void transferVector(const Stream& stream, const PinnedHostVector& vectorHost, PRECISION* deviceMemoryPosition) {
const cudaStream_t& cudaStream = stream.getCudaStream();
const int numberOfRows = vectorHost.getNumberOfRows();
const PRECISION* hostVectorPointer = vectorHost.getMemoryPointer();
handleCublasStatus(
cublasSetVectorAsync(numberOfRows, sizeof(PRECISION), hostVectorPointer, 1, deviceMemoryPosition, 1, cudaStream),
"Error when transferring vector from host to device point: ");
}
void magma_setvector_async(
magma_int_t n, size_t elemSize,
void const* hx_src, magma_int_t incx,
void* dy_dst, magma_int_t incy,
cudaStream_t stream )
{
cublasStatus_t status;
status = cublasSetVectorAsync(
n, elemSize,
hx_src, incx,
dy_dst, incy, stream );
check_error( status );
}
DeviceVector* transferVector(const Stream& stream, const PinnedHostVector& vectorHost) {
const cudaStream_t& cudaStream = stream.getCudaStream();
const int numberOfRows = vectorHost.getNumberOfRows();
DeviceVector* deviceVector = new DeviceVector(numberOfRows);
PRECISION* deviceVectorPointer = deviceVector->getMemoryPointer();
const PRECISION* hostVectorPointer = vectorHost.getMemoryPointer();
handleCublasStatus(
cublasSetVectorAsync(numberOfRows, sizeof(PRECISION), hostVectorPointer, 1, deviceVectorPointer, 1, cudaStream),
"Error when transferring vector from host to device: ");
return deviceVector;
}
void magma_setvector_async_internal(
magma_int_t n, magma_int_t elemSize,
void const* hx_src, magma_int_t incx,
void* dy_dst, magma_int_t incy,
cudaStream_t stream,
const char* func, const char* file, int line )
{
cublasStatus_t status;
status = cublasSetVectorAsync(
n, elemSize,
hx_src, incx,
dy_dst, incy, stream );
check_xerror( status, func, file, line );
}
// --------------------
extern "C" void
magma_zsetvector_async_internal(
magma_int_t n,
magmaDoubleComplex const* hx_src, magma_int_t incx,
magmaDoubleComplex_ptr dy_dst, magma_int_t incy,
magma_queue_t queue,
const char* func, const char* file, int line )
{
cublasStatus_t status;
status = cublasSetVectorAsync(
n, sizeof(magmaDoubleComplex),
hx_src, incx,
dy_dst, incy, queue );
check_xerror( status, func, file, line );
}
/***************************************************************************//**
@fn magma_setvector( n, elemSize, hx_src, incx, dy_dst, incy, queue )
Copy vector hx_src on CPU host to dy_dst on GPU device.
Elements may be arbitrary size.
Type-safe versions set elemSize appropriately.
This version synchronizes the queue after the transfer.
See magma_setvector_async() for an asynchronous version.
@param[in]
n Number of elements in vector.
@param[in]
elemSize Size of each element, e.g., sizeof(double).
@param[in]
hx_src Source array of dimension (1 + (n-1))*incx, on CPU host.
@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_setvector
*******************************************************************************/
extern "C" void
magma_setvector_q_internal(
magma_int_t n, magma_int_t elemSize,
void const* hx_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 )
{
assert( queue != NULL );
cublasStatus_t status;
status = cublasSetVectorAsync(
int(n), int(elemSize),
hx_src, int(incx),
dy_dst, int(incy), queue->cuda_stream() );
cudaStreamSynchronize( queue->cuda_stream() );
check_xerror( status, func, file, line );
}
/***************************************************************************//**
@fn magma_setvector_async( n, elemSize, hx_src, incx, dy_dst, incy, queue )
Copy vector hx_src on CPU host to dy_dst on GPU device.
Elements may be arbitrary size.
Type-safe versions set elemSize appropriately.
This version is asynchronous: it may return before the transfer finishes,
if hx_src is pinned CPU memory.
See magma_setvector() 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]
hx_src Source array of dimension (1 + (n-1))*incx, on CPU host.
@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_setvector
*******************************************************************************/
extern "C" void
magma_setvector_async_internal(
magma_int_t n, magma_int_t elemSize,
void const* hx_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__ );
}
cublasStatus_t status;
status = cublasSetVectorAsync(
int(n), int(elemSize),
hx_src, int(incx),
dy_dst, int(incy), stream );
check_xerror( status, func, file, line );
}
