Winograd-OpenCL
Winograd-OpenCL copied to clipboard
Winograd-based convolution implementation in OpenCL
Because this was a toy project when I was an undergraduate student, the codes are broken here and there. Please use it for study purposes only.
Yout work is good try for cnn developers. But when I test the your code , I found that the winograd resulst is not equal with the traditional conv, Why?
I see input transform, filter transform, inverse transform. But where is GEMM in the kernel? https://github.com/csehydrogen/Winograd-OpenCL/blob/master/kernel.cl#L265
I notice that there are much variables that don't correspond to the original paper; I have also read the technic blog of intel, haven't found any relationship with the unknown...