Closed knsong closed 6 years ago
Thanks for your interest in our project. Unfortunately, we focused on CPU speedups and GPU code is not maintained. The GPU code path is mostly a remnant of Wei's https://github.com/wenwei202/caffe that we forked from. If you're interested in getting speedups in GPU from structured sparsity, please refer to that project.
Thank you for the reply. BTW, is there an efficient sparse winograd inference implementation in SkimCaffe
? I noticed that WinogradLayer<Dtype>::Forward_cpu()
didn't take advantage of sparsity.
Sorry, looks like I didn't add the optimized sparse sparse Winograd forward implementation before I left Intel.
@jspark1105 Ok, thanks.
hi, thank you for the work first. But I notice that in gpu mode,
BaseConvolutionLayer
inskimcaffe
usescusparseScsrmm()
to performim2col+gemm
like convolution and treatsweights
andcol_buff
as a dense, sparse matrix respectively. I'm confused here and shouldn't it be the inverse case thatweights
is sparse andcol_buff
is dense?