ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-06-30 11:47:48 +00:00
Code Issues Packages Projects Releases Wiki Activity
2,002 Commits 776 Branches 40 Tags
116a5032745fd2671af283c8a1a7c1493facb532
Commit Graph

255 Commits

Author SHA1 Message Date
Jakub Piasecki
116a503274 resolve conflicts 2025-06-16 15:33:24 +00:00
Bartlomiej Kocot
1996c99bee rebase fixes 2025-06-16 15:03:09 +00:00
Jakub Piasecki
60bd2a4fdf resolved conflicts 2025-06-16 12:22:53 +00:00
Bartlomiej Kocot
6e2b32a58a Merge branch 'develop' of github.com:ROCm/composable_kernel into barkocot/ck-tile-grouped-conv-fwd 2025-06-16 11:49:21 +00:00
Bartlomiej Kocot
a44fe5f477 refactor 2025-06-16 11:48:36 +00:00
Thomas Ning
d996bc78be fix the flatmm (#2349) 2025-06-16 02:17:53 -07:00
ruanjm
b34c234f51 Add support for specifying valid flag when fetching elements for tile_scatter_gather (#2332) 
* Add support for specifying valid flag when fetching elements for tile_scatter_gather

Add constexpr for operator[] of TrueGenerator

* Use different path when valid is enabled
 2025-06-16 17:17:03 +08:00
carlushuang
fb97f75099 hot fix block_gemm fail with pipeline_problem by adding NumWaveGroups inside block gemm problem (#2348) 2025-06-15 22:49:04 -07:00
Mateusz Ozga
bd96ac9742 [CK_TILE] Multiple-D GEMM example (#2219) 
* Multiple d, initial commit

* Check Ds Layout

* Readme and clang format

* Update branch & conflicts

* Multiple D - fix clang-formatter

* Rename elemetwise_op

* Fix CI

* Code review part1

* Remove printf

* Remove unnecessary comment

* Add new tests with Col layout

* Review part 2

* Added support for Multiple D GEMM

* Update comment

* Remove maybe_unused

* Clang-format

* Review part 3

* Add comment to function

* Add comment to function: another

* Take number of params for a refrence function

* Remove additional d param for 0 tensor

* Change name of function

* Fix CI fails
 2025-06-13 19:39:11 +02:00
Bartlomiej Kocot
0a3b31a6f3 Merge branch 'develop' of github.com:ROCm/composable_kernel into barkocot/ck-tile-grouped-conv-fwd 2025-06-13 13:22:14 +00:00
kylasa
5f1ad09b61 Code drop for 2 warp ping pong scheduler along K dimension. (#2276) 
* Code drop for 2 warp ping pong scheduler along K dimension.

* Addressing code review comments.

* Addressing Clang formatting issues.

* Addressing build issues.

* Addressing build issues of other GEMM pipelines with ping pong scheduler code drop.

* Fix for LDS memory size for GEMM pipelines.

* Addressing code review feedback comments.

* Change log update.

* Addressing code review comments and build issues.

* Added new policy for pipeline specific logic about LDS needs.

* Clang Fix during build.
 2025-06-12 18:24:02 -07:00
Thomas Ning
f59b8c7d3d OCP FP8 Macro restructure (#2331) 
* solved the problem
 2025-06-12 09:46:33 -07:00
carlushuang
8aff45a8af [CK_TILE] moe sorting optimization : refactor subtoken logic to let more kernel pickup mp kernel (#2327) 
* refactor subtoken logic to let more kernel pickup mp kernel

* typo
 2025-06-12 11:44:22 +08:00
Thomas Ning
06e0b8436c Epilogue cshuffle Improvement (#2312) 
* add cshuffle's mxdlperwavepershuffle support, not finished

* add epilogue functions

* add cshuffle's mxdlperwavepershuffle support, not finished

* add epilogue functions

* update cshuffle logic

* update cshuffle_logics

* add some change within review

* update some codes following the code review

* update epilogue logic

* remove from problem

* update codes following review.

