Commits · a2f8e035987c8c49b51c88d71f3b04e2a0bc8fda · gaoqiong / MIGraphX

07 Dec, 2023 1 commit
- FP8 2D forward convolution using rocMLIR (#2507) · 6a72e8fc
  Umang Yadav authored Dec 06, 2023
  
  6a72e8fc
06 Dec, 2023 8 commits
- revert changes for nested converts · c52d1f69
  Umang Yadav authored Dec 06, 2023
  
  c52d1f69
- Enable standalone convolution for mlir by default (#2503) · a09dc502
  Paul Fultz II authored Dec 05, 2023
  
  a09dc502
- Improve dpp reductions on navi (#2439) · 8c73c72e
  Paul Fultz II authored Dec 05, 2023
  
  8c73c72e
- Set MIGRAPHX_USE_COMPOSABLEKERNEL=1 in cmake (#2518) · 7d4b1719
  turneram authored Dec 05, 2023
  
  7d4b1719
- Device kernels using FP8 (#2510) · 98ef0abb
  Umang Yadav authored Dec 05, 2023
  
  98ef0abb
- Add FP8 rocblas gemm support (#2473) · 6d0b6bcf
  Umang Yadav authored Dec 05, 2023
  
  6d0b6bcf
- add local fix · 8c5678e0
  Umang Yadav authored Dec 06, 2023
  
  8c5678e0
- merge changes · d97fb02a
  Umang Yadav authored Dec 06, 2023
  
  d97fb02a
05 Dec, 2023 12 commits
- add DCE · 4315a991
  Umang Yadav authored Dec 05, 2023
  
  4315a991
- remove unnecessary changes · 351b43b7
  Umang Yadav authored Dec 05, 2023
  
  351b43b7
- use eliminate_data_type to avoid lossy downcast · 9e2946a4
  Umang Yadav authored Dec 05, 2023
  
  9e2946a4
- add quant_dot as unsupported fp8 op · 39bf5dc5
  Umang Yadav authored Dec 05, 2023
  
  39bf5dc5
- quant dot · 5d240fa4
  Umang Yadav authored Dec 05, 2023
  
  5d240fa4
- add quant_dot support for fp8 · d5fa82db
  Umang Yadav authored Dec 02, 2023
  
  d5fa82db
- remove else · aeaac20f
  Umang Yadav authored Dec 05, 2023
  
  aeaac20f
- fix typo · a9db2bf4
  Umang Yadav authored Dec 05, 2023
  
  a9db2bf4
- add comments · 82f98478
  Umang Yadav authored Dec 05, 2023
  
  82f98478
- add changes for the eliminate_data_type pass · cf91c2b1
  Umang Yadav authored Dec 05, 2023
  
  cf91c2b1
- remove header · b099a7da
  Umang Yadav authored Dec 05, 2023
  
  b099a7da
- use eliminate_data_type pass instead of eliminate_fp8 pass · 402c66ab
  Umang Yadav authored Dec 05, 2023
  
  402c66ab
04 Dec, 2023 2 commits
- [MLIR] Updates needed for general stride support (#2428) · e3e00547
  Krzysztof Drewniak authored Dec 04, 2023
  
  e3e00547
- Remove unused device kernels (#2509) · 609ad966
  Umang Yadav authored Dec 04, 2023
  
  609ad966
03 Dec, 2023 10 commits
- formatting · fc093b02
  Umang Yadav authored Dec 03, 2023
  
  fc093b02
- Try removing fusing converts · 119a6b86
  Umang Yadav authored Dec 03, 2023
  
  119a6b86
- fix naming · 51ac4fdd
  Umang Yadav authored Dec 03, 2023
  
  51ac4fdd
- use helper function to determine gfx940 · 994d24b6
  Umang Yadav authored Dec 03, 2023
  
  994d24b6
- Disable FP8 tests for the non-gfx940 arches · fe585d42
  Umang Yadav authored Dec 03, 2023
  
  fe585d42
- add quant_conv tests · 370d18c1
  Umang Yadav authored Dec 03, 2023
  
  370d18c1
- revert some changes · 050184cb
  Umang Yadav authored Dec 03, 2023
  
  050184cb
- add conv tests supported by MLIR · 88eb3559
  Umang Yadav authored Nov 28, 2023
  
  88eb3559
- add pooling to unsupported ops · f18418be
  Umang Yadav authored Nov 28, 2023
  
  f18418be
- mlir fp8 · 575fc04a
  Umang Yadav authored Nov 26, 2023
  
  575fc04a
02 Dec, 2023 1 commit
- add quant_dot support for fp8 · ce61ea6b
  Umang Yadav authored Dec 02, 2023
  
  ce61ea6b
01 Dec, 2023 4 commits
- few changes · 38218edc
  Umang Yadav authored Dec 01, 2023
  
  38218edc
- merge changes · f155b0e6
  Umang Yadav authored Dec 01, 2023
  
  f155b0e6
- FP8 GPU implementation (#2455) · eafd55de
  Umang Yadav authored Dec 01, 2023
  
  eafd55de
- Add a tuning limit for mlir (#2442) · b47dd32b
  Paul Fultz II authored Dec 01, 2023
  
  b47dd32b
30 Nov, 2023 1 commit
- [6.1] Add support for dot-(mul)-softmax-dot offloads to MLIR (#2345) · ff485c7a
  Manupa Karunaratne authored Nov 30, 2023
  
  ff485c7a
29 Nov, 2023 1 commit
- use miopen for fp8 · da7717ce
  Umang Yadav authored Nov 29, 2023
  
  da7717ce