GitLab

* (2/5) bilinear gemm pass, perf bug: skip a lds has lower performance than skip b lds

* (3/5) batched gemm pass, perf bug: skip a lds has lower performance than skip b lds

* (4/5) grouped conv pass

* (5/5) attention pass, todo: debug lds perf bug

* AIT Attention API refactor (#8)

* sanity pass

* sanity pass 2

* confirm significant performance regression.

* turn on all instances

* turn off instance format

* Fix bug & tunning & format

* DML meta, self_attn+cross_attn

* sanity pass

* remove useless flag

* update tile and problem size used in AIT attention

* bug fix in grouped conv supporting check

* deprecate inline asm wmma

* Bug fix: double lds skip

* clang-format

* Fix errors in
1. example, fmha
2. gridwise pipeline
3. deviceop, fmha, change some containers from vector to array

* part2 of previous commit

* clang format

* API fix of gridwisegemmpipeline

* separate array base and vector base attention tensor transformation

* fix gemm

* clang format

* add gemm fp16 instances

* Temp save

* fpAintB kernel compile pass

* Sanity pass.

* Temp save

* debug code enabled

* Fp16AInt8B_GEMM sanity

* MQA implementation

* GQA-4 example

* tempsave

* Compile pass

* New implementation of fp16Aint8B Gemm, Acheieve similar math throughput with native fp16 Gemm

* Bump rocm-docs-core from 0.24.0 to 0.29.0 in /docs/sphinx

Bumps [rocm-docs-core](https://github.com/RadeonOpenCompute/rocm-docs-core) from 0.24.0 to 0.29.0.
- [Release notes](https://github.com/RadeonOpenCompute/rocm-docs-core/releases)
- [Changelog](https://github.com/RadeonOpenCompute/rocm-docs-core/blob/develop/CHANGELOG.md)
- [Commits](https://github.com/RadeonOpenCompute/rocm-docs-core/compare/v0.24.0...v0.29.0

)

---
updated-dependencies:
- dependency-name: rocm-docs-core
  dependency-type: direct:production
  update-type: version-update:semver-minor
...
Signed-off-by: dependabot[bot] <support@github.com>

* initial enablement of gfx950

* fix clang format

* disable examples 31 and 41 int8 on gfx950

* initial navi4x enablement

* remove extra endif

* enabled dl_gemm

* update s_barrier and s_waitcnt for gfx12

* fix the gfx12 assembly syntax

* fixed block_sync_lds

* add support for more dl kernels on navi4

* add wmma

* format

* Todo: fix gemm_bilinear_wmma instances compilation bug

* Solve a bug when K1=16

* remove unnecessary changes

* Remove tensor layout limitation to LDS usage in tesnor contraction

* fixed block_sync_lds

* merge navi3_ref

* update self-attention and cross-attention

* fix a typo of name

* fixed layout

* debugging

* Add arch limiter for fp8 gemm

* fixed wmma

* enable fp8 gemm_xdl for all gfx9 targets

* temporarily disable gemm_xdl_fp16_fp8 on MI100/200

* fix the cmake logic for gemm_xdl_fp16_fp8

* fixed c_output

* re-enable the gemm_xdl_fp16_fp8 on MI100/200

* fixed gfx12

* fixed

* fixed

* seperate gfx12 blockwise_gemm

* fixed

* enable fwd conv on navi4x

* enable gridwise

* enabled gemm

* fixed merge

* remove empty example fold

* fixed conflicts

* some small changes

* Update cmake-ck-dev.sh

* Update cmake-ck-dev.sh

* enabled other types

* fixed register loads

* test fa

* enable gfx12

* clean up

* enable some instances on gfx12

* add gfx1201 macro in amd_wmma header

* fix clang format

* enable batched_gemm_softmax_gemm_perm_wmma for gfx12

* disable instances with blocksize=256 in attention examples

* debuggging

* debug

* fixed lds_enabled

* debugging

* Fix and add limit to skiplds feature

* Enable skipLds feature and fix compilation bugs

* add ck_tile definitions for gfx12

* fix clang format and test/wmma_op

* updage instances cmake for gfx12

* disable the test_wmma_op on gfx12

* fix the builds for gfx950

* add gfx12 and gfx950 to default target list

* clean-up cmake file

* Initial introduction of OFP8 data types.

