Add check for fusable ops to attention fusion
Not all pointwise ops are mlir fusable, and so this check is required. Fixes #3479
Codecov Report
All modified and coverable lines are covered by tests :white_check_mark:
Project coverage is 92.02%. Comparing base (
209951a) to head (0a3b385). Report is 150 commits behind head on develop.
Additional details and impacted files
@@ Coverage Diff @@
## develop #3480 +/- ##
========================================
Coverage 92.02% 92.02%
========================================
Files 509 509
Lines 21005 21005
========================================
Hits 19330 19330
Misses 1675 1675
| Flag | Coverage Δ | |
|---|---|---|
92.02% <ø> (ø) |
Flags with carried forward coverage won't be shown. Click here to find out more.
:umbrella: View full report in Codecov by Sentry.
:loudspeaker: Have feedback on the report? Share it here.
| Test | Batch | Rate new 0a3b38 |
Rate old 1ab830 |
Diff | Compare |
|---|---|---|---|---|---|
| torchvision-resnet50 | 64 | 3,254.61 | 3,258.24 | -0.11% | :white_check_mark: |
| torchvision-resnet50_fp16 | 64 | 6,969.90 | 6,986.28 | -0.23% | :white_check_mark: |
| torchvision-densenet121 | 32 | 2,432.09 | 2,435.48 | -0.14% | :white_check_mark: |
| torchvision-densenet121_fp16 | 32 | 4,101.67 | 4,094.65 | 0.17% | :white_check_mark: |
| torchvision-inceptionv3 | 32 | 1,636.12 | 1,636.30 | -0.01% | :white_check_mark: |
| torchvision-inceptionv3_fp16 | 32 | 2,753.87 | 2,745.30 | 0.31% | :white_check_mark: |
| cadene-inceptionv4 | 16 | 779.64 | 779.12 | 0.07% | :white_check_mark: |
| cadene-resnext64x4 | 16 | 807.34 | 808.04 | -0.09% | :white_check_mark: |
| slim-mobilenet | 64 | 7,465.71 | 7,462.28 | 0.05% | :white_check_mark: |
| slim-nasnetalarge | 64 | 208.04 | 208.07 | -0.02% | :white_check_mark: |
| slim-resnet50v2 | 64 | 3,435.50 | 3,438.00 | -0.07% | :white_check_mark: |
| bert-mrpc-onnx | 8 | 1,145.42 | 1,154.53 | -0.79% | :white_check_mark: |
| bert-mrpc-tf | 1 | 318.06 | 320.51 | -0.76% | :white_check_mark: |
| pytorch-examples-wlang-gru | 1 | 412.24 | 391.03 | 5.42% | :high_brightness: |
| pytorch-examples-wlang-lstm | 1 | 384.12 | 383.67 | 0.12% | :white_check_mark: |
| torchvision-resnet50_1 | 1 | 752.56 | 786.92 | -4.37% | :red_circle: |
| cadene-dpn92_1 | 1 | 402.89 | 400.89 | 0.50% | :white_check_mark: |
| cadene-resnext101_1 | 1 | 383.08 | 382.29 | 0.21% | :white_check_mark: |
| onnx-taau-downsample | 1 | 344.40 | 343.92 | 0.14% | :white_check_mark: |
| dlrm-criteoterabyte | 1 | 35.03 | 35.06 | -0.09% | :white_check_mark: |
| dlrm-criteoterabyte_fp16 | 1 | 58.20 | 58.15 | 0.09% | :white_check_mark: |
| agentmodel | 1 | 9,766.29 | 8,244.43 | 18.46% | :high_brightness: |
| unet_fp16 | 2 | 57.89 | 58.03 | -0.24% | :white_check_mark: |
| resnet50v1_fp16 | 1 | 934.57 | 935.05 | -0.05% | :white_check_mark: |
| resnet50v1_int8 | 1 | 972.77 | 972.22 | 0.06% | :white_check_mark: |
| bert_base_cased_fp16 | 64 | 1,169.22 | 1,172.19 | -0.25% | :white_check_mark: |
| bert_large_uncased_fp16 | 32 | 363.07 | 362.83 | 0.07% | :white_check_mark: |
| bert_large_fp16 | 1 | 213.13 | 211.45 | 0.80% | :white_check_mark: |
| distilgpt2_fp16 | 16 | 2,204.04 | 2,204.90 | -0.04% | :white_check_mark: |
| yolov5s | 1 | 533.21 | 527.91 | 1.00% | :white_check_mark: |
| tinyllama | 1 | 43.41 | 43.44 | -0.08% | :white_check_mark: |
| vicuna-fastchat | 1 | 178.11 | 172.02 | 3.54% | :high_brightness: |
| whisper-tiny-encoder | 1 | 417.78 | 415.69 | 0.50% | :white_check_mark: |
| whisper-tiny-decoder | 1 | 436.67 | 424.19 | 2.94% | :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