/***************************************************************************//**
@fn magma_copyvector_async( n, elemSize, dx_src, incx, dy_dst, incy, queue )
Copy vector dx_src on GPU device to dy_dst on GPU device.
Elements may be arbitrary size.
Type-safe versions set elemSize appropriately.
With CUDA unified addressing, dx and dy can be on different GPUs.
This version is asynchronous: it may return before the transfer finishes.
See magma_copyvector() for a synchronous version.
@param[in]
n Number of elements in vector.
@param[in]
elemSize Size of each element, e.g., sizeof(double).
@param[in]
dx_src Source array of dimension (1 + (n-1))*incx, on GPU device.
@param[in]
incx Increment between elements of hx_src. incx > 0.
@param[out]
dy_dst Destination array of dimension (1 + (n-1))*incy, on GPU device.
@param[in]
incy Increment between elements of dy_dst. incy > 0.
@param[in]
queue Queue to execute in.
@ingroup magma_copyvector
*******************************************************************************/
extern "C" void
magma_copyvector_async_internal(
magma_int_t n, magma_int_t elemSize,
magma_const_ptr dx_src, magma_int_t incx,
magma_ptr dy_dst, magma_int_t incy,
magma_queue_t queue,
const char* func, const char* file, int line )
{
// for backwards compatability, accepts NULL queue to mean NULL stream.
cudaStream_t stream = NULL;
if ( queue != NULL ) {
stream = queue->cuda_stream();
}
else {
fprintf( stderr, "Warning: %s got NULL queue\n", __func__ );
}
if ( incx == 1 && incy == 1 ) {
cudaError_t status;
status = cudaMemcpyAsync(
dy_dst,
dx_src,
int(n*elemSize), cudaMemcpyDeviceToDevice, stream );
check_xerror( status, func, file, line );
}
else {
magma_copymatrix_async_internal(
1, n, elemSize, dx_src, incx, dy_dst, incy, queue, func, file, line );
}
}
void magma_copyvector_async_internal(
magma_int_t n, magma_int_t elemSize,
void const* dx_src, magma_int_t incx,
void* dy_dst, magma_int_t incy,
cudaStream_t stream,
const char* func, const char* file, int line )
{
if ( incx == 1 && incy == 1 ) {
cudaError_t status;
status = cudaMemcpyAsync(
dy_dst,
dx_src,
n*elemSize, cudaMemcpyDeviceToDevice, stream );
check_xerror( status, func, file, line );
}
else {
magma_copymatrix_async_internal(
1, n, elemSize, dx_src, incx, dy_dst, incy, stream, func, file, line );
}
}