* Renamed FP8 and BF8 tests into FP8_FNUZ and BF8_FNUZ.

* Implementation of ConvertFP32Nearest in test_fp8_ocp.

* Remove dependence on possibly undeclared alias.

* Implement FP8OCP test for stochastic rounding mode.

* Implement FP8OCP tests for half_t type conversions.

* enable bf16 atomic add on gfx950

* Implement ConvertFP32Nearest test.

* Implement ConvertFP32Stochastic test.

* Implement ConvertFP16Nearest and ConvertFP16Stochastic tests.

* Refactoring. Move FP8 definitions into a separate header file.

* Enable easy switching between architectures.

* Fix compilation error for gfx942 architecture.

* only builf gfx950 branch for gfx950 target by default

* Enable OCP build of example_gemm_xdl_fp8.

* Fix formatting.

* fix the build logic for gfx950

* Improve GEMM example verbosity.

* Add constexpr where applicable.

* fix the logic of enabling XDL and WMMA instances

* Improve GEMM example verbosity.

* Enable build of example_gemm_xdl_fp8_bf8 test.

* Fix tests for gfx1101 architecture.

* Build DPP examples only on gfx103 and gfx11 architectures.

* Optionaly run either CPU or GPU verifications with GEMM examples.

* Extend GeneratorTensor_Sequential to produce values of prescribed data types.

* Add missing constructor.

* Improve infrastructure for OFP8 data type support.

* BUGFIX. Should not use FP8 as Compute/Accum data type.

* Add custom target for grouped_convnd_bwd_weight tests.

* Can build `tests` target on gfx950.

* Bugfixes on gfx1101 architecture.

* Fix dependencies.

* Provide single point of truth for FP8 INF and NAN checks

* Prevent instantiation of operators that are not supported by FP8 data types

* Add FP8 type selection into client_axample CMakeLists.txt

* Prevent sccache server from shutting down during build

* Fix test success reporting logic

* Change default verification method to CPU.

GPU verification takes too much time to complete on the emulator.

* Make sure all tests and examples are built for gfx950

* Facilitate testing of FP8 data types on the emulator

* Introduce two new tensor generators

* Enable instances built for gfx94 to be built on gfx950

* Verify 35_splitk_gemm on floating point numbers.

splitk gemm appears to be losing precision VS reference implementation when FP numbers are involved.

* Verify 04_gemm_add_add_fastgelu on floating point numbers

* Verify 20_grouped_conv_bwd_weight on floating point numbers

* Verify 38_grouped_conv_bwd_data_multiple_d on floating point numbers

* Verify more tests on floating point data

* Fix data types and improve testing verbocity.

* Upgrade to NPI 573 build docker.

* Skip on gemm_universal tests.

The tests take too long to complete on the emulator.
Need to see if it is possible to reduce the scope of the testing to just FP8 data types.

* Fix gfx1101 build

* Document test availability

* Re-enable fp8 gemms for gfx94/95

* Cherry-pick GEMM Universal tests for FP8 data types

* Cleanup

* CK_USE_GFX94 has already been set on this branch

* Address formatting issues and leftovers

* Make fail/pass logic consistent within 01_gemm folder

Removed multiple negations in fail/pass logic to propagate `true` as the success indicator.

* Fix GPU verification reporting logic.

* Update year in copyright notice.

