ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-06-07 00:04:37 +00:00
Code Issues Packages Projects Releases Wiki Activity
1,419 Commits 766 Branches 40 Tags
ca4b208b601094bba588d2a1be6e53ee3cfcf5df
Commit Graph

54 Commits

Author SHA1 Message Date
PoYen, Chen
ca4b208b60 Fix wrong grid size 2024-07-23 14:20:52 +00:00
PoYen, Chen
52b47810bb Rename more tile size constants 2024-07-23 09:30:05 +00:00
PoYen, Chen
99c1d463de Align naming of some tile size constants 2024-07-23 09:24:38 +00:00
PoYen, Chen
ce5e0f1d67 Re-order parameters 2024-07-23 09:02:41 +00:00
PoYen, Chen
fb80c7b2cb Extract rotary embedding logic out 2024-07-23 08:51:59 +00:00
PoYen, Chen
2192bbc68a Rename RotaryEmbeddingEnum 2024-07-23 07:50:50 +00:00
PoYen, Chen
d4606cf3c3 Rename header 2024-07-23 07:45:25 +00:00
PoYen, Chen
b275732128 Remove always true static_assert() 2024-07-23 07:25:50 +00:00
PoYen, Chen
eb649a2f25 Move thread locating logics into policy 2024-07-23 07:21:20 +00:00
PoYen, Chen
0e5cb6f913 Skip code if # of block is more than needed 2024-07-23 06:53:24 +00:00
PoYen, Chen
0925c0e941 Use better naming for tile indices 2024-07-23 06:40:53 +00:00
PoYen, Chen
bc7c7ee0c5 Fix wrong knew/vnew appending positions 2024-07-23 04:46:53 +00:00
PoYen, Chen
56df4d6397 Remove debug print code in kernel 2024-07-23 04:01:55 +00:00
PoYen, Chen
48c70720b5 Apply RoPE to q_tile 2024-07-23 03:54:11 +00:00
PoYen, Chen
e88253a2f4 Add code blocks for q_tile 2024-07-23 03:28:40 +00:00
PoYen, Chen
1dbed18555 Remove constness from q_ptr 2024-07-23 03:11:31 +00:00
PoYen, Chen
c26c60db4c Unify parameter/variable naming style 2024-07-23 02:59:17 +00:00
PoYen, Chen
c0bc097758 Apply elementwise function to the loaded tiles 2024-07-23 02:50:07 +00:00
PoYen, Chen
df352f955a Add comment 2024-07-23 02:31:45 +00:00
PoYen, Chen
d1ecfdc700 Support 8x rotary_dim under half-rotated RoPE 2024-07-23 02:19:16 +00:00
PoYen, Chen
631f29d527 Handle RoPE half-rotated logics 2024-07-22 08:50:03 +00:00
PoYen, Chen
01865d2ae4 Clean-up pipeline 2024-07-22 03:14:10 +00:00
PoYen, Chen
fffd6799e6 Instantiate multiple kernels for RoPE approaches 2024-07-20 02:28:21 +00:00
PoYen, Chen
27b5141706 Fix wrong thread starting offset 2024-07-18 20:02:06 +00:00
PoYen, Chen
23450526c0 Only apply interleaved RoPE on Knew for now 2024-07-18 19:42:14 +00:00
PoYen, Chen
85bfed07fa Add dram distribution for rotary_cos/rotary_sin (interleaved) 2024-07-18 09:11:22 +00:00
PoYen, Chen
99f863e4cd Fix rotary cos/sin tensor/tile size 2024-07-16 06:31:17 +00:00
PoYen, Chen
b32fd8d3f4 Rename variables used in distributio encoding 2024-07-16 06:27:28 +00:00
PoYen, Chen
879710a495 Fix wrong seqlen_k for kvcache 2024-07-16 03:42:51 +00:00
PoYen, Chen
b0925bb7f6 Create Rotary Cos/Sin tile windows in kernel 2024-07-14 23:47:40 +00:00
PoYen, Chen
391210ed9e Pass RoPE kernel args 2024-07-14 23:18:32 +00:00
PoYen, Chen
b34ddf5f71 Merge remote-tracking branch 'origin/feature/cond-add-splitkv' into feature/fmha-fwd-appendkv 2024-07-12 04:42:45 +00:00
PoYen, Chen
8c733fb3be Fix compilation errors 2024-07-10 10:53:58 +00:00
carlushuang
8182976c37 [CK_TILE] wa prec, remove sgpr offset for inline asm (#1356) 
* wa prec, remove sgpr offset for inline asm

