Skip to content

GitLab

Switch branch/tag
  1. 28 Aug, 2024 1 commit
    • Po Yen Chen's avatar
      [CK_TILE] Add PagedAttention kernels (#1387) · c1569892
      Po Yen Chen authored 
      

* Use dictionary to config all the functions

* Add init codegen logic for fmha fwd appendkv

* Call HIP_CHECK_ERROR() macro to get real source info

* Setup meaningfull arguments

* Sync kernel name with the codegen

* Add knew/vnew tensors to the kernel argument

* Fix wrong K values after appending

* Fix vnew append errro

* Extract common logics

* Fix Vnew tile dstr for row major case

* Conditionally add fwd_splitkv API in fmha_fwd example

* Conditionally add call to fmha_fwd_splitkv()

* Remove "EXAMPLE_" prefix of cmake variables

* Regsiter API handlers automatically

* Early return if 0 < s_k_new is not supported

* Show message if we are ignoring option

* Unify CMakeLists.txt coding style

* Set num_splits=1 if split-kv is not supported

* Add length/stride getters for HostTensor

* Add RoPE example utilities

* Add reference_rotary_position_embedding() (not implemented)

* Finish reference_rotary_position_embedding() impl

* Fix typo of HostTensor<>::get_length()

* Fix compilation errors

* Fix wrong answer when interleaved=false

* Fix wrong answer when interleaved=true

* Append K/V in the host verification code

* Simplify K appending logics

* Simplify v_host_ref definition

* Reduce input/output dimensions

* Rename function: add "batched" prefix

* Apply RoPE on host side

* Rename RoPE utility function

* Fix wrong tensor size

* Avoid invoking deprecated method 'find_module'

* Pass RoPE kernel args

* Create Rotary Cos/Sin tile windows in kernel

* Add compute data type alias for RoPE

* Randomly generate seqlen_knew if needed

* Fix seqlen_knew enabling check logic

* Add minimum seqlen_k to generate compliance kvcache

* Fix compilation error in debug mode

* Fix wrong boundaries

* Fix wrong seqlen_k for kvcache

* Rename variables used in distributio encoding

* Fix rotary cos/sin tensor/tile size

* Add constraint to the rotary_dim option

* Remove unused inner namespace

* Add dram distribution for rotary_cos/rotary_sin (interleaved)

* Only apply interleaved RoPE on Knew for now

* Fix wrong thread starting offset

* Instantiate multiple kernels for RoPE approaches

* Clean-up pipeline

* Fix error in RoPE host reference

* Handle RoPE half-rotated logics

* Support 8x rotary_dim under half-rotated RoPE

* Add comment

* Apply elementwise function to the loaded tiles

* Unify parameter/variable naming style

* Remove constness from q_ptr

* Add code blocks for q_tile

* Apply RoPE to q_tile

* Remove debug print code in kernel

* Fix wrong knew/vnew appending positions

* Use better naming for tile indices

* Add make_tile_window() for adding distribution only

* Skip code if # of block is more than needed

* Move thread locating logics into policy

* Remove always true static_assert()

* Rename header

* Rename RotaryEmbeddingEnum

* Extract rotary embedding logic out

* Re-order parameters

* Align naming of some tile size constants

* Rename more tile size constants

* Fix wrong grid size

* Fix wrong shape of knew_host/vnew_host

* Fix wrong index into knew_host/vnew_host

* Fix wrong rotary_cos/rotary_sin memory size for Q

* Extract Q/Knew vector size to helper methods

* Use different rotary_cos/rotary_sin distr for Q/Knew

* Update host/device specifiers

* Fix wrong data type for Q rotary_cos/rotary_sin

* Remove RoPEComputeDataType type alias

* Shift rotary_cos/rotary_sin by cache_seqlen_k

* Add comment for why I just 't' for all padding flags

* Align commit message to the real comment

* Fix wrong pipeline

* Rename utility function

* Disable host verification if API not exist

* Fix wrong rope key for fp8 pipeline

* Allow only apply RoPE on Q (without append KV)

* Add append-kv smoke tests

* Remove debug statements

* Remove more debug statements

* Re-arrange the 'set +x' command

* Remove no-longer used method in pipeline

* Add missing init code

* Refine pipeline padding settings

* Enlarge rotary_dim limit (8 -> 16)

