16) in the number of arguments, (3) forwarded arguments must cast to pointer types, which complicates debugging. This patch avoids those issues by aggregating captured arguments in a struct to pass to the fork_call. Reviewed By: jdoerfert, jhuber6 Differential Revision: https://reviews.llvm.org/D102107
2021-09-20 17:12:14 -07:00
//
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK3-SAME: (i32 noundef [[AA:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[AA_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK3-NEXT: store i16 [[TMP0]], ptr [[AA_CASTED]], align 2
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[AA_CASTED]], align 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110.omp_outlined, i32 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[AA:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP0]] to i32
// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[CONV]], 1
// CHECK3-NEXT: [[CONV1:%.*]] = trunc i32 [[ADD]] to i16
// CHECK3-NEXT: store i16 [[CONV1]], ptr [[AA_ADDR]], align 2
// CHECK3-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK3-NEXT: [[TMP3:%.*]] = call i32 @__kmpc_cancel(ptr @[[GLOB1]], i32 [[TMP2]], i32 1)
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[TMP4:%.*]] = icmp ne i32 [[TMP3]], 0
// CHECK3-NEXT: br i1 [[TMP4]], label [[DOTCANCEL_EXIT:%.*]], label [[DOTCANCEL_CONTINUE:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3: .cancel.exit:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_cancel_barrier(ptr @[[GLOB2:[0-9]+]], i32 [[TMP2]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: br label [[DOTCANCEL_CONTINUE]]
// CHECK3: .cancel.continue:
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK3-SAME: (i32 noundef [[A:%.*]], i32 noundef [[AA:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[A_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK3-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK3-NEXT: store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK3-NEXT: store i16 [[TMP2]], ptr [[AA_CASTED]], align 2
// CHECK3-NEXT: [[TMP3:%.*]] = load i32, ptr [[AA_CASTED]], align 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 2, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119.omp_outlined, i32 [[TMP1]], i32 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]], i32 noundef [[AA:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK3-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP0]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP1]] to i32
// CHECK3-NEXT: [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
// CHECK3-NEXT: [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
// CHECK3-NEXT: store i16 [[CONV2]], ptr [[AA_ADDR]], align 2
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-SAME: (i32 noundef [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[BN:%.*]], ptr noundef nonnull align 4 dereferenceable(400) [[C:%.*]], i32 noundef [[VLA1:%.*]], i32 noundef [[VLA3:%.*]], ptr noundef nonnull align 4 dereferenceable(8) [[CN:%.*]], ptr noundef nonnull align 4 dereferenceable(12) [[D:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[BN_ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[VLA_ADDR2:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[VLA_ADDR4:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[CN_ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[D_ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[A_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK3-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK3-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4
// CHECK3-NEXT: store ptr [[BN]], ptr [[BN_ADDR]], align 4
// CHECK3-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK3-NEXT: store i32 [[VLA1]], ptr [[VLA_ADDR2]], align 4
// CHECK3-NEXT: store i32 [[VLA3]], ptr [[VLA_ADDR4]], align 4
// CHECK3-NEXT: store ptr [[CN]], ptr [[CN_ADDR]], align 4
// CHECK3-NEXT: store ptr [[D]], ptr [[D_ADDR]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[VLA_ADDR]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = load ptr, ptr [[BN_ADDR]], align 4
// CHECK3-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[VLA_ADDR2]], align 4
// CHECK3-NEXT: [[TMP5:%.*]] = load i32, ptr [[VLA_ADDR4]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = load ptr, ptr [[CN_ADDR]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = load ptr, ptr [[D_ADDR]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK3-NEXT: store i32 [[TMP8]], ptr [[A_CASTED]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = load i32, ptr [[A_CASTED]], align 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 9, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144.omp_outlined, i32 [[TMP9]], ptr [[TMP0]], i32 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], i32 [[TMP4]], i32 [[TMP5]], ptr [[TMP6]], ptr [[TMP7]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[BN:%.*]], ptr noundef nonnull align 4 dereferenceable(400) [[C:%.*]], i32 noundef [[VLA1:%.*]], i32 noundef [[VLA3:%.*]], ptr noundef nonnull align 4 dereferenceable(8) [[CN:%.*]], ptr noundef nonnull align 4 dereferenceable(12) [[D:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[BN_ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[VLA_ADDR2:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[VLA_ADDR4:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[CN_ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[D_ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK3-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK3-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4
// CHECK3-NEXT: store ptr [[BN]], ptr [[BN_ADDR]], align 4
// CHECK3-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK3-NEXT: store i32 [[VLA1]], ptr [[VLA_ADDR2]], align 4
// CHECK3-NEXT: store i32 [[VLA3]], ptr [[VLA_ADDR4]], align 4
// CHECK3-NEXT: store ptr [[CN]], ptr [[CN_ADDR]], align 4
// CHECK3-NEXT: store ptr [[D]], ptr [[D_ADDR]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[VLA_ADDR]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = load ptr, ptr [[BN_ADDR]], align 4
// CHECK3-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[VLA_ADDR2]], align 4
// CHECK3-NEXT: [[TMP5:%.*]] = load i32, ptr [[VLA_ADDR4]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = load ptr, ptr [[CN_ADDR]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = load ptr, ptr [[D_ADDR]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP8]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], ptr [[TMP0]], i32 0, i32 2
// CHECK3-NEXT: [[TMP9:%.*]] = load float, ptr [[ARRAYIDX]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[CONV:%.*]] = fpext float [[TMP9]] to double
// CHECK3-NEXT: [[ADD5:%.*]] = fadd double [[CONV]], 1.000000e+00
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[CONV6:%.*]] = fptrunc double [[ADD5]] to float
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store float [[CONV6]], ptr [[ARRAYIDX]], align 4
// CHECK3-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds float, ptr [[TMP2]], i32 3
// CHECK3-NEXT: [[TMP10:%.*]] = load float, ptr [[ARRAYIDX7]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[CONV8:%.*]] = fpext float [[TMP10]] to double
// CHECK3-NEXT: [[ADD9:%.*]] = fadd double [[CONV8]], 1.000000e+00
// CHECK3-NEXT: [[CONV10:%.*]] = fptrunc double [[ADD9]] to float
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store float [[CONV10]], ptr [[ARRAYIDX7]], align 4
// CHECK3-NEXT: [[ARRAYIDX11:%.*]] = getelementptr inbounds [5 x [10 x double]], ptr [[TMP3]], i32 0, i32 1
// CHECK3-NEXT: [[ARRAYIDX12:%.*]] = getelementptr inbounds [10 x double], ptr [[ARRAYIDX11]], i32 0, i32 2
// CHECK3-NEXT: [[TMP11:%.*]] = load double, ptr [[ARRAYIDX12]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[ADD13:%.*]] = fadd double [[TMP11]], 1.000000e+00
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store double [[ADD13]], ptr [[ARRAYIDX12]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[TMP12:%.*]] = mul nsw i32 1, [[TMP5]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[ARRAYIDX14:%.*]] = getelementptr inbounds double, ptr [[TMP6]], i32 [[TMP12]]
// CHECK3-NEXT: [[ARRAYIDX15:%.*]] = getelementptr inbounds double, ptr [[ARRAYIDX14]], i32 3
// CHECK3-NEXT: [[TMP13:%.*]] = load double, ptr [[ARRAYIDX15]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[ADD16:%.*]] = fadd double [[TMP13]], 1.000000e+00
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store double [[ADD16]], ptr [[ARRAYIDX15]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[X:%.*]] = getelementptr inbounds nuw [[STRUCT_TT:%.*]], ptr [[TMP7]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[TMP14:%.*]] = load i64, ptr [[X]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[ADD17:%.*]] = add nsw i64 [[TMP14]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i64 [[ADD17]], ptr [[X]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[Y:%.*]] = getelementptr inbounds nuw [[STRUCT_TT]], ptr [[TMP7]], i32 0, i32 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[TMP15:%.*]] = load i8, ptr [[Y]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[CONV18:%.*]] = sext i8 [[TMP15]] to i32
// CHECK3-NEXT: [[ADD19:%.*]] = add nsw i32 [[CONV18]], 1
// CHECK3-NEXT: [[CONV20:%.*]] = trunc i32 [[ADD19]] to i8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i8 [[CONV20]], ptr [[Y]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@_Z3bari
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK3-SAME: (i32 noundef [[N:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[A:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[S:%.*]] = alloca [[STRUCT_S1:%.*]], align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK3-NEXT: store i32 0, ptr [[A]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load i32, ptr [[N_ADDR]], align 4
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK3-NEXT: [[CALL:%.*]] = call noundef i32 @_Z3fooi(i32 noundef [[TMP0]])
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[A]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], [[CALL]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[ADD]], ptr [[A]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[N_ADDR]], align 4
// CHECK3-NEXT: [[CALL1:%.*]] = call noundef i32 @_ZN2S12r1Ei(ptr noundef nonnull align 4 dereferenceable(8) [[S]], i32 noundef [[TMP2]])
// CHECK3-NEXT: [[TMP3:%.*]] = load i32, ptr [[A]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[ADD2:%.*]] = add nsw i32 [[TMP3]], [[CALL1]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[ADD2]], ptr [[A]], align 4
// CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[N_ADDR]], align 4
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK3-NEXT: [[CALL3:%.*]] = call noundef i32 @_ZL7fstatici(i32 noundef [[TMP4]])
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[TMP5:%.*]] = load i32, ptr [[A]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[ADD4:%.*]] = add nsw i32 [[TMP5]], [[CALL3]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[ADD4]], ptr [[A]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = load i32, ptr [[N_ADDR]], align 4
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK3-NEXT: [[CALL5:%.*]] = call noundef i32 @_Z9ftemplateIiET_i(i32 noundef [[TMP6]])
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[TMP7:%.*]] = load i32, ptr [[A]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP7]], [[CALL5]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[ADD6]], ptr [[A]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = load i32, ptr [[A]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret i32 [[TMP8]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZN2S12r1Ei
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-SAME: (ptr noundef nonnull align 4 dereferenceable(8) [[THIS:%.*]], i32 noundef [[N:%.*]]) #[[ATTR0]] comdat align 2 {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[B:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[SAVED_STACK:%.*]] = alloca ptr, align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[__VLA_EXPR0:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[B_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [5 x ptr], align 4
