#include <malloc_quda.h>

Include dependency graph for blas_cublas.cu:

Namespaces
	quda

	quda::cublas

Macros
#define	FMULS_GETRF(m_, n_)

#define	FADDS_GETRF(m_, n_)

#define	FLOPS_ZGETRF(m_, n_) (6. * FMULS_GETRF((double)(m_), (double)(n_)) + 2.0 * FADDS_GETRF((double)(m_), (double)(n_)) )

#define	FLOPS_CGETRF(m_, n_) (6. * FMULS_GETRF((double)(m_), (double)(n_)) + 2.0 * FADDS_GETRF((double)(m_), (double)(n_)) )

#define	FMULS_GETRI(n_) ( (n_) * ((5. / 6.) + (n_) * ((2. / 3.) * (n_) + 0.5)) )

#define	FADDS_GETRI(n_) ( (n_) * ((5. / 6.) + (n_) * ((2. / 3.) * (n_) - 1.5)) )

#define	FLOPS_ZGETRI(n_) (6. * FMULS_GETRI((double)(n_)) + 2.0 * FADDS_GETRI((double)(n_)) )

#define	FLOPS_CGETRI(n_) (6. * FMULS_GETRI((double)(n_)) + 2.0 * FADDS_GETRI((double)(n_)) )

Functions
void	quda::cublas::init ()
	Create the CUBLAS context. More...

void	quda::cublas::destroy ()
	Destroy the CUBLAS context. More...

template<typename T >
__global__ void	quda::cublas::set_pointer (T *output_array_a, T input_a, T *output_array_b, T input_b, int batch_offset)

long long	quda::cublas::BatchInvertMatrix (void Ainv, void A, const int n, const int batch, QudaPrecision precision, QudaFieldLocation location)

Macro Definition Documentation

◆ FADDS_GETRF

#define FADDS_GETRF	(	m_,
		n_
	)

Value:

( ((m_) < (n_)) \
    ? (0.5 * (m_) * ((m_) * ((n_) - (1./3.) * (m_)      ) - (n_)) + (1. / 6.) * (m_)) \
    : (0.5 * (n_) * ((n_) * ((m_) - (1./3.) * (n_)      ) - (m_)) + (1. / 6.) * (n_)) )

Definition at line 10 of file blas_cublas.cu.

◆ FADDS_GETRI

#define FADDS_GETRI ( n_ ) ( (n_) * ((5. / 6.) + (n_) * ((2. / 3.) * (n_) - 1.5)) )

Definition at line 18 of file blas_cublas.cu.

◆ FLOPS_CGETRF

#define FLOPS_CGETRF	(	m_,
		n_
	)	(6. * FMULS_GETRF((double)(m_), (double)(n_)) + 2.0 * FADDS_GETRF((double)(m_), (double)(n_)) )

Definition at line 15 of file blas_cublas.cu.

Referenced by quda::cublas::BatchInvertMatrix().

◆ FLOPS_CGETRI

#define FLOPS_CGETRI ( n_ ) (6. * FMULS_GETRI((double)(n_)) + 2.0 * FADDS_GETRI((double)(n_)) )

Definition at line 21 of file blas_cublas.cu.

Referenced by quda::cublas::BatchInvertMatrix().

◆ FLOPS_ZGETRF

#define FLOPS_ZGETRF	(	m_,
		n_
	)	(6. * FMULS_GETRF((double)(m_), (double)(n_)) + 2.0 * FADDS_GETRF((double)(m_), (double)(n_)) )

Definition at line 14 of file blas_cublas.cu.

◆ FLOPS_ZGETRI

#define FLOPS_ZGETRI ( n_ ) (6. * FMULS_GETRI((double)(n_)) + 2.0 * FADDS_GETRI((double)(n_)) )

Definition at line 20 of file blas_cublas.cu.

◆ FMULS_GETRF

#define FMULS_GETRF	(	m_,
		n_
	)

Value:

( ((m_) < (n_)) \
    ? (0.5 * (m_) * ((m_) * ((n_) - (1./3.) * (m_) - 1. ) + (n_)) + (2. / 3.) * (m_)) \
    : (0.5 * (n_) * ((n_) * ((m_) - (1./3.) * (n_) - 1. ) + (m_)) + (2. / 3.) * (n_)) )

Definition at line 7 of file blas_cublas.cu.

◆ FMULS_GETRI

#define FMULS_GETRI ( n_ ) ( (n_) * ((5. / 6.) + (n_) * ((2. / 3.) * (n_) + 0.5)) )

Definition at line 17 of file blas_cublas.cu.

Namespaces

Macros

Functions