Add rocWMMA to TheRock by bsyrowik · Pull Request #1938 · ROCm/TheRock

bsyrowik · 2025-10-25T17:33:47Z

Motivation

This PR adds the rocWMMA library to TheRock with testing.

Technical Details

We pull from rocm-libraries, and enable support and testing for Windows build.

Test Plan

Pull-Request CI run.

Test Result

Should check CI run when it completes.

Submission Checklist

Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

## Motivation These changes are required to integrate rocWMMA into TheRock. This PR subsumes the patches from my PR for TheRock here: ROCm/TheRock#1938 ## Technical Details Mostly changes to testing and CMake configuration. Notably: - Update googletest dependency. - Create new regression and smoke test ctest files. - Suggest using Ninja. - Export the project. ## Test Plan TheRock CI: https://github.com/ROCm/TheRock/actions/runs/18804601794 An equivalent PR from ROCm/rocWMMA passed all checks as well: https://github.com/ROCm/rocWMMA/pull/601/checks ## Test Result Tests passed ## Submission Checklist - [x] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests. --------- Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

[rocWMMA] CMake changes for integration into TheRock ## Motivation These changes are required to integrate rocWMMA into TheRock. This PR subsumes the patches from my PR for TheRock here: ROCm/TheRock#1938 ## Technical Details Mostly changes to testing and CMake configuration. Notably: - Update googletest dependency. - Create new regression and smoke test ctest files. - Suggest using Ninja. - Export the project. ## Test Plan TheRock CI: https://github.com/ROCm/TheRock/actions/runs/18804601794 An equivalent PR from ROCm/rocWMMA passed all checks as well: https://github.com/ROCm/rocWMMA/pull/601/checks ## Test Result Tests passed ## Submission Checklist - [x] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests.

marbre

Please excuse the delayed review. Some comments.

Co-authored-by: Marius Brehler <marius.brehler@amd.com>

Copilot

Pull Request Overview

This PR adds the rocWMMA library to TheRock, integrating it as a new math library component with comprehensive testing support on both Linux and Windows platforms.

Integrates rocWMMA from rocm-libraries repository with CMake configuration
Adds test infrastructure and CI configuration for rocWMMA validation
Excludes rocWMMA from unsupported GPU targets (older gfx906, gfx101X, gfx103X families)

Reviewed Changes

Copilot reviewed 13 out of 13 changed files in this pull request and generated 4 comments.

Show a summary per file

File	Description
`math-libs/artifact-rocwmma.toml`	Defines artifact packaging for rocWMMA library including headers, tests, and samples
`math-libs/CMakeLists.txt`	Adds rocWMMA subproject configuration with optional rocBLAS and benchmark dependencies
`examples/cpp-sdk-user/CMakeLists.txt`	Adds find_package configuration for rocWMMA in SDK user examples
`examples/clean_configure_test_project.cmake`	Propagates THEROCK_ENABLE_ROCWMMA flag to test project configuration
`examples/CMakeLists.txt`	Passes THEROCK_ENABLE_ROCWMMA flag to test configuration
`docs/development/windows_support.md`	Documents rocWMMA Windows support status
`docs/development/installing_artifacts.md`	Adds --rocwmma flag documentation for artifact installation
`cmake/therock_amdgpu_targets.cmake`	Excludes rocWMMA from unsupported GPU targets (gfx906, gfx101X, gfx103X, gfx1103)
`build_tools/install_rocm_from_artifacts.py`	Implements --rocwmma flag for artifact retrieval script
`build_tools/github_actions/test_executable_scripts/test_rocwmma.py`	Adds test execution script with sharding support for rocWMMA tests
`build_tools/github_actions/fetch_test_configurations.py`	Configures CI test job for rocWMMA with 4 shards on Linux and Windows
`README.md`	Documents THEROCK_ENABLE_ROCWMMA configuration flag
`CMakeLists.txt`	Defines rocWMMA feature with optional rocBLAS dependency

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

stellaraccident

Generally looks good. I will review in detail after meetings today.

marbre

LGTM, assuming that the newly added rocWMMA tests all pass (as they already did on Windows). Let's wait for the CI signal prior to merging.

