SL_03_001 - XNNPACK

SL_03_001 - XNNPACK

About

XNNPACK is a highly optimized solution for neural network inference.

Status

Development

STALLED

 

Development Timeline

Q3-Q4 2023

 

Upstreaming

ONGOING

 

Upstream Version

https://github.com/google/XNNPACK

 

Contacts

@Simon Harvey (Imagination)

Frank Barchard (Google)

 

 

Looks like there's some initial work on this: https://github.com/google/android-riscv64/issues/42

Trying to clarify current state as we commence activity on this

@Elliott Hughes

Dependencies

 

 

 

Updates

Dec 11, 2024 

  • Simon Harvey (Imagination) - spreadsheet summarising RVV support for various XNNPACK kernels: XNNPACK Status

Nov 17, 2023

  • Frank Barchard (Google) - fa64a71: Add F16-VCLAMP-RVV for RISC-V clamping

Jan 8, 2024

  • Frank Barchard (Google) - 821b84f: RISCV build fix for RVV tests on Linux.

Jan 8, 2024

  • Bruce Lai (SiFive) - XNNPACK PR 5876: RVV F32-Softmax Patch 2: Add RVV f32-raddstoreexpminusmax

Dec 20, 2023

  • Frank Barchard (Google) - 7ca6a3b: Generator tool support u4v notation for RVV

Dec 15, 2023

  • Bruce Lai (SiFive) - XNNPACK PR 5869: RVV F32-Softmax Patch 1: Add RVV reduced min & max & minmax

Dec 4, 2023

Nov 17, 2023

  • Frank Barchard (Google) - cf5f791: RVV binary ops use m2 versions of microkernels 

Nov 7, 2023

  • Frank Barchard (Google) - 6516f2a: vmul tests update for RISC-V

Oct 17, 2023

  • Manoj Bhosale (Imagination) - XNNPACK PR 5638: Add QS8/QU8 vmul/vmulc RVV microkernel implementations

Aug 4, 2023 

  • Reported commencement of development effort