ROCmSoftwarePlatform
Fuse Split-Reduce with MLIR
Part of #3212
Depends on #3097 #3299 and https://github.com/ROCm/rocMLIR/pull/1590
Codecov Report
Attention: Patch coverage is 28.57143% with 5 lines in your changes missing coverage. Please review.
Project coverage is 92.23%. Comparing base (
05b2ff4) to head (94e112a). Report is 152 commits behind head on develop.
| Files with missing lines | Patch % | Lines |
|---|---|---|
| src/module.cpp | 28.57% | 5 Missing :warning: |
Additional details and impacted files
@@ Coverage Diff @@
## develop #3319 +/- ##
===========================================
- Coverage 92.26% 92.23% -0.03%
===========================================
Files 500 500
Lines 20057 20062 +5
===========================================
- Hits 18506 18505 -1
- Misses 1551 1557 +6
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
🚨 Try these New Features:
- Flaky Tests Detection - Detect and resolve failed and flaky tests
![codecov[bot] avatar](https://avatars.githubusercontent.com/in/254?v=4 "Published by a pub.dev verified publisher"){kind=small}
We should also have a compiler pass test for the new fusion, right?
Yeah. They are a bit tricky to write. Let me add a one/two. I have verify test otherwise.
We should also have a compiler pass test for the new fusion, right?
Added tests
| Test | Batch | Rate new 94e112 |
Rate old 05b2ff |
Diff | Compare |
|---|---|---|---|---|---|
| torchvision-resnet50 | 64 | 3,233.91 | 3,238.29 | -0.14% | :white_check_mark: |
| torchvision-resnet50_fp16 | 64 | 6,887.36 | 6,890.63 | -0.05% | :white_check_mark: |
| torchvision-densenet121 | 32 | 2,428.79 | 2,427.57 | 0.05% | :white_check_mark: |
| torchvision-densenet121_fp16 | 32 | 4,081.01 | 4,070.04 | 0.27% | :white_check_mark: |
| torchvision-inceptionv3 | 32 | 1,633.94 | 1,634.43 | -0.03% | :white_check_mark: |
| torchvision-inceptionv3_fp16 | 32 | 2,742.24 | 2,737.22 | 0.18% | :white_check_mark: |
| cadene-inceptionv4 | 16 | 770.98 | 771.30 | -0.04% | :white_check_mark: |
| cadene-resnext64x4 | 16 | 807.25 | 806.92 | 0.04% | :white_check_mark: |
| slim-mobilenet | 64 | 7,437.40 | 7,442.09 | -0.06% | :white_check_mark: |
| slim-nasnetalarge | 64 | 207.38 | 207.44 | -0.03% | :white_check_mark: |
| slim-resnet50v2 | 64 | 3,340.00 | 3,342.32 | -0.07% | :white_check_mark: |
| bert-mrpc-onnx | 8 | 1,148.01 | 1,152.95 | -0.43% | :white_check_mark: |
| bert-mrpc-tf | 1 | 309.91 | 309.74 | 0.06% | :white_check_mark: |
| pytorch-examples-wlang-gru | 1 | 418.38 | 512.77 | -18.41% | :red_circle: |
| pytorch-examples-wlang-lstm | 1 | 388.16 | 387.70 | 0.12% | :white_check_mark: |
| torchvision-resnet50_1 | 1 | 767.53 | 804.05 | -4.54% | :red_circle: |
| cadene-dpn92_1 | 1 | 431.92 | 395.66 | 9.16% | :high_brightness: |
| cadene-resnext101_1 | 1 | 379.02 | 374.54 | 1.20% | :white_check_mark: |
| onnx-taau-downsample | 1 | 343.93 | 344.49 | -0.16% | :white_check_mark: |
| dlrm-criteoterabyte | 1 | 35.08 | 35.05 | 0.07% | :white_check_mark: |
| dlrm-criteoterabyte_fp16 | 1 | 57.25 | 57.31 | -0.11% | :white_check_mark: |
| agentmodel | 1 | 8,174.68 | 8,142.79 | 0.39% | :white_check_mark: |
| unet_fp16 | 2 | 57.77 | 57.75 | 0.04% | :white_check_mark: |
| resnet50v1_fp16 | 1 | 933.75 | 929.86 | 0.42% | :white_check_mark: |
| resnet50v1_int8 | 1 | 945.60 | 922.95 | 2.45% | :white_check_mark: |
| bert_base_cased_fp16 | 64 | 1,141.42 | 1,142.41 | -0.09% | :white_check_mark: |
| bert_large_uncased_fp16 | 32 | 351.78 | 351.90 | -0.03% | :white_check_mark: |
| bert_large_fp16 | 1 | 211.18 | 208.73 | 1.18% | :white_check_mark: |
| distilgpt2_fp16 | 16 | 2,153.21 | 2,155.12 | -0.09% | :white_check_mark: |
| yolov5s | 1 | 503.72 | 503.82 | -0.02% | :white_check_mark: |
| tinyllama | 1 | 43.34 | 43.36 | -0.04% | :white_check_mark: |
| vicuna-fastchat | 1 | 177.12 | 175.40 | 0.98% | :white_check_mark: |
| whisper-tiny-encoder | 1 | 409.80 | 410.24 | -0.11% | :white_check_mark: |
| whisper-tiny-decoder | 1 | 427.53 | 426.66 | 0.20% | :white_check_mark: |
This build is not recommended to merge :red_circle:
:red_circle:bert_large_uncased_fp16: FAILED: MIGraphX is not within tolerance - check verbose output