* macro for set tile

* ignore unused param if no kernel instances in host API

* fix more prec issue

* cache buffer resource

* fix

* support pre-nop

* clear tile by vector type members

* add workaround to reduce scratch memory

* conditionally enable workaround code

* enable workaround start from certain build version

* fallback set_tile() implementation from certain build version

* undo template argument changes

* put dummy asm in load_raw()

* fix comments, refactor s_nop inside buffer_load

---------

Co-authored-by: PoYen, Chen <PoYen.Chen@amd.com>
 2024-07-08 11:09:55 -07:00
PoYen, Chen
1c070380fa Merge branch 'feature/cond-add-splitkv' into feature/fmha-fwd-appendkv 2024-07-08 07:13:34 +00:00
PoYen, Chen
34a3ff849f Fix Vnew tile dstr for row major case 2024-06-27 09:55:35 +00:00
PoYen, Chen
c40c1daff0 Extract common logics 2024-06-26 18:02:28 +00:00
PoYen, Chen
8fb567c286 Fix vnew append errro 2024-06-26 17:00:07 +00:00
Po Yen Chen
0cb2e06ddc [CK_TILE] fmha forward split-kv + combine kernels (#1338) 
* FA fwd dropout

* FA bwd

* epilogue reuse

* CMakeLists update

* [CK_TILE] support alibi (#1269)

* add alibi support

* fix code

* update code based on comment

* Support more hdim

* fix fp8 bias

* support seqlen_k=0 case

* remove unused printf

* fix format

---------

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

* now fwd/bwd can build

* bwd alibi

* add bwd validation stream_config

* update generated filenames

* update bwd kernel launch

* CK_TILE_HOST_DEVICE in philox

* Transpose -> transpose

* format

* format

* format

* Generate the instance for FA required

* format

* fix error in WarpGemm

* Add num_splits option and dummy split-kv api method

* Generate fmha_fwd_splitkv()

* Add SplitKV kernel codegen logics

* Add SplitKV combine kernel codegen logics

* Fix mismatched return type

* Clean-up code

* Replace sentinel value before storing

* Fix wrong layout of LSE/LSEacc/Oacc

* Format codes

* Fix o_acc memory error

* Fix wrong kBlockSize used in policy

* Reduce # of combine kernels

* Fix split-kv combine kernel name

* Fix wrong LDS indexing logics

* Fix wrong loop counter step logic

* Undo vector size changes

* Remove no-longer used field

* Remove in-consistent comment

* Remove debug statements in example

* Remove more debug statements

* Add constness to local variables

* Clearn up generate.py

* Fix unstable clang-format comment

* Remove unused include directive

* Use shorter template parameter name

* Enable non-split-kv blobs

* Update license date

* Print num_splits conditionally

* Undo disabling data types

* Remove unnessary tile size for fp8

* Fix wrong pipeline args for fp8

* Fix example output format

* Remove more debug code in combine pipeline

* Add stride kernel arguments for LSE/O acc workspace

* Re-order split-kv pipeline call operator arguments

* Pass LSE/O strides in kernel argument

* Re-order pipeline call operator arguments

* Use tensor_descriptor to locate LSEacc elements

* Support providing invalid element for tensor view

* Set invalid element value for LSEacc tensor view

* Remove hand-written store_tile() code

* Remove necessary value-overwrite logic

* Add transposed lds descriptor

* Support load_tile() for tile_window_with_static_lengths<>

* Undo removing necessary value-overwrite logic

* Use read descriptor to locate lds elements

* Simplify pipeline source code

* Add constraint to kMaxSplits

* Default use kMaxSplits=64 in generate.py

* Revert "Add constraint to kMaxSplits"

This reverts commit 0a2132d758.

* Revert "Default use kMaxSplits=64 in generate.py"

This reverts commit c7d9c80b77.

