Skip to content

[RVV] Add missing maxpool and avgpool rvv kernels #9622

Open
ken-unger wants to merge 2 commits intogoogle:masterfrom
ken-unger:maxpool-rvv
Open

[RVV] Add missing maxpool and avgpool rvv kernels #9622
ken-unger wants to merge 2 commits intogoogle:masterfrom
ken-unger:maxpool-rvv

Conversation

@ken-unger
Copy link
Copy Markdown
Contributor

  • Add rvv script to generate f32-avgpool and f16-avgpool rvv kernels
  • Rewrite rvv script to generate f32-maxpool and add f16-maxpool, s8-maxpool, u8-maxpool rvv kernels
  • Script now closely follows the simd version.

Relevant tests and benchmarks executed and pass (BPI-F3).

@ken-unger
Copy link
Copy Markdown
Contributor Author

RE: rvv hardware detection support for fp16

@dsharlet just fyi, I've opened a PR in pytorch/cpuinfo for this purpose. Once that is reviewed and merged I can make the needed changes to hardware-config.c. pytorch/cpuinfo#375

While other changes in cpuinfo are desirable (e.g cache info for riscv uarchs) I tried to keep the changes to a minimum this round in the hope that the review goes smoothly.

@ken-unger
Copy link
Copy Markdown
Contributor Author

@fbarchard please also give this PR a lookover when you have a few minutes. Thank you.

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