void Hemv
( UpperOrLower uplo,
T alpha, const Matrix<T>& A, const Matrix<T>& x, T beta, Matrix<T>& y )
{
DEBUG_CSE
Symv( uplo, alpha, A, x, beta, y, true );
}
void Hemv
( UpperOrLower uplo,
T alpha, const Matrix<T>& A, const Matrix<T>& x, T beta, Matrix<T>& y )
{
DEBUG_ONLY(CallStackEntry cse("Hemv"))
Symv( uplo, alpha, A, x, beta, y, true );
}
inline void HermitianTridiagU( Matrix<R>& A )
{
#ifndef RELEASE
PushCallStack("HermitianTridiagU");
if( A.Height() != A.Width() )
throw std::logic_error( "A must be square." );
#endif
// Matrix views
Matrix<R>
ATL, ATR, A00, a01, A02, a01T,
ABL, ABR, a10, alpha11, a12, alpha01B,
A20, a21, A22;
// Temporary matrices
Matrix<R> w01;
PushBlocksizeStack( 1 );
PartitionUpDiagonal
( A, ATL, ATR,
ABL, ABR, 0 );
while( ABR.Height()+1 < A.Height() )
{
RepartitionUpDiagonal
( ATL, /**/ ATR, A00, a01, /**/ A02,
/**/ a10, alpha11, /**/ a12,
/*************/ /**********************/
ABL, /**/ ABR, A20, a21, /**/ A22 );
PartitionUp
( a01, a01T,
alpha01B, 1 );
w01.ResizeTo( a01.Height(), 1 );
//--------------------------------------------------------------------//
const R tau = Reflector( alpha01B, a01T );
const R epsilon1 = alpha01B.Get(0,0);
alpha01B.Set(0,0,R(1));
Symv( UPPER, tau, A00, a01, R(0), w01 );
const R alpha = -tau*Dot( w01, a01 )/R(2);
Axpy( alpha, a01, w01 );
Syr2( UPPER, R(-1), a01, w01, A00 );
alpha01B.Set(0,0,epsilon1);
//--------------------------------------------------------------------//
SlidePartitionUpDiagonal
( ATL, /**/ ATR, A00, /**/ a01, A02,
/*************/ /**********************/
/**/ a10, /**/ alpha11, a12,
ABL, /**/ ABR, A20, /**/ a21, A22 );
}
PopBlocksizeStack();
#ifndef RELEASE
PopCallStack();
#endif
}
void Hemv
( UpperOrLower uplo,
T alpha, const AbstractDistMatrix<T>& A,
const AbstractDistMatrix<T>& x,
T beta, AbstractDistMatrix<T>& y,
const SymvCtrl<T>& ctrl )
{
DEBUG_ONLY(CallStackEntry cse("Hemv"))
Symv( uplo, alpha, A, x, beta, y, true, ctrl );
}
void Hemv
( UpperOrLower uplo,
T alpha, const ElementalMatrix<T>& A,
const ElementalMatrix<T>& x,
T beta, ElementalMatrix<T>& y,
const SymvCtrl<T>& ctrl )
{
DEBUG_ONLY(CSE cse("Hemv"))
Symv( uplo, alpha, A, x, beta, y, true, ctrl );
}
void Hemv
( UpperOrLower uplo,
T alpha, const AbstractDistMatrix<T>& A,
const AbstractDistMatrix<T>& x,
T beta, AbstractDistMatrix<T>& y,
const SymvCtrl<T>& ctrl )
{
DEBUG_CSE
Symv( uplo, alpha, A, x, beta, y, true, ctrl );
}
// Describes how to run the CLBlast routine
static StatusCode RunRoutine(const Arguments<T> &args, Buffers<T> &buffers, Queue &queue) {
#ifdef OPENCL_API
auto queue_plain = queue();
auto event = cl_event{};
auto status = Symv(args.layout, args.triangle,
args.n, args.alpha,
buffers.a_mat(), args.a_offset, args.a_ld,
buffers.x_vec(), args.x_offset, args.x_inc, args.beta,
buffers.y_vec(), args.y_offset, args.y_inc,
&queue_plain, &event);
if (status == StatusCode::kSuccess) { clWaitForEvents(1, &event); clReleaseEvent(event); }
#elif CUDA_API
auto status = Symv(args.layout, args.triangle,
args.n, args.alpha,
buffers.a_mat(), args.a_offset, args.a_ld,
buffers.x_vec(), args.x_offset, args.x_inc, args.beta,
buffers.y_vec(), args.y_offset, args.y_inc,
queue.GetContext()(), queue.GetDevice()());
cuStreamSynchronize(queue());
#endif
return status;
}
inline void
Hemv
( UpperOrLower uplo,
T alpha, const Matrix<T>& A, const Matrix<T>& x, T beta, Matrix<T>& y )
{
#ifndef RELEASE
PushCallStack("Hemv");
#endif
Symv( uplo, alpha, A, x, beta, y, true );
#ifndef RELEASE
PopCallStack();
#endif
}
