daisytuner / docc / 27837920800

62%

Build Type

push

github

Committed by web-flow

Commit Message

Offload grouped convolutions (#782)

- The target specific expansion for CUDA and ROCm used im2row which does not support grouped convolutions. The fallback was im2col which could not be offloaded. Now, the fallback for im2row is the naïve expansion such that a grouped convolution can be offloaded successfully.
- Extended/added unit tests for target specific expansion of convolutions
- Fixed a small bug in the segformer component benchmarks
- Adapted the harness to test PyTorch with CUDA/ROCm

Coverage Stats

204 of 204 new or added lines in 1 file covered. (100.0%)

3 existing lines in 1 file now uncovered.

36906 of 59722 relevant lines covered (61.8%)

1019.61 hits per line