Comment Comment by p1esk You'll be able to do low level research (e.g. deformable convolutions) in a high level language (Swift), rather than needing to write CUDANot sure I understand - will Swift automatically generate efficient GPU kernels for these low level ops, or will it be making calls to CuDNN, etc? parent p1esk 6 years ago
Not sure I understand - will Swift automatically generate efficient GPU kernels for these low level ops, or will it be making calls to CuDNN, etc?