* fix some issues

* solve the previous PR error, refine the code

* Update include/ck_tile/ops/epilogue/cshuffle_epilogue.hpp

Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>

* Comment addressed

* handling tile_engine failing case

* handling tile_engine failing case

---------

Co-authored-by: joyeamd <John.Ye@amd.com>
Co-authored-by: joye <joye@amd.com>
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>
Co-authored-by: khushbu agarwal <khuagarw@amd.com>
 2025-06-10 22:44:50 -07:00
Thomas Ning
14d229d6c8 fix on the typo (#2326) 2025-06-10 16:34:33 -07:00
Khushbu Agarwal
bd270fe4bc fix flatmm kernel for bigger size for fp16 datatype (#2302) 2025-06-10 11:13:40 -07:00
Eisuke Kawashima
4e586ca958 chore: unset executable permission (#2303) 
Co-authored-by: Eisuke Kawashima <e-kwsm@users.noreply.github.com>
 2025-06-10 09:13:59 -07:00
John Afaganis
6635d1bb88 Remove usage of 'warpSize' variable as it has been deprecated (#2295) 
* SWDEV-535598 - remove usage of 'warpSize' variable as it has been deprecated. Ideally get_warp_size() should not be constexpr but this is just a workaround

* SWDEV-535598 - remove comment from get_warp_size as constexpr is required for this repo

---------

Co-authored-by: Gerardo Hernandez <gerardo.hernandez@amd.com>
 2025-06-10 07:34:54 -07:00
carlushuang
2e0536269e hot fix (#2315) 2025-06-10 20:35:28 +08:00
MHYangAMD
9fcf21a4ec Fix fmha fwd precision issue on MI3XX series (#2285) 
* Fix fmha fwd precision issue on MI3XX series

For fmha fwd fp16 cases, we found that using
impl::cast_tile_pk_fp16_fp32 for casting P would lead to precision
issues, since it uses __builtin_amdgcn_cvt_pkrtz, which is round to zero.

For examaple, fixing K,V to be all 1, and Q is random, which outputs are
expected to be all 1. But we found that it would have some incorrect
outputs 0.9995, which are smaller than the atol 0.001. (1 - 0.9995 =
0.0005 < 0.001) Thus, ck do not report this error.

* Add option to switch rtn/rtz for fmha fwd
 2025-06-10 15:03:23 +08:00
carlushuang
65835c0bbb MUST USE INLINE FOR ANY NON TEMPLATE FUNCTION IN HEADER!!! (#2305) 2025-06-10 10:40:54 +08:00
Aviral Goel
5a0bd157db Code Refactor for check_err.hpp (#2284) 
* refactor & add documentation

* removed return datatype from doxygen comments

* Update include/ck_tile/host/check_err.hpp

Co-authored-by: John Afaganis <john.afaganis@amd.com>

* Update include/ck_tile/host/check_err.hpp

Co-authored-by: spolifroni-amd <Sandra.Polifroni@amd.com>

* Update include/ck_tile/host/check_err.hpp

Co-authored-by: spolifroni-amd <Sandra.Polifroni@amd.com>

* Update include/ck_tile/host/check_err.hpp

Co-authored-by: spolifroni-amd <Sandra.Polifroni@amd.com>

* Update include/ck_tile/host/check_err.hpp

Co-authored-by: spolifroni-amd <Sandra.Polifroni@amd.com>

---------

Co-authored-by: John Afaganis <john.afaganis@amd.com>
Co-authored-by: spolifroni-amd <Sandra.Polifroni@amd.com>
 2025-06-08 13:41:27 -07:00
Sami Remes
1c6f83df6c [CK_TILE] Tileloop persistent gemm - resubmit (#2299) 
* Reapply "[CK_TILE] Tile loop persistent gemm kernel (#2191)" (#2293)

This reverts commit 233e274077.

* Add missing header for kentry

---------

Co-authored-by: Thomas Ning <Thomas.Ning@amd.com>
 2025-06-06 14:18:49 -07:00
valarLip
8482977a37 extend buffer load to support load 32 bf16/fp16 at same time (#2291) 2025-06-06 17:21:19 +08:00
Andriy Roshchenko
00247e3c29 Optimized GEMMs for MX FP4/8 (#2294) 
Adds V3 GEMM pipeline for MX FP4 and MX FP8 
Adds V3 GEMM pipeline for MX FP4 with preshuffling
Adds MXFP4 GEMM tests (#2275)
Adds MXFP4 GEMM examples
Adds MXFP4 GEMMs to ckProfiler




Co-authored-by: Andriy Roshchenko <107577548+andriy-ca@users.noreply.github.com>
Co-authored-by: Andriy Roshchenko <andriy.roshchenko@amd.com>
Co-authored-by: aska-0096 <haocwang@amd.com>
Co-authored-by: lalala-sh <Jiaxing.Wen@amd.com>
Co-authored-by: OscarXu <huaiguxu@amd.com>
Co-authored-by: mtgu0705 <mtgu@amd.com>
Co-authored-by: Ding, Yi <yi.ding@amd.com>
Co-authored-by: feifei14119 <feiw@amd.com>
Co-authored-by: Lin, Qun <qlin@amd.com>
Co-authored-by: joye <joye@amd.com>
Co-authored-by: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>
 2025-06-05 13:54:15 -06:00
Illia Silin
233e274077 Revert "[CK_TILE] Tile loop persistent gemm kernel (#2191)" (#2293) 
This reverts commit ffb52783d0.
 2025-06-05 09:24:00 -07:00
Sami Remes
7ea1508b59 [CK_TILE] Move GEMM pipeline tail handling logic to pipelines (#2222) 
* Add TailHandler for V3, V4 and Mem pipelines

* Adapt examples and tests to use TailHandler

* move tail-handling logic to pipeline in persistent grouped gemm

* Fix Mem pipeline dispatching, add CompV4 dispatching

* Use a macro for handling the many tails of Mem pipeline

* Fix formatting again

* Use const-ref RunFunction, remove unnecessary try_run
 2025-06-04 11:50:21 +03:00
Sami Remes
ffb52783d0 [CK_TILE] Tile loop persistent gemm kernel (#2191) 
* Implement tile loop persistent gemm kernel

* Enable timing

* Add tests for persistent gemm

* Fix formatting

* Fix gemm_basic

* Rename True/False to Persistent/NonPersistent

* Use only one set of layouts for persistent tests

* Fix gemm example persistent template parameter

* Fix formatting
 2025-06-04 11:46:28 +03:00
Khushbu Agarwal
59a85cb4bc [CK_Tile] Fix gemm kernel for 4,64,16 and 64,4,16 warp tile sizes (#2262) 
* debugging issue

* debugging issue

* debugging

* debugging

* reverting debugging code

* clang formatted

* updating default_config.json

* fix ci failure

* clang formatted
 2025-06-03 20:16:10 -07:00
Khushbu Agarwal
2e38eb4f1c Rotating buffer PR CI fix (#2257) 
* Revert "Revert "[CK_tile] Add rotating buffer feature for universal gemm (#2200)" (#2256)"

This reverts commit bbdaf79a52.

* fix regression
 2025-06-02 10:25:01 -07:00
valarLip
0fdbf6bcd1 extend buffer load for fp16/bf16x16 (#2270) 
* extend buffer load for fp16/bf16x16

* format
 2025-06-02 10:29:54 +08:00
Illia Silin
4e561af18c Revert "add CShuffleM/NXdlPerWavePerShuffle in cshuffle_epilogue (#2185)" (#2260) 
This reverts commit fd6a859b44.
 2025-05-29 16:22:16 -07:00
joyeamd
fd6a859b44 add CShuffleM/NXdlPerWavePerShuffle in cshuffle_epilogue (#2185) 
* add cshuffle's mxdlperwavepershuffle support, not finished

* add epilogue functions

* add cshuffle's mxdlperwavepershuffle support, not finished

* add epilogue functions

* update cshuffle logic

* update cshuffle_logics

* add some change within review

* update some codes following the code review

* update epilogue logic

* remove from problem

* update codes following review.

