Relax CUDA device requirement
This change removes the cudaDevAttrIntegrated check and only requires pageable memory access. This change allows to run GPU accelerated channel convolution on GH machines.
Showing
Please register or sign in to comment