master-061f5f8
061f5f8d · CUDA: add __restrict__ to mul mat vec kernels (#2140) ·