void CUDABLAS1::scal( IndexType n, const double alpha, double* x_d, const IndexType incx, SyncToken* syncToken )
{
LAMA_CHECK_CUDA_ACCESS
cudaStream_t stream = NULL;
if ( syncToken )
{
CUDAStreamSyncToken* cudaStreamSyncToken = dynamic_cast<CUDAStreamSyncToken*>( syncToken );
LAMA_ASSERT_DEBUG( cudaStreamSyncToken, "no cuda stream sync token provided" )
stream = cudaStreamSyncToken->getCUDAStream();
}
cublasSetKernelStream( stream );
LAMA_CHECK_CUBLAS_ERROR
cublasDscal( n, alpha, x_d, incx );
// No error check here possible as kernel is started asynchronously
if ( !syncToken )
{
cudaStreamSynchronize( 0 );
LAMA_CHECK_CUDA_ERROR
}
void trsm_magmatask( int m, int t, REAL *A, REAL *B) {
REAL one = 1.0;
cudaStream_t stream = nanos_get_kernel_execution_stream();
cublasSetKernelStream(stream);
MAGMABLASTRSM('R', 'L', 'T', 'N', m, m, one, A, m, B, m );
}
void gemm_magmatask( int m, int t, REAL *A, REAL *B, REAL *C) {
REAL mone = -1.0;
REAL one = 1.0;
cudaStream_t stream = nanos_get_kernel_execution_stream();
cublasSetKernelStream(stream);
MAGMABLASGEMM('N', 'T', m, m, m, mone, A, m, B, m, one, C, m);
}
void syrk_magmatask(int m, REAL *A, REAL *C) {
REAL mone = -1.0;
REAL one = 1.0;
cudaStream_t stream = nanos_get_kernel_execution_stream();
cublasSetKernelStream(stream);
MAGMABLASSYRK('L', 'N', m, m, mone, A, m, one, C, m );
}
