Skip to content

[Feature Request] Optimized Convolution kernels for aarch64 #24790

Open
@aviralagrawal

Description

@aviralagrawal

Describe the feature request

Similar to how x86 has optimized convolution kernels written using AVX, for example - SconvKernelAvx512F.S, it would be nice to have SVE/NEON convolution kernels for aarch64.
This would greatly benefit ONNX users running CV models on Mac M-series chips, NVIDIA Grace, AWS Graviton, Fujitsu-Monaka, etc.
I want to understand from Arm ISA experts if this is a planned activity or not; @MaajidKhan @milpuz01 @chenfucn @snnn @yufenglee

Describe scenario use case

Models containing convolutional operators will benefit from these optimized kernels if they are running on aarch64 machines.

Metadata

Metadata

Assignees

No one assigned

    Labels

    feature requestrequest for unsupported feature or enhancement

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions