GitLab

* add batched_transpose implement

---------
Co-authored-by: root <root@ctr-ubbsmc16.amd.com>
Co-authored-by: ThruptiRajLakshmanaGowda <tlakshma@amd.com>
Co-authored-by: ThomasNing <thomas.ning@amd.com>

* Add OCP FP8 to CK_TILE

* Validate OCP FP8 in FMHA FWD under VALID=1

* Refactor universal gemm policy.

* Adapt example to refactor changes.

* Introduce static encoding pattern

* Adding shuffled encoding patterns.

* Fix err in reverse tuple.

* Add transpose_tile2d

* Small refactoring + doc

* Enable reading on contiguous dimension in all layouts.

* Transpose A/B register tile if needed for comp v3 pipeline.

* Take contiguous dim size when calculating dram vector load size.

* A/B smem pack size taken from WarpGemm attributes

* Update B LDS layout and setup tile distribution pattern at class level.

* Fix static assert.

* Fix errors in examples.

* Formatting & fix IsTranspose

* Fix VectorSize & refactor.

* Add error loging messages.

* Fix VecLoadSize and TranspseC for mem pipeline.

* Update unit-tests & disable mem pipeline.

* Clang format

* Update include/ck_tile/core/tensor/tile_window.hpp
Co-authored-by: jakpiase <jakub.piasecki@amd.com>

* Fix compilation and reviewers comments.

* Refactor unit-test. Fallback to non-universal gemm.

Need to use GemmPipelineAGmemBGmemCRegV1 for now,
since GemmKernel is now supporting also non-K major vector reads.

---------
Co-authored-by: jakpiase <jakub.piasecki@amd.com>

* not using structures under ck_tile/ops for ck_tile/host

* update as constexpr function

* Rename fn

* Update other examples.

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

* Grouped gemm simple code refactor

* Offset invoker

* Invoke generic Run, and replace name of parrtitioner variable

* Tests fix type

* Removed namespaces

* Add template param to avoid implicit cast

* Remove generic function

* Constant value

* underline enum to int16_t

* Generalize partitioner function

* Remove whitespaces

* Rename function

* Using support

* Clang-format

* Clang-format

* Fn-partitioner description fn

* Typo

* Typo 2

* Better description

* Better description

* Refactor after review

* Use ctr instead of set fn

* Inovke ctr and typo

* Comments

* Remove unnecessary comment

* Review, remove modulo

* smoke and regression targets working with tests

* test filters work for both examples and test

* removed uneccesary comments

* added a missing comment

* added a missing comment

* fixed typo in the comments

* updated README

* Update PULL_REQUEST_TEMPLATE.md

updating the template for future addition of test cases

* Update PULL_REQUEST_TEMPLATE.md

* fix mock token id

* prepare host for g1u1

* reformat inline-asm

* restructure uk_0

* restructure gate_up

* done

* change default to init=1

* update readme

* fix a bug in interleave pipeline

* rcp for silu

* Add rounding for float to bf16 conversion

* Add bhalf test

* Add inf test bhalf

* Refactor

* update cmake

* Fixes

* Add shortcut to RMSNorm

* Modify test for adding shortcut for RMSNorm

* Add fused parameter into tests

* 1. Add YDataType. 2. rmsnorm2d_fwd_traits_ from rmsnorm2d_fwd.hpp to rmsnorm2d_fwd_api.cpp and rmsnorm2d_fwd_instance_common.hpp

* 1. Supports various stride and percisions.