// CHECK3-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [5 x ptr], align 4
// CHECK3-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [5 x ptr], align 4
Revert "[OPENMP]Fix PR50347: Mapping of global scope deep object fails." This reverts commit 638938117aeae5518d6cacd066ffd9830ef4fc9a. Need to fix reported fail https://lab.llvm.org/buildbot/#/builders/193/builds/7496
2022-02-24 12:04:39 -08:00
// CHECK3-NEXT: [[DOTOFFLOAD_SIZES:%.*]] = alloca [5 x i64], align 4
[Clang][OpenMP] Insert alloca for kernel args at function entry block instead of the launch point. If an inlined kernel is called in a loop, the launch point alloca would lead to increasing stack usage every time the kernel is invoked. This could make the application run out of stack space and crash. This problem is fixed by using the alloca insertion point while creating the alloca instruction. Fixes https://github.com/llvm/llvm-project/issues/60602 Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D145820
2023-03-17 12:16:17 -04:00
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// CHECK3-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK3-NEXT: [[THIS1:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load i32, ptr [[N_ADDR]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP0]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[ADD]], ptr [[B]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[N_ADDR]], align 4
Intrinsics: Add type overload to stacksave and stackstore This allows use with non-0 address space stacks. llvm_ptr_ty should never be used. This could use some more percolation up through mlir, but this is enough to fix existing tests. https://reviews.llvm.org/D156666
2023-07-29 09:15:31 -04:00
// CHECK3-NEXT: [[TMP2:%.*]] = call ptr @llvm.stacksave.p0()
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[TMP2]], ptr [[SAVED_STACK]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[TMP3:%.*]] = mul nuw i32 2, [[TMP1]]
// CHECK3-NEXT: [[VLA:%.*]] = alloca i16, i32 [[TMP3]], align 2
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[TMP1]], ptr [[__VLA_EXPR0]], align 4
// CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[B]], align 4
// CHECK3-NEXT: store i32 [[TMP4]], ptr [[B_CASTED]], align 4
// CHECK3-NEXT: [[TMP5:%.*]] = load i32, ptr [[B_CASTED]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = load i32, ptr [[N_ADDR]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP6]], 60
// CHECK3-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK3: omp_if.then:
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[A:%.*]] = getelementptr inbounds nuw [[STRUCT_S1:%.*]], ptr [[THIS1]], i32 0, i32 0
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[TMP7:%.*]] = mul nuw i32 2, [[TMP1]]
// CHECK3-NEXT: [[TMP8:%.*]] = mul nuw i32 [[TMP7]], 2
// CHECK3-NEXT: [[TMP9:%.*]] = sext i32 [[TMP8]] to i64
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: call void @llvm.memcpy.p0.p0.i32(ptr align 4 [[DOTOFFLOAD_SIZES]], ptr align 4 @.offload_sizes.5, i32 40, i1 false)
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK3-NEXT: store ptr [[THIS1]], ptr [[TMP10]], align 4
// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: store ptr [[A]], ptr [[TMP11]], align 4
// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
// CHECK3-NEXT: store i32 [[TMP5]], ptr [[TMP13]], align 4
// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
// CHECK3-NEXT: store i32 [[TMP5]], ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 4
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
// CHECK3-NEXT: store i32 2, ptr [[TMP16]], align 4
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
// CHECK3-NEXT: store i32 2, ptr [[TMP17]], align 4
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 3
// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP19]], align 4
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 3
// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP20]], align 4
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 3
// CHECK3-NEXT: store ptr null, ptr [[TMP21]], align 4
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 4
// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP22]], align 4
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 4
// CHECK3-NEXT: store ptr [[VLA]], ptr [[TMP23]], align 4
// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 4
// CHECK3-NEXT: store i64 [[TMP9]], ptr [[TMP24]], align 4
// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 4
// CHECK3-NEXT: store ptr null, ptr [[TMP25]], align 4
// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds [5 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds [5 x i64], ptr [[DOTOFFLOAD_SIZES]], i32 0, i32 0
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
[OpenMP] Increment kernel args version, used by runtime for detecting dyn_ptr. (#85363) A kernel implicit parameter (dyn_ptr) was introduced some time back. This patch increments the kernel args version for a compiler supporting dyn_ptr. The version will be used by the runtime to determine whether the implicit parameter is generated by the compiler. The versioning is required to support use cases where code generated by an older compiler is linked with a newer runtime. If approved, this patch should be backported to release 18.
2024-03-19 16:40:22 -07:00
// CHECK3-NEXT: store i32 3, ptr [[TMP29]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 5, ptr [[TMP30]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[TMP26]], ptr [[TMP31]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[TMP27]], ptr [[TMP32]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[TMP28]], ptr [[TMP33]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP34:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: store ptr @.offload_maptypes.6, ptr [[TMP34]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP35:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr null, ptr [[TMP35]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP36:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr null, ptr [[TMP36]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP37:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i64 0, ptr [[TMP37]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP38:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i64 0, ptr [[TMP38]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP39:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP39]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP40:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP40]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP41:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i32 0, ptr [[TMP41]], align 4
// CHECK3-NEXT: [[TMP42:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.region_id, ptr [[KERNEL_ARGS]])
// CHECK3-NEXT: [[TMP43:%.*]] = icmp ne i32 [[TMP42]], 0
// CHECK3-NEXT: br i1 [[TMP43]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3: omp_offload.failed:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK3: omp_if.else:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216(ptr [[THIS1]], i32 [[TMP5]], i32 2, i32 [[TMP1]], ptr [[VLA]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: [[TMP44:%.*]] = mul nsw i32 1, [[TMP1]]
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[VLA]], i32 [[TMP44]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[ARRAYIDX2:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i32 1
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: [[TMP45:%.*]] = load i16, ptr [[ARRAYIDX2]], align 2
// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP45]] to i32
// CHECK3-NEXT: [[TMP46:%.*]] = load i32, ptr [[B]], align 4
// CHECK3-NEXT: [[ADD3:%.*]] = add nsw i32 [[CONV]], [[TMP46]]
// CHECK3-NEXT: [[TMP47:%.*]] = load ptr, ptr [[SAVED_STACK]], align 4
Intrinsics: Add type overload to stacksave and stackstore This allows use with non-0 address space stacks. llvm_ptr_ty should never be used. This could use some more percolation up through mlir, but this is enough to fix existing tests. https://reviews.llvm.org/D156666
2023-07-29 09:15:31 -04:00
// CHECK3-NEXT: call void @llvm.stackrestore.p0(ptr [[TMP47]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret i32 [[ADD3]]
//
//
// CHECK3-LABEL: define {{[^@]+}}@_ZL7fstatici
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK3-SAME: (i32 noundef [[N:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[A:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA:%.*]] = alloca i16, align 2
// CHECK3-NEXT: [[AAA:%.*]] = alloca i8, align 1
// CHECK3-NEXT: [[B:%.*]] = alloca [10 x i32], align 4
// CHECK3-NEXT: [[A_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AAA_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [4 x ptr], align 4
// CHECK3-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [4 x ptr], align 4
// CHECK3-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [4 x ptr], align 4
[Clang][OpenMP] Insert alloca for kernel args at function entry block instead of the launch point. If an inlined kernel is called in a loop, the launch point alloca would lead to increasing stack usage every time the kernel is invoked. This could make the application run out of stack space and crash. This problem is fixed by using the alloca insertion point while creating the alloca instruction. Fixes https://github.com/llvm/llvm-project/issues/60602 Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D145820
2023-03-17 12:16:17 -04:00
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK3-NEXT: store i32 0, ptr [[A]], align 4
// CHECK3-NEXT: store i16 0, ptr [[AA]], align 2
// CHECK3-NEXT: store i8 0, ptr [[AAA]], align 1
// CHECK3-NEXT: [[TMP0:%.*]] = load i32, ptr [[A]], align 4
// CHECK3-NEXT: store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA]], align 2
// CHECK3-NEXT: store i16 [[TMP2]], ptr [[AA_CASTED]], align 2
// CHECK3-NEXT: [[TMP3:%.*]] = load i32, ptr [[AA_CASTED]], align 4
// CHECK3-NEXT: [[TMP4:%.*]] = load i8, ptr [[AAA]], align 1
// CHECK3-NEXT: store i8 [[TMP4]], ptr [[AAA_CASTED]], align 1
// CHECK3-NEXT: [[TMP5:%.*]] = load i32, ptr [[AAA_CASTED]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = load i32, ptr [[N_ADDR]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP6]], 50
// CHECK3-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK3: omp_if.then:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// CHECK3-NEXT: store ptr null, ptr [[TMP9]], align 4
// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP10]], align 4
// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP11]], align 4
// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
// CHECK3-NEXT: store ptr null, ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
// CHECK3-NEXT: store i32 [[TMP5]], ptr [[TMP13]], align 4
// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
// CHECK3-NEXT: store i32 [[TMP5]], ptr [[TMP14]], align 4
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
// CHECK3-NEXT: store ptr null, ptr [[TMP15]], align 4
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 3
// CHECK3-NEXT: store ptr [[B]], ptr [[TMP16]], align 4
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 3
// CHECK3-NEXT: store ptr [[B]], ptr [[TMP17]], align 4
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 3
// CHECK3-NEXT: store ptr null, ptr [[TMP18]], align 4
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds [4 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
[OpenMP] Increment kernel args version, used by runtime for detecting dyn_ptr. (#85363) A kernel implicit parameter (dyn_ptr) was introduced some time back. This patch increments the kernel args version for a compiler supporting dyn_ptr. The version will be used by the runtime to determine whether the implicit parameter is generated by the compiler. The versioning is required to support use cases where code generated by an older compiler is linked with a newer runtime. If approved, this patch should be backported to release 18.
