ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-06-07 08:15:04 +00:00
Code Issues Packages Projects Releases Wiki Activity
1,576 Commits 766 Branches 40 Tags
d52278a5efbaff8f412f33e591976ac4061cb049
Commit Graph

60 Commits

Author SHA1 Message Date
PoYen, Chen
43b8100b7f Support cache_batch_idx in example 2024-08-16 16:27:56 +00:00
PoYen, Chen
9c904b0e4c Pass cache_batch_idx to kernels 2024-08-16 15:32:24 +00:00
PoYen, Chen
2523c8e36c Fix more format 2024-08-16 10:32:17 +00:00
PoYen, Chen
5805f5aa73 Remove group mode from appendkv kernel 2024-08-16 10:04:48 +00:00
PoYen, Chen
9de0f35ebc Remove unused template paremeter 2024-08-13 09:29:20 +00:00
PoYen, Chen
370babc996 Make tile window directly via PageBlockNavigator 2024-08-13 09:18:24 +00:00
PoYen, Chen
3dd6ef61ef Re-order pipeline paremeters 2024-08-13 07:38:41 +00:00
PoYen, Chen
c54de6416a Rename TileWindowNavigator to PageBlockNavigator 2024-08-13 07:23:40 +00:00
PoYen, Chen
a0d2163045 Remove dropout code in splitkv kernel 2024-08-08 10:21:34 +00:00
PoYen, Chen
cef9da0a76 Remove debug macro usages 2024-08-07 15:26:43 +00:00
PoYen, Chen
f265742b63 Handle cross-page-block write 2024-08-07 09:33:41 +00:00
PoYen, Chen
26ed468ac6 Pass re-created tile window to pipeline 2024-08-07 06:00:17 +00:00
PoYen, Chen
78209c7326 Fix wrong tensor descriptor lengths 2024-08-07 05:59:26 +00:00
PoYen, Chen
7789b53e15 Add tile navigators to the appendkv kernel 2024-08-07 04:51:21 +00:00
PoYen, Chen
443a528adc Add block_table kernel args for appendkv kernel 2024-08-07 04:27:15 +00:00
PoYen, Chen
15d0034a64 Add paged-kv codegen logic for appendkv kernels 2024-08-07 04:19:45 +00:00
PoYen, Chen
bd0d2f3975 Add batch_stride_k/batch_stride_v in group mode 2024-08-06 08:02:43 +00:00
PoYen, Chen
4fed268723 Move code after decide seqlen_q/seqlen_k 2024-08-06 01:39:49 +00:00
PoYen, Chen
77dac7775c Move V tile through TileWindowNavigator 2024-08-05 22:36:52 +00:00
PoYen, Chen
bb78353264 Remove ununnecessary data members 2024-08-05 21:52:59 +00:00
PoYen, Chen
8fea4139df Fix tile window navigation bugs 2024-08-05 21:34:15 +00:00
PoYen, Chen
ecaaa6f136 Simplify TileWindowNavigator interfaces 2024-08-05 16:31:31 +00:00
PoYen, Chen
1c9d77b606 Introduce 'TileWindowNavigator' types 2024-08-05 15:58:41 +00:00
PoYen, Chen
baf4a612f0 Fix wrong kernel name 2024-08-02 10:26:47 +00:00
PoYen, Chen
e7969b9fd2 Add template argument 'kIsPagedKV' for splitkv kernels 2024-08-02 10:14:51 +00:00
PoYen, Chen
5c733dc568 Remove debug statements 2024-07-24 06:47:52 +00:00
PoYen, Chen
59e1d9b84f Shift rotary_cos/rotary_sin by cache_seqlen_k 2024-07-24 05:06:47 +00:00
PoYen, Chen
3348131699 Fix wrong data type for Q rotary_cos/rotary_sin 2024-07-24 04:10:43 +00:00
PoYen, Chen
eb4ea3ac2a Fix wrong rotary_cos/rotary_sin memory size for Q 2024-07-23 16:22:25 +00:00
PoYen, Chen
b11f92dc4c Fix wrong shape of knew_host/vnew_host 2024-07-23 14:52:42 +00:00
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
2192bbc68a Rename RotaryEmbeddingEnum 2024-07-23 07:50:50 +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
1dbed18555 Remove constness from q_ptr 2024-07-23 03:11:31 +00:00
PoYen, Chen
fffd6799e6 Instantiate multiple kernels for RoPE approaches 2024-07-20 02:28:21 +00:00
PoYen, Chen
99f863e4cd Fix rotary cos/sin tensor/tile size 2024-07-16 06:31:17 +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
8c733fb3be Fix compilation errors 2024-07-10 10:53:58 +00:00
PoYen, Chen
1c070380fa Merge branch 'feature/cond-add-splitkv' into feature/fmha-fwd-appendkv 2024-07-08 07:13:34 +00:00
PoYen, Chen
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