* Cleanup

* Use `enum class` instead of `enum`

* Remove set_property for FP8 tests

* Narrowing the scope of PR to OCP FP8 enablement only

* Add tests for OCP FP8 vector_type storage

* Enable gemm kernel on all gfx9 architectures (#227)

* clean-up

* Implement `non_native_vector_base` with `ext_vector_type` array. (#232)

* Enable support of 1, 2, 4, and 8-byte custom types in CK.

* Fix pool tests for OCP FP8 data type

* fix jenkins file

* restore cron trigger

---------
Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: aska-0096 <haocwang@amd.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Co-authored-by: Jing Zhang <jizhan@amd.com>
Co-authored-by: zjing14 <zhangjing14@gmail.com>
Co-authored-by: Jun Liu <Liu.Jun@amd.com>
Co-authored-by: Andriy Roshchenko <andriy.roshchenko@amd.com>
Co-authored-by: Andriy Roshchenko <107577548+andriy-ca@users.noreply.github.com>

* [CK Tile] Batched GEMM Example

* [CK Tile] Batched GEMM Example - minor refactor

* [CK Tile] Batched GEMM Example - README update

* [CK Tile] Batched Gemm Example - review changes

- Added tensor data layours as input parameters
- Changed structure of Host and Kernel args
- Removed bug with invalid vector read on non-contiguous memory

* [CK Tile] Batched Gemm Example - remove comment

* [CK Tile] Batched Gemm Example - Add GTests part1

* [CK Tile] Batched Gemm Example - GTests part2 + review changes

* [CK TILE] Batched GEMM post merge fixes

* [CK Tile] Batched GEMM Example - fix pad views

* [CK TILE] Add gemm compute pipeline v3

* Enable universal gemm compute pipeline.

* Rename example and add compute pipeline.

* Introduce ag bg cr pipeline impl base.

* Refactor to reuse code.

* Cleaning

* Formatting.

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

* add interwave scheduler for gemm mem pipeline

* Fix merge artifacts.

* Refactor unit tests.

* Switch to interwave scheduler for mem example

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

* Few small fixes.

* New GroupedGemm instances (BF16)

* Unify and refactor GroupedGEMM device API.

* Adapt changes to new API.

* Adapt grouped gemm profiler.

* Accept multiple kbatches for grouped gemm profiler.

- delete obsolete two stage as it is now covered by grouped gemm

* Update unit test for grouped gemm.

* Fix thresholds for BF16 and F8. Unblock tests.

* Fix few instances.

* Multiple small fixes.

* Adapt to new API, check dynamic casting.

* Uncomment few data types in grouped gemm profiler.

* Fix call to SetDeviceArgs.

* Fix profile grouped gemm multiply tile loop.

* Fix grouped gemm tile loop kernel args in client examples.

* Review comments.

* Fix cmake example build

* Support max3 in smoothquant one pass

* support max3 in two pass

* support max3 in add_rmsnorm_rdquant

* Allow getting batch size from splitkv tile partitioner

* Fix wrong paged-kvcache impl for group mode

* Fix wrong example code for page-kvcache

* Undo changes in fmha_fwd.cpp

* Always use 2D block table

* Add is_gappy kernel argument for paged-kvcache

The is_gappy argument is used for differentiating seqstart_k_ptr usage
in flash-attention & xformers

* Remove out-of-date comments

* Remove no-longer used method

* Fix wrong # page-block calculation

* Fix wrong comment

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

* Block universal gemm.

* Universal block gemm with interwave scheduler - draft.

* Refactoring

* Move a/b_warp_tiles into BlockGemmImpl
* set BlockGemmImpl as a class member

* Change tile size for more suitable to memory bound cases.

* Introduce kKPerThread to WarpGemm

* Add documentation comment.

* Fix Interwave scheduler block gemm.

* Add compute/memory friendly tile configuration.

* Clean

* New tile configurations in gemm mem example.

* Add more static checks and fix loop order in block gemm.

* Add more static checks and use warp gemm mfma dispatcher.

* Add default scheduler block gemm.

* Remove logging in example.

* moe pipeline

* update code

* compile OK

* update

* update cpu reference

* update pipeline_gemm0

* compiler ok

* update pipeline

* rename to ex pipeline

* block-asm

* update

* update

* update first gemm ok

* compute correct

* update file structure

* update README

* update

* update

* update code

* update API

* return unsupport case

* add comment

* update readme

* update

* uncomment

* update

* fix build err

---------
Co-authored-by: valarLip <340077269@qq.com>

* Fix mis-matched tuple<> elem types

* Rename MakeKargs() as MakeKargsImpl()

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

* update MOCK_ID for moe-sorting

* add moe-smoothquant

* update a comment

* fix format

* hot fix

* update topk in overflow case

* update comments

* update bf16 cvt

---------
Co-authored-by: valarLip <340077269@qq.com>

* Change in fwd-splitkv kernel to support num_splits=1 case

* Update in codegen fwd-splitkv to make num_splits > 1 cases pass

* Specify instance traits in dispatch

* Fix link error for fp8 kernels

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

* universal streamk fp8 changes & ckprofiler instances

* revert strides to -1 and verification options

* fp8 exclusion on pre-gfx94 for universal_streamk

* PR review based revisions: permissions reverted,  removed hip err checks


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

* Generate group mode paged-attn kernel

* Enable paged-kvcache + group mode support

* Add missing header: fused_moe.hpp

* Add comment to explain kernel arg usage

* Make error message more clear

* Add comment for confusing data member names

* Add more comment for confusing variable names

* Fix typo in option description

* add bf16 gemms for gfx11/gfx12

* reduce the input values in test_gemm

* add int8 wmma gemm instances for gfx11/gfx12

* add example gemm_wmma_int8

* fix bug in gemm_wmma_int8 test

* increase bf16 gemm test tolerance

* update the dates and clean-up commented-out instances

* Batched GEMM Multiple D based on Universal GEMM
Co-authored-by: Jing Zhang <jizhan@fb.com>

* CI fixes
Co-authored-by: Jing Zhang <jizhan@fb.com>

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

* Improve test verbosity.

* BUGFIX: Add missing initialization for reduction buffer

* Change default initialization method

Performance may be affected for fp32 and int8 examples.

* Improve test verbosity

* Cleanup

* Finished the feature

* Modified the test file

* Test case update

* addresss comment

* Addressed the review comment

* Fixed the CI error

* [CK_TILE] add more stride for layernorm to support un-continuous Tensor

* align CK coding style

* extend strides to layernrom expample

* clang-format...

* add moe_sorting & check ok

* fix comments & typo

* Run remod.py under include/ck_tile & example/ck_tile directories

* format codes

* fix output ci check bug

* fix moe sorting readme and error commit file

* use magiv div to accelerate compute

* add an loop unroll for moe lds ops

* add extblocksnel to set zeros for moebufs

* [Ck_tile] moe set zero run ok, add size check and fix ref check

* [Ck_tile]fix moe_sorting fuse set_zero remod

* [Ck_tile] change name style, fix zero buffer size err, change folder

* [Ck_tile] moe_sorting: fix name style

* [Ck_tile] moe_sorting, remove useless params in traits

* [Ck_tile] change outputtile cnt * unit_size; change output buf alloc

---------
Co-authored-by: dummycoderfe <noplydummmycoder@163.com>
Co-authored-by: Po Yen, Chen <PoYen.Chen@amd.com>
Co-authored-by: carlushuang <carlus.huang@amd.com>

* optimze small N case using vec io and using rcp div

* [Ck_tile] layernorm, add param to control fastdiv; change generate codes and test pass

* [Ck_tile] fix blockSize compute in Generic2dBlockShape

* [Ck_tile]fix kfastfdiv template style

* [Ck_tile] layernorm, fix stype in review

---------
Co-authored-by: dummycoderfe <noplydummmycoder@163.com>

* make sure cmake can handle xnack targets

* dont build xdl instances for gfx906:xnack-

* dont build xdl tests for gfx906:xnack-

Before, generate.py appended the list at the end of the output file.
When running the cmake configuration steps multiple times on the
examples, the blob list (such as fwd_blob_list.txt) would grow at every
configuration.
`library/src/tensor_operation_instance/gpu/mha/CMakeLists.txt` worked around
this issue by removing the output file if it exists.

