PLASMA  2.4.5
PLASMA - Parallel Linear Algebra for Scalable Multi-core Architectures
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups
Macros | Functions
pssyrk.c File Reference
#include "common.h"
Include dependency graph for pssyrk.c:

Go to the source code of this file.

Macros

#define A(m, n)   BLKADDR(A, float, m, n)
#define C(m, n)   BLKADDR(C, float, m, n)

Functions

void plasma_pssyrk (plasma_context_t *plasma)
void plasma_pssyrk_quark (PLASMA_enum uplo, PLASMA_enum trans, float alpha, PLASMA_desc A, float beta, PLASMA_desc C, PLASMA_sequence *sequence, PLASMA_request *request)

Detailed Description

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

Version:
2.4.5
Author:
Mathieu Faverge
Jakub Kurzak
Date:
2010-11-15 s Tue Nov 22 14:35:39 2011

Definition in file pssyrk.c.

Macro Definition Documentation

#define A (   m,
  n 
)    BLKADDR(A, float, m, n)

Definition at line 18 of file pssyrk.c.

#define C (   m,
  n 
)    BLKADDR(C, float, m, n)

Definition at line 19 of file pssyrk.c.

Function Documentation

void plasma_pssyrk ( plasma_context_t *  plasma)

Parallel tile symmetric rank-k update - static scheduling

Definition at line 23 of file pssyrk.c.

References A, BLKLDD, C, CORE_sgemm(), CORE_ssyrk(), plasma_desc_t::m, plasma_desc_t::mb, plasma_desc_t::mt, plasma_desc_t::n, plasma_desc_t::nb, plasma_desc_t::nt, PLASMA_RANK, PLASMA_SIZE, PLASMA_SUCCESS, plasma_unpack_args_8, PlasmaLower, PlasmaNoTrans, PlasmaTrans, plasma_sequence_t::status, trans, and uplo.

{
PLASMA_enum uplo;
PLASMA_enum trans;
float alpha;
PLASMA_desc A;
float beta;
PLASMA_desc C;
PLASMA_sequence *sequence;
PLASMA_request *request;
int m, n, k;
int next_m;
int next_n;
int ldam, ldan, ldak, ldcm, ldcn;
int tempkn, tempkm, tempmm, tempnn;
float zbeta;
float zone = (float)1.0;
plasma_unpack_args_8(uplo, trans, alpha, A, beta, C, sequence, request);
if (sequence->status != PLASMA_SUCCESS)
return;
n = 0;
m = PLASMA_RANK;
while (m >= C.mt && n < C.nt) {
n++;
m = m-C.mt+n;
}
while (n < C.nt) {
next_n = n;
next_m = m + PLASMA_SIZE;
while (next_m >= C.mt && next_n < C.nt) {
next_n++;
next_m = next_m - C.mt + next_n;
}
tempmm = m == C.mt-1 ? C.m-m*C.mb : C.mb;
tempnn = n == C.nt-1 ? C.n-n*C.nb : C.nb;
if (m == n) {
ldcm = BLKLDD(C, m);
/*
* PlasmaNoTrans
*/
if (trans == PlasmaNoTrans) {
ldam = BLKLDD(A, m);
for (k = 0; k < A.nt; k++) {
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
zbeta = k == 0 ? beta : zone;
CORE_ssyrk(
uplo, trans,
tempnn, tempkn,
alpha, A(m, k), ldam,
zbeta, C(m, n), ldcm);
}
}
/*
* Plasma[Conj]Trans
*/
else {
for (k = 0; k < A.mt; k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
ldak = BLKLDD(A, k);
zbeta = k == 0 ? beta : zone;
CORE_ssyrk(
uplo, trans,
tempnn, tempkm,
alpha, A(k, m), ldak,
zbeta, C(m, n), ldcm);
}
}
}
else {
if (trans == PlasmaNoTrans) {
ldam = BLKLDD(A, m);
ldan = BLKLDD(A, n);
/*
* PlasmaNoTrans / PlasmaLower
*/
if (uplo == PlasmaLower) {
ldcm = BLKLDD(C, m);
for (k = 0; k < A.nt; k++) {
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
zbeta = k == 0 ? beta : zone;
CORE_sgemm(
trans, PlasmaTrans,
tempmm, tempnn, tempkn,
alpha, A(m, k), ldam,
A(n, k), ldan,
zbeta, C(m, n), ldcm);
}
}
/*
* PlasmaNoTrans / PlasmaUpper
*/
else {
ldcn = BLKLDD(C, n);
for (k = 0; k < A.nt; k++) {
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
zbeta = k == 0 ? beta : zone;
CORE_sgemm(
trans, PlasmaTrans,
tempnn, tempmm, tempkn,
alpha, A(n, k), ldan,
A(m, k), ldam,
zbeta, C(n, m), ldcn);
}
}
}
else {
/*
* Plasma[Conj]Trans / PlasmaLower
*/
if (uplo == PlasmaLower) {
ldcm = BLKLDD(C, m);
for (k = 0; k < A.mt; k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
ldak = BLKLDD(A, k);
zbeta = k == 0 ? beta : zone;
CORE_sgemm(
trans, PlasmaNoTrans,
tempmm, tempnn, tempkm,
alpha, A(k, m), ldak,
A(k, n), ldak,
zbeta, C(m, n), ldcm);
}
}
/*
* Plasma[Conj]Trans / PlasmaUpper
*/
else {
ldcn = BLKLDD(C, n);
for (k = 0; k < A.mt; k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
ldak = BLKLDD(A, k);
zbeta = k == 0 ? beta : zone;
CORE_sgemm(
trans, PlasmaNoTrans,
tempnn, tempmm, tempkm,
alpha, A(k, n), ldak,
A(k, m), ldak,
zbeta, C(n, m), ldcn);
}
}
}
}
m = next_m;
n = next_n;
}
}

Here is the call graph for this function:

Here is the caller graph for this function:

void plasma_pssyrk_quark ( PLASMA_enum  uplo,
PLASMA_enum  trans,
float  alpha,
PLASMA_desc  A,
float  beta,
PLASMA_desc  C,
PLASMA_sequence *  sequence,
PLASMA_request *  request 
)

Parallel tile symmetric rank-k update - dynamic scheduling

Definition at line 180 of file pssyrk.c.