* Enlarge KPerThread for rotary_interleaved=false

* Update rotary_dim range in smoke_test_fwd.sh

* Add template argument 'kIsPagedKV' for splitkv kernels

* Launch splitkv kernel if given page_block_size

* Fix wrong kernel name

* Fix seqlen_k_min for pre-fill case (1 -> 0)

* Add copy_const<> type trait

* Add another make_tile_window()

* Introduce 'TileWindowNavigator' types

* Simplify TileWindowNavigator interfaces

* Fix tile window navigation bugs

* Disable calling fmha_fwd()

* Remove ununnecessary data members

* Simplify more make_tile_window() overloads

* Move V tile through TileWindowNavigator

* Fix uneven split checking logic

* Move code after decide seqlen_q/seqlen_k

* Make sure we always start reading complete tile

* Use 128 as minimus page_block_size

* Fix wrong origin for bias

* Add batch_stride_k/batch_stride_v in group mode

* Unify origin

* Add missing kernel arguments for group mode

* Add paged-kv codegen logic for appendkv kernels

* Add block_table kernel args for appendkv kernel

* Add tile navigators to the appendkv kernel

* Fix wrong tensor descriptor lengths

* Pass re-created tile window to pipeline

* Fix wrong strides for appendkv kernel

* Allow transit tile_window to another page-block

* Handle cross-page-block write

* Donot perform write again if already in last page-block

* Always add fmha_fwd() api

* Add missing group mode argument

* Remove debug macro usages

* Rename option s_k_new to s_knew

* Separate splitkv/non-splitkv args/traits

* Remove fmha_fwd_dispatch()

* Fix compilation errors

* Remove dropout code in splitkv kernel

* Allow problem types without define kHasDropout attr

* Use generic lambda to init traits objects

* Separate more non-splitkv & splitkv traits/args

* Display more info for specific kernels

* Show more detailed warning message

* Rename 'max_num_blocks' to 'max_num_page_blocks'

* Remove no-longer used pipeline files

* Wrap code by #if directives

* Move functors to the begining of validation code

* Use generic lambda to init all the api traits/args

* Fix wrong seqlen for kvcache

* Add missing comment

* Rename TileWindowNavigator to PageBlockNavigator

* Only expose necessary methods (not attributes)

* Re-order pipeline paremeters

* Refine smoke_test_fwd.sh

* Fix wrong arugment count

* Make tile window directly via PageBlockNavigator

* Remove unused template paremeter

* Remove group mode from appendkv kernel

* Fix skcheck logic

* Fix wrong syntax in skcheck expr

* Use meaningful options in smoke test

* Remove options

* Fix formatting

* Fix more format

* Re-organize bash functions

* Pass cache_batch_idx to kernels

* Support cache_batch_idx in example

* Fix compilation error

* Add more appendkv test

* Add more case for appendkv

* Fix unexisted attribute

* Remove 0 < seqlen_knew constraint

* Clarify the case in warning message

* Remove macro checking

* Force batch mode when invoking appendkv & splitkv apis

* Fix mode overriding logics

* Fix wrong parameter name

* Randomize seqlen_k if use kvcache

* Use randomized seqlen_k for kvcache

* Avoid using too small rotary_cos & rotary_sin

* Rename parameter

* Add seqlen_q & seqlen_k rules

* Add comment

* Add more comments

* Fix compilation errors

* Fix typo in comment

* Remove type argument

* Avoid seqlen_k=0 for kvcache

* Revert "Avoid seqlen_k=0 for kvcache"

This reverts commit 21c4df89e416182e8e9bc78e67bd4b98dbb6c88d.

* Fix wrong uneven split checking logics

* Only randomize kvcache seqlen_k if 1 < batch

* Return earlier if split is empty

* Revert "Only randomize kvcache seqlen_k if 1 < batch"

This reverts commit b9a4ab0d7e3c2beecc0fccafd2a13259dd06299c.

* Re-order seqlen_k_start adjustment logics

* Fix compilation errors

* Re-format script

* Find executable from folder automatically

* Fix kvcache seqlen_k generating logic

* Make comment more clear

* Fix wrong knew/vew appending logic on host

* Add s_barrier to sync threads

* Revert "Add s_barrier to sync threads"

This reverts commit d3f550f30c0a4d9df15c613015d5dff268d6746d.

