AMDMIGraphX
AMDMIGraphX copied to clipboard
Move auto_contiguous before lowering
This should fix the perf isssues.
Codecov Report
All modified and coverable lines are covered by tests :white_check_mark:
Project coverage is 92.02%. Comparing base (
6899adc) to head (a15f67e). Report is 144 commits behind head on develop.
Additional details and impacted files
@@ Coverage Diff @@
## develop #3428 +/- ##
========================================
Coverage 92.02% 92.02%
========================================
Files 509 509
Lines 21014 21014
========================================
Hits 19339 19339
Misses 1675 1675
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
Perf report looks pretty bad. Is there something else needed?
Perf regression should be fixed by #3494.
Perf regression should be fixed by #3494.
Do we close this out then or get 3494 in before this?
| Test | Batch | Rate new a15f67 |
Rate old e2e921 |
Diff | Compare |
|---|---|---|---|---|---|
| torchvision-resnet50 | 64 | 3,262.43 | 3,255.97 | 0.20% | :white_check_mark: |
| torchvision-resnet50_fp16 | 64 | 6,997.35 | 6,971.30 | 0.37% | :white_check_mark: |
| torchvision-densenet121 | 32 | 2,431.81 | 2,431.20 | 0.03% | :white_check_mark: |
| torchvision-densenet121_fp16 | 32 | 4,062.83 | 4,087.45 | -0.60% | :white_check_mark: |
| torchvision-inceptionv3 | 32 | 1,638.86 | 1,635.71 | 0.19% | :white_check_mark: |
| torchvision-inceptionv3_fp16 | 32 | 2,761.55 | 2,755.82 | 0.21% | :white_check_mark: |
| cadene-inceptionv4 | 16 | 776.36 | 780.33 | -0.51% | :white_check_mark: |
| cadene-resnext64x4 | 16 | 808.36 | 806.94 | 0.18% | :white_check_mark: |
| slim-mobilenet | 64 | 7,530.11 | 7,464.88 | 0.87% | :white_check_mark: |
| slim-nasnetalarge | 64 | 211.46 | 208.51 | 1.42% | :white_check_mark: |
| slim-resnet50v2 | 64 | 3,500.88 | 3,435.02 | 1.92% | :white_check_mark: |
| bert-mrpc-onnx | 8 | 1,152.16 | 1,146.59 | 0.49% | :white_check_mark: |
| bert-mrpc-tf | 1 | 466.71 | 318.44 | 46.56% | :high_brightness: |
| pytorch-examples-wlang-gru | 1 | 423.53 | 421.02 | 0.60% | :white_check_mark: |
| pytorch-examples-wlang-lstm | 1 | 400.90 | 374.33 | 7.10% | :high_brightness: |
| torchvision-resnet50_1 | 1 | 800.89 | 803.63 | -0.34% | :white_check_mark: |
| cadene-dpn92_1 | 1 | 399.27 | 395.85 | 0.87% | :white_check_mark: |
| cadene-resnext101_1 | 1 | 383.24 | 379.14 | 1.08% | :white_check_mark: |
| onnx-taau-downsample | 1 | 342.91 | 366.52 | -6.44% | :red_circle: |
| dlrm-criteoterabyte | 1 | 33.33 | 35.01 | -4.81% | :red_circle: |
| dlrm-criteoterabyte_fp16 | 1 | 52.72 | 58.19 | -9.40% | :red_circle: |
| agentmodel | 1 | 10,092.80 | 8,063.61 | 25.16% | :high_brightness: |
| unet_fp16 | 2 | 58.78 | 58.95 | -0.30% | :white_check_mark: |
| resnet50v1_fp16 | 1 | 930.75 | 950.25 | -2.05% | :white_check_mark: |
| resnet50v1_int8 | 1 | 973.74 | 971.97 | 0.18% | :white_check_mark: |
| bert_base_cased_fp16 | 64 | 1,172.13 | 1,170.73 | 0.12% | :white_check_mark: |
| bert_large_uncased_fp16 | 32 | 363.97 | 363.46 | 0.14% | :white_check_mark: |
| bert_large_fp16 | 1 | 199.40 | 213.45 | -6.58% | :red_circle: |
| distilgpt2_fp16 | 16 | 2,204.48 | 2,199.49 | 0.23% | :white_check_mark: |
| yolov5s | 1 | 546.55 | 539.25 | 1.35% | :white_check_mark: |
| tinyllama | 1 | 43.50 | 43.47 | 0.07% | :white_check_mark: |
| vicuna-fastchat | 1 | 173.56 | 174.86 | -0.74% | :white_check_mark: |
| whisper-tiny-encoder | 1 | 418.72 | 417.66 | 0.25% | :white_check_mark: |
| whisper-tiny-decoder | 1 | 429.41 | 427.92 | 0.35% | :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