* Add support of Epilogue

* Add fuse and epilogue support to rmsnorm ref

* Modify rmsnorm example

* Refactor tests/examples

* Bug fix for newly added tests/examples

* Bug fix for new tests 2

* Modify smoke test scripts

remove dbg code

* Supports non-smooth dyanmic quant

* Update Rmsnorm2dFwd::GetName()

* rename xscale and prec_sx to smoothscale and prec_sm

Bug fix after rename

Remove files

* change example_rmsnorm2d_fwd.cpp

* update performance calculator

* Fix issue in two-pass when fuse add is enabled

* Remove comment of beta

---------
Co-authored-by: rocking <ChunYu.Lai@amd.com>

* refactor the block_gemm_areg_breg_creg_v1 and add the v2 policy with 2x2 warp gemm

* Finished the 2x2 warp gemm policy and the block selection mechanism

* Clang format

* address poyen's comment

* Address feedbacks

* Fixed the compilation issue

* Change the function name

* Observed a 2x perf improvement with kBlockSize = 256
* Using 512 threads may lead to redundant computations

* Finished adding the performance benchmark for ck tile gemm

* Fix the executable rename problem

* fix the executable name error

* delete the unsupported layout combinations

* Update run_full_test.sh

* Update benchmark_mem_pipeline.sh

* Update benchmark_basic.sh

* change the executable of gemm_universal

* change ck_tile_gemm script permissions

* Addressed the comment

* Addressed the comment

* Fixed the comments

* Fixed Comment

* roll back the malfunctioned change

* Fix the Typo

* finalize the tile_gemm_fp16 performance monitoring

* fix the stash names for ck_tile gemm logs

* change the stashing logic

* change stashing syntax

---------
Co-authored-by: Illia Silin <98187287+illsilin@users.noreply.github.com>
Co-authored-by: illsilin <Illia.Silin@amd.com>

* Disable building DPP kernels by default

* Disable building dpp instances, examples, or tests if DPP_KERNELS is not set

* Add new DPP_KERNELS flag to readme

* 1. enable bias feature that add bias before adding residual; 2. change block size from 128->64 when m<64 in fp16

* delete comment

* 1.remove fmha change 2.change buffer name from bias to xbias

* Now bias can be used independently from fadd

* change kbias to kxbias

---------
Co-authored-by: feli <felix.li@amd.com>

* Update license year

* Add initial code to override decode problem

* Fix splitkv traits/args overriding error

* Reshape and transpose lse for decode

* Remove debug code

* Prettify example code

* Use better function name

* Add kMergeNumHeadGroupsSeqLenQ flag

Kernel user can use this switch to turn on/off optimization for
some problem sizes

* Add missing flag declarations

* Default turn off kMergeNumHeadGroupsSeqLenQ in codegen

* Group similar statements together

* Remove assumption of seqlen_q=1

* Remove kMergeNumHeadGroupsSeqLenQ from splitkv combine kernel

* Support kMergeNumHeadGroupsSeqLenQ=true in fmha splitkv kernel

* Run kMergeNumHeadGroupsSeqLenQ=true kernels when need

* Fix group mode block skip logics

* Undo changes of normal fwd kernel

* Update in GridSize() and using GridSize() for splitkv kernel (#1799)

---------
Co-authored-by: Qianfeng <qianfeng.zhang@amd.com>

* quant

* fix bug

* simple smoothquant after softmax

* update kv-quant

* update stride

* fix fp8-pertoken-kvcache

* update int8/fp8 quant support

---------

Co-authored-by: so <a.com>
Co-authored-by: Po Yen Chen <PoYen.Chen@amd.com>

* enable int4 scale (weight only) kernel

* format some files

* Add unit test for int4 weight only

* fixed and formatted code

* fixed

* formated

* formated

* fixed

* fixed a bug in the ckProfiler, and formatted the code

---------
Co-authored-by: mtgu0705 <mtgu@amd.com>

* add no welford

* enable output raw

* raw of int8

* fix build

* fix smoke test err

* [ck_tile]layernorm: fix welford ok, set int8 and bf16 small N as default and others open by generate

* [cktile]layernorm, fix err commit files and remove uselss

* fix quant 8192 err & change norm_reduce class and file name

---------
Co-authored-by: coderfeli <coderfeli@163.com>
Co-authored-by: carlushuang <carlus.huang@amd.com>

