This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
[MXNET-97] implement DepthwiseConv2dBackwardFilterKernel from tensorflow codebase #10098
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
implement DepthwiseConv2dBackwardFilterKernel from tensorflow codebase, which is over 5x faster
Description
The current mxnet implementation calls __syncthreads() function too much, which is extemely slow.
The new code comes from tensorflow, but the variable names are adjusted for consistency.
My model uses depthwise conv heavily, and now its training time per iteration is over 5x faster on single P40 gpu. ( old 92s vs new 18s )
Checklist
Essentials
make lint
)