* fix some issues
 2025-05-29 14:31:14 +02:00
Illia Silin
bbdaf79a52 Revert "[CK_tile] Add rotating buffer feature for universal gemm (#2200)" (#2256) 
This reverts commit 99857e10e6.
 2025-05-28 09:46:52 -06:00
Khushbu Agarwal
99857e10e6 [CK_tile] Add rotating buffer feature for universal gemm (#2200) 
* Add rotating buffer feature for universal gemm

* adding changes in tile_engine

* Updated code to merge kernel_launch

* removing comments

* Enable rotating buffer changes to flatmm

* Created diff launch_kernel function for rotating buffer

* Simplfied calculation using macros

* merge code with new changes in tile_engine

* clang formatted

* Redefine macros
 2025-05-27 23:00:58 -07:00
Casey-Shi
128f5a1eab [Tile Engine] Add benchmark for tile engine gemm. (#2193) 
* initial commit -m benchmark

* only support profile

* fix

* fix doc

* add default config

* add ci

* fix cmake

* tmp save for gen blobs

* fix bug

* merge

* range config

* test success

* fix

* fix

* move struct

* remove config property

* fix config

* remove comment

* add cmake option & modify

* add changelog

* fix

* format

* add pydantic module to the docker image

* fix

* add benchmark for cold and warmp up

* python format

* add asm cache control

* fix README

* remove pydantic module

* modify changelog

* fix config

* recover benchmark_gemm and fix

* format python

* refactor profiler

* fix csv bug

* fix codegen bug

* add kernel instance object

* add benchmark gemm executable

* fix jenkins & delete extra header

* disable warning output & enable default config

* Disable sparsity for invalid warp tile combinations

* fix gemm host template func

* refactor gemm profiler

* filter out some inmstances

* default config test & fix codegen bug

* add sparse flag to gen more instances

---------

Co-authored-by: illsilin <Illia.Silin@amd.com>
Co-authored-by: khuagarw <khuagarw@amd.com>
Co-authored-by: Thomas Ning <Thomas.Ning@amd.com>
 2025-05-26 22:32:36 -07:00
Po Yen Chen
c42b957d65 [CK_TILE] For FMHA forward kernels, assign block indices reversely if using mask (#2209) 
* Assign block indices reversely if kHasMask=true

* Assign block indices reversely for splitkv kernel
 2025-05-27 10:58:58 +08:00
Zzz9990
ece38b9d7a [VLLM V1] Add chunked prefill for FA to pass seq with small seqlen_q (#2221) 
* fix splitkv compiler issue since lse is used to select kernel instances

* bypass seqlen == 1

* add chunked prefill into mha varlen

This reverts commit aa9847e42d.

* skip compile when receipt 2-4 and add comments

* fix

---------

Co-authored-by: fsx950223 <fsx950223@outlook.com>
 2025-05-26 19:17:18 +08:00
Illia Silin
8146e471f1 fix the buffer intrinsic names for clang >=20 (#2228) 2025-05-23 14:58:25 -07:00
Illia Silin
1b846143c6 Revert "Update the buffer load/store intrinsic names for clang>=20. (#2192)" (#2227) 
This reverts commit 58f9e9ffbc.
 2025-05-22 15:41:17 -07:00
Aviral Goel
534d4594d0 Refactor tile_window.hpp, tile_window_linear.hpp into a CK Tile Hierarchy (#2214) 
* window_origin variable now in base class

* abstracted more functions

* consolidated tile_window_static_distribution and tile_window_static_lengths

* clang format

* skeleton code for tile_window and tile_window_linear consolidation

* more abstraction

* moved variables from child to parent

* clang format

* removed comments

* removed debug code

* removed debug code

* abstracting traits WIP

* consolidated traits

* removed comments and clang formatted
 2025-05-21 23:28:00 -07:00
Aviral Goel
fa39c4e798 Add Doxygen Documentation for HostTesnor, HostTensorDescriptor, DeviceMem, FillUniformDistribution (#2160) 
* added documentation for HostTensorDescriptor

* added documentation for DeviceMem and FillUniformDistribution

* fixed merging error

* fixed host_tensor_descriptor error

* clang format
 2025-05-21 10:34:30 -07:00
Sami Remes
d1e6f0982d [CK_TILE] Grouped GEMM tile loop (#2146) 
* Add trait to use a persistent kernel and split the entrypoints in grouped gemm

