This repository has been archived by the owner on Nov 17, 2023. It is now read-only.
[MKLDNN] Remove overhead of sg_mkldnn_fullyconnected op #17707
Merged
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Outdated suggestions cannot be applied.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.
Description
This PR is mainly focus on removing the overhead of sg_mkldnn_fullyconnected especially in the case of channel-wise quantization mode (no much change for FP32 or tensor-wise quantization mode) described in #17705, via removing extra condition check in the current logic (weights version and calibrated data value), but leaving an ENV of
MXNET_MKLDNN_QFC_DYNAMIC_PARAMS
for the scenario of changing those values on the fly (we don't meet such usage but in case there is).Checklist
Essentials
Please feel free to remove inapplicable items for your PR.
Changes
Comments
@TaoLv @eric-haibin-lin @pengzhao-intel