2024-03-19 16:40:22 -07:00
// CHECK3-NEXT: store i32 3, ptr [[TMP21]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 4, ptr [[TMP22]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[TMP19]], ptr [[TMP23]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[TMP20]], ptr [[TMP24]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: store ptr @.offload_sizes.7, ptr [[TMP25]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: store ptr @.offload_maptypes.8, ptr [[TMP26]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr null, ptr [[TMP27]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr null, ptr [[TMP28]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP29:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i64 0, ptr [[TMP29]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP30:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i64 0, ptr [[TMP30]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP31:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP31]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP32:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP32]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP33:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i32 0, ptr [[TMP33]], align 4
// CHECK3-NEXT: [[TMP34:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198.region_id, ptr [[KERNEL_ARGS]])
// CHECK3-NEXT: [[TMP35:%.*]] = icmp ne i32 [[TMP34]], 0
// CHECK3-NEXT: br i1 [[TMP35]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3: omp_offload.failed:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], ptr [[B]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK3: omp_if.else:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198(i32 [[TMP1]], i32 [[TMP3]], i32 [[TMP5]], ptr [[B]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: [[TMP36:%.*]] = load i32, ptr [[A]], align 4
// CHECK3-NEXT: ret i32 [[TMP36]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
//
//
// CHECK3-LABEL: define {{[^@]+}}@_Z9ftemplateIiET_i
[Clang/Test]: Rename enable_noundef_analysis to disable-noundef-analysis and turn it off by default Turning on `enable_noundef_analysis` flag allows better codegen by removing freeze instructions. I modified clang by renaming `enable_noundef_analysis` flag to `disable-noundef-analysis` and turning it off by default. Test updates are made as a separate patch: D108453 Reviewed By: eugenis Differential Revision: https://reviews.llvm.org/D105169
2022-01-16 18:53:11 +09:00
// CHECK3-SAME: (i32 noundef [[N:%.*]]) #[[ATTR0]] comdat {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[N_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[A:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA:%.*]] = alloca i16, align 2
// CHECK3-NEXT: [[B:%.*]] = alloca [10 x i32], align 4
// CHECK3-NEXT: [[A_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[DOTOFFLOAD_BASEPTRS:%.*]] = alloca [3 x ptr], align 4
// CHECK3-NEXT: [[DOTOFFLOAD_PTRS:%.*]] = alloca [3 x ptr], align 4
// CHECK3-NEXT: [[DOTOFFLOAD_MAPPERS:%.*]] = alloca [3 x ptr], align 4
[Clang][OpenMP] Insert alloca for kernel args at function entry block instead of the launch point. If an inlined kernel is called in a loop, the launch point alloca would lead to increasing stack usage every time the kernel is invoked. This could make the application run out of stack space and crash. This problem is fixed by using the alloca insertion point while creating the alloca instruction. Fixes https://github.com/llvm/llvm-project/issues/60602 Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D145820
2023-03-17 12:16:17 -04:00
// CHECK3-NEXT: [[KERNEL_ARGS:%.*]] = alloca [[STRUCT___TGT_KERNEL_ARGUMENTS:%.*]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[N]], ptr [[N_ADDR]], align 4
// CHECK3-NEXT: store i32 0, ptr [[A]], align 4
// CHECK3-NEXT: store i16 0, ptr [[AA]], align 2
// CHECK3-NEXT: [[TMP0:%.*]] = load i32, ptr [[A]], align 4
// CHECK3-NEXT: store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA]], align 2
// CHECK3-NEXT: store i16 [[TMP2]], ptr [[AA_CASTED]], align 2
// CHECK3-NEXT: [[TMP3:%.*]] = load i32, ptr [[AA_CASTED]], align 4
// CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[N_ADDR]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[CMP:%.*]] = icmp sgt i32 [[TMP4]], 40
// CHECK3-NEXT: br i1 [[CMP]], label [[OMP_IF_THEN:%.*]], label [[OMP_IF_ELSE:%.*]]
// CHECK3: omp_if.then:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[TMP5:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP5]], align 4
// CHECK3-NEXT: [[TMP6:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
// CHECK3-NEXT: store i32 [[TMP1]], ptr [[TMP6]], align 4
// CHECK3-NEXT: [[TMP7:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 0
// CHECK3-NEXT: store ptr null, ptr [[TMP7]], align 4
// CHECK3-NEXT: [[TMP8:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 1
// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP8]], align 4
// CHECK3-NEXT: [[TMP9:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 1
// CHECK3-NEXT: store i32 [[TMP3]], ptr [[TMP9]], align 4
// CHECK3-NEXT: [[TMP10:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 1
// CHECK3-NEXT: store ptr null, ptr [[TMP10]], align 4
// CHECK3-NEXT: [[TMP11:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 2
// CHECK3-NEXT: store ptr [[B]], ptr [[TMP11]], align 4
// CHECK3-NEXT: [[TMP12:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 2
// CHECK3-NEXT: store ptr [[B]], ptr [[TMP12]], align 4
// CHECK3-NEXT: [[TMP13:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_MAPPERS]], i32 0, i32 2
// CHECK3-NEXT: store ptr null, ptr [[TMP13]], align 4
// CHECK3-NEXT: [[TMP14:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_BASEPTRS]], i32 0, i32 0
// CHECK3-NEXT: [[TMP15:%.*]] = getelementptr inbounds [3 x ptr], ptr [[DOTOFFLOAD_PTRS]], i32 0, i32 0
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP16:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 0
[OpenMP] Increment kernel args version, used by runtime for detecting dyn_ptr. (#85363) A kernel implicit parameter (dyn_ptr) was introduced some time back. This patch increments the kernel args version for a compiler supporting dyn_ptr. The version will be used by the runtime to determine whether the implicit parameter is generated by the compiler. The versioning is required to support use cases where code generated by an older compiler is linked with a newer runtime. If approved, this patch should be backported to release 18.
