ARM-software / ComputeLibrary

The Compute Library is a set of computer vision and machine learning functions optimised for both Arm CPUs and GPUs using SIMD technologies.
MIT License
2.76k stars 767 forks source link

NC4HW4 layout support? #990

Closed daoxian closed 1 year ago

daoxian commented 1 year ago

I wonder if there's any plan on the NC4HW4 layout? This layout can obviously improve the convolution operator's performance. So why don't compute library support it ?

morgolock commented 1 year ago

Hi @daoxian

ACL aligns with the data types found major APIs like tflite and nnapi.

Could you please provide to a model or use-case where NC4HW4 brings a considerable improvement?

daoxian commented 1 year ago

https://arxiv.org/pdf/2002.12418.pdf