ScottTodd · 2025-11-13T17:48:30Z

This caused build failures on the gfx101X and gfx103X GPU families on both Linux and Windows: #2127

2025-11-13T07:55:19.5029690Z [rocWMMA] /therock/src/rocm-libraries/projects/rocwmma/library/include/rocwmma/internal/config.hpp:79:15: error: static assertion failed: Unsupported architecture
2025-11-13T07:55:19.5030400Z [rocWMMA] 79 | static_assert(0, "Unsupported architecture");
2025-11-13T07:55:19.5030686Z [rocWMMA] | ^
2025-11-13T07:55:19.5030944Z [rocWMMA] 1 error generated when compiling for gfx906.
2025-11-13T07:55:19.5031232Z [rocWMMA] Elapsed time (seconds): 5.43847
2025-11-13T07:55:19.5031624Z [rocWMMA] [4/1054] Building CXX object samples/CMakeFiles/hipRTC_gemm.dir/hipRTC_gemm.cpp.o
2025-11-13T07:55:19.5032104Z [rocWMMA] FAILED: samples/CMakeFiles/hipRTC_gemm.dir/hipRTC_gemm.cpp.o

Looking at the code in rocWMMA at https://github.com/ROCm/rocm-libraries/blob/07b9a2d3c3c7b3642bd41ec07a4096e877c7b729/projects/rocwmma/library/include/rocwmma/internal/config.hpp#L29-L80, unsupported architectures are compilation errors. At a minimum for now I think we'll need to exclude rocWMMA from certain target families in https://github.com/ROCm/TheRock/blob/main/cmake/therock_amdgpu_targets.cmake, but ideally the project would at least build in some form (fallback behavior, etc.) for all architectures.

cc @stellaraccident @sa-faizal

bsyrowik · 2025-11-13T19:16:11Z

ideally the project would at least build in some form (fallback behavior, etc.) for all architectures.

I am hesitant to do this because rocWMMA facilitates utilization of hardware constructs that are not available on the gfx10** devices, so it doesn't make much sense to provide support for them.

rocWMMA is excluded for all gfx10** families in https://github.com/ROCm/TheRock/blob/main/cmake/therock_amdgpu_targets.cmake, so I'm not sure why we're still trying to build against them?

bsyrowik · 2025-11-13T19:18:39Z

I opened #1944 to cover this. The only device we don't support but ideally should support is gfx1103.

…ck after ROCm/TheRock#1938

* DGGML_HIP_FORCE_ROCWMMA_FATTN_GFX12 is not needed anymore * More updates for README and manual_instructions * Add gfx1150 to the build-llamacpp-rocm worflow * Enabling rocWMMA_FATTN builds since rocWMMA headers are part of TheRock after ROCm/TheRock#1938 * rocWMMA patch is not needed anymore

## Motivation This PR adds the rocWMMA library to TheRock with testing. ## Technical Details We pull from rocm-libraries, and enable support and testing for Windows build. ## Test Plan Pull-Request CI run. ## Test Result Should check CI run when it completes. ## Submission Checklist - [x] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests. --------- Co-authored-by: Marius Brehler <marius.brehler@amd.com>

## Motivation These changes are required to integrate rocWMMA into TheRock. This PR subsumes the patches from my PR for TheRock here: ROCm/TheRock#1938 ## Technical Details Mostly changes to testing and CMake configuration. Notably: - Update googletest dependency. - Create new regression and smoke test ctest files. - Suggest using Ninja. - Export the project. ## Test Plan TheRock CI: https://github.com/ROCm/TheRock/actions/runs/18804601794 An equivalent PR from ROCm/rocWMMA passed all checks as well: https://github.com/ROCm/rocWMMA/pull/601/checks ## Test Result Tests passed ## Submission Checklist - [x] Look over the contributing guidelines at https://github.com/ROCm/ROCm/blob/develop/CONTRIBUTING.md#pull-requests. --------- Co-authored-by: Copilot <175728472+Copilot@users.noreply.github.com>

github-project-automation Bot added this to TheRock Triage Oct 25, 2025

github-project-automation Bot moved this to TODO in TheRock Triage Oct 25, 2025

bsyrowik commented Oct 25, 2025

View reviewed changes

Comment thread build_tools/github_actions/test_executable_scripts/test_rocwmma.py Outdated

bsyrowik marked this pull request as ready for review October 27, 2025 23:31

bsyrowik requested review from geomin12 and marbre October 28, 2025 00:24

amd-aakash reviewed Oct 28, 2025

View reviewed changes

Comment thread patches/amd-mainline/rocWMMA/0001-Check-for-Ninja-and-use-newer-gtest.patch Outdated

bsyrowik mentioned this pull request Oct 29, 2025

[rocWMMA] CMake changes for integration into TheRock ROCm/rocm-libraries#2322

Merged

1 task

geomin12 reviewed Oct 29, 2025

View reviewed changes

bsyrowik force-pushed the users/bsyrowik/add_rocwmma branch from 44e4193 to 7bd48ac Compare October 29, 2025 19:53

bsyrowik commented Oct 30, 2025

View reviewed changes

Comment thread build_tools/github_actions/test_executable_scripts/test_rocwmma.py Outdated

catan2001 mentioned this pull request Nov 5, 2025

[Feature]: Add llama.cpp integration to TheRock #1652

Open

5 tasks

bsyrowik force-pushed the users/bsyrowik/add_rocwmma branch from bfb3c5f to c2a5287 Compare November 6, 2025 18:40

bsyrowik force-pushed the users/bsyrowik/add_rocwmma branch from c2a5287 to bd804f7 Compare November 7, 2025 21:49

bsyrowik requested review from amd-aakash and geomin12 November 7, 2025 21:51

marbre reviewed Nov 10, 2025

View reviewed changes

bsyrowik and others added 10 commits November 10, 2025 13:01

Add rocWMMA to cmake files

5fe4ce2

Add artifact-rocwmma.toml and update artifact handling

09a36d8

Add rocWMMA testing

875dd94

Add rocWMMA patches

498b2c3

Update rocWMMA GPU target compatibility

39baa13

Remove patches now that rocm-libraries is updated

98004c9

Update windows_support.md

c1cb86f

Apply suggestions from code review

f217dcc

Co-authored-by: Marius Brehler <marius.brehler@amd.com>

Lower test timeout limits to be more reasonable

ba6606d

Update rocWMMA config

867413d

bsyrowik force-pushed the users/bsyrowik/add_rocwmma branch from e7b6c39 to 867413d Compare November 10, 2025 19:09

Fix list append

d31bfd5

bsyrowik requested review from Copilot and marbre and removed request for amd-aakash, geomin12 and marbre November 10, 2025 23:23

Copilot started reviewing on behalf of bsyrowik November 11, 2025 01:01 View session

Copilot finished reviewing on behalf of bsyrowik November 11, 2025 01:03

Copilot AI reviewed Nov 11, 2025

View reviewed changes

Comment thread math-libs/CMakeLists.txt

Comment thread math-libs/CMakeLists.txt

Comment thread build_tools/github_actions/test_executable_scripts/test_rocwmma.py

Comment thread build_tools/github_actions/test_executable_scripts/test_rocwmma.py

Clean up rocWMMA options

b40c1b3

bsyrowik requested review from amd-aakash, geomin12 and marbre November 11, 2025 21:38

stellaraccident reviewed Nov 11, 2025

View reviewed changes

marbre approved these changes Nov 11, 2025

View reviewed changes

jammm mentioned this pull request Nov 12, 2025

[Feature]: Please add rocWMMA to hip-sdk ROCm/rocWMMA#593

Closed

marbre merged commit d24cccc into main Nov 12, 2025
57 checks passed

marbre deleted the users/bsyrowik/add_rocwmma branch November 12, 2025 17:24

github-project-automation Bot moved this from TODO to Done in TheRock Triage Nov 12, 2025

This was referenced Nov 12, 2025

Add rocWMMA hipTensor to theROCK #1887

Closed

[Feature]: Add support for projects: rocWMMA, AITER, rocGDB #1835

Open

ScottTodd mentioned this pull request Nov 13, 2025

[issue] rocWMMA failing on Linux, windows nightlies, -"static assertion failed: Unsupported architecture" #2127

Closed

slojosic-amd added a commit to lemonade-sdk/llamacpp-rocm that referenced this pull request Dec 5, 2025

Enabling rocWMMA_FATTN builds since rocWMMA headers are part of TheRo…

d88d535

…ck after ROCm/TheRock#1938

jammm mentioned this pull request Dec 27, 2025

SageAttention AMD version thu-ml/SageAttention#332

Open

Conversation

bsyrowik commented Oct 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Motivation

Technical Details

Test Plan

Test Result

Submission Checklist

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

marbre left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull Request Overview

Reviewed Changes

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

stellaraccident left a comment

Choose a reason for hiding this comment

Uh oh!

marbre left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

ScottTodd commented Nov 13, 2025

Uh oh!

bsyrowik commented Nov 13, 2025

Uh oh!

bsyrowik commented Nov 13, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

7 participants

bsyrowik commented Oct 25, 2025 •

edited

Loading

bsyrowik commented Nov 13, 2025 •

edited

Loading