2024-03-19 16:40:22 -07:00
// CHECK3-NEXT: store i32 3, ptr [[TMP16]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP17:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 3, ptr [[TMP17]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP18:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 2
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[TMP14]], ptr [[TMP18]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP19:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 3
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[TMP15]], ptr [[TMP19]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP20:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: store ptr @.offload_sizes.9, ptr [[TMP20]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP21:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 5
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: store ptr @.offload_maptypes.10, ptr [[TMP21]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP22:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 6
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr null, ptr [[TMP22]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP23:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 7
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr null, ptr [[TMP23]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP24:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i64 0, ptr [[TMP24]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP25:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 9
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i64 0, ptr [[TMP25]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP26:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 10
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store [3 x i32] [i32 1, i32 0, i32 0], ptr [[TMP26]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP27:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 11
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store [3 x i32] zeroinitializer, ptr [[TMP27]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[TMP28:%.*]] = getelementptr inbounds nuw [[STRUCT___TGT_KERNEL_ARGUMENTS]], ptr [[KERNEL_ARGS]], i32 0, i32 12
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: store i32 0, ptr [[TMP28]], align 4
// CHECK3-NEXT: [[TMP29:%.*]] = call i32 @__tgt_target_kernel(ptr @[[GLOB1]], i64 -1, i32 1, i32 0, ptr @.{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181.region_id, ptr [[KERNEL_ARGS]])
// CHECK3-NEXT: [[TMP30:%.*]] = icmp ne i32 [[TMP29]], 0
// CHECK3-NEXT: br i1 [[TMP30]], label [[OMP_OFFLOAD_FAILED:%.*]], label [[OMP_OFFLOAD_CONT:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3: omp_offload.failed:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: br label [[OMP_OFFLOAD_CONT]]
// CHECK3: omp_offload.cont:
// CHECK3-NEXT: br label [[OMP_IF_END:%.*]]
// CHECK3: omp_if.else:
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-NEXT: call void @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181(i32 [[TMP1]], i32 [[TMP3]], ptr [[B]]) #[[ATTR3]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: br label [[OMP_IF_END]]
// CHECK3: omp_if.end:
[OpenMP] Modernize the kernel launching interface and APIs We already created a versioned `__tgt_kernel_arguments` struct but it was only briefly used and its content was passed in isolation anyway. This makes it hard to add more information in the future. With this patch we fully embrace the struct as means to pass information from the compiler to the plugin as part of a kernel launch. The patch also extends and renames the struct, bumping the version number to 2. Version 1 entries are auto-upgraded. This is in preparation for "bare" kernel launches, per kernel dynamic shared memory, CUDA/HIP lowering, etc. The `__tgt_target_kernel_nowait` interface was deprecated as it was unused. Once we actually implement support for something like that, we can add an appropriate API. Note: Only plugins with the `launch_kernel` interface are now supported. That means that a new clang won't be able to use an old runtime. An old clang can still use the new runtime since the libomptarget interface did not change. Differential Revision: https://reviews.llvm.org/D141232
2023-01-19 13:40:58 -08:00
// CHECK3-NEXT: [[TMP31:%.*]] = load i32, ptr [[A]], align 4
// CHECK3-NEXT: ret i32 [[TMP31]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-SAME: (ptr noundef [[THIS:%.*]], i32 noundef [[B:%.*]], i32 noundef [[VLA:%.*]], i32 noundef [[VLA1:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[C:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[B_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[VLA_ADDR2:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[B_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// CHECK3-NEXT: store i32 [[B]], ptr [[B_ADDR]], align 4
// CHECK3-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4
// CHECK3-NEXT: store i32 [[VLA1]], ptr [[VLA_ADDR2]], align 4
// CHECK3-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[VLA_ADDR]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[VLA_ADDR2]], align 4
// CHECK3-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[B_ADDR]], align 4
// CHECK3-NEXT: store i32 [[TMP4]], ptr [[B_CASTED]], align 4
// CHECK3-NEXT: [[TMP5:%.*]] = load i32, ptr [[B_CASTED]], align 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.omp_outlined, ptr [[TMP0]], i32 [[TMP5]], i32 [[TMP1]], i32 [[TMP2]], ptr [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]], i32 noundef [[B:%.*]], i32 noundef [[VLA:%.*]], i32 noundef [[VLA1:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[C:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[B_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[VLA_ADDR2:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// CHECK3-NEXT: store i32 [[B]], ptr [[B_ADDR]], align 4
// CHECK3-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4
// CHECK3-NEXT: store i32 [[VLA1]], ptr [[VLA_ADDR2]], align 4
// CHECK3-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[VLA_ADDR]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[VLA_ADDR2]], align 4
// CHECK3-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[B_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP4]] to double
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: [[ADD:%.*]] = fadd double [[CONV]], 1.500000e+00
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[A:%.*]] = getelementptr inbounds nuw [[STRUCT_S1:%.*]], ptr [[TMP0]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store double [[ADD]], ptr [[A]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK3-NEXT: [[A3:%.*]] = getelementptr inbounds nuw [[STRUCT_S1]], ptr [[TMP0]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[TMP5:%.*]] = load double, ptr [[A3]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[INC:%.*]] = fadd double [[TMP5]], 1.000000e+00
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store double [[INC]], ptr [[A3]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[CONV4:%.*]] = fptosi double [[INC]] to i16
// CHECK3-NEXT: [[TMP6:%.*]] = mul nsw i32 1, [[TMP2]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[TMP3]], i32 [[TMP6]]
// CHECK3-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i32 1
// CHECK3-NEXT: store i16 [[CONV4]], ptr [[ARRAYIDX5]], align 2
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-SAME: (i32 noundef [[A:%.*]], i32 noundef [[AA:%.*]], i32 noundef [[AAA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AAA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[A_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AAA_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK3-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK3-NEXT: store i32 [[AAA]], ptr [[AAA_ADDR]], align 4
// CHECK3-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK3-NEXT: store i32 [[TMP1]], ptr [[A_CASTED]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[A_CASTED]], align 4
// CHECK3-NEXT: [[TMP3:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK3-NEXT: store i16 [[TMP3]], ptr [[AA_CASTED]], align 2
// CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[AA_CASTED]], align 4
// CHECK3-NEXT: [[TMP5:%.*]] = load i8, ptr [[AAA_ADDR]], align 1
// CHECK3-NEXT: store i8 [[TMP5]], ptr [[AAA_CASTED]], align 1
// CHECK3-NEXT: [[TMP6:%.*]] = load i32, ptr [[AAA_CASTED]], align 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198.omp_outlined, i32 [[TMP2]], i32 [[TMP4]], i32 [[TMP6]], ptr [[TMP0]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]], i32 noundef [[AA:%.*]], i32 noundef [[AAA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AAA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK3-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK3-NEXT: store i32 [[AAA]], ptr [[AAA_ADDR]], align 4
// CHECK3-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP2]] to i32
// CHECK3-NEXT: [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
// CHECK3-NEXT: [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
// CHECK3-NEXT: store i16 [[CONV2]], ptr [[AA_ADDR]], align 2
// CHECK3-NEXT: [[TMP3:%.*]] = load i8, ptr [[AAA_ADDR]], align 1
// CHECK3-NEXT: [[CONV3:%.*]] = sext i8 [[TMP3]] to i32
// CHECK3-NEXT: [[ADD4:%.*]] = add nsw i32 [[CONV3]], 1
// CHECK3-NEXT: [[CONV5:%.*]] = trunc i32 [[ADD4]] to i8
// CHECK3-NEXT: store i8 [[CONV5]], ptr [[AAA_ADDR]], align 1
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 2
// CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
// CHECK3-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP4]], 1
// CHECK3-NEXT: store i32 [[ADD6]], ptr [[ARRAYIDX]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret void
//
//
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-SAME: (i32 noundef [[A:%.*]], i32 noundef [[AA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[A_CASTED:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK3-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK3-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK3-NEXT: store i32 [[TMP1]], ptr [[A_CASTED]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = load i32, ptr [[A_CASTED]], align 4
// CHECK3-NEXT: [[TMP3:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK3-NEXT: store i16 [[TMP3]], ptr [[AA_CASTED]], align 2
// CHECK3-NEXT: [[TMP4:%.*]] = load i32, ptr [[AA_CASTED]], align 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181.omp_outlined, i32 [[TMP2]], i32 [[TMP4]], ptr [[TMP0]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK3-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK3-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]], i32 noundef [[AA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR2]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK3-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 4
// CHECK3-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK3-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK3-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK3-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK3-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK3-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK3-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK3-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK3-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK3-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK3-NEXT: [[CONV:%.*]] = sext i16 [[TMP2]] to i32
// CHECK3-NEXT: [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
// CHECK3-NEXT: [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
// CHECK3-NEXT: store i16 [[CONV2]], ptr [[AA_ADDR]], align 2
// CHECK3-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 2
// CHECK3-NEXT: [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
// CHECK3-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP3]], 1
// CHECK3-NEXT: store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK3-NEXT: ret void
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-SAME: (ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0:[0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: entry:
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-NEXT: [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK9-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1:[0-9]+]], i32 0, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100.omp_outlined)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: ret void
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], i64 noundef [[AA:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: entry:
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-NEXT: [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[AA_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-NEXT: store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK9-NEXT: store i16 [[TMP0]], ptr [[AA_CASTED]], align 2
// CHECK9-NEXT: [[TMP1:%.*]] = load i64, ptr [[AA_CASTED]], align 8
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK9-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110.omp_outlined, i64 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[AA:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK9-NEXT: [[CONV:%.*]] = sext i16 [[TMP0]] to i32
// CHECK9-NEXT: [[ADD:%.*]] = add nsw i32 [[CONV]], 1
// CHECK9-NEXT: [[CONV1:%.*]] = trunc i32 [[ADD]] to i16
// CHECK9-NEXT: store i16 [[CONV1]], ptr [[AA_ADDR]], align 2
// CHECK9-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK9-NEXT: [[TMP3:%.*]] = call i32 @__kmpc_cancel(ptr @[[GLOB1]], i32 [[TMP2]], i32 1)
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[TMP4:%.*]] = icmp ne i32 [[TMP3]], 0
// CHECK9-NEXT: br i1 [[TMP4]], label [[DOTCANCEL_EXIT:%.*]], label [[DOTCANCEL_CONTINUE:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9: .cancel.exit:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_cancel_barrier(ptr @[[GLOB2:[0-9]+]], i32 [[TMP2]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: br label [[DOTCANCEL_CONTINUE]]
// CHECK9: .cancel.continue:
// CHECK9-NEXT: ret void
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], i64 noundef [[A:%.*]], i64 noundef [[AA:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: entry:
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-NEXT: [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[AA_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-NEXT: store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK9-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK9-NEXT: store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK9-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK9-NEXT: store i16 [[TMP2]], ptr [[AA_CASTED]], align 2
// CHECK9-NEXT: [[TMP3:%.*]] = load i64, ptr [[AA_CASTED]], align 8
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK9-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 2, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119.omp_outlined, i64 [[TMP1]], i64 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]], i64 noundef [[AA:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK9-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP0]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK9-NEXT: [[TMP1:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK9-NEXT: [[CONV:%.*]] = sext i16 [[TMP1]] to i32
// CHECK9-NEXT: [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
// CHECK9-NEXT: [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
// CHECK9-NEXT: store i16 [[CONV2]], ptr [[AA_ADDR]], align 2
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 1d66649adf28d48ae1731516d87fb899426e3349. Revert to fix AMG GPU issue.
2021-09-21 13:20:39 -07:00
// CHECK9-NEXT: ret void
//
//
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], i64 noundef [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[BN:%.*]], ptr noundef nonnull align 8 dereferenceable(400) [[C:%.*]], i64 noundef [[VLA1:%.*]], i64 noundef [[VLA3:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[CN:%.*]], ptr noundef nonnull align 8 dereferenceable(16) [[D:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: entry:
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-NEXT: [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[BN_ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: [[VLA_ADDR2:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[VLA_ADDR4:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[CN_ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: [[D_ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-NEXT: store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK9-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK9-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8
// CHECK9-NEXT: store ptr [[BN]], ptr [[BN_ADDR]], align 8
// CHECK9-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK9-NEXT: store i64 [[VLA1]], ptr [[VLA_ADDR2]], align 8
// CHECK9-NEXT: store i64 [[VLA3]], ptr [[VLA_ADDR4]], align 8
// CHECK9-NEXT: store ptr [[CN]], ptr [[CN_ADDR]], align 8
// CHECK9-NEXT: store ptr [[D]], ptr [[D_ADDR]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK9-NEXT: [[TMP1:%.*]] = load i64, ptr [[VLA_ADDR]], align 8
// CHECK9-NEXT: [[TMP2:%.*]] = load ptr, ptr [[BN_ADDR]], align 8
// CHECK9-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
// CHECK9-NEXT: [[TMP4:%.*]] = load i64, ptr [[VLA_ADDR2]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = load i64, ptr [[VLA_ADDR4]], align 8
// CHECK9-NEXT: [[TMP6:%.*]] = load ptr, ptr [[CN_ADDR]], align 8
// CHECK9-NEXT: [[TMP7:%.*]] = load ptr, ptr [[D_ADDR]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK9-NEXT: store i32 [[TMP8]], ptr [[A_CASTED]], align 4
// CHECK9-NEXT: [[TMP9:%.*]] = load i64, ptr [[A_CASTED]], align 8
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK9-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 9, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144.omp_outlined, i64 [[TMP9]], ptr [[TMP0]], i64 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], i64 [[TMP4]], i64 [[TMP5]], ptr [[TMP6]], ptr [[TMP7]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]], i64 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[BN:%.*]], ptr noundef nonnull align 8 dereferenceable(400) [[C:%.*]], i64 noundef [[VLA1:%.*]], i64 noundef [[VLA3:%.*]], ptr noundef nonnull align 8 dereferenceable(8) [[CN:%.*]], ptr noundef nonnull align 8 dereferenceable(16) [[D:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[BN_ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[VLA_ADDR2:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[VLA_ADDR4:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[CN_ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: [[D_ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK9-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK9-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8
// CHECK9-NEXT: store ptr [[BN]], ptr [[BN_ADDR]], align 8
// CHECK9-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK9-NEXT: store i64 [[VLA1]], ptr [[VLA_ADDR2]], align 8
// CHECK9-NEXT: store i64 [[VLA3]], ptr [[VLA_ADDR4]], align 8
// CHECK9-NEXT: store ptr [[CN]], ptr [[CN_ADDR]], align 8
// CHECK9-NEXT: store ptr [[D]], ptr [[D_ADDR]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK9-NEXT: [[TMP1:%.*]] = load i64, ptr [[VLA_ADDR]], align 8
// CHECK9-NEXT: [[TMP2:%.*]] = load ptr, ptr [[BN_ADDR]], align 8
// CHECK9-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
// CHECK9-NEXT: [[TMP4:%.*]] = load i64, ptr [[VLA_ADDR2]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = load i64, ptr [[VLA_ADDR4]], align 8
// CHECK9-NEXT: [[TMP6:%.*]] = load ptr, ptr [[CN_ADDR]], align 8
// CHECK9-NEXT: [[TMP7:%.*]] = load ptr, ptr [[D_ADDR]], align 8
// CHECK9-NEXT: [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP8]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK9-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], ptr [[TMP0]], i64 0, i64 2
// CHECK9-NEXT: [[TMP9:%.*]] = load float, ptr [[ARRAYIDX]], align 4
// CHECK9-NEXT: [[CONV:%.*]] = fpext float [[TMP9]] to double
// CHECK9-NEXT: [[ADD5:%.*]] = fadd double [[CONV]], 1.000000e+00
// CHECK9-NEXT: [[CONV6:%.*]] = fptrunc double [[ADD5]] to float
// CHECK9-NEXT: store float [[CONV6]], ptr [[ARRAYIDX]], align 4
// CHECK9-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds float, ptr [[TMP2]], i64 3
// CHECK9-NEXT: [[TMP10:%.*]] = load float, ptr [[ARRAYIDX7]], align 4
// CHECK9-NEXT: [[CONV8:%.*]] = fpext float [[TMP10]] to double
// CHECK9-NEXT: [[ADD9:%.*]] = fadd double [[CONV8]], 1.000000e+00
// CHECK9-NEXT: [[CONV10:%.*]] = fptrunc double [[ADD9]] to float
// CHECK9-NEXT: store float [[CONV10]], ptr [[ARRAYIDX7]], align 4
// CHECK9-NEXT: [[ARRAYIDX11:%.*]] = getelementptr inbounds [5 x [10 x double]], ptr [[TMP3]], i64 0, i64 1
// CHECK9-NEXT: [[ARRAYIDX12:%.*]] = getelementptr inbounds [10 x double], ptr [[ARRAYIDX11]], i64 0, i64 2
// CHECK9-NEXT: [[TMP11:%.*]] = load double, ptr [[ARRAYIDX12]], align 8
// CHECK9-NEXT: [[ADD13:%.*]] = fadd double [[TMP11]], 1.000000e+00
// CHECK9-NEXT: store double [[ADD13]], ptr [[ARRAYIDX12]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[TMP12:%.*]] = mul nsw i64 1, [[TMP5]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[ARRAYIDX14:%.*]] = getelementptr inbounds double, ptr [[TMP6]], i64 [[TMP12]]
// CHECK9-NEXT: [[ARRAYIDX15:%.*]] = getelementptr inbounds double, ptr [[ARRAYIDX14]], i64 3
// CHECK9-NEXT: [[TMP13:%.*]] = load double, ptr [[ARRAYIDX15]], align 8
// CHECK9-NEXT: [[ADD16:%.*]] = fadd double [[TMP13]], 1.000000e+00
// CHECK9-NEXT: store double [[ADD16]], ptr [[ARRAYIDX15]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK9-NEXT: [[X:%.*]] = getelementptr inbounds nuw [[STRUCT_TT:%.*]], ptr [[TMP7]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[TMP14:%.*]] = load i64, ptr [[X]], align 8
// CHECK9-NEXT: [[ADD17:%.*]] = add nsw i64 [[TMP14]], 1
// CHECK9-NEXT: store i64 [[ADD17]], ptr [[X]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK9-NEXT: [[Y:%.*]] = getelementptr inbounds nuw [[STRUCT_TT]], ptr [[TMP7]], i32 0, i32 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[TMP15:%.*]] = load i8, ptr [[Y]], align 8
// CHECK9-NEXT: [[CONV18:%.*]] = sext i8 [[TMP15]] to i32
// CHECK9-NEXT: [[ADD19:%.*]] = add nsw i32 [[CONV18]], 1
// CHECK9-NEXT: [[CONV20:%.*]] = trunc i32 [[ADD19]] to i8
// CHECK9-NEXT: store i8 [[CONV20]], ptr [[Y]], align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: ret void
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], i64 noundef [[A:%.*]], i64 noundef [[AA:%.*]], i64 noundef [[AAA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: entry:
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-NEXT: [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[AAA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[AA_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[AAA_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-NEXT: store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK9-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK9-NEXT: store i64 [[AAA]], ptr [[AAA_ADDR]], align 8
// CHECK9-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK9-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK9-NEXT: store i32 [[TMP1]], ptr [[A_CASTED]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK9-NEXT: [[TMP3:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK9-NEXT: store i16 [[TMP3]], ptr [[AA_CASTED]], align 2
// CHECK9-NEXT: [[TMP4:%.*]] = load i64, ptr [[AA_CASTED]], align 8
// CHECK9-NEXT: [[TMP5:%.*]] = load i8, ptr [[AAA_ADDR]], align 1
// CHECK9-NEXT: store i8 [[TMP5]], ptr [[AAA_CASTED]], align 1
// CHECK9-NEXT: [[TMP6:%.*]] = load i64, ptr [[AAA_CASTED]], align 8
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK9-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198.omp_outlined, i64 [[TMP2]], i64 [[TMP4]], i64 [[TMP6]], ptr [[TMP0]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]], i64 noundef [[AA:%.*]], i64 noundef [[AAA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[AAA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK9-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK9-NEXT: store i64 [[AAA]], ptr [[AAA_ADDR]], align 8
// CHECK9-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK9-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK9-NEXT: [[CONV:%.*]] = sext i16 [[TMP2]] to i32
// CHECK9-NEXT: [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
// CHECK9-NEXT: [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
// CHECK9-NEXT: store i16 [[CONV2]], ptr [[AA_ADDR]], align 2
// CHECK9-NEXT: [[TMP3:%.*]] = load i8, ptr [[AAA_ADDR]], align 1
// CHECK9-NEXT: [[CONV3:%.*]] = sext i8 [[TMP3]] to i32
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: [[ADD4:%.*]] = add nsw i32 [[CONV3]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[CONV5:%.*]] = trunc i32 [[ADD4]] to i8
// CHECK9-NEXT: store i8 [[CONV5]], ptr [[AAA_ADDR]], align 1
// CHECK9-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 2
// CHECK9-NEXT: [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
// CHECK9-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP4]], 1
// CHECK9-NEXT: store i32 [[ADD6]], ptr [[ARRAYIDX]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: ret void
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef [[THIS:%.*]], i64 noundef [[B:%.*]], i64 noundef [[VLA:%.*]], i64 noundef [[VLA1:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[C:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: entry:
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-NEXT: [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[VLA_ADDR2:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[B_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-NEXT: store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK9-NEXT: store i64 [[B]], ptr [[B_ADDR]], align 8
// CHECK9-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8
// CHECK9-NEXT: store i64 [[VLA1]], ptr [[VLA_ADDR2]], align 8
// CHECK9-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK9-NEXT: [[TMP1:%.*]] = load i64, ptr [[VLA_ADDR]], align 8
// CHECK9-NEXT: [[TMP2:%.*]] = load i64, ptr [[VLA_ADDR2]], align 8
// CHECK9-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
// CHECK9-NEXT: [[TMP4:%.*]] = load i32, ptr [[B_ADDR]], align 4
// CHECK9-NEXT: store i32 [[TMP4]], ptr [[B_CASTED]], align 4
// CHECK9-NEXT: [[TMP5:%.*]] = load i64, ptr [[B_CASTED]], align 8
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK9-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.omp_outlined, ptr [[TMP0]], i64 [[TMP5]], i64 [[TMP1]], i64 [[TMP2]], ptr [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]], i64 noundef [[B:%.*]], i64 noundef [[VLA:%.*]], i64 noundef [[VLA1:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[C:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[B_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[VLA_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[VLA_ADDR2:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 8
// CHECK9-NEXT: store i64 [[B]], ptr [[B_ADDR]], align 8
// CHECK9-NEXT: store i64 [[VLA]], ptr [[VLA_ADDR]], align 8
