[rocm-libraries] ROCm/rocm-libraries#4302 (commit e62bd8a)

[CK_TILE] add tf32 support
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

## Proposed changes

TF32 is added in CK on gfx942 and gfx950. This PR is to initiate tf32 in
CK_TILE on gfx942 and gfx950.

## Checklist

Please put an into the boxes that apply. You can also fill these out
after creating the PR. If you're not sure, please don't hesitate to ask.

- [ ] I have added tests relevant to the introduced functionality, and
the unit tests are passing locally
- [ ] I have added the test to REGRESSION_TESTS list defined at the top
of CMakeLists.txt in tests/CMakeLists.txt, **IF** the test takes more
than 30 seconds to run.
- [ ] I have added inline documentation which enables the maintainers
with understanding the motivation
- [ ] I have removed the stale documentation which is no longer relevant
after this pull request
- [ ] (If this change is user-facing) I have added release notes which
provide the end users with a brief summary of the improvement from this
pull request
- [x] I have run  on all changed files
- [ ] Any dependent changes have been merged

## Discussion
This commit is contained in:
yinglu
2026-03-19 09:19:06 +00:00
committed by assistant-librarian[bot]
parent 652d3456ca
commit d460ab35b6
30 changed files with 1164 additions and 260 deletions

View File

@@ -48,6 +48,28 @@ using WarpGemmMfmaF32F32F32M16N16K16TransposedCDistribution =
4,
AttrNumAccess>>;
// tf32
// On gfx950: uses 3x bf16 MFMA emulation (no native xf32 support)
#if defined(CK_GFX950_SUPPORT)
// gfx950: tf32 emulated using 3x bf16 MFMA
using WarpGemmMfmaTf32Tf32F32M32N32K16Native = WarpGemmImpl<WarpGemmAttributeMfma<
WarpGemmAttributeMfmaImplF32F32F32M32N32K16Tf32Gfx950<WGAttrCtlEnum::Default_>>>;
using WarpGemmMfmaTf32Tf32F32M16N16K32Native = WarpGemmImpl<WarpGemmAttributeMfma<
WarpGemmAttributeMfmaImplF32F32F32M16N16K32Tf32Gfx950<WGAttrCtlEnum::Default_>>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfmaTf32Tf32F32M32N32K16 = WarpGemmImpl<WarpGemmAttributeMfma<
WarpGemmAttributeMfmaImplF32F32F32M32N32K16Tf32Gfx950<WGAttrCtlEnum::Default_>,
AttrNumAccess>>;
template <WGAttrNumAccessEnum AttrNumAccess = WGAttrNumAccessEnum::Single>
using WarpGemmMfmaTf32Tf32F32M16N16K32 = WarpGemmImpl<WarpGemmAttributeMfma<
WarpGemmAttributeMfmaImplF32F32F32M16N16K32Tf32Gfx950<WGAttrCtlEnum::Default_>,
AttrNumAccess>>;
#endif
// fp16
using WarpGemmMfmaF16F16F32M32N32K8 = WarpGemmImpl<