mirror of
https://github.com/saymrwulf/onnxruntime.git
synced 2026-06-04 23:59:56 +00:00
352b745deb
### Description This PR is to enhance the profiling information. With the PR, the profiling result is like below: ``` [profiling] kernel "[Split] 51288384" input[0]: 1,256,64,64, output[0]: 1,256,64,64, execution time: 37135 ns program-manager.ts:114 [profiling] kernel "[Concat] 52361040" input[0]: 1,256,64,64, output[0]: 1,256,64,64, execution time: 50833 ns program-manager.ts:114 [profiling] kernel "[Transpose] 52375264" input[0]: 1,256,64,64, output[0]: 1,64,64,256, execution time: 99791 ns program-manager.ts:114 [profiling] kernel "[Sub] 51098472" input[0]: , input[1]: 1, output[0]: 1, execution time: 7448 ns program-manager.ts:114 [profiling] kernel "[Mul] 51344440" input[0]: 1, input[1]: 1,256,1,1, output[0]: 1,256,1,1, execution time: 8334 ns ``` Without this PR, the profiling result is like below: ``` [profiling] kernel "52097928|[Split] 52097928" execution time: 37760 ns program-manager.ts:105 [profiling] kernel "41898328|[Concat] 41898328" execution time: 51666 ns program-manager.ts:105 [profiling] kernel "41915648|[Transpose] 41915648" execution time: 95416 ns program-manager.ts:105 [profiling] kernel "49757856|[Sub] 49757856" execution time: 7969 ns program-manager.ts:105 [profiling] kernel "51680504|[Mul] 51680504" execution time: 8906 ns ``` With the new information, we can easily know what kind of shape ops have poor performance. Also it can help us to check whether too small shape ops run on gpu. |
||
|---|---|---|
| .. | ||
| onnxjs | ||
| wasm | ||
| backend-onnxjs.ts | ||
| backend-wasm.ts | ||
| build-def.d.ts | ||
| index.ts | ||
| version.ts | ||