* Some helper functions for persistent kernel case

* Get max occupancy grid using device properties

* Implement tile loop in main entry point to grouped gemm

* Enable GridSize() on device

* Handle offset tile index using real current block index

* Add persistent kernel choice to grouped gemm example

* Use a for-loop for iterating over the group

* Reduce VGPR spills by early-exit

* Enable persistent kernel choice in grouped_gemm example

* Add persistent kernel option to grouped_gemm test

* Fix formatting with remod.py

* Remove GridUpdateBlocks as blocks are now iteratively computed

* Add comment about VGPR spilling

* Fix formatting

* Use CK_TILE_HOST instead of __host__

* Enable all Row/Col combinations in grouped gemm unit test

* Add some KBatch=2 cases to grouped gemm tests

* Fix SplitK for grouped gemm

* Enable pipeline hotloop/tailnumber selection in-kernel for grouped gemm

* Add type traits

* Split examples to regular and tileloop

* Formatting

* Use hipExtStreamGetCUMask to get current active CUs for the given stream

* Align test and example kernel config, and disable validation for splitk repeats

* Remove debug options from CMakeLists.txt

* Separate the code paths for persistent/non-persistent in test

* Fix formatting

* Address review comments

---------

Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>
 2025-05-20 17:18:57 +03:00
Po Yen Chen
791802b381 [CK_TILE] fMHA batch_prefill block index & logits soft-capping optimizations (#2198) 
* Write soft-sign in inline asm

* Change tile idx computation

* Add macro to turn off soft-sign asm opt

* Use simple for loop to avoid register spill

* Only do block id transform for masking cases
 2025-05-16 15:14:46 +08:00
Khushbu Agarwal
3d8d6e75e4 Adding validation for tile sizes in Tile Engine (#2189) 
* Adding validation for tile sizes

* Add architecture in config, and shuffle lines of code in warp_gemm.hpp

* Enable MFMA for gfx950, and invalid tile handling
 2025-05-15 10:28:31 -07:00
BingYuan.Zhou
41c17d0a95 fix moe sorting build fail (#2190) 
* fix moe sorting build fail

* refile code

---------

Co-authored-by: solin <bingzhou@amd.com>
 2025-05-14 09:31:26 +08:00
Illia Silin
58f9e9ffbc Update the buffer load/store intrinsic names for clang>=20. (#2192) 
* fix the buffer load/store intrinsic names

* fix clang format
 2025-05-13 10:18:14 -07:00
Bartlomiej Kocot
976e32ccfa custom vector size 2025-05-13 14:38:36 +00:00
Bartlomiej Kocot
aaca2e2b08 [CK TILE] Grouped Convolution Forward Kernel 2025-05-13 11:52:11 +00:00
Po Yen Chen
2920604786 [CK_TILE] Add logits soft-capping & customization support to the FMHA forward kernel/pipelines (#2163) 
* hack for cap logits

* fix bug

* Re-format files

* Allow specifying logits_soft_cap through APIs

* Support turn on/off logits_soft_cap in async pipeline

* Do not generate non-verified kernels

* Align receipt used in Aiter

* Sync logits soft-capping across pipelines

* Re-enable some hdim pipelines

* fix perf

* Add attention variant for logits_soft_cap

* Add newline at end-of-file

* Fix performance

* Add comment to explain logits_soft_cap pre-processing

* Unify code

* Unify floating-point literal style

* Use class data member to slience the compilation error

* [CK_TILE] Update attention customizaton interface: add LogitsMask() (#2133)

* Send 'mask' along with variant params to the LogitsMask()

* Send block indices to the variant

* Add indices parameters in variant interface

* Fix fmha bwd codegen error

* Allow switch logits_soft_cap impl

* Eliminate register spills

* Fix compilation errors

* Fix wrong LSE

* Fix LSE for splitkv kernel

* Sync splitkv pipeline changes

* Add batch_prefill kernel/pipeline

* Fix codegen error

* Undo changes in CMakeLists.txt

* Merge pipeline filtering check

* Use different code path if kHasLogitsSoftCap=false

* Remove [[maybe_unused]] attribute

* Use pre-existing compile-time flag to instantiate templates

* Sync pipeline changes

* Update CHANGELOG.md

---------

Co-authored-by: Bernard <bernaliu@amd.com>
Co-authored-by: coderfeli <coderfeli@163.com>
 2025-05-13 12:19:25 +08:00