Commits · transformer_opts · gaoqiong / MIGraphX

"...include/git@developer.sourcefind.cn:gaoqiong/migraphx.git" did not exist on "cbf4c8d64e274bc92b54dc2b6f6527266b0e5635"

20 May, 2022 22 commits
- Merge remote-tracking branch 'origin/jit-vector-reduce' into transformer_opts · 4b83b5a1
  turneram authored May 20, 2022
  
  4b83b5a1
- Merge remote-tracking branch 'origin/fuse-gemm-add-pointwise' into transformer_opts · 7f387483
  turneram authored May 20, 2022
  
  7f387483
- Merge remote-tracking branch 'origin/gpu-batch-gemm-bert' into transformer_opts · f7fe5508
  turneram authored May 20, 2022
  
  f7fe5508
- Merge remote-tracking branch 'origin/jit-softmax' into transformer_opts · e7e24981
  turneram authored May 20, 2022
  
  e7e24981
- Merge remote-tracking branch 'origin/jit-contiguous' into transformer_opts · 8b649bf9
  turneram authored May 20, 2022
  
  8b649bf9
- Add contiguous before reshape · eff3d2d3
  turneram authored May 20, 2022
  
  eff3d2d3
- Formatting · cd96c1c8
  turneram authored May 20, 2022
  
  cd96c1c8
- Remove transpose kernels · 37351ed6
  turneram authored May 20, 2022
  
  37351ed6
- Formatting · 48187e79
  turneram authored May 20, 2022
  
  48187e79
- Add attention verify_onnx test · 6202ea15
  turneram authored May 20, 2022
  
  6202ea15
- Formatting · b745f416
  turneram authored May 20, 2022
  
  b745f416
- Fix layernorm verify test · a16adb42
  turneram authored May 20, 2022
  
  a16adb42
- Formatting · d41f0d66
  turneram authored May 20, 2022
  
  d41f0d66
- Add transposectx and transposeqkv ref tests · 095e49a3
  turneram authored May 20, 2022
  
  095e49a3
- Remove non-inference portions of parse_attention · 7757cfd0
  turneram authored May 20, 2022
  
  7757cfd0
- Generate layernorm onnx file · 5a62e9e7
  turneram authored May 20, 2022
  
  5a62e9e7
- Formatting · 05e8bfde
  turneram authored May 20, 2022
  
  05e8bfde
- Add attention, layernorm op, transposectx, and transposeqkv · 3ea9fe4c
  turneram authored May 20, 2022
  
  3ea9fe4c
- Rename pointwise ops (#1145) · 4a312201
  kahmed10 authored May 20, 2022
```
For clarity on kernel names found when profiling. The new names are set to the order of the ops being compiled. For example: add + relu = add_relu_kernel.
```
  4a312201
- Merge branch 'develop' into fuse-gemm-add-pointwise · 8354be18
  Umang Yadav authored May 20, 2022
  
  8354be18
- Merge branch 'develop' into gpu-batch-gemm-bert · 3d2cdf03
  Umang Yadav authored May 20, 2022
  
  3d2cdf03
- Improve matching with has_value when there are convert operators (#1212) · 27af0170
  Paul Fultz II authored May 19, 2022
  
  27af0170
19 May, 2022 2 commits
- Remove check for standard since it is already done in the matcher · 9d12bc5c
  Paul authored May 19, 2022
  
  9d12bc5c
- Fix perf regression · c84154b8
  Paul authored May 19, 2022
  
  c84154b8
18 May, 2022 9 commits
- Fix tidy issue · 7133eee6
  Paul authored May 18, 2022
  
  7133eee6
- Merge branch 'develop' into gpu-batch-gemm-bert · 4a4f537e
  Paul Fultz II authored May 18, 2022
  
  4a4f537e
- Format · adc03be6
  Paul authored May 18, 2022
  
  adc03be6
- Fix single dim · 6ba5859b
  Paul authored May 18, 2022
  
  6ba5859b
- Format · e70d4269
  Paul authored May 18, 2022
  
  e70d4269
- Fix unit test · 03ae8013
  Paul authored May 18, 2022
  
  03ae8013
- Remove copy · cbd73ae7
  Paul authored May 18, 2022
  
  cbd73ae7
- Format · fd801fee
  Paul authored May 18, 2022
  
  fd801fee
- Fuse gemm add with pointwise fusions · 449df648
  Paul authored May 18, 2022
  
  449df648
17 May, 2022 7 commits
- Format · 835cc1e2
  Paul authored May 17, 2022
  
  835cc1e2
- Fuse contiguous · 77be2528
  Paul authored May 17, 2022
  
  77be2528
- Format · 9426aae5
  Paul authored May 17, 2022
  
  9426aae5
- Dont hinder eliminate_contiguous · e83dc134
  Paul authored May 17, 2022
  
  e83dc134
- Format · 8e49a9f2
  Paul authored May 17, 2022
  
  8e49a9f2
- Jit contiguous · 407acb7d
  Paul authored May 17, 2022
  
  407acb7d
- Merge branch 'develop' into jit-vector-reduce · d9a5acbd
  Paul Fultz II authored May 17, 2022
  
  d9a5acbd