* initial

* Cmake file

* successfull compilation but validation failed

* Cmake

* update

* gpu validation

* gemm universal

* gemm universal sk update

* sk bf16 universal instance

* gemm_universal_streamk.hpp

* only build for gfx94

* Cmakelist

* profiler update, bf16 sk only works at gfx42

* clang

* clang

* clang all

* no need flags

* cmake script

* delete comment

* gemm universal sk fix

* clang

* profiler fix

* clang

* update

* update

* delete comment

* code formatting

* cmake

* fix instance

* clang

* argument supported

* argument supported and clang

* update

* fix

* removing unnecessary comments

* clang formatting

* Update library/src/tensor_operation_instance/gpu/CMakeLists.txt
Co-authored-by: afagaj <john.afaganis@gmail.com>

* CopyRight Comment 2025

* clang reformatting

* copy right 2025

---------
Co-authored-by: Emin Ozturk <ozturk.27@osu.edu>
Co-authored-by: root <root@ctr-ubbsmc16.amd.com>
Co-authored-by: Muhammed Emin Ozturk <meozturk@t004-008.hpcfund>
Co-authored-by: root <root@splinter-126-wr-d3.amd.com>
Co-authored-by: Muhammed Emin Ozturk <meozturk@t006-001.hpcfund>
Co-authored-by: Muhammed Emin Ozturk <meozturk@login1.hpcfund>
Co-authored-by: Muhammed Emin Ozturk <meozturk@t004-004.hpcfund>
Co-authored-by: Emin Ozturk <emin.ozturk@utah.edu>
Co-authored-by: Muhammed Emin Ozturk <meozturk@t008-001.hpcfund>
Co-authored-by: afagaj <john.afaganis@gmail.com>

* add a prototype of int4

* clean

* debug

* clean

* clean

* move packed into dynamic_buffer

* fixed coord reset

* add fast pki4 to half conversion

* fix

* fixed reference and host_tensor

* fixed tensor init

* format

* debug i4_to_f16_convert

* format

* fixed splitk

* weight permute

* add b tile permute

* clean

* weight permute with splitki

* format

* improve weight layout

* add and_or_b32

* fixed splitk crush

* add permute switch as a template

* recover v3r1

* clean

* failure with intrawave v2

* fixed

* fixed

* add ckProfiler

* add bfp16 support

* add bf16 example

* fixed int4 to bhalf_t conversion

* format

* fixed int4 to bf16 conversion

* clean

* add instances for mem

* clean

* fixed host tensor size

* fixed

* debug

* fixed

* add pk_i4_t as a struct

* fix

* Update example/01_gemm/gemm_xdl_bf16_pk_i4_v3.cpp
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>

* Update example/01_gemm/gemm_xdl_bf16_pk_i4_v3.cpp
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>

* Update example/01_gemm/gemm_xdl_bf16_pk_i4_v3.cpp
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>

* revert

* Update example/01_gemm/gemm_xdl_bf16_pk_i4_v3.cpp
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>

* Update example/01_gemm/gemm_xdl_fp16_pk_i4_v3.cpp
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>

* Update example/01_gemm/gemm_xdl_fp16_pk_i4_v3.cpp
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>

* Update example/01_gemm/gemm_xdl_fp16_pk_i4_v3.cpp
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>

* Update example/01_gemm/gemm_xdl_fp16_pk_i4_v3.cpp
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>

* fixed comments

* revert

* clean

* revert

* revert

* fixed

* Update CMakeLists.txt

* Update script/cmake-ck-dev.sh
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>

* Update include/ck/tensor_operation/gpu/element/unary_element_wise_operation.hpp
Co-authored-by: Adam Osewski <19374865+aosewski@users.noreply.github.com>

