Skip to content

Add Xtensa HiFi depthwise conv int8/int16/float32 kernels - #3699

Open
unmeshna017 wants to merge 1 commit into
tensorflow:mainfrom
unmeshna017:xtensa-depthwise-conv-hifi
Open

unmeshna017 wants to merge 1 commit into
tensorflow:mainfrom
unmeshna017:xtensa-depthwise-conv-hifi

Conversation

@unmeshna017

Copy link
Copy Markdown
Contributor

Adds HiFi-optimized depthwise conv kernels for int8, int16 and float32, split into per-type helper files with reference fallbacks. Registers the new helper sources in xtensa.inc and adds XTENSA-specific registrations in the shared depthwise_conv.h.

@cad-audio @joshih-cad

Adds HiFi NNLib optimized depthwise conv paths (int8/int16/float32, dilation) split into per-type eval and reference helpers, dispatched from depthwise_conv.cc. Registers the new helper sources in xtensa_makefile.inc. No new NNLib release required.
@unmeshna017
unmeshna017 requested a review from a team as a code owner September 10, 2026 12:31

This branch has not been deployed

No deployments
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant