Commits · 48dbbd11af2d3bf5b5d968d10febe7dd29e2fc36 · gaoqiong / MIGraphX

04 Jul, 2022 6 commits
- Merge branch 'bert-opt3' of github.com:ROCmSoftwarePlatform/AMDMIGraphX into bert-opt3 · 48dbbd11
  Paul authored Jul 04, 2022
  
  48dbbd11
- Merge · 621df96d
  Paul authored Jul 04, 2022
  
  621df96d
- Merge branch 'jit-layernorm' into bert-opt2 · 365d3df1
  Paul authored Jul 04, 2022
  
  365d3df1
- Merge branch 'jit-vector-softmax' into bert-opt2 · f4b87a36
  Paul authored Jul 04, 2022
  
  f4b87a36
- Dont divide by vec.size · d9867f64
  Paul authored Jul 04, 2022
  
  d9867f64
- Dont divide by vec.size · e01224f5
  Paul authored Jul 04, 2022
  
  e01224f5
03 Jul, 2022 2 commits
- Format · 0a5e9b99
  Paul authored Jul 03, 2022
  
  0a5e9b99
- Add license header · 97dc231b
  Paul authored Jul 03, 2022
  
  97dc231b
02 Jul, 2022 20 commits
- Add missing header · 92ab5fe9
  Paul authored Jul 02, 2022
  
  92ab5fe9
- Add missing header · 1e893131
  Paul authored Jul 02, 2022
  
  1e893131
- Add debug header · 5adb8185
  Paul authored Jul 02, 2022
  
  5adb8185
- Merge · 121bd661
  Paul authored Jul 02, 2022
  
  121bd661
- Merge branch 'jit-vector-softmax' into bert-opt2 · 637b483c
  Paul authored Jul 02, 2022
  
  637b483c
- Merge · e22cf227
  Paul authored Jul 02, 2022
  
  e22cf227
- Merge branch 'jit-layernorm' into bert-opt2 · 7147acea
  Paul authored Jul 02, 2022
  
  7147acea
- Divide by vec.size · 84d0f5c9
  Paul authored Jul 02, 2022
  
  84d0f5c9
- Improve calculation with vectorization · 94e983ad
  Paul authored Jul 02, 2022
  
  94e983ad
- Fix merg conflicts · 598cd71a
  Paul authored Jul 02, 2022
  
  598cd71a
- Merge branch 'jit-layernorm' into bert-opt2 · 16fee68f
  Paul authored Jul 02, 2022
  
  16fee68f
- Fix layernorm schedule · 047162bb
  Paul authored Jul 02, 2022
  
  047162bb
- Dont set global as multiple of local · 307c2024
  Paul authored Jul 02, 2022
  
  307c2024
- Format · 864e1b8d
  Paul authored Jul 02, 2022
  
  864e1b8d
- Update block size calculation · 94256bc4
  Paul authored Jul 02, 2022
  
  94256bc4
- Merge branch 'jit-layernorm-merge' into bert-opt3 · 0ee486c5
  Paul authored Jul 02, 2022
  
  0ee486c5
- Merge branch 'dot-add' into bert-opt2 · b1d86d7c
  Paul authored Jul 02, 2022
  
  b1d86d7c
- Format · 9cb9bc09
  Paul authored Jul 02, 2022
  
  9cb9bc09
- Const fold adds for gemms · c27a6376
  Paul authored Jul 02, 2022
  
  c27a6376
- Merge branch 'jit-layernorm' into jit-layernorm-merge · 5e8d91c3
  Paul authored Jul 01, 2022
  
  5e8d91c3
01 Jul, 2022 8 commits
- Merge branch 'dot-add' into bert-opt2 · 3b8ae098
  Paul authored Jul 01, 2022
  
  3b8ae098
- Format · cf9cec1c
  Paul authored Jul 01, 2022
  
  cf9cec1c
- Only gemm used once not c matrix · 48cf58f6
  Paul authored Jul 01, 2022
  
  48cf58f6
- Merge branch 'bert-opt-fastsotfmax' into bert-opt2 · ba4b69a7
  Paul authored Jul 01, 2022
  
  ba4b69a7
- Merge branch 'jit-layernorm' into bert-opt2 · 477e0162
  Paul authored Jul 01, 2022
  
  477e0162
- Merge branch 'jit-improve' into bert-opt2 · 0db1af37
  Paul authored Jul 01, 2022
  
  0db1af37
- Fix vectorization of layernorm · 7b332efc
  Paul authored Jul 01, 2022
  
  7b332efc
- Add const attribute to improve optimizations · 6deee23b
  Paul authored Jun 30, 2022
  
  6deee23b
30 Jun, 2022 4 commits
- Format · 551c2e45
  Paul authored Jun 30, 2022
  
  551c2e45
- Improve loops · 23c97fa9
  Paul authored Jun 30, 2022
  
  23c97fa9
- Only run __syncthreads when there is data to preload · 7ddeb944
  Paul authored Jun 30, 2022
  
  7ddeb944
- Merge branch 'jit-layernorm' into jit-layernorm-merge · 024de2a5
  Paul authored Jun 30, 2022
  
  024de2a5