Now, generate.py overrides the content of the output file.
There is no need for the workaround in the CMakeLists.txt;
and the issue is solved for the example projects too.

* more accurate residual

* modify comment

* Fix literal case in README.md

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

* fix compile error

* fix typo of padding

* Add smoothquant op

* Add smoothquant instance library

* refine type

* add test script

* Re-generate smoothquant.hpp

* Always use 'current year' in copyright

* use Generic2dBlockShape instead

* Add vector = 8 instance back

* Find exe path automatically

* Simplify the api condition

* Remove debugging code

* update year

* Add blank line between function declaration

* explicitly cast return value to dim3

* refine return value

* Fix default warmup and repeat value

* Add comment

* refactor sommthquant cmake

* Add README

* Fix typo

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

* hot fix ln

* some rename

* add prenorm/postnorm support, refactor using generate.py

* update README

* update README

* fix format

* update some description and fix format

* update format

* format

* use non-raw for loading

* format and update n4096

* dynamic-quant ready

* update readme

* support fused dynamic-quant

* update fused-quant, with smooth

* update README

* update args

* update some based on comment

* CK-Tile GEMM with memory bound pipeline.

* Memory bound gemm pipeline.

* Fix not closed namespace.

* Block gemm mem pipeline draft.

* Do not use ck_tile:: within ck_tile namespace.

* Refactoring & Move Layout info to pipeline problem.

* Get hot loop and TailNum information before lunching kernel.

* Fixes in pipeline.

* Add comment to load_tile_raw and change variable naming style.

* Few small changes & formatting.

* Do not use macro.

* Add gtests.

* Use AccDataType for Output of MFMA instruction.

* Formatting.

* Refactor gemm examples.

* Switch over to current block gemm.

* Use currently available pipeline policy.

* Refactoring and review comment.s

* Fixes after merge.

* Add missing include.

* Add load tile overload which accepts output tensor as parameter.

* This give 8% perf boost at the cost of using more registers.

* Rename example.

* Small changes.

* Fix compilation err and lower K.

* Support different layouts for A/B

* Fix vector size for different layouts.

* Rename Alignment into VectorSize

* Unblock tests.

* Add reduce2d new api

* Prevent user use cross warp reduction

* Fix bug of std caculation

* Add rmsnorm2d

* Add rmsnorm small example

* Remove static assert to prevent compile fail

* Add script to test performance and correctness

* Add missing cmake change

* refine naming

* refine example of rmsnorm

* Fix bug of rmsnorm

* Refine naming

* Fix cmake

* clang format

* Refine pipeline name

* Add add_rmsnorm2d_rdquant kernel

* Add reduce op

* host verification

* Fix bug of one pass pipeline

* Refine tile size

* Add two pass pipeline

* Rename two pass to three pass

* Fix bug of kSaveX == false

* Add instance library

* Add test script

* Fix bug of x verification

* Add save_x to trait

* Add README

* Move reduce2d into reduce folder

* Fix bug of welford when number of m warp > 1

* remove reduncant comment

* 1. move 06_rmsnorm2d to 10_rmsnorm2d
2. move 07_add_rmsnorm2d_rdquant to 11_add_rmsnorm2d_rdquant

* clang format and add missing header

* Add host validation of add + layernorm2d + rsquant

* Revert "Add host validation of add + layernorm2d + rsquant"

This reverts commit 936cb457978b928b90eff89a08fcdb7dc8bbed67.

