ROCm/composable_kernel

mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-14 18:17:44 +00:00

Files

History

金黄色葡萄球君君 b5b3ba728d projects/composablekernel: add SwigluStep support for MoE blockscale (#6118 )

## Summary
- add `swiglustep_and_mul` to the composablekernel MoE blockscale
activation enum
- implement the corresponding blockscale epilogue path for `SwigluStep`
- keep existing `silu` and `gelu` paths unchanged

## Scope
This PR covers the classic composablekernel blockscale MoE path under
`projects/composablekernel`.

This is separate from the `ck_tile` / FlatMM path being discussed in
ROCm/rocm-libraries#5992.

## Motivation
`Step-3.5-Flash-FP8` uses `SwigluStep` in its MoE MLP path. The
dependent AITER change needs native support for this activation in the
classic composablekernel MoE blockscale path.

## Validation
- patch is limited to two composablekernel files under
`projects/composablekernel`
- existing `silu` / `gelu` paths are unchanged
- dependent AITER runtime validation hit the classic CK 2-stage path
with AITER MoE enabled

2026-04-21 07:24:48 +00:00

..

[MIOpen][CK] Fix bwd weight conv test failures by disabling one block-GEMM V5 instance for 3D convs (#6421 )

2026-04-17 09:16:32 +03:00

library/utility

Add a README.md file to ck/library/util (#4277 )

2026-02-10 21:26:45 +00:00

problem_transform

CK: Remove 41 commented-out dead code blocks (~200 lines) (#6302 )

2026-04-10 11:17:11 -04:00

[Compiler] Addressing new compiler warnings (#3640 )

2026-02-02 09:39:48 -08:00

tensor_description

[CK] fix clang lifetimebound errors with staging compiler (#5921 )

2026-03-30 07:19:32 -07:00

tensor_operation

projects/composablekernel: add SwigluStep support for MoE blockscale (#6118 )

2026-04-21 07:24:48 +00:00

CK: Remove 41 commented-out dead code blocks (~200 lines) (#6302 )

2026-04-10 11:17:11 -04:00

Add generate_identity_sequences helper and replace lambdas with named functors (#4828 )

2026-02-28 12:10:11 -08:00

BUILD_TIME_OPTIMIZATION.md

Remove concrete performance numbers from BUILD_TIME_OPTIMIZATION.md (#3702 )

2026-02-03 03:54:18 -07:00

ck.hpp

Add missing gfx1033 to gfx103 group definition in ck (#5141 )

2026-04-03 13:44:38 -06:00

config.h.in

ck: add CK_USE_GFX950 macro (#3636 )

2026-01-26 11:38:45 -08:00

filesystem.hpp

chore(copyright): update copyright header for include directory (#3224 )

2025-11-18 10:17:18 -08:00

README.md

Add Grouped Convolution and GEMM documentation (#1719 )

2025-02-04 16:41:49 +01:00

stream_config.hpp

Remove unnecessary hip_fp16 include from stream_config (#3549 )

2026-01-16 10:40:05 -08:00

version.h.in

[HotFix] add config and version files to pass on build info (#856 )

2023-08-23 11:36:17 -07:00

README.md

Back to the main page

Composable Kernel supported operations

Supported device operations