Closed dsharlet closed 4 years ago
@jiawen found that adding CUDA support just required adding __host__ __device__ to some of the functions in array.h.
__host__ __device__
I think we should do a few things here:
shape
array_ref
__device__
cudaMalloc
cudaFree
array
cudaMemcpy*
copy
(1) by itself would be really useful and a good first step.
Fixed by #35.
@jiawen found that adding CUDA support just required adding
__host__ __device__
to some of the functions in array.h.I think we should do a few things here:
__host__ __device__
to appropriate helper functions in array.h to enableshape
andarray_ref
to be used in__device__
functions.cudaMalloc
/cudaFree
to enablearray
to be used on the host.cudaMemcpy*
, ideally transparently withcopy
.(1) by itself would be really useful and a good first step.