// CHECK9-NEXT: store i64 [[VLA1]], ptr [[VLA_ADDR2]], align 8
// CHECK9-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 8
// CHECK9-NEXT: [[TMP1:%.*]] = load i64, ptr [[VLA_ADDR]], align 8
// CHECK9-NEXT: [[TMP2:%.*]] = load i64, ptr [[VLA_ADDR2]], align 8
// CHECK9-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 8
// CHECK9-NEXT: [[TMP4:%.*]] = load i32, ptr [[B_ADDR]], align 4
// CHECK9-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP4]] to double
// CHECK9-NEXT: [[ADD:%.*]] = fadd double [[CONV]], 1.500000e+00
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK9-NEXT: [[A:%.*]] = getelementptr inbounds nuw [[STRUCT_S1:%.*]], ptr [[TMP0]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: store double [[ADD]], ptr [[A]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK9-NEXT: [[A3:%.*]] = getelementptr inbounds nuw [[STRUCT_S1]], ptr [[TMP0]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[TMP5:%.*]] = load double, ptr [[A3]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[INC:%.*]] = fadd double [[TMP5]], 1.000000e+00
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: store double [[INC]], ptr [[A3]], align 8
// CHECK9-NEXT: [[CONV4:%.*]] = fptosi double [[INC]] to i16
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[TMP6:%.*]] = mul nsw i64 1, [[TMP2]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[TMP3]], i64 [[TMP6]]
// CHECK9-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i64 1
// CHECK9-NEXT: store i16 [[CONV4]], ptr [[ARRAYIDX5]], align 2
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: ret void
//
//
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], i64 noundef [[A:%.*]], i64 noundef [[AA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: entry:
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-NEXT: [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 8
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[A_CASTED:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[AA_CASTED:%.*]] = alloca i64, align 8
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK9-NEXT: store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK9-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK9-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK9-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK9-NEXT: store i32 [[TMP1]], ptr [[A_CASTED]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = load i64, ptr [[A_CASTED]], align 8
// CHECK9-NEXT: [[TMP3:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK9-NEXT: store i16 [[TMP3]], ptr [[AA_CASTED]], align 2
// CHECK9-NEXT: [[TMP4:%.*]] = load i64, ptr [[AA_CASTED]], align 8
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK9-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181.omp_outlined, i64 [[TMP2]], i64 [[TMP4]], ptr [[TMP0]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK9-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK9-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i64 noundef [[A:%.*]], i64 noundef [[AA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[A_ADDR:%.*]] = alloca i64, align 8
// CHECK9-NEXT: [[AA_ADDR:%.*]] = alloca i64, align 8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 8
// CHECK9-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 8
// CHECK9-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 8
// CHECK9-NEXT: store i64 [[A]], ptr [[A_ADDR]], align 8
// CHECK9-NEXT: store i64 [[AA]], ptr [[AA_ADDR]], align 8
// CHECK9-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 8
// CHECK9-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 8
// CHECK9-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK9-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK9-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK9-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK9-NEXT: [[CONV:%.*]] = sext i16 [[TMP2]] to i32
// CHECK9-NEXT: [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
// CHECK9-NEXT: [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
// CHECK9-NEXT: store i16 [[CONV2]], ptr [[AA_ADDR]], align 2
// CHECK9-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i64 0, i64 2
// CHECK9-NEXT: [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
// CHECK9-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP3]], 1
// CHECK9-NEXT: store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK9-NEXT: ret void
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-SAME: (ptr noalias noundef [[DYN_PTR:%.*]]) #[[ATTR0:[0-9]+]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: entry:
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-NEXT: [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK11-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1:[0-9]+]], i32 0, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100.omp_outlined)
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l100.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK11-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: ret void
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], i32 noundef [[AA:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: entry:
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-NEXT: [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[AA_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-NEXT: store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK11-NEXT: [[TMP0:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK11-NEXT: store i16 [[TMP0]], ptr [[AA_CASTED]], align 2
// CHECK11-NEXT: [[TMP1:%.*]] = load i32, ptr [[AA_CASTED]], align 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK11-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 1, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110.omp_outlined, i32 [[TMP1]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l110.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[AA:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK11-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK11-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK11-NEXT: [[TMP0:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK11-NEXT: [[CONV:%.*]] = sext i16 [[TMP0]] to i32
// CHECK11-NEXT: [[ADD:%.*]] = add nsw i32 [[CONV]], 1
// CHECK11-NEXT: [[CONV1:%.*]] = trunc i32 [[ADD]] to i16
// CHECK11-NEXT: store i16 [[CONV1]], ptr [[AA_ADDR]], align 2
// CHECK11-NEXT: [[TMP1:%.*]] = load ptr, ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = load i32, ptr [[TMP1]], align 4
// CHECK11-NEXT: [[TMP3:%.*]] = call i32 @__kmpc_cancel(ptr @[[GLOB1]], i32 [[TMP2]], i32 1)
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[TMP4:%.*]] = icmp ne i32 [[TMP3]], 0
// CHECK11-NEXT: br i1 [[TMP4]], label [[DOTCANCEL_EXIT:%.*]], label [[DOTCANCEL_CONTINUE:%.*]]
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11: .cancel.exit:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[TMP5:%.*]] = call i32 @__kmpc_cancel_barrier(ptr @[[GLOB2:[0-9]+]], i32 [[TMP2]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: br label [[DOTCANCEL_CONTINUE]]
// CHECK11: .cancel.continue:
// CHECK11-NEXT: ret void
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], i32 noundef [[A:%.*]], i32 noundef [[AA:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: entry:
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-NEXT: [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[A_CASTED:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[AA_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-NEXT: store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK11-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK11-NEXT: [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK11-NEXT: store i32 [[TMP0]], ptr [[A_CASTED]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_CASTED]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK11-NEXT: store i16 [[TMP2]], ptr [[AA_CASTED]], align 2
// CHECK11-NEXT: [[TMP3:%.*]] = load i32, ptr [[AA_CASTED]], align 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK11-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 2, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119.omp_outlined, i32 [[TMP1]], i32 [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l119.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]], i32 noundef [[AA:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK11-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK11-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK11-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK11-NEXT: [[TMP0:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP0]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK11-NEXT: [[CONV:%.*]] = sext i16 [[TMP1]] to i32
// CHECK11-NEXT: [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
// CHECK11-NEXT: [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
// CHECK11-NEXT: store i16 [[CONV2]], ptr [[AA_ADDR]], align 2
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: ret void
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], i32 noundef [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[BN:%.*]], ptr noundef nonnull align 4 dereferenceable(400) [[C:%.*]], i32 noundef [[VLA1:%.*]], i32 noundef [[VLA3:%.*]], ptr noundef nonnull align 4 dereferenceable(8) [[CN:%.*]], ptr noundef nonnull align 4 dereferenceable(12) [[D:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: entry:
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-NEXT: [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[BN_ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: [[VLA_ADDR2:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[VLA_ADDR4:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[CN_ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: [[D_ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[A_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-NEXT: store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK11-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK11-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4
// CHECK11-NEXT: store ptr [[BN]], ptr [[BN_ADDR]], align 4
// CHECK11-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK11-NEXT: store i32 [[VLA1]], ptr [[VLA_ADDR2]], align 4
// CHECK11-NEXT: store i32 [[VLA3]], ptr [[VLA_ADDR4]], align 4
// CHECK11-NEXT: store ptr [[CN]], ptr [[CN_ADDR]], align 4
// CHECK11-NEXT: store ptr [[D]], ptr [[D_ADDR]], align 4
// CHECK11-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = load i32, ptr [[VLA_ADDR]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = load ptr, ptr [[BN_ADDR]], align 4
// CHECK11-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK11-NEXT: [[TMP4:%.*]] = load i32, ptr [[VLA_ADDR2]], align 4
// CHECK11-NEXT: [[TMP5:%.*]] = load i32, ptr [[VLA_ADDR4]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = load ptr, ptr [[CN_ADDR]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = load ptr, ptr [[D_ADDR]], align 4
// CHECK11-NEXT: [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK11-NEXT: store i32 [[TMP8]], ptr [[A_CASTED]], align 4
// CHECK11-NEXT: [[TMP9:%.*]] = load i32, ptr [[A_CASTED]], align 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK11-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 9, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144.omp_outlined, i32 [[TMP9]], ptr [[TMP0]], i32 [[TMP1]], ptr [[TMP2]], ptr [[TMP3]], i32 [[TMP4]], i32 [[TMP5]], ptr [[TMP6]], ptr [[TMP7]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z3fooi_l144.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]], i32 noundef [[VLA:%.*]], ptr noundef nonnull align 4 dereferenceable(4) [[BN:%.*]], ptr noundef nonnull align 4 dereferenceable(400) [[C:%.*]], i32 noundef [[VLA1:%.*]], i32 noundef [[VLA3:%.*]], ptr noundef nonnull align 4 dereferenceable(8) [[CN:%.*]], ptr noundef nonnull align 4 dereferenceable(12) [[D:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[BN_ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[VLA_ADDR2:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[VLA_ADDR4:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[CN_ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: [[D_ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK11-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK11-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK11-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK11-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4
// CHECK11-NEXT: store ptr [[BN]], ptr [[BN_ADDR]], align 4
// CHECK11-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK11-NEXT: store i32 [[VLA1]], ptr [[VLA_ADDR2]], align 4
// CHECK11-NEXT: store i32 [[VLA3]], ptr [[VLA_ADDR4]], align 4
// CHECK11-NEXT: store ptr [[CN]], ptr [[CN_ADDR]], align 4
// CHECK11-NEXT: store ptr [[D]], ptr [[D_ADDR]], align 4
// CHECK11-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = load i32, ptr [[VLA_ADDR]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = load ptr, ptr [[BN_ADDR]], align 4
// CHECK11-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK11-NEXT: [[TMP4:%.*]] = load i32, ptr [[VLA_ADDR2]], align 4
// CHECK11-NEXT: [[TMP5:%.*]] = load i32, ptr [[VLA_ADDR4]], align 4
// CHECK11-NEXT: [[TMP6:%.*]] = load ptr, ptr [[CN_ADDR]], align 4
// CHECK11-NEXT: [[TMP7:%.*]] = load ptr, ptr [[D_ADDR]], align 4
// CHECK11-NEXT: [[TMP8:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP8]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK11-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x float], ptr [[TMP0]], i32 0, i32 2
// CHECK11-NEXT: [[TMP9:%.*]] = load float, ptr [[ARRAYIDX]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[CONV:%.*]] = fpext float [[TMP9]] to double
// CHECK11-NEXT: [[ADD5:%.*]] = fadd double [[CONV]], 1.000000e+00
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: [[CONV6:%.*]] = fptrunc double [[ADD5]] to float
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store float [[CONV6]], ptr [[ARRAYIDX]], align 4
// CHECK11-NEXT: [[ARRAYIDX7:%.*]] = getelementptr inbounds float, ptr [[TMP2]], i32 3
// CHECK11-NEXT: [[TMP10:%.*]] = load float, ptr [[ARRAYIDX7]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[CONV8:%.*]] = fpext float [[TMP10]] to double
// CHECK11-NEXT: [[ADD9:%.*]] = fadd double [[CONV8]], 1.000000e+00
// CHECK11-NEXT: [[CONV10:%.*]] = fptrunc double [[ADD9]] to float
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store float [[CONV10]], ptr [[ARRAYIDX7]], align 4
// CHECK11-NEXT: [[ARRAYIDX11:%.*]] = getelementptr inbounds [5 x [10 x double]], ptr [[TMP3]], i32 0, i32 1
// CHECK11-NEXT: [[ARRAYIDX12:%.*]] = getelementptr inbounds [10 x double], ptr [[ARRAYIDX11]], i32 0, i32 2
// CHECK11-NEXT: [[TMP11:%.*]] = load double, ptr [[ARRAYIDX12]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[ADD13:%.*]] = fadd double [[TMP11]], 1.000000e+00
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store double [[ADD13]], ptr [[ARRAYIDX12]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[TMP12:%.*]] = mul nsw i32 1, [[TMP5]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[ARRAYIDX14:%.*]] = getelementptr inbounds double, ptr [[TMP6]], i32 [[TMP12]]
// CHECK11-NEXT: [[ARRAYIDX15:%.*]] = getelementptr inbounds double, ptr [[ARRAYIDX14]], i32 3
// CHECK11-NEXT: [[TMP13:%.*]] = load double, ptr [[ARRAYIDX15]], align 8
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[ADD16:%.*]] = fadd double [[TMP13]], 1.000000e+00
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store double [[ADD16]], ptr [[ARRAYIDX15]], align 8
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK11-NEXT: [[X:%.*]] = getelementptr inbounds nuw [[STRUCT_TT:%.*]], ptr [[TMP7]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[TMP14:%.*]] = load i64, ptr [[X]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[ADD17:%.*]] = add nsw i64 [[TMP14]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store i64 [[ADD17]], ptr [[X]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK11-NEXT: [[Y:%.*]] = getelementptr inbounds nuw [[STRUCT_TT]], ptr [[TMP7]], i32 0, i32 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[TMP15:%.*]] = load i8, ptr [[Y]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[CONV18:%.*]] = sext i8 [[TMP15]] to i32
// CHECK11-NEXT: [[ADD19:%.*]] = add nsw i32 [[CONV18]], 1
// CHECK11-NEXT: [[CONV20:%.*]] = trunc i32 [[ADD19]] to i8
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store i8 [[CONV20]], ptr [[Y]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: ret void
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], i32 noundef [[A:%.*]], i32 noundef [[AA:%.*]], i32 noundef [[AAA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: entry:
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-NEXT: [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[AAA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[A_CASTED:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[AA_CASTED:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[AAA_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-NEXT: store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK11-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK11-NEXT: store i32 [[AAA]], ptr [[AAA_ADDR]], align 4
// CHECK11-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK11-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK11-NEXT: store i32 [[TMP1]], ptr [[A_CASTED]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = load i32, ptr [[A_CASTED]], align 4
// CHECK11-NEXT: [[TMP3:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK11-NEXT: store i16 [[TMP3]], ptr [[AA_CASTED]], align 2
// CHECK11-NEXT: [[TMP4:%.*]] = load i32, ptr [[AA_CASTED]], align 4
// CHECK11-NEXT: [[TMP5:%.*]] = load i8, ptr [[AAA_ADDR]], align 1
// CHECK11-NEXT: store i8 [[TMP5]], ptr [[AAA_CASTED]], align 1
// CHECK11-NEXT: [[TMP6:%.*]] = load i32, ptr [[AAA_CASTED]], align 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK11-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 4, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198.omp_outlined, i32 [[TMP2]], i32 [[TMP4]], i32 [[TMP6]], ptr [[TMP0]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZL7fstatici_l198.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]], i32 noundef [[AA:%.*]], i32 noundef [[AAA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[AAA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK11-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK11-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK11-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK11-NEXT: store i32 [[AAA]], ptr [[AAA_ADDR]], align 4
// CHECK11-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK11-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK11-NEXT: [[CONV:%.*]] = sext i16 [[TMP2]] to i32
// CHECK11-NEXT: [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
// CHECK11-NEXT: [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
// CHECK11-NEXT: store i16 [[CONV2]], ptr [[AA_ADDR]], align 2
// CHECK11-NEXT: [[TMP3:%.*]] = load i8, ptr [[AAA_ADDR]], align 1
// CHECK11-NEXT: [[CONV3:%.*]] = sext i8 [[TMP3]] to i32
// CHECK11-NEXT: [[ADD4:%.*]] = add nsw i32 [[CONV3]], 1
// CHECK11-NEXT: [[CONV5:%.*]] = trunc i32 [[ADD4]] to i8
// CHECK11-NEXT: store i8 [[CONV5]], ptr [[AAA_ADDR]], align 1
// CHECK11-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 2
// CHECK11-NEXT: [[TMP4:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
// CHECK11-NEXT: [[ADD6:%.*]] = add nsw i32 [[TMP4]], 1
// CHECK11-NEXT: store i32 [[ADD6]], ptr [[ARRAYIDX]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: ret void
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], ptr noundef [[THIS:%.*]], i32 noundef [[B:%.*]], i32 noundef [[VLA:%.*]], i32 noundef [[VLA1:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[C:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: entry:
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-NEXT: [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: [[B_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[VLA_ADDR2:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[B_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-NEXT: store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// CHECK11-NEXT: store i32 [[B]], ptr [[B_ADDR]], align 4
// CHECK11-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4
// CHECK11-NEXT: store i32 [[VLA1]], ptr [[VLA_ADDR2]], align 4
// CHECK11-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK11-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = load i32, ptr [[VLA_ADDR]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = load i32, ptr [[VLA_ADDR2]], align 4
// CHECK11-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK11-NEXT: [[TMP4:%.*]] = load i32, ptr [[B_ADDR]], align 4
// CHECK11-NEXT: store i32 [[TMP4]], ptr [[B_CASTED]], align 4
// CHECK11-NEXT: [[TMP5:%.*]] = load i32, ptr [[B_CASTED]], align 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK11-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 5, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.omp_outlined, ptr [[TMP0]], i32 [[TMP5]], i32 [[TMP1]], i32 [[TMP2]], ptr [[TMP3]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__ZN2S12r1Ei_l216.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], ptr noundef [[THIS:%.*]], i32 noundef [[B:%.*]], i32 noundef [[VLA:%.*]], i32 noundef [[VLA1:%.*]], ptr noundef nonnull align 2 dereferenceable(2) [[C:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: [[THIS_ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[B_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[VLA_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[VLA_ADDR2:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[C_ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK11-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK11-NEXT: store ptr [[THIS]], ptr [[THIS_ADDR]], align 4
// CHECK11-NEXT: store i32 [[B]], ptr [[B_ADDR]], align 4
// CHECK11-NEXT: store i32 [[VLA]], ptr [[VLA_ADDR]], align 4
// CHECK11-NEXT: store i32 [[VLA1]], ptr [[VLA_ADDR2]], align 4
// CHECK11-NEXT: store ptr [[C]], ptr [[C_ADDR]], align 4
// CHECK11-NEXT: [[TMP0:%.*]] = load ptr, ptr [[THIS_ADDR]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = load i32, ptr [[VLA_ADDR]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = load i32, ptr [[VLA_ADDR2]], align 4
// CHECK11-NEXT: [[TMP3:%.*]] = load ptr, ptr [[C_ADDR]], align 4
// CHECK11-NEXT: [[TMP4:%.*]] = load i32, ptr [[B_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[CONV:%.*]] = sitofp i32 [[TMP4]] to double
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: [[ADD:%.*]] = fadd double [[CONV]], 1.500000e+00
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK11-NEXT: [[A:%.*]] = getelementptr inbounds nuw [[STRUCT_S1:%.*]], ptr [[TMP0]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store double [[ADD]], ptr [[A]], align 4
[IRBuilder] Generate nuw GEPs for struct member accesses (#99538) Generate nuw GEPs for struct member accesses, as inbounds + non-negative implies nuw. Regression tests are updated using update scripts where possible, and by find + replace where not.