References A, BLKLDD, C, plasma_desc_t::m, plasma_desc_t::mb, plasma_desc_t::mt, plasma_desc_t::n, plasma_desc_t::nb, plasma_desc_t::nt, plasma_context_self(), PLASMA_SUCCESS, PlasmaLower, PlasmaNoTrans, PlasmaTrans, plasma_context_struct::quark, QUARK_CORE_sgemm(), QUARK_CORE_ssyrk(), plasma_sequence_t::quark_sequence, QUARK_Task_Flag_Set(), Quark_Task_Flags_Initializer, plasma_sequence_t::status, and TASK_SEQUENCE.

{
plasma_context_t *plasma;
Quark_Task_Flags task_flags = Quark_Task_Flags_Initializer;
int m, n, k;
int ldak, ldam, ldan, ldcm, ldcn;
int tempnn, tempmm, tempkn, tempkm;
float zbeta;
float zone = (float)1.0;
plasma = plasma_context_self();
if (sequence->status != PLASMA_SUCCESS)
return;
QUARK_Task_Flag_Set(&task_flags, TASK_SEQUENCE, (intptr_t)sequence->quark_sequence);
for (n = 0; n < C.nt; n++) {
tempnn = n == C.nt-1 ? C.n-n*C.nb : C.nb;
ldan = BLKLDD(A, n);
ldcn = BLKLDD(C, n);
/*
* PlasmaNoTrans
*/
if (trans == PlasmaNoTrans) {
for (k = 0; k < A.nt; k++) {
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
zbeta = k == 0 ? beta : zone;
QUARK_CORE_ssyrk(
plasma->quark, &task_flags,
uplo, trans,
tempnn, tempkn, A.mb,
alpha, A(n, k), ldan, /* ldan * K */
zbeta, C(n, n), ldcn); /* ldc * N */
}
/*
* PlasmaNoTrans / PlasmaLower
*/
if (uplo == PlasmaLower) {
for (m = n+1; m < C.mt; m++) {
tempmm = m == C.mt-1 ? C.m-m*C.mb : C.mb;
ldam = BLKLDD(A, m);
ldcm = BLKLDD(C, m);
for (k = 0; k < A.nt; k++) {
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
zbeta = k == 0 ? beta : zone;
QUARK_CORE_sgemm(
plasma->quark, &task_flags,
trans, PlasmaTrans,
tempmm, tempnn, tempkn, A.mb,
alpha, A(m, k), ldam, /* ldam * K */
A(n, k), ldan, /* ldan * K */
zbeta, C(m, n), ldcm); /* ldc * N */
}
}
}
/*
* PlasmaNoTrans / PlasmaUpper
*/
else {
for (m = n+1; m < C.mt; m++) {
tempmm = m == C.mt-1 ? C.m-m*C.mb : C.mb;
ldam = BLKLDD(A, m);
for (k = 0; k < A.nt; k++) {
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
zbeta = k == 0 ? beta : zone;
QUARK_CORE_sgemm(
plasma->quark, &task_flags,
trans, PlasmaTrans,
tempnn, tempmm, tempkn, A.mb,
alpha, A(n, k), ldan, /* ldan * K */
A(m, k), ldam, /* ldam * M */
zbeta, C(n, m), ldcn); /* ldc * M */
}
}
}
}
/*
* PlasmaTrans
*/
else {
for (k = 0; k < A.mt; k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
ldak = BLKLDD(A, k);
zbeta = k == 0 ? beta : zone;
QUARK_CORE_ssyrk(
plasma->quark, &task_flags,
uplo, trans,
tempnn, tempkm, A.mb,
alpha, A(k, n), ldak, /* lda * N */
zbeta, C(n, n), ldcn); /* ldc * N */
}
/*
* PlasmaTrans / PlasmaLower
*/
if (uplo == PlasmaLower) {
for (m = n+1; m < C.mt; m++) {
tempmm = m == C.mt-1 ? C.m-m*C.mb : C.mb;
ldcm = BLKLDD(C, m);
for (k = 0; k < A.mt; k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
ldak = BLKLDD(A, k);
zbeta = k == 0 ? beta : zone;
QUARK_CORE_sgemm(
plasma->quark, &task_flags,
trans, PlasmaNoTrans,
tempmm, tempnn, tempkm, A.mb,
alpha, A(k, m), ldak, /* lda * M */
A(k, n), ldak, /* lda * N */
zbeta, C(m, n), ldcm); /* ldc * N */
}
}
}
/*
* PlasmaTrans / PlasmaUpper
*/
else {
for (m = n+1; m < C.mt; m++) {
tempmm = m == C.mt-1 ? C.m-m*C.mb : C.mb;
for (k = 0; k < A.mt; k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
ldak = BLKLDD(A, k);
zbeta = k == 0 ? beta : zone;
QUARK_CORE_sgemm(
plasma->quark, &task_flags,
trans, PlasmaNoTrans,
tempnn, tempmm, tempkm, A.mb,
alpha, A(k, n), ldak, /* lda * K */
A(k, m), ldak, /* lda * M */
zbeta, C(n, m), ldcn); /* ldc * M */
}
}
}
}
}
}

Here is the call graph for this function: