NNPACK caffe_nnp_convolution_output using a lot of memory and the speed is not fast.

Maratyszcza / NNPACK

Acceleration package for neural networks on multi-core CPUs

BSD 2-Clause "Simplified" License

1.68k stars 316 forks source link

NNPACK caffe_nnp_convolution_output using a lot of memory and the speed is not fast. #104

Closed victorygogogo closed 7 years ago

victorygogogo commented 7 years ago

I test it on msm8953.

NNPACK caffe_nnp_convolution_output is not fast.

is there some switch in NNPACK to speed up ?

I want to use neon.

Maratyszcza commented 7 years ago

Why do you use nnp_convolution_output? It is optimized for training and large batch size. On mobile use-cases, typical batch size is 1, and nnp_convolution_inference handles this case much better.

victorygogogo commented 7 years ago

I use nnp_convolution_inference ,it is the same problem.