Ville Pietilä fbf53fb970 [CK_TILE] Split-K autodeduction (#3351) ...

* First version of split-K autodeduction.

* Fix circular dependency and kernel construction.

* Fix tolerance calculation for bwd weight example.

* Simplify kernel construction.

* Fix kernel launching bug for split-K autodeduce.

* Add split-K autodeduction support for the two stage example.

* Fix a corner case.

* Fix clang-format.

* Fix clang-format for inc files.

* Add missing header.

* Prevent too large split-K values.

* Fix formatting.

* Add unit tests for IsSupportedArgument in grouped bwd conv.

* clang-format.

* Fix merge conflicts.

* Address feedback from code review.

* clang-format

* Fix new tests after merge.

---------

Co-authored-by: Ville Pietilä <>

[ROCm/composable_kernel commit: fc22320d78]

2025-12-10 09:30:30 +02:00

..

add_rmsnorm2d_rdquant

chore(copyright): update copyright header for include directory (#3293)

2025-11-26 11:00:05 -07:00

batched_contraction

[CK Tile] batched contraction kernel generalizing (#3126)

2025-12-02 13:30:27 +01:00

batched_transpose

chore(copyright): update copyright header for include directory (#3293)

2025-11-26 11:00:05 -07:00

common

feat(block_scale_gemm): Support RRR-R, CRR-R and CCR-C layout for aquant quant mode (#3193)

2025-12-02 14:59:07 -08:00

elementwise

Fix and improve the gemm quant pipeline infrastructure (#3245)

2025-11-26 18:04:27 -08:00

epilogue

chore(copyright): update copyright header for include directory (#3293)

2025-11-26 11:00:05 -07:00

flatmm

[CK_TILE MOE] add NT & preshuffle permute to cktile MOE (#3377)

2025-12-10 10:03:28 +08:00

fmha

Add a workaround for a compiler issue for bwd on gfx90a and ROCm 7.1.1 (#3369)

2025-12-08 07:44:17 -08:00

fused_moe

chore(copyright): update copyright header for include directory (#3293)

2025-11-26 11:00:05 -07:00

gemm

[CK_TILE MOE] add NT & preshuffle permute to cktile MOE (#3377)

2025-12-10 10:03:28 +08:00

gemm_quant

[CK_TILE] Support more layouts for BQuant GEMM (#3349)

2025-12-08 13:05:56 -08:00

grouped_convolution

[CK_TILE] Split-K autodeduction (#3351)

2025-12-10 09:30:30 +02:00

image_to_column

chore(copyright): update copyright header for include directory (#3293)

2025-11-26 11:00:05 -07:00

layernorm2d

chore(copyright): update copyright header for include directory (#3293)

2025-11-26 11:00:05 -07:00

norm_reduce

chore(copyright): update copyright header for include directory (#3293)

2025-11-26 11:00:05 -07:00

permute

chore(copyright): update copyright header for include directory (#3293)

2025-11-26 11:00:05 -07:00

pooling

[CK Tile] enable building examples by default (#3259)

2025-11-26 16:24:44 -08:00

reduce

chore(copyright): update copyright header for include directory (#3293)

2025-11-26 11:00:05 -07:00

rmsnorm2d

chore(copyright): update copyright header for include directory (#3293)

2025-11-26 11:00:05 -07:00

smoothquant

chore(copyright): update copyright header for include directory (#3293)

2025-11-26 11:00:05 -07:00

softmax/block

chore(copyright): update copyright header for include directory (#3293)

2025-11-26 11:00:05 -07:00

topk/block

chore(copyright): update copyright header for include directory (#3293)

2025-11-26 11:00:05 -07:00

topk_softmax

chore(copyright): update copyright header for include directory (#3293)

2025-11-26 11:00:05 -07:00

add_rmsnorm2d_rdquant.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00

batched_contraction.hpp

Update pre-commit to fixed versions, run remod for ck_tile (#2895)

2025-10-16 15:29:17 -07:00

batched_transpose.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00

common.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00

elementwise.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00

epilogue.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00

flatmm.hpp

[CK_TILE] Add mxfp4 flatmm (#3080)

2025-10-31 11:29:05 +08:00

fmha.hpp

Support fp8 dynamic quantization for fmha (#3206)

2025-11-24 16:28:25 +08:00

fused_moe.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00

gemm_quant.hpp

formatting (#3182)

2025-11-11 07:42:26 -08:00

gemm.hpp

Replace CK_TILE_PIPELINE macros with a common enum

2025-11-03 09:35:05 -07:00

grouped_convolution.hpp

[CK_TILE] Split-K autodeduction (#3351)

2025-12-10 09:30:30 +02:00

image_to_column.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00

layernorm2d.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00

moe_flatmm.hpp

chore(copyright): update copyright header for include directory (#3293)

2025-11-26 11:00:05 -07:00

norm_reduce.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00

permute.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00

pooling.hpp

Update pre-commit to fixed versions, run remod for ck_tile (#2895)

2025-10-16 15:29:17 -07:00

reduce.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00

rmsnorm2d.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00

smoothquant.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00

softmax.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00

topk_softmax.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00

topk.hpp

Update include path to break the remod's cyclic dep issue (#2978)

2025-10-13 13:24:47 +02:00