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

* fixed

* fixed

* fixed

* revert

* revert

* add comments

* format

* fixed assert

* fixed

* Fix I4 define in ckProfiler

* Fixed example_gemm_xdl_bf16_pk_i4_v3 test failed issue

---------
Co-authored-by: Jing Zhang <jizhan@fb.com>
Co-authored-by: zjing14 <zhangjing14@gmail.com>
Co-authored-by: mtgu0705 <mtgu@amd.com>

* Remove using tile partitioner for fmha_fwd_kernel

* Remove using tile partitioner for fmha_fwd_splitkv and splitkv-combine kernels

* Remove using tile partitioner for fmha_fwd_appendkv kernel

* Unify the format of GetTileIndex

* [CK TILE] Add split K support in GEMM

* Updates

* Fixes

* rebase

* fix

* Fix

* fixes

* support for batched gemm

* opt moe sorting

* remove commented code

* Add check for zero values

* Add static assertions

* Remove invalid option '-e' in smoke_test.sh

* Use correct path of smoke_test.sh

* Avoid zero-sized shared memory array

* Add warning comment

* Replace expr by integer_divide_ceil() call

* Use more readable constant names

* Write down assumption as static assertion

* Add more diagnostic error messages

* Fix wrong BlockWarps when using default pipeline policy

* Add more static assertions for A LDS desc

* Allow using vector size < 8 for data type fp16/bf16

* Align vector size between DRAM dist & LDS desc

* Remove no-longer used func decl

* Fix wrong displayed piepline name

* Undo policy template changes for tile_example_gemm_basic

* Add missing space and make error message stands out

* Unify print precision

* Add missing include directive <iomanip>

* Replace constant 64 by get_warp_size() call

* Replace constant 128 by named variable: BankLength

* Add kAMBlock/kBNBlock attributes

* Allow usig different A/B warp dist for multiple blocks

* Add helper function to get warp dist encodings

* Add 4x64x4 fp16 warp gemm attribute impl

* Complete the A/B warp dist encoding logic

* Fix wrong thread mapping for C matrix

* Use smaller vector size for small tile

* Add static assert to block unsupported warp gemm impl

* Extract common code out as helper method

* Add 4x64x16 fp16 warp gemm type alias

* Add comment to warning developers

* Undo WarpGemmAtrributeMfma<> changes

* Use more clear static assertion error message

* Add trivial wrapper to get warp dstr encodings

* Only transpose warp gemm result if it's square

* Fix compilation error

* Support multi-block warp gemm (on N direction)

