core_zplgsy.c File Reference

#include "common.h"

Include dependency graph for core_zplgsy.c:

Go to the source code of this file.

Macros
#define	COMPLEX
#define	Rnd64_A   6364136223846793005ULL
#define	Rnd64_C   1ULL
#define	RndF_Mul   5.4210108624275222e-20f
#define	RndD_Mul   5.4210108624275222e-20
#define	NBELEM   2

Functions
void	CORE_zplgsy (PLASMA_Complex64_t bump, int m, int n, PLASMA_Complex64_t *A, int lda, int bigM, int m0, int n0, unsigned long long int seed)
void	QUARK_CORE_zplgsy (Quark *quark, Quark_Task_Flags *task_flags, PLASMA_Complex64_t bump, int m, int n, PLASMA_Complex64_t *A, int lda, int bigM, int m0, int n0, unsigned long long int seed)
void	CORE_zplgsy_quark (Quark *quark)

---

Detailed Description

PLASMA core_blas kernel PLASMA is a software package provided by Univ. of Tennessee, Univ. of California Berkeley and Univ. of Colorado Denver

Version:

2.4.5

Author:

Piotr Luszczek

Pierre Lemarinier

Mathieu Faverge

Date:

2010-11-15 normal z -> c d s

Definition in file core_zplgsy.c.

---

Macro Definition Documentation

#define COMPLEX

Definition at line 19 of file core_zplgsy.c.

#define NBELEM   2

Definition at line 36 of file core_zplgsy.c.

#define Rnd64_A   6364136223846793005ULL

Definition at line 30 of file core_zplgsy.c.

#define Rnd64_C   1ULL

Definition at line 31 of file core_zplgsy.c.

#define RndD_Mul   5.4210108624275222e-20

Definition at line 33 of file core_zplgsy.c.

#define RndF_Mul   5.4210108624275222e-20f

Definition at line 32 of file core_zplgsy.c.

---

Function Documentation

void CORE_zplgsy	(	PLASMA_Complex64_t	bump,
		int	m,
		int	n,
		PLASMA_Complex64_t *	A,
		int	lda,
		int	bigM,
		int	m0,
		int	n0,
		unsigned long long int	seed
	)

Definition at line 64 of file core_zplgsy.c.

References A, NBELEM, Rnd64_A, Rnd64_C, and RndF_Mul.

{
    PLASMA_Complex64_t *tmp = A;
    int64_t i, j;
    unsigned long long int ran, jump;

    jump = m0 + n0 * bigM;

    /*
     * Tile diagonal
     */
    if ( m0 == n0 ) {
        for (j = 0; j < n; j++) {
            ran = Rnd64_jump( NBELEM * jump, seed );

            for (i = j; i < m; i++) {
                *tmp = 0.5f - ran * RndF_Mul;
                ran  = Rnd64_A * ran + Rnd64_C;
#ifdef COMPLEX
                *tmp += I*(0.5f - ran * RndF_Mul);
                ran   = Rnd64_A * ran + Rnd64_C;
#endif
                tmp++;
            }
            tmp  += (lda - i + j + 1);
            jump += bigM + j;
        }

        for (j = 0; j < n; j++) {
            A[j+j*lda] += bump;

            for (i=0; i<j; i++) {
                A[lda*j+i] = A[lda*i+j];
            }
        }
    } 
    /*
     * Lower part
     */
    else if ( m0 > n0 ) {
        for (j = 0; j < n; j++) {
            ran = Rnd64_jump( NBELEM * jump, seed );

            for (i = 0; i < m; i++) {
                *tmp = 0.5f - ran * RndF_Mul;
                ran  = Rnd64_A * ran + Rnd64_C;
#ifdef COMPLEX
                *tmp += I*(0.5f - ran * RndF_Mul);
                ran   = Rnd64_A * ran + Rnd64_C;
#endif
                tmp++;
            }
            tmp  += (lda - i);
            jump += bigM;
        }
    }
    /*
     * Upper part
     */
    else if ( m0 < n0 ) {
        /* Overwrite jump */
        jump = n0 + m0 * bigM;

        for (i = 0; i < m; i++) {
            ran = Rnd64_jump( NBELEM * jump, seed );

            for (j = 0; j < n; j++) {
                A[j*lda+i] = 0.5f - ran * RndF_Mul;
                ran = Rnd64_A * ran + Rnd64_C;
#ifdef COMPLEX
                A[j*lda+i] += I*(0.5f - ran * RndF_Mul);
                ran = Rnd64_A * ran + Rnd64_C;
#endif
            }
            jump += bigM;
        }
    }
}

Here is the caller graph for this function:

void CORE_zplgsy_quark

(

Quark *

quark

)

Definition at line 172 of file core_zplgsy.c.

References A, CORE_zplgsy(), and quark_unpack_args_9.

{
    PLASMA_Complex64_t bump;
    int m;
    int n;
    PLASMA_Complex64_t *A;
    int lda;
    int bigM;
    int m0;
    int n0;
    unsigned long long int seed;

    quark_unpack_args_9( quark, bump, m, n, A, lda, bigM, m0, n0, seed );
    CORE_zplgsy( bump, m, n, A, lda, bigM, m0, n0, seed );
}

Here is the call graph for this function:

Here is the caller graph for this function:

void QUARK_CORE_zplgsy	(	Quark *	quark,
		Quark_Task_Flags *	task_flags,
		PLASMA_Complex64_t	bump,
		int	m,
		int	n,
		PLASMA_Complex64_t *	A,
		int	lda,
		int	bigM,
		int	m0,
		int	n0,
		unsigned long long int	seed
	)

Definition at line 147 of file core_zplgsy.c.

References CORE_zplgsy_quark(), DAG_CORE_PLGSY, OUTPUT, QUARK_Insert_Task(), and VALUE.

{
    DAG_CORE_PLGSY;
    QUARK_Insert_Task(quark, CORE_zplgsy_quark, task_flags,
        sizeof(PLASMA_Complex64_t),       &bump, VALUE,
        sizeof(int),                      &m,    VALUE,
        sizeof(int),                      &n,    VALUE,
        sizeof(PLASMA_Complex64_t)*lda*n, A,         OUTPUT,
        sizeof(int),                      &lda,  VALUE,
        sizeof(int),                      &bigM, VALUE,
        sizeof(int),                      &m0,   VALUE,
        sizeof(int),                      &n0,   VALUE,
        sizeof(unsigned long long int),   &seed, VALUE,
        0);
}

Here is the call graph for this function:

Here is the caller graph for this function: