[SYCL][CUDA][HIP] Implement support for AMD and NVIDIA architectures as argument to fsycl-targets #7348

mmoadeli · 2022-11-10T09:30:25Z

Update if_architecture_is extension to include NVIDIA and AMD architectures.
Moves if_architecture_is code header file from intel to oneapi.
*Updates experimental/sycl_ext_intel_device_architecture.asciidoc with the contents from proposed/sycl_ext_oneapi_device_architecture.asciidoc.
Rename sycl_ext_intel_device_architecture.asciidoc to sycl_ext_oneapi_device_architecture.asciidoc.
Delete proposed/ycl_ext_oneapi_device_architecture.asciidoc.
Renames nvidia_gpu_smxx to nvidia_gpu_sm_xx.
Remove DPC++ un-supported architectures for nvidia (sm_20 to sm_37)

…IA and AMD architectures - Reflect earlier updates in DeviceIf.md to the implementation.

- Reflect above update into the related unit test.

…sciidoc with the contents from proposed/sycl_ext_oneapi_device_architecture.asciidoc

…l_ext_oneapi_device_architecture.asciidoc. - Delete proposed/ycl_ext_oneapi_device_architecture.asciidoc

gmlueck · 2022-11-10T12:03:52Z

Don't we also need changes to the compiler driver to predefine macros like __SYCL_TARGET_NVIDIA_GPU_SM20__?

dm-vodopyanov · 2022-11-10T12:17:52Z

@mmoadeli thanks for the PR! Please don't forget to update tests in intel/llvm-test-suite: https://github.com/intel/llvm-test-suite/blob/intel/SYCL/DeviceArchitecture/device_architecture.cpp

mmoadeli · 2022-11-10T12:22:21Z

@mmoadeli thanks for the PR! Please don't forget to update tests in intel/llvm-test-suite: https://github.com/intel/llvm-test-suite/blob/intel/SYCL/DeviceArchitecture/device_architecture.cpp

thanks @dm-vodopyanov , i'll update the tests.

mmoadeli · 2022-11-10T12:22:56Z

Don't we also need changes to the compiler driver to predefine macros like __SYCL_TARGET_NVIDIA_GPU_SM20__?

thanks @gmlueck , I address this.

- Fixes an isssue in getGenDeviceMacro returning StringRef to locally created string. - Updates user mantual to add supported nvidia and amd gpus.

clang/lib/Driver/Driver.cpp

sycl/doc/UsersManual.md

…x64-nvidia-gpu - Rename nvidia_gpu_smxx into nvidia_gpu_sm_xx to be more conformant to existing code. - Reflect updates into sycl-intel-gpu.cpp lit tests

- Rename and generalise isIntelGPUTarget - Add gpu lit tests

mmoadeli · 2022-11-21T10:46:26Z

@mmoadeli thanks for the PR! Please don't forget to update tests in intel/llvm-test-suite: https://github.com/intel/llvm-test-suite/blob/intel/SYCL/DeviceArchitecture/device_architecture.cpp
@dm-vodopyanov done 1402

mmoadeli · 2022-11-21T10:47:15Z

Don't we also need changes to the compiler driver to predefine macros like __SYCL_TARGET_NVIDIA_GPU_SM20__?

@gmlueck, thanks for your comment. It's addressed.

dm-vodopyanov · 2022-11-21T12:02:34Z

/verify with intel/llvm-test-suite#1402

…ure support.

gmlueck

Doc and spec changes LGTM, but you should still get approvals for the driver and header changes.

clang/lib/Driver/ToolChains/SYCL.h

clang/test/Driver/sycl-oneapi-gpu.cpp

sycl/doc/extensions/experimental/sycl_ext_oneapi_device_architecture.asciidoc

mmoadeli · 2022-11-24T15:53:42Z

/verify with intel/llvm-test-suite#1402

mmoadeli · 2022-11-24T21:02:31Z

@v-klochkov @mdtoguchi @dm-vodopyanov
would anyone please review this PR? One approval needed to have it merged.

clang/test/Driver/sycl-oneapi-gpu.cpp

bader

sycl/doc/UsersManual.md changes look good to me.

clang/test/Driver/sycl-oneapi-gpu.cpp

mmoadeli · 2022-12-08T12:31:09Z

/verify with intel/llvm-test-suite#1402

mmoadeli · 2022-12-09T11:13:08Z

@AlexeySachkov any chance to have this PR merged please?

AlexeySachkov · 2022-12-09T11:22:37Z

@mmoadeli, from what I see, this PR still lacks an approval from @intel/llvm-reviewers-runtime and I don't have enough permissions to merge without approvals.

mmoadeli · 2022-12-09T11:37:23Z

@mmoadeli, from what I see, this PR still lacks an approval from @intel/llvm-reviewers-runtime and I don't have enough permissions to merge without approvals.
@AlexeySachkov, thanks to @dm-vodopyanov, we have approval on behalf of @intel/llvm-reviewers-runtime

mmoadeli · 2022-12-09T20:52:10Z

@v-klochkov would you mind reviewing this PR please?

pvchupin · 2022-12-09T21:06:12Z

@mmoadeli, there is a fail on opencl for modified test and some documentation issue. Can you take a look?

mmoadeli · 2022-12-09T21:21:42Z

@mmoadeli, there is a fail on opencl for modified test and some documentation issue. Can you take a look?

