Skip to content

Add gfx1250 support to CK tile group GEMM#626

Open
aris134 wants to merge 3 commits into
devfrom
amartin/ck-group-gemm-gfx1250-clean-v2
Open

Add gfx1250 support to CK tile group GEMM#626
aris134 wants to merge 3 commits into
devfrom
amartin/ck-group-gemm-gfx1250-clean-v2

Conversation

@aris134

@aris134 aris134 commented Jun 16, 2026

Copy link
Copy Markdown
Contributor

Description

Forwarding of #576 to dev with minor updates.

Fixes # (16490)

Type of change

  • Documentation change (change only to the documentation, either a fix or a new content)
  • Bug fix (non-breaking change which fixes an issue)
  • New feature (non-breaking change which adds functionality)
  • Breaking change (fix or feature that would cause existing functionality to not work as expected)
  • Infra/Build change
  • Code refactoring

Changes

Please list the changes introduced in this PR:

  • Add gfx1250 WMMA config for fp8/fp16
  • Add arch level dispatch to CK tile fp16

Checklist:

  • I have read and followed the contributing guidelines
  • The functionality is complete
  • I have commented my code, particularly in hard-to-understand areas
  • I have made corresponding changes to the documentation
  • My changes generate no new warnings
  • I have added tests that prove my fix is effective or that my feature works
  • New and existing unit tests pass locally with my changes

@aris134 aris134 requested a review from matthiasdiener June 16, 2026 14:39
@aris134 aris134 self-assigned this Jun 16, 2026
@aris134 aris134 added the ci-level 1 CI test level 1 label Jun 16, 2026
Comment on lines +481 to +484
#if defined(__gfx942__)
case GPUArch::GFX942:
return ck_tile_grouped_gemm_fp8_dispatch_arch<GPUArch::GFX942>(a_dtype, b_dtype, d_dtype, ctx);
#endif

Copy link
Copy Markdown
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please re-check that these additional ifdefs are required.

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

You're right, they are not. Fixed in 2cb357b

bool need_k_pad,
const GroupedGemmRunContext& ctx) {
switch (detect_gpu_arch()) {
#if defined(__gfx942__)

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

It is device compile define, it should not be used in host code

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed in 2cb357b


// Templated dispatch on A/B layouts, shared by all layout-specific .cpp files.
template <typename ALayout, typename BLayout>
template <GPUArch Arch, typename ALayout, typename BLayout>

Copy link
Copy Markdown
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

For host code runtime arch detection should be used rather than building multiple template instances

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed in 2cb357b

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

ci-level 1 CI test level 1

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants