Commits · aea5a3d0b809614ac89b618bda96b6b0b3ed7acd · gaoqiong / composable_kernel

27 Dec, 2022 35 commits
- restore old file · aea5a3d0
  Anthony Chang authored Dec 27, 2022
  
  aea5a3d0
- rename · 0cf81fde
  Anthony Chang authored Dec 27, 2022
  
  0cf81fde
- remove unused variables · 394f9207
  Anthony Chang authored Dec 27, 2022
  
  394f9207
- fix compiler warnings · fc7e83ee
  Anthony Chang authored Dec 27, 2022
  
  fc7e83ee
- clean up · 5eec2aef
  Anthony Chang authored Dec 27, 2022
  
  5eec2aef
- post-merge fix · 8079a1b0
  Anthony Chang authored Dec 27, 2022
  
  8079a1b0
- post-refactor fix for output coord in dV/dK · 39f5d9d4
  Anthony Chang authored Dec 27, 2022
  
  39f5d9d4
- calculate correct flops and bytes · ad93b411
  Anthony Chang authored Dec 26, 2022
  
  ad93b411
- dK validates · ec9c2b5e
  Anthony Chang authored Dec 26, 2022
  
  ec9c2b5e
- refactor Gemm2 · 2d55c14c
  Anthony Chang authored Dec 21, 2022
  
  2d55c14c
- rearrange gemm0/gemm1 · 383211ef
  Anthony Chang authored Dec 21, 2022
  
  383211ef
- remove printf's · d13c92bd
  Anthony Chang authored Dec 21, 2022
  
  d13c92bd
- refactor Gemm1 · 825f7f02
  Anthony Chang authored Dec 20, 2022
  
  825f7f02
- refactor gemm0 · c798cff9
  Anthony Chang authored Dec 20, 2022
  
  c798cff9
- dQ validates · db7f7bed
  Anthony Chang authored Dec 20, 2022
  
  db7f7bed
- debugging dQ; suspected K mat not properly loaded · 7a302cc9
  Anthony Chang authored Dec 20, 2022
  
  7a302cc9
- format · b637c77d
  Anthony Chang authored Dec 19, 2022
  
  b637c77d
- finish dS; not yet validated · 8dad40d0
  Anthony Chang authored Dec 19, 2022
  
  8dad40d0
- accumulators are zeroed in gemm_pipeline · ee6c4ff7
  Anthony Chang authored Dec 14, 2022
  
  ee6c4ff7
- fix typo: ygrad instead of vgrad · 7f177f07
  Anthony Chang authored Dec 14, 2022
  
  7f177f07
- dP · 3db3fe42
  Anthony Chang authored Dec 14, 2022
  
  3db3fe42
- format · c26b46de
  Anthony Chang authored Dec 13, 2022
  
  c26b46de
- compute y dot dy · 15f1d4ad
  Anthony Chang authored Dec 13, 2022
  
  15f1d4ad
- add description in example code · a3e487ca
  Anthony Chang authored Dec 13, 2022
  
  a3e487ca
- format · f1b2e521
  Anthony Chang authored Dec 07, 2022
  
  f1b2e521
- strictly follow natural indexing for traversing P tile to avoid jumping accesses (no snake pattern) · 4ae9919e
  Anthony Chang authored Dec 07, 2022
  
  4ae9919e
- can validate dV with relaxed error tolerance · b67a58c0
  Anthony Chang authored Nov 29, 2022
  
  b67a58c0
- start with dY · 8551dd43
  Anthony Chang authored Nov 21, 2022
```
start with dY
```
  8551dd43
- comment LDS bank conflict considerations · ecd5f7c9
  Anthony Chang authored Nov 25, 2022
  
  ecd5f7c9
- ready to plug in kernel · b1e544e2
  Anthony Chang authored Nov 16, 2022
  
  b1e544e2
- add transpose const counterpart · 4f6d52c1
  Anthony Chang authored Nov 16, 2022
  
  4f6d52c1
- host softmax can run with pre-calculated stats for debug purposes · 3908c88b
  Anthony Chang authored Nov 15, 2022
  
  3908c88b
- host attention seems to validate · 0aafc6be
  Anthony Chang authored Nov 15, 2022
  
  0aafc6be
- serialize tensor object in readable format · 25e26104
  Anthony Chang authored Nov 13, 2022
  
  25e26104
- helper function for transposing host tensor · a82804a4
  Anthony Chang authored Nov 11, 2022
  
  a82804a4
15 Dec, 2022 4 commits

Add MNK padding, M = 0 support into grouped_gemm (#539) · 0345963e

zjing14 authored Dec 15, 2022



* add mnk padding, support m=0

* clean code

* clean code
Co-authored-by: Rostyslav Geyyer <46627076+geyyer@users.noreply.github.com>

0345963e

disable the attention test that fails on MI100 (#540) · 11151175
Illia Silin authored Dec 15, 2022

11151175
Add interface GetTypeIdName() and GetTypeIdHashCode() for Device Op (#533) · 10c72ace
Qianfeng authored Dec 15, 2022

10c72ace

Add padding device_gemm_add_add_fastgelu_xdl_c_shuffle instances to enable... · 9a1f2475

Rostyslav Geyyer authored Dec 14, 2022

Add padding device_gemm_add_add_fastgelu_xdl_c_shuffle instances to enable arbitrary problem size (#535)

* Add padding device_gemm_add_add_fastgelu_xdl_c_shuffle instances

* Add padding device_gemm_add_fastgelu_xdl_c_shuffle instances

* Add gemm_add_fastgelu profiler impl

* Add padding device_gemm_fastgelu_xdl_c_shuffle instances

* Add gemm_fastgelu profiler impl

9a1f2475

14 Dec, 2022 1 commit
- Add a docker hub doc file (#538) · 74744cab
  Rostyslav Geyyer authored Dec 14, 2022
  
  74744cab