|
@@ -668,7 +668,6 @@ static int copy_ram_to_cuda(void *src_interface, unsigned src_node __attribute__
|
|
|
size_t elemsize = src_block->elemsize;
|
|
|
|
|
|
cudaError_t cures;
|
|
|
- int ret;
|
|
|
|
|
|
/* We may have a contiguous buffer for the entire block, or contiguous
|
|
|
* plans within the block, we can avoid many small transfers that way */
|
|
@@ -678,13 +677,13 @@ static int copy_ram_to_cuda(void *src_interface, unsigned src_node __attribute__
|
|
|
if (((nx*ny) == src_block->ldz) && (src_block->ldz == dst_block->ldz))
|
|
|
{
|
|
|
cures = cudaMemcpy((char *)dst_block->ptr, (char *)src_block->ptr,
|
|
|
- nx*ny*nz*elemsize, cudaMemcpyHostToDevice, *stream);
|
|
|
+ nx*ny*nz*elemsize, cudaMemcpyHostToDevice);
|
|
|
}
|
|
|
else {
|
|
|
/* Are all plans contiguous */
|
|
|
cures = cudaMemcpy2D((char *)dst_block->ptr, dst_block->ldz*elemsize,
|
|
|
(char *)src_block->ptr, src_block->ldz*elemsize,
|
|
|
- nx*ny*elemsize, nz, cudaMemcpyHostToDevice, *stream);
|
|
|
+ nx*ny*elemsize, nz, cudaMemcpyHostToDevice);
|
|
|
}
|
|
|
if (STARPU_UNLIKELY(cures))
|
|
|
STARPU_CUDA_REPORT_ERROR(cures);
|