Thanks @pvchupin
The failure on opencl relates to the issue which is fixed in this PR. Running the llvm-test-suite yesterday with this PR did not show the failure.

Would you please clarify on the issues with documentations?

pvchupin · 2022-12-10T02:39:04Z

Would you please clarify on the issues with documentations?

Generate Doxygen CI task fails. It seems it was fixed in #7524, but
CI picks up older base

pvchupin · 2022-12-10T07:13:33Z

@mmoadeli, can you take a look into post-commit issue on Windows please?
https://github.com/intel/llvm/actions/runs/3662283607/jobs/6191278254

 Clang :: Driver/sycl-oneapi-gpu.cpp

mmoadeli · 2022-12-12T09:08:15Z

@mmoadeli, can you take a look into post-commit issue on Windows please? https://github.com/intel/llvm/actions/runs/3662283607/jobs/6191278254
 Clang :: Driver/sycl-oneapi-gpu.cpp

@pvchupin, sorry, seems some tests are assumed linux build. I address them shortly.

A shorter syntax to specify device targets has been introduced in intel/llvm#7348: * `-DSYCL_CXX_FLAGS_EXTRA='-fsycl-targets=nvidia_gpu_sm_86'`, or * `-DSYCL_CXX_FLAGS_EXTRA='-fsycl-targets=amd_gpu_gfx906'`. Now, our CMake correctly parses these flags too. Refs #4716

mmoadeli added 4 commits November 9, 2022 15:03

[SYCL][CUDA][HIP] Update if_architecture_is extension to include NVID…

be33adc

…IA and AMD architectures - Reflect earlier updates in DeviceIf.md to the implementation.

[SYCL] Moves if_architecture_is code header file from intel to oneapi.

124a5d3

- Reflect above update into the related unit test.

[SYCL][Doc] Updates experimental/sycl_ext_intel_device_architecture.a…

9ee8129

…sciidoc with the contents from proposed/sycl_ext_oneapi_device_architecture.asciidoc

[SYCL][Doc] Rename sycl_ext_intel_device_architecture.asciidoc to syc…

a5adf7c

…l_ext_oneapi_device_architecture.asciidoc. - Delete proposed/ycl_ext_oneapi_device_architecture.asciidoc

mmoadeli requested review from a team as code owners November 10, 2022 09:30

mmoadeli requested a review from v-klochkov November 10, 2022 09:30

Merge branch 'intel:sycl' into nvida-amd-imp-ext

c361c79

[SYCL][CUDA] Adds driver support for nvidia gpus.

2ea6ff7

- Fixes an isssue in getGenDeviceMacro returning StringRef to locally created string. - Updates user mantual to add supported nvidia and amd gpus.

mmoadeli requested review from a team as code owners November 14, 2022 17:08

mdtoguchi reviewed Nov 15, 2022

View reviewed changes

clang/lib/Driver/Driver.cpp Outdated Show resolved Hide resolved

clang/lib/Driver/Driver.cpp Outdated Show resolved Hide resolved

gmlueck reviewed Nov 15, 2022

View reviewed changes

sycl/doc/UsersManual.md Outdated Show resolved Hide resolved

mmoadeli added 5 commits November 15, 2022 23:39

[SYCL][CUDA] Modify nvidia_gpu_* device support to be treated as nvpt…

71775fd

…x64-nvidia-gpu - Rename nvidia_gpu_smxx into nvidia_gpu_sm_xx to be more conformant to existing code. - Reflect updates into sycl-intel-gpu.cpp lit tests

[SYCL] Add support for gpu archs as argument for -fsycl-targets

53ed132

- Rename and generalise isIntelGPUTarget - Add gpu lit tests

[SYCL] Rename sycl-intel-gpu.cpp to sycl-oneapli-gpu.cpp

cd21176

[SYCL] Minor required update in a comment to reflect modifications.

4a20b80

[SYCL] Rename TranslateAMDGPUTargetOpt to TranslateGPUTargetOpt

0cf57e0

mmoadeli requested a review from Naghasan November 21, 2022 10:08

mmoadeli mentioned this pull request Nov 21, 2022

[SYCL][CUDA][HIP] Reflect namespace and path updates to relevant test. intel/llvm-test-suite#1402

Merged

[SYCL] Updates User Manual to include AMD and NVIDIA device architect…

ff41b65

…ure support.

gmlueck approved these changes Nov 21, 2022

View reviewed changes

Naghasan requested changes Nov 21, 2022

View reviewed changes

clang/lib/Driver/ToolChains/SYCL.h Outdated Show resolved Hide resolved

clang/test/Driver/sycl-oneapi-gpu.cpp Show resolved Hide resolved

sycl/doc/extensions/experimental/sycl_ext_oneapi_device_architecture.asciidoc Outdated Show resolved Hide resolved

[SYCL] Removes unused TranslateGPUTargetOpt function argument

aeab924

mmoadeli requested a review from Naghasan November 23, 2022 10:42

[SYCL] Fix clang code style

e6d89a7

Naghasan approved these changes Nov 24, 2022

View reviewed changes

mmoadeli requested a review from mdtoguchi November 24, 2022 20:53

mdtoguchi reviewed Nov 28, 2022

View reviewed changes

clang/test/Driver/sycl-oneapi-gpu.cpp Show resolved Hide resolved

mmoadeli requested a review from mdtoguchi November 29, 2022 23:46

bader approved these changes Dec 1, 2022

View reviewed changes

mdtoguchi reviewed Dec 1, 2022

View reviewed changes

clang/test/Driver/sycl-oneapi-gpu.cpp Show resolved Hide resolved

mmoadeli requested a review from mdtoguchi December 7, 2022 22:15

mdtoguchi approved these changes Dec 7, 2022

View reviewed changes

dm-vodopyanov approved these changes Dec 9, 2022

View reviewed changes

pvchupin merged commit e5de913 into intel:sycl Dec 10, 2022

mmoadeli mentioned this pull request Dec 12, 2022

[SYCL][CUDA] Fix a test which assumed build on Linux machines. #7734

Merged

al42and mentioned this pull request Jan 14, 2023

[HIP] ICE when compiling for gfx1034 #8016

Closed

al42and mentioned this pull request Jan 26, 2023

Allow specifying arbitrary target architecture via backend flags #8112

Closed

mmoadeli deleted the nvida-amd-imp-ext branch July 7, 2023 10:43

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[SYCL][CUDA][HIP] Implement support for AMD and NVIDIA architectures as argument to fsycl-targets #7348

[SYCL][CUDA][HIP] Implement support for AMD and NVIDIA architectures as argument to fsycl-targets #7348

mmoadeli commented Nov 10, 2022 •

edited

Loading

gmlueck commented Nov 10, 2022

dm-vodopyanov commented Nov 10, 2022

mmoadeli commented Nov 10, 2022

mmoadeli commented Nov 10, 2022

mmoadeli commented Nov 21, 2022

mmoadeli commented Nov 21, 2022

dm-vodopyanov commented Nov 21, 2022

gmlueck left a comment

mmoadeli commented Nov 24, 2022

mmoadeli commented Nov 24, 2022 •

edited

Loading

bader left a comment

mmoadeli commented Dec 8, 2022

mmoadeli commented Dec 9, 2022

AlexeySachkov commented Dec 9, 2022

mmoadeli commented Dec 9, 2022

mmoadeli commented Dec 9, 2022

pvchupin commented Dec 9, 2022

mmoadeli commented Dec 9, 2022

pvchupin commented Dec 10, 2022 •

edited

Loading

pvchupin commented Dec 10, 2022

mmoadeli commented Dec 12, 2022 •

edited

Loading

[SYCL][CUDA][HIP] Implement support for AMD and NVIDIA architectures as argument to fsycl-targets #7348

[SYCL][CUDA][HIP] Implement support for AMD and NVIDIA architectures as argument to fsycl-targets #7348

Conversation

mmoadeli commented Nov 10, 2022 • edited Loading

gmlueck commented Nov 10, 2022

dm-vodopyanov commented Nov 10, 2022

mmoadeli commented Nov 10, 2022

mmoadeli commented Nov 10, 2022

mmoadeli commented Nov 21, 2022

mmoadeli commented Nov 21, 2022

dm-vodopyanov commented Nov 21, 2022

gmlueck left a comment

Choose a reason for hiding this comment

mmoadeli commented Nov 24, 2022

mmoadeli commented Nov 24, 2022 • edited Loading

bader left a comment

Choose a reason for hiding this comment

mmoadeli commented Dec 8, 2022

mmoadeli commented Dec 9, 2022

AlexeySachkov commented Dec 9, 2022

mmoadeli commented Dec 9, 2022

mmoadeli commented Dec 9, 2022

pvchupin commented Dec 9, 2022

mmoadeli commented Dec 9, 2022

pvchupin commented Dec 10, 2022 • edited Loading

pvchupin commented Dec 10, 2022

mmoadeli commented Dec 12, 2022 • edited Loading

mmoadeli commented Nov 10, 2022 •

edited

Loading

mmoadeli commented Nov 24, 2022 •

edited

Loading

pvchupin commented Dec 10, 2022 •

edited

Loading

mmoadeli commented Dec 12, 2022 •

edited

Loading