* Remove deprecated flag

* Add ceil_to_qualified_tile_length()

* Rename kK0BlockLength to kQKHeaddim

* Add kSubQKHeaddim concept to support headdim96

* Fix in math.hpp to avoid using __half interfaces

* Add LdsBufferSequence instance for headdim96

* Update in fmha_fwd/fmha_fwd_splitkv codegen to support hd96 testing

* Disable hd96 instance generation in codegen fmha_fwd and fmha_fwd_splitkv to save compiling time

* Reformat one file

* Fix text alignment in fmha_fwd_splitkv.py

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

* topk_softmax

* remove some file

* fix atomix linear_offset

* address various comment, and change sfc get_index api to static(tuple)

* Add dynamic elementwise op
Co-authored-by: ThruptiRajLakshmanaGowda <thruptiraj.lakshmanagowda@amd.com>

* CI issues fix

* Custom parameter value for dynamic functions - Comments addressed

---------
Co-authored-by: ThruptiRajLakshmanaGowda <thruptiraj.lakshmanagowda@amd.com>
Co-authored-by: ThruptiRajLakshmanaGowda <tlakshma@amd.com>

* Use pre-defined constants for readability

* Use vector write for o_acc tensor

* Remove no-longer used policy method

* Deprecate no-longer used policy/pipeline

* Specify gemm0/gemm1 block warps separately in codegen

* Fix wrong ps_idx creation logic

* Add single-warp block gemm

* Supoprt single-warp gemm0

* Make MakeCBlockTile() as static method

* Use MakeCBlockTile() to get underlying tile distribution

* Use kNumGemm1Warps to compute # threads for gemm1

* Put normal case in the if clause

* Refine fmha splitkv block mapping

* Refine & fix the lse_acc/o_acc layout

* Fix wrong LDS size for K tile

* Use kK0=64 for hdim=128,256 fmha splitkv kernels

* Use kK1=64 for hdim=32,64,128 fmha splitkv kernels

* Undo kK0/kK1 changes

* Use more reasonable GetAlignmentV() computation

* Using store_tile() in fmha splitkv kernel epilogue

* add int8 gemm multiply multiply a8w8

* uncomment

* clang-format-12

* Add example_gemm_multiply_multiply_xdl_int8

* Remove shell scripts

* update preprocess number for mi308; bring back printout in ckprofiler

* format

---------
Co-authored-by: chenjun <junchen2@amd.com>
Co-authored-by: Haocong WANG <haocwang@amd.com>
Co-authored-by: carlushuang <carlus.huang@amd.com>

* Update inits

* Update static_cast to type_convert

* Add verification option selection

* port layernorm

* change warp_welford.hpp

* Update warpshuffle

* 1. Add save mean and save std back
2. Move construction of tensor_view and tile_window to operator()

* refine welford max count calculation

* unify layernorm api

* Rename file

* Remove save mean and inv std

* Revert "refine welford max count calculation"

This reverts commit 02236580

.

* Fix order of parameter

* refine welford max count calculation again

* Remove fp32 instances

* Fix bug of padding

* refactor api

* Support bf16

* Extract common function

* Refine arg of operator()

* Add kMThreadPerBlock to template parameter

* clang format

* Refine variable name

* Refine file name

* remove redundant line

* refactor layernorm2d pipeline and add block-per-block utility

* fix name

* rename more

* add more block-per-tile instance

* remove duplicated define

* update instance for 2048, 1024 case

* support up to 2048 now

* opt loading

* add n1536

* Add two pass pipeline

* format

* Fix incorrect type

* parallel compilation

* Use smaller N

* fix 2p pass

* Support Repeat_M in distribution

* Refine nameing

* Add reduce example

---------
Co-authored-by: letaoqin <letaoqin@amd.com>
Co-authored-by: aska-0096 <haocwang@amd.com>
Co-authored-by: rocking <ChunYu.Lai@amd.com>
Co-authored-by: carlushuang <carlus.huang@amd.com>