* Remove duplicated code

* Fix output encoding of warp gemm

* Fix wrong shape of WarpGemmAtrributeMfmaIterateK<>

* Remove unused code

* Fix wrong shape of WarpGemmAttributeMfmaImplF16F16F32M4N64K4

* Add type config for bf16_t

* Add 4x64x16 bf16 warp gemm

* Update WarpGemmAtrributeMfmaIterateKAndTransposedCDistribution

* Add 64x4x4 fp16/bf16 warp gemm impl

* Add 64x4x16 fp16/bf16 warp gemm

* Add static assertion for better error diagnostic

* Get Q dram dstr directly form block gemm

* Add missing header: fused_moe.hpp

* Allow specifying different warp-gemm for gemm0 & gemm1

* Store P matrix into LDS before gemm1

* Fix inconsistant kernel name

* Remove constraint on gemm0 & gemm1 block warps

* Remove unsupported vector size from checking list

* Allow using 4x64x16 warp gemm for gemm0

* Finish policy customization

* Finish pipeline modification
F#

* Use block warps in codegen

* Fix wrong rank of m_lds_window origin

* Use better distributed tensor

* Make P-store earlier

* Remove duplicated experssions

* Remove unnecessary tile window

* Create new files for new splitkv pipeline

* Separate old/new pipeline codegen logic

* Sync changes form develop

* Undo gemm kernel/pipeline changes

* Undo gemm example changes

* Remove blank lines

* Fix typo

* Use new warp gemm interface

* Fix link error

* Fix wrong pipeline tag

* Fix more link error

* Avoid unnecessary padding

* Always use vector load for K

* Padding on fastest dimension when necessary

* Force padding Q on hdim_q

* Set high dimension padding flag to false

* Re-format headers

* Use warps=<1, 4, 1> for both gemm0 & gemm1

* Fix complilation errors

* Remove m/l shuffle logics

* Ignore duplicate data when write lse_acc

* Use gemm0 block warps as lds tile width

* Remove hard-coded numbers

* Fix wrong distribution width

* Remove unnecessary code

* Add s_barrier before writing to LDS

* Store Q into LDS before gemm0

* Fix wrong Q tile size

* Use simple Q lds descriptor for debuging

* Use more realistic Q lds descriptor

* Add comment & use better variable name

* Make Q lds space not overlapped with others

* Remove unnecessary block_tile_reduce_sync() call

* Move Q load statements

* Move block_sync_lds() right before use

* Re-order instructions

* Remove necessary lambda expression

* Use 8 threads on kMaxSplits direction while doing reduction

* Tiny correction for using 8 threads on kMaxSplits direction for combine kernel

* Padding num_split direction of o_acc tile window to 4x

* Update splitkv combine pipeline design

* Add kN1 back to splitkv combine pipeline problem

* Fix compilation errors

* Add missing template parameter

* Fix wrong splitkv combine kernel name

* Fix wrong origin

* Fix wrong LDS descriptor shape

* Fix sync & reduction logics

* Remove unnecessary static assertions

* Extract tile size computation logics

* Make sure we can reuse padding flags in combine kernels

* Rename variables

* Use OaccDataType in BlockFmhaSplitKVCombinePipelineTileSizes<>

* Remove unnecessary static assertion

* Fix function name typo

* Add constraint on kN1 template parameter

* Hide K tile loading latency in earlier iteration

* Fix wrong splitkv kernel name

* Use s_shuffling to replace p_shuffling which removes the needs of cross-warp reduction

* Rename pipeline

* Fix wrong pipeline name attribute

* Add GetAlignmentQ() for NWarpSShuffle pipeline

* Separate Q tile into dram tile & register tile concepts

* Remove non-squre warp gemm transpose c type alias

* Fallback tile size changes for fmha fwd splitkv

* Remove redundant change

* Refine naming for the S tile

* Use better naming of the S tile dstr (read from lds)

* Share Q lds with K lds

* Tiny change

* Fix with using static_for for passing CI checking

---------
Co-authored-by: Qianfeng Zhang <Qianfeng.Zhang@amd.com>

* Parser for a vector was added. Additionaly we valid correctnes of numbers

* Remove unnecessary comments

* Review part 1

* Review part 2

* Add const to variadic lambda

* Rename C->K

* Gemm Kernel Refactor part1

* Gemm Kernel Refactor common gemm pipeline part2

* [CK TILE] Refactor batched gemm to reuse GemmKernel

* [CK TILE] Refactor GemmKernel - review changes part1

* [CK TILE] Refactor GemmKernel - references fix

* [CK TILE] Refactor GemmKernel - naming changes, add problem

* [CK_TILE] Refactor GemmKernel - update tests

* [CK_TILE] Refactor GemmKernel - review changes

* [CK_TILE] Refactor GemmKernel - update test

* [CK_TILE] Refactor GemmKernel - constness fixes

* [CK_TILE] Refactor GemmKernel - update tests

* [CK_TILE] Move hipmalloc/memcpy calls out of gpu reference gemm

* [CK_TILE] Move hipmalloc/memcpy calls out of gpu reference gemm - review changes

* [CK_TILE] Move hipmalloc/memcpy calls out of gpu reference gemm - review fix

* add bmm api

* add bf16 multi_d

* add ckProfiler for bf16

* add ckProfiler files

* add more instance; fixed 64bit index issue

* fixed naming

* enabled batched Ds

* use long_index for ds offsets

* clean

* add bmm fp8 ckProfiler

* Update example/24_batched_gemm/batched_gemm_xdl_bf16_v3.cpp
Co-authored-by: Bartłomiej Kocot <bartlomiejkocot98@gmail.com>

* Update example/24_batched_gemm/batched_gemm_xdl_fp8_rowwise_v3.cpp
Co-authored-by: Bartłomiej Kocot <bartlomiejkocot98@gmail.com>

* Update example/24_batched_gemm/run_batched_gemm_example_rowwise.inc
Co-authored-by: Bartłomiej Kocot <bartlomiejkocot98@gmail.com>

* Update library/src/tensor_operation_instance/gpu/gemm_universal_batched/device_batched_gemm_xdl_universal_bf16_bf16_bf16/device_batched_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn.hpp
Co-authored-by: Bartłomiej Kocot <bartlomiejkocot98@gmail.com>

* Update library/src/tensor_operation_instance/gpu/gemm_universal_batched/device_batched_gemm_xdl_universal_bf16_bf16_bf16/device_batched_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn_mem_v1_default_instance.cpp
Co-authored-by: Bartłomiej Kocot <bartlomiejkocot98@gmail.com>

* Update library/src/tensor_operation_instance/gpu/gemm_universal_batched/device_batched_gemm_xdl_universal_bf16_bf16_bf16/device_batched_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn_mem_v2_default_instance.cpp
Co-authored-by: Bartłomiej Kocot <bartlomiejkocot98@gmail.com>

* Update profiler/src/profile_gemm_universal_batched.cpp
Co-authored-by: Bartłomiej Kocot <bartlomiejkocot98@gmail.com>

* Update profiler/include/profiler/profile_gemm_universal_batched_impl.hpp
Co-authored-by: Bartłomiej Kocot <bartlomiejkocot98@gmail.com>

* clean

* Update include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_xdl_cshuffle_v3.hpp

* Update include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_xdl_cshuffle_v3.hpp

* Update library/src/tensor_operation_instance/gpu/gemm_universal_batched/device_batched_gemm_xdl_universal_bf16_bf16_bf16/device_batched_gemm_xdl_universal_bf16_bf16_bf16_mk_nk_mn_comp_default_instance.cpp

* Update include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_xdl_cshuffle_v3.hpp

* Update include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_xdl_cshuffle_v3.hpp

* Update include/ck/tensor_operation/gpu/device/impl/device_batched_gemm_multiple_d_xdl_cshuffle_v3.hpp

* refactor batch offset func

* add splitk suppport into bmm_v3

* clean

* clean

* format

* fixed

* fix

---------
Co-authored-by: Jing Zhang <jizhan@fb.com>
Co-authored-by: zjing14 <zhangjing14@gmail.com>

* add ck_tile/smoothquant out stride parameter

* Remove the default stride value

---------

Co-authored-by: so <a.com>

* add reference attention fwd

* refactor addresser

* update

* paged, and i8 reflect-quant

* lets call it forward-quant

* fix error in decode variation

* update naive-attn

* fix page table

* fix build err

* Add data type config, Prepare to add mix precision in the future

* Fix compile error

* upgrade to rocm6.3 compiler

* Proposed solution to convnd test failures in ROCm 6.3

---------
Co-authored-by: Andriy Roshchenko <andriy.roshchenko@amd.com>

* add IsSupportedArgument to gemm_kernel

* add ut and do some refactoring

* switched to ck_tile's integral_constant

* Ck-tile, impl. grouped gemm

* Workspace is allocated by user, and is passed to the function

* Prepare test to new api design

* Unify GemTransKernelArgs, removing N0 param

* Add 1 to dim3 in paritioner

* Typo: gem - > gemm

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

* Add basic documentation structure

* Add terminology placeholder

* Add codegen placeholder

* Create template for each page