Commits · 3990522d93dd61f51cda8e948bf34019bfe8971c · gaoqiong / composable_kernel

17 Feb, 2021 1 commit

overhauling DynamicTensorDescriptor and dynamic multi-index transform in... · 3990522d

Chao Liu authored Feb 17, 2021

overhauling DynamicTensorDescriptor and dynamic multi-index transform in preparation for partially compile-time and partially run-time tensor descriptor

3990522d

14 Feb, 2021 1 commit
- fixed GetSrcCoordinateResetStep and GetDstCoordinateResetStep in v1r3 and v3 · 2dea900b
  Chao Liu authored Feb 14, 2021
  
  2dea900b
12 Feb, 2021 2 commits
- move iterator hack to host · 6fe9682a
  Chao Liu authored Feb 12, 2021
  
  6fe9682a
- refactor · 0d475c27
  Chao Liu authored Feb 11, 2021
  
  0d475c27
10 Feb, 2021 1 commit
- clean up threadwise copy · 1c704471
  Chao Liu authored Feb 09, 2021
  
  1c704471
08 Feb, 2021 1 commit
- pass tensor descriptor from host to device by reference, pointer and void* · 8ce8f734
  Chao Liu authored Feb 07, 2021
  
  8ce8f734
07 Feb, 2021 3 commits
- changing class to POD · e1eea81a
  Chao Liu authored Feb 07, 2021
  
  e1eea81a
- changing class to POD: remove explicit keyword · 43d5e05f
  Chao Liu authored Feb 07, 2021
  
  43d5e05f
- changing class to POD · e00de11e
  Chao Liu authored Feb 07, 2021
  
  e00de11e
03 Feb, 2021 1 commit
- adding n-D RunWrite for threadwise copy v3 · f16356d4
  Chao Liu authored Feb 02, 2021
  
  f16356d4
01 Feb, 2021 1 commit
- N-D copy for threadwise copy v3 · 1e347018
  Chao Liu authored Jan 31, 2021
  
  1e347018
31 Jan, 2021 1 commit
- N-D tensor copy for threadwise copy v1r3 · b8d64385
  Chao Liu authored Jan 31, 2021
  
  b8d64385
29 Jan, 2021 1 commit
- refactor, clean up · 39821a90
  Chao Liu authored Jan 29, 2021
  
  39821a90
28 Jan, 2021 2 commits
- fix control flow issue for padding case · 01e94729
  Chao Liu authored Jan 28, 2021
  
  01e94729
- add vector load and access order to threadwise copy v1r3 · 68ea43b1
  Chao Liu authored Jan 27, 2021
  
  68ea43b1
21 Jan, 2021 1 commit
- added vector load, but ISA is much worse · 5fdccfce
  Chao Liu authored Jan 20, 2021
  
  5fdccfce
19 Jan, 2021 1 commit
- adding vector load · 44ddcdcb
  Chao Liu authored Jan 18, 2021
  
  44ddcdcb
17 Jan, 2021 2 commits
- improving index calculation: change to UpdateIndexDiff() · c1ed17f8
  Chao Liu authored Jan 17, 2021
  
  c1ed17f8
- improving index calculation · 77c81617
  Chao Liu authored Jan 16, 2021
  
  77c81617
11 Jan, 2021 1 commit
- add hack for merge transformation · f2f35201
  Chao Liu authored Jan 11, 2021
  
  f2f35201
08 Jan, 2021 1 commit
- modify gridwise dynamic gemm looping · fa479ce4
  Chao Liu authored Jan 07, 2021
  
  fa479ce4
07 Jan, 2021 7 commits
- refactor · 8e35a579
  Chao Liu authored Jan 06, 2021
  
  8e35a579
- refactor · 89123dd7
  Chao Liu authored Jan 06, 2021
  
  89123dd7
- update script · 76df7392
  Chao Liu authored Jan 06, 2021
  
  76df7392
- update dynamic gemm · 50a6b8d7
  Chao Liu authored Jan 06, 2021
  
  50a6b8d7
- update threadwise copy (for static tensor descriptor) for updated buffer_load oob · b90cccf7
  Chao Liu authored Jan 06, 2021
  
  b90cccf7
- update config macro for buffer_load oob feature · f8a869a6
  Chao Liu authored Jan 06, 2021
  
  f8a869a6
- revise buffer_load oob for fp32, it will not modify offset, but it will modify... · 6f0ef227
  Chao Liu authored Jan 06, 2021
```
revise buffer_load oob for fp32, it will not modify offset, but it will modify data loaded by buffer_load
```
  6f0ef227
03 Dec, 2020 2 commits
- revise comment · 82bf5de2
  Chao Liu authored Dec 03, 2020
  
  82bf5de2
- use StaticallyIndexedArray for buffer in threadwise copy, in order to get rid of alloca in IR · 760a234f
  Chao Liu authored Dec 03, 2020
  
  760a234f
30 Nov, 2020 1 commit
- fixing useless instruction issue · 70d06fa9
  Chao Liu authored Nov 30, 2020
  
  70d06fa9
28 Nov, 2020 1 commit
- use readfirstlane to force result into SGPR to reduce VGPR usage · 7733dd88
  Chao Liu authored Nov 28, 2020
  
  7733dd88
27 Nov, 2020 1 commit
- add blockwise copy that doesn't has thread buffer as member to avoid alloca... · 3b3cfae5
  Chao Liu authored Nov 27, 2020
```
add blockwise copy that doesn't has thread buffer as member to avoid alloca and therefore scratch mem
```
  3b3cfae5
26 Nov, 2020 2 commits
- update script · 54138dc8
  Chao Liu authored Nov 25, 2020
  
  54138dc8
- add dynamic forward v4r4, but see lots of scrach mem · be4e3133
  Chao Liu authored Nov 25, 2020
  
  be4e3133
20 Nov, 2020 2 commits
- remove copy_driver · dab53610
  Chao Liu authored Nov 19, 2020
  
  dab53610
- fix scratch memory issue · 8e5085fa
  Chao Liu authored Nov 19, 2020
  
  8e5085fa
08 Nov, 2020 1 commit
- tweak bwd-data-v4r1 · b50fa980
  Chao Liu authored Nov 07, 2020
  
  b50fa980
26 Oct, 2020 1 commit
- debugging scatch mem issue · 92661018
  Chao Liu authored Oct 25, 2020
  
  92661018
25 Oct, 2020 1 commit

Merge branch 'dynamic_tensor_descriptor' of... · adc4f1bf

Chao Liu authored Oct 25, 2020

Merge branch 'dynamic_tensor_descriptor' of github.com:asroy/modular_convolution into dynamic_tensor_descriptor

adc4f1bf