* Decide alignment by the padding parameter

* Remove no-longer used utility functions

* Remove not-working code

* Add comment & remove no-longer used code

* Fix computation errors

* Add heuristic to override num_splits option

* Add constraint to kMaxSplits

* Fix compilation error

* Clean up pipeline code

* Wrap pointer access as lambda function

* Rename confusing methods

* Use kLogMasSplits as template parameter

* Finish splitkv combine kernel codegen

* Update kMaxSplits limit

* Use smaller kM0 for splitkv combine kernel

* Ignore droupout flag in splitkv pipeline

* Unify flag usage

* Add back flag kStoreLSE

* Merge lambda calls in pipeline

* Fix compilation errors

* Avoid all empty splits

* Always check for empty loop in splitkv pipelines

* Re-order parameters

* Remove redundant p_drop option check

* Add traits/problem for fwd splitkv kernel

* Conditionally enable uneven split boundary checks

* Add comment for the splitkv traits field

* Change even split criteria

* Re-order statements

* Refine occupancy value for hdim=128&256

* Refine occupancy value for hdim=32&64

* Remove redundant kernel argument

* Separate fmha bwd codegen logics

* Separate fmha fwd codegen logics

* Remove redundant direction parameter in fwd&bwd codegen logics

* Support generate multiple APIs for an example

* Let 'api' an alias of 'direction' option

* Remove choices for the 'direction' option

* Use dictionary to config all the functions

* Move fmha splitkv codegen logics to other file

* Add fwd_splitkv api for tile_example_fmha_fwd

---------

Co-authored-by: danyao12 <danyao12>
Co-authored-by: carlushuang <carlus.huang@amd.com>
Co-authored-by: rocking <ChunYu.Lai@amd.com>
Co-authored-by: Jing Zhang <jizhan@amd.com>
 2024-06-26 17:41:15 +08:00
PoYen, Chen
4e6c28522c Fix wrong K values after appending 2024-06-25 10:12:13 +00:00
PoYen, Chen
1ac17dae50 Add knew/vnew tensors to the kernel argument 2024-06-25 07:56:36 +00:00
PoYen, Chen
344902732a Sync kernel name with the codegen 2024-06-24 14:50:25 +00:00
PoYen, Chen
bace0e5df0 Add init codegen logic for fmha fwd appendkv 2024-06-24 12:33:51 +00:00
rocking
cb13839425 layernorm2d forward (#1339) 
* Add layernorm2d forward

* Refind file path

* clang format

* Exclude ck_tile op from all

* use add_executable instead

* refactor layernorm2d_fwd example

---------

Co-authored-by: carlushuang <carlus.huang@amd.com>
 2024-06-24 08:45:52 +08:00
Dan Yao
1da802bdf2 Fix FA bwd alibi+causal NaN errors (#1352) 
* fix bwd alibi nan error

* fix datatype

---------

Co-authored-by: danyao12 <danyao12>
 2024-06-20 09:50:53 -05:00
Qianfeng
e3f44659cf Fix in dropout lambda to avoid the compiling issue on some docker/compiler envs (#1350) 2024-06-20 11:36:42 +08:00
Qianfeng
1973903f49 Hacking ck_tile fmha Dropout facility (#1344) 
* Add NullBlockDropout to be used when kHasDropout is false

* Change to BlockDropout::Run() for forward to reduce conditional checkings

* Re-format files

---------

Co-authored-by: PoYen, Chen <PoYen.Chen@amd.com>
 2024-06-19 10:37:22 +08:00
carlushuang
17ed368f58 [CK_TILE][FA] using pk f16_f32 (#1343) 
* [CK_TILE][FA] using pk f16_f32

* correct a error
 2024-06-17 17:16:46 +08:00
Dan Yao
2cab8d39e3 CK Tile FA Training kernels (#1286) 
* FA fwd dropout

* FA bwd

* epilogue reuse

* CMakeLists update

* [CK_TILE] support alibi (#1269)

* add alibi support

* fix code

* update code based on comment

* Support more hdim

* fix fp8 bias

* support seqlen_k=0 case

* remove unused printf

* fix format

---------

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

* now fwd/bwd can build

* bwd alibi

* add bwd validation stream_config

* update generated filenames

* update bwd kernel launch

* CK_TILE_HOST_DEVICE in philox

* Transpose -> transpose

* format

* format

* format

* Generate the instance for FA required

* format

* fix error in WarpGemm

---------

Co-authored-by: danyao12 <danyao12>
Co-authored-by: carlushuang <carlus.huang@amd.com>
Co-authored-by: rocking <ChunYu.Lai@amd.com>
Co-authored-by: Po Yen Chen <PoYen.Chen@amd.com>
Co-authored-by: Jing Zhang <jizhan@amd.com>
 2024-06-04 13:12:45 -05:00
carlushuang
5055b3bdcb [CK_TILE] support group from cmdline (#1295) 
* support cmdline seqlen decode

* silent print

* update readme

* update kernel launch 3d

* update tile partitioner

* fix spill for bf16

* modify based on comment

* modify payload_t

* fix bug for alibi mode

* fix alibi test err

* refactor kernel launch, support select timer

* add missing file

* remove useless code

* add some comments
 2024-05-28 11:13:21 +08:00