When we use torch convolution layer, there are 2 kernels. One is “void implicit_convolve_sgemm” and the other is “void at::native::unrolled_elementwise_kernel”. I want to use only “void at::native::unrolled_elementwise_kernel”.
Is there any way to use this function seperatively? I tried to do this by bring some pytorch source libraries, but “void at::native::unrolled_elementwise_kernel” is related to PyTorch’s build system, so it is very hard to use this independently.
If anyone know about this, please let me know the steps to use.
I suggest asking pytorch questions on a pytorch forum, such as discuss.pytorch.org