Commits · afe127651664fc334a20788834516c26d9aaa2e3 · gaoqiong / MIGraphX

27 Nov, 2023 1 commit
- mlir fp8 · afe12765
  Umang Yadav authored Nov 26, 2023
  
  afe12765
26 Nov, 2023 9 commits
- formatting · 9a9e9648
  Umang Yadav authored Nov 26, 2023
  
  9a9e9648
- Move pass before optimize module · 83ce487a
  Umang Yadav authored Nov 26, 2023
  
  83ce487a
- remove convert from lowering · 8734ffa3
  Umang Yadav authored Nov 26, 2023
  
  8734ffa3
- add eliminate_fp8 pass · ad9c25ea
  Umang Yadav authored Nov 26, 2023
  
  ad9c25ea
- add tests · 4604f2e1
  Umang Yadav authored Nov 26, 2023
  
  4604f2e1
- add tests and fix cmake · d473b802
  Umang Yadav authored Nov 26, 2023
  
  d473b802
- add ifdef to compile · 87548b5d
  Umang Yadav authored Nov 26, 2023
  
  87548b5d
- Works now · 037205c5
  Umang Yadav authored Nov 26, 2023
  
  037205c5
- compiles all right · 3aa465fd
  Umang Yadav authored Nov 26, 2023
  
  3aa465fd
21 Nov, 2023 2 commits
- add float8 for tensorview · a6c57726
  Umang Yadav authored Nov 21, 2023
  
  a6c57726
- use void for highest to match template candidates · 480288f8
  Umang Yadav authored Nov 21, 2023
  
  480288f8
20 Nov, 2023 9 commits
- Put numeric_max and numeeric lowest into float8 · f9542d5b
  Umang Yadav authored Nov 20, 2023
  
  f9542d5b
- Remove MIN max overloads · 836e201e
  Umang Yadav authored Nov 20, 2023
  
  836e201e
- Remove DPP · 2acd265b
  Umang Yadav authored Nov 20, 2023
  
  2acd265b
- Remove old cast · 6fbd9970
  Umang Yadav authored Nov 20, 2023
  
  6fbd9970
- add comment about c++20 extensions · 86605723
  Umang Yadav authored Nov 20, 2023
  
  86605723
- add test for rsqrt and remove old-styple-cast · 13ef4148
  Umang Yadav authored Nov 20, 2023
  
  13ef4148
- use __builtin_is_constant_evaluated · 6155c782
  Umang Yadav authored Nov 20, 2023
  
  6155c782
- add layernorm, remove constexpr for 1/r · f550f814
  Umang Yadav authored Nov 20, 2023
  
  f550f814
- roialign, softmax, pow, acosh, atanh,pad tests are enabled now · 13403ab2
  Umang Yadav authored Nov 20, 2023
  
  13403ab2
18 Nov, 2023 4 commits
- tidy fix · 1be95870
  Umang Yadav authored Nov 18, 2023
  
  1be95870
- add nolint · 249464c1
  Umang Yadav authored Nov 18, 2023
  
  249464c1
- Fix tidy and other errors · e2b0c406
  Umang Yadav authored Nov 18, 2023
  
  e2b0c406
- Add math and reduce tests · 91cc9c7c
  Umang Yadav authored Nov 18, 2023
  
  91cc9c7c
17 Nov, 2023 15 commits
- revert unwanted changes · ef425d04
  Umang Yadav authored Nov 17, 2023
  
  ef425d04
- formatting · 60dd1f46
  Umang Yadav authored Nov 17, 2023
  
  60dd1f46
- disable DPP For FP8 · e88d46a6
  Umang Yadav authored Nov 17, 2023
  
  e88d46a6
- Add back floatequal · 32033d85
  Umang Yadav authored Nov 17, 2023
  
  32033d85
- Remove clang diagnostics · f975c633
  Umang Yadav authored Nov 17, 2023
  
  f975c633
- Remove unnecessary pragmas · aed1922b
  Umang Yadav authored Nov 17, 2023
  
  aed1922b
- constructor from float works with constexpr · 85ba819b
  Umang Yadav authored Nov 17, 2023
  
  85ba819b
- Update float8_impl · b36f72d3
  Umang Yadav authored Nov 17, 2023
  
  b36f72d3
- remove using for dtypes · 5da68df6
  Umang Yadav authored Nov 17, 2023
  
  5da68df6
- Remove FP8_Lowest/Max · d2c25a07
  Umang Yadav authored Nov 17, 2023
  
  d2c25a07
- remove non-JIT related code · 3411649c
  Umang Yadav authored Nov 17, 2023
  
  3411649c
- only compile for device · 78ec77ec
  Umang Yadav authored Nov 17, 2023
  
  78ec77ec
- Make FNUZ template param and add numeric limits · 60942349
  Umang Yadav authored Nov 17, 2023
  
  60942349
- Ref implementation of FP8 (#2438) · 7f93a818
  Umang Yadav authored Nov 17, 2023
```
Handles all 4 Fp8 dtypes listed here : https://onnx.ai/onnx/technical/float8.html
Follows saturation/clipping logic from table there as well : https://onnx.ai/onnx/technical/float8.html#cast
Only adding fp8e4m3fnuz in MIGraphX IR for now.
```
  7f93a818
- use bit cast · d7339e8a
  Umang Yadav authored Nov 17, 2023
  
  d7339e8a