2024-08-09 13:25:04 +01:00
// CHECK11-NEXT: [[A3:%.*]] = getelementptr inbounds nuw [[STRUCT_S1]], ptr [[TMP0]], i32 0, i32 0
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[TMP5:%.*]] = load double, ptr [[A3]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[INC:%.*]] = fadd double [[TMP5]], 1.000000e+00
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store double [[INC]], ptr [[A3]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[CONV4:%.*]] = fptosi double [[INC]] to i16
// CHECK11-NEXT: [[TMP6:%.*]] = mul nsw i32 1, [[TMP2]]
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds i16, ptr [[TMP3]], i32 [[TMP6]]
// CHECK11-NEXT: [[ARRAYIDX5:%.*]] = getelementptr inbounds i16, ptr [[ARRAYIDX]], i32 1
// CHECK11-NEXT: store i16 [[CONV4]], ptr [[ARRAYIDX5]], align 2
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: ret void
//
//
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-SAME: (ptr noalias noundef [[DYN_PTR:%.*]], i32 noundef [[A:%.*]], i32 noundef [[AA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: entry:
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-NEXT: [[DYN_PTR_ADDR:%.*]] = alloca ptr, align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[A_CASTED:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[AA_CASTED:%.*]] = alloca i32, align 4
[OpenMP] Introduce the KernelLaunchEnvironment as implicit argument (#70401) The KernelEnvironment is for compile time information about a kernel. It allows the compiler to feed information to the runtime. The KernelLaunchEnvironment is for dynamic information *per* kernel launch. It allows the rutime to feed information to the kernel that is not shared with other invocations of the kernel. The first use case is to replace the globals that synchronize teams reductions with per-launch versions. This allows concurrent teams reductions. More uses cases will follow, e.g., per launch memory pools. Fixes: https://github.com/llvm/llvm-project/issues/70249
2023-10-31 19:38:43 -07:00
// CHECK11-NEXT: store ptr [[DYN_PTR]], ptr [[DYN_PTR_ADDR]], align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK11-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK11-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK11-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
// CHECK11-NEXT: store i32 [[TMP1]], ptr [[A_CASTED]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = load i32, ptr [[A_CASTED]], align 4
// CHECK11-NEXT: [[TMP3:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK11-NEXT: store i16 [[TMP3]], ptr [[AA_CASTED]], align 2
// CHECK11-NEXT: [[TMP4:%.*]] = load i32, ptr [[AA_CASTED]], align 4
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK11-NEXT: call void (ptr, i32, ptr, ...) @__kmpc_fork_call(ptr @[[GLOB1]], i32 3, ptr @{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181.omp_outlined, i32 [[TMP2]], i32 [[TMP4]], ptr [[TMP0]])
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: ret void
//
//
[OpenMP] Prefix outlined and reduction func names with original func's name This patch prefixes omp outlined helpers and reduction funcs with the original function's name. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D140722
2022-12-23 11:35:47 +02:00
// CHECK11-LABEL: define {{[^@]+}}@{{__omp_offloading_[0-9a-z]+_[0-9a-z]+}}__Z9ftemplateIiET_i_l181.omp_outlined
[OpenMP] Unify the min/max thread/teams pathways We used to pass the min/max threads/teams values through different paths from the frontend to the middle end. This simplifies the situation by passing the values once, only when we will create the KernelEnvironment, which contains the values. At that point we also manifest the metadata, as appropriate. Some footguns have also been removed, e.g., our target check is now triple-based, not calling convention-based, as the latter is dependent on the ordering of operations. The types of the values have been unified to int32_t.
2023-10-25 16:44:15 -07:00
// CHECK11-SAME: (ptr noalias noundef [[DOTGLOBAL_TID_:%.*]], ptr noalias noundef [[DOTBOUND_TID_:%.*]], i32 noundef [[A:%.*]], i32 noundef [[AA:%.*]], ptr noundef nonnull align 4 dereferenceable(40) [[B:%.*]]) #[[ATTR0]] {
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: entry:
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[DOTGLOBAL_TID__ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: [[DOTBOUND_TID__ADDR:%.*]] = alloca ptr, align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[A_ADDR:%.*]] = alloca i32, align 4
// CHECK11-NEXT: [[AA_ADDR:%.*]] = alloca i32, align 4
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: [[B_ADDR:%.*]] = alloca ptr, align 4
// CHECK11-NEXT: store ptr [[DOTGLOBAL_TID_]], ptr [[DOTGLOBAL_TID__ADDR]], align 4
// CHECK11-NEXT: store ptr [[DOTBOUND_TID_]], ptr [[DOTBOUND_TID__ADDR]], align 4
// CHECK11-NEXT: store i32 [[A]], ptr [[A_ADDR]], align 4
// CHECK11-NEXT: store i32 [[AA]], ptr [[AA_ADDR]], align 4
// CHECK11-NEXT: store ptr [[B]], ptr [[B_ADDR]], align 4
// CHECK11-NEXT: [[TMP0:%.*]] = load ptr, ptr [[B_ADDR]], align 4
// CHECK11-NEXT: [[TMP1:%.*]] = load i32, ptr [[A_ADDR]], align 4
Revert "[OpenMP] Codegen aggregate for outlined function captures" This reverts commit 7539e9cf811e590d9f12ae39673ca789e26386b4.
2022-09-15 03:08:46 +00:00
// CHECK11-NEXT: [[ADD:%.*]] = add nsw i32 [[TMP1]], 1
[OpenMP] Convert more tests to opaque pointers (NFC) These were converted using the script at https://gist.github.com/nikic/98357b71fd67756b0f064c9517b62a34 followed by a re-run of update_cc_test_checks.py.
2022-10-07 15:28:44 +02:00
// CHECK11-NEXT: store i32 [[ADD]], ptr [[A_ADDR]], align 4
// CHECK11-NEXT: [[TMP2:%.*]] = load i16, ptr [[AA_ADDR]], align 2
// CHECK11-NEXT: [[CONV:%.*]] = sext i16 [[TMP2]] to i32
// CHECK11-NEXT: [[ADD1:%.*]] = add nsw i32 [[CONV]], 1
// CHECK11-NEXT: [[CONV2:%.*]] = trunc i32 [[ADD1]] to i16
// CHECK11-NEXT: store i16 [[CONV2]], ptr [[AA_ADDR]], align 2
// CHECK11-NEXT: [[ARRAYIDX:%.*]] = getelementptr inbounds [10 x i32], ptr [[TMP0]], i32 0, i32 2
// CHECK11-NEXT: [[TMP3:%.*]] = load i32, ptr [[ARRAYIDX]], align 4
// CHECK11-NEXT: [[ADD3:%.*]] = add nsw i32 [[TMP3]], 1
// CHECK11-NEXT: store i32 [[ADD3]], ptr [[ARRAYIDX]], align 4
[OpenMP][NFC] Refactor Clang OpenMP tests using update_cc_test_checks This patch refactors a subset of Clang OpenMP tests, generating checklines using the update_cc_test_checks script. This refactoring facilitates updating the Clang OpenMP code generation codebase by automating test generation. Reviewed By: jdoerfert Differential Revision: https://reviews.llvm.org/D101849
2021-05-05 15:13:14 -07:00
// CHECK11-NEXT: ret void
//
Reference in New Issue Copy Permalink