* Support only using 1 row of rotary_cos/rotary_sin

* Rotate Q in different way

* Unify tensor view creation logics

* Fix wrong argument

* Add mask to switch how we use the rotary_cos/sin

* Move attr from traits to problem

* Move has_mask to fmha_fwd_appendkv_args

* Support use uint32_t as SAD operand in Alibi<>

* Use sad_u32() in splitkv kernels

* Store tensor views in PageBlockNavigator

* Use stored tensor view to update tile windows

* Enlarge tensor view size

* Remove debug code

* Fix wrong tensor view size

* Wrap tensor view into PageBlockNavigator

* Add DataType member to PageBlockNavigator

* Remove unnecessary member functions

* Refind macro use

* Fix typo

* Add blank line between directives and actual code

* Re-format files

* Remove type in comment

---------
Co-authored-by: default avatarcarlushuang <carlus.huang@amd.com>
Co-authored-by: default avatarrocking <ChunYu.Lai@amd.com>
      c1569892
  3. 26 Aug, 2024 1 commit
    • Illia Silin's avatar
      Enable daily ninja build traces. (#1487) · 19d22e60
      Illia Silin authored 
      * add ninja trace to CI builds

* fix ninja trace logic

* update the ninja trace logic in jenkins file

* limit the number of threads to run ninja build

* use ninja for installation after build

* update the path to ninjatracing tool

* use ninja to run check when using build trace

* fix jenkins logic

* fix typos

* set proper setup_args for all stages

* fix ninja syntax

* replace ninja check with ninja test

* enable ninja tracing with mainline and staging compilers
      19d22e60
  5. 23 Aug, 2024 1 commit
  7. 22 Aug, 2024 5 commits
  9. 21 Aug, 2024 2 commits
    • Andriy Roshchenko's avatar
      Adding Instances and Examples for FP8-based Scaled Convolution and AMAX Reduction. (#1473) · c3515f27
      Andriy Roshchenko authored 
      * Enable CMakePresets build

* Verify Convolution, Scaling and ReLU algorithms.

* Add tensor element-wise scale and type cast operation.

* Reduction implemented but does not work.

* Exploration of Reduction functionality.

* Completed example for Convolution scaled with ReLu activation and AMAX reduction.

* WIP: Add required instances for convolution.

* WIP: Create client example. Implement convolution stage.

* Add elementwise instances.

* Add elementwise scale + convert example.

* Add reduction instances.

* WIP: Client example for AMAX reduction.

* WIP: Add instances for multistage reduction.

* WIP: Implementation of multistage reduction.

* Refactoring.

* Clean up.

* Add CMakePresets.json

* Guard off FP8 instances when the data type is not available.

* Add example for Scaled FP8 Convolution with AMAX reduction.

* Refactor CombConvScaleRelu instances.

* Add CombConvScale instances.

* Add client example for Scaled FP8 Convolution with AMAX reduction.

* Cleanup.
      c3515f27
    • Rostyslav Geyyer's avatar
      Set RNE fp8 conversion as a default (#1458) · e20f20ef
      Rostyslav Geyyer authored 
      * Set RNE fp8 conversion as a default

* Update f8 tests

* Disable failing test on gfx11

* Update bf8 tests

* Add a flag

* Fix the flag

* Raise flag for gfx10 as well

* Temp commit for tolerance testing

* Update tolerances
      e20f20ef
  11. 20 Aug, 2024 3 commits
  13. 19 Aug, 2024 1 commit
  15. 16 Aug, 2024 3 commits
    • Illia Silin's avatar
      Re-enable fp8 types for all architectures. (#1470) · c8b6b642
      Illia Silin authored 
      * re-enable fp8 and bf8 for all targets

* restore the fp8 gemm instances

* re-enable conv_3d fp8 on all architectures

* diasble several fp8 gemm instances on all architectures except gfx94

* clang format fix
      c8b6b642
    • Dan Yao's avatar
      [CK_TILE] FA bwd kernels optimization (#1397) · 79a5d9c1
      Dan Yao authored 
      

* tmp save

* fix batch deterministic bugs

* fix group deterministic bugs

* codegen update

* reorder files

* bias support

* hd256 bias support

* bwd smoke test update

* simplify convert dq

* fix hd256 dropout scratch

* do{}while() -> while(){}

* comments

* remove FmhaBwdTilePartitioner

* save clear_tile

* refactor dropout

* code cleanup

* code cleanup

* comments

* fix epilogue problem

* fix fwd dropout

* group convert_dq opt

* fix dq alignment

* Do not store storerandval in bwd for flash attention integration

* fix hd32 error and boost performance

* revert

* Remove duplicated WarpGemm definitions in the policy file

* dropout patch for mrepeat 16*16

* code sync up

* dq_acc stride

* dq_acc stride stuff

* codegen update

* fwd dropout revert

* fix hd128 scratches and boost performance

* receipt 3 for simplified smoke test

* more strides for fa integration

* fix hd64 scratches and boost performance

* non-iglp pipeline for headdim padding cases

* dpad same as dvpad for flash attention integration

* unpadded lse&d for group mode

* Support unpad layout for group lse

* Support unpad lse layout for splitkv

* Fix stride for splitkv kernel

* fix unpadded lse issue in fwd splitkv

* comment

* solve lds read&write conflicts

* rename

* bias rename

* tile index revert

---------

Co-authored-by: danyao12 <danyao12>
Co-authored-by: default avatarrocking <ChunYu.Lai@amd.com>
Co-authored-by: default avatarQianfeng Zhang <Qianfeng.Zhang@amd.com>
      79a5d9c1
    • Bartłomiej Kocot's avatar
      Add performance and large tensor tests for grouped conv (#1456) · 2581727d
      Bartłomiej Kocot authored 
      

* Add performance and large tensor tests for grouped conv

* Resize tests

* Resize tests

* update the python script to parse the grouped_conv results

* Remove int8 tests

* change bwd wei layout

---------
Co-authored-by: default avatarillsilin <Illia.Silin@amd.com>
      2581727d
  17. 15 Aug, 2024 2 commits
  19. 14 Aug, 2024 1 commit
    • Haocong WANG's avatar
      [GEMM] gemm_universal related optimization (#1453) · 3049b546
      Haocong WANG authored 
      

* replace buffer_atomic with global_atomic

* fixed global_atomic_add

* added bf16 atomic_add

* format

* clang-format-12

* clean

* clean

* add guards

* Update gtest.cmake

* enabled splitk_gemm_multi_d

* format

* add ckProfiler

* format

* fixed naming

* format

* clean

* clean

* add guards

* fix clang format

* format

* add kbatch printout

* clean

* Add rocm6.2 related gemm optimization

* Limit bf16 atomic usage

* remove redundant RCR gemm_universal instance

* Add RRR fp8 gemm universal instance

* Bug fix

* Add GPU_TARGET guard to FP8/BF8 target

* bug fix

* update cmake

* remove all fp8/bf8 example if arch not support

* Enable fp8 RRR support in ckProfiler

* limit greedy-reverse flag to gemm_universal in ckProfiler

---------
Co-authored-by: default avatarJing Zhang <jizhan@fb.com>
Co-authored-by: default avatarJing Zhang <jizhan@meta.com>
Co-authored-by: default avatarzjing14 <zhangjing14@gmail.com>
Co-authored-by: default avatarIllia Silin <98187287+illsilin@users.noreply.github.com>
Co-authored-by: default avatarillsilin <Illia.Silin@amd.com>
      3049b546
  21. 13 Aug, 2024 2 commits
    • AngryLoki's avatar
      Fix compilation errors with libc++ (#1461) · 50c42348
      AngryLoki authored 
      

This fixes 2 issues when compiled with libc++.

First issue is attempt to call std::numeric_limits<ranges::range_value_t<_Float16>>::min().
_Float16 is extension of libstdc++, it does not exist in C++ standard[2].
Luckily, there is NumericLimits class in composable_kernel, which does everything needed.

Second issue with call to 'check_err' is ambiguous: there are 2 candidates.
It happens because composable_kernel relies on idea that f8_t (defined as _BitInt(8)) does not pass is_integral trait.
However, libc++ treats _BitInt(N) as integral (per standard "any implementation-defined extended integer types" can be integral).

Closes: #1460
Signed-off-by: default avatarSv. Lockal <lockalsash@gmail.com>
      50c42348
    • Mateusz Ozga's avatar
      Support large: 12d tensor size for reduction kenrel (#1465) · 0606e549
      Mateusz Ozga authored
      0606e549
  23. 12 Aug, 2024 2 commits
  25. 10 Aug, 2024 1 commit
  27. 09 Aug, 2024 2 commits
  29. 08 Aug, 2024 3 commits
  31. 07 Aug, 2024 4 commits
  33. 06 Aug, 2024 6 commits