/** Perform triangular matrix-vector product.
\f$ x = A x \f$ (trans == MagmaNoTrans), or \n
\f$ x = A^T x \f$ (trans == MagmaTrans), or \n
\f$ x = A^H x \f$ (trans == MagmaConjTrans).
@param[in]
uplo Whether the upper or lower triangle of A is referenced.
@param[in]
trans Operation to perform on A.
@param[in]
diag Whether the diagonal of A is assumed to be unit or non-unit.
@param[in]
n Number of rows and columns of A. n >= 0.
@param[in]
dA COMPLEX array of dimension (ldda,n), ldda >= max(1,n).
The n-by-n matrix A, on GPU device.
@param[in]
ldda Leading dimension of dA.
@param[in]
dx COMPLEX array on GPU device.
The n element vector x of dimension (1 + (n-1)*incx).
@param[in]
incx Stride between consecutive elements of dx. incx != 0.
@ingroup magma_cblas2
*/
extern "C" void
magma_ctrmv(
magma_uplo_t uplo, magma_trans_t trans, magma_diag_t diag,
magma_int_t n,
magmaFloatComplex_const_ptr dA, size_t dA_offset, magma_int_t ldda,
magmaFloatComplex_ptr dx, size_t dx_offset, magma_int_t incx,
magma_queue_t queue )
{
if ( n <= 0 )
return;
magmaFloatComplex_ptr dwork;
magma_cmalloc( &dwork, (1 + (n-1)*abs(incx)) );
cl_int err = clblasCtrmv(
clblasColumnMajor,
clblas_uplo_const( uplo ),
clblas_trans_const( trans ),
clblas_diag_const( diag ),
n,
dA, dA_offset, ldda,
dx, dx_offset, incx,
dwork,
1, &queue, 0, NULL, g_event );
clFlush(queue);
check_error( err );
magma_free( dwork );
}
void
xTrmv<cl_float2>::
call_func()
{
timer.Start(timer_id);
clblasCtrmv(order_, buffer_.uplo_, buffer_.trans_a_,
buffer_.diag_, buffer_.m_, buffer_.buf_a_, 0,
buffer_.lda_, buffer_.buf_x_, 0, 1, buffer_.scratch_,
1, &queue_, 0, NULL, &event_);
clWaitForEvents(1, &event_);
timer.Stop(timer_id);
}
