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
pdsyrbt.c File Reference
#include "common.h"
Include dependency graph for pdsyrbt.c:

Go to the source code of this file.

Macros

#define A(m, n)   BLKADDR(A, double, m, n)
#define T(m, n)   BLKADDR(T, double, m, n)

Functions

void plasma_pdsyrbt_quark (PLASMA_enum uplo, PLASMA_desc A, PLASMA_desc T, 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:
Hatem Ltaief
Azzam Haidar
Date:
2010-11-15 d Tue Nov 22 14:35:43 2011

Definition in file pdsyrbt.c.

Macro Definition Documentation

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

Definition at line 18 of file pdsyrbt.c.

#define T (   m,
  n 
)    BLKADDR(T, double, m, n)

Definition at line 19 of file pdsyrbt.c.

Function Documentation

void plasma_pdsyrbt_quark ( PLASMA_enum  uplo,
PLASMA_desc  A,
PLASMA_desc  T,
PLASMA_sequence *  sequence,
PLASMA_request *  request 
)

Parallel tile BAND Tridiagonal Reduction - dynamic scheduler

Definition at line 23 of file pdsyrbt.c.

References A, BLKLDD, 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_IB, PLASMA_SUCCESS, PlasmaLeft, PlasmaLower, PlasmaNoTrans, PlasmaRight, PlasmaTrans, PlasmaUpper, plasma_context_struct::quark, QUARK_CORE_dgelqt(), QUARK_CORE_dgeqrt(), QUARK_CORE_dormlq(), QUARK_CORE_dormqr(), QUARK_CORE_dsyrfb(), QUARK_CORE_dtslqt(), QUARK_CORE_dtsmlq(), QUARK_CORE_dtsmlq_corner(), QUARK_CORE_dtsmlq_sytra1(), QUARK_CORE_dtsmqr(), QUARK_CORE_dtsmqr_corner(), QUARK_CORE_dtsmqr_sytra1(), QUARK_CORE_dtsqrt(), plasma_sequence_t::quark_sequence, QUARK_Task_Flag_Set(), Quark_Task_Flags_Initializer, plasma_sequence_t::status, T, and TASK_SEQUENCE.

{
plasma_context_t *plasma;
Quark_Task_Flags task_flags = Quark_Task_Flags_Initializer;
int k, m, n, i, j;
int ldak, ldam, ldan, ldaj, ldai;
int tempkn, tempmm, tempnn, tempjj;
int ib;
plasma = plasma_context_self();
if (sequence->status != PLASMA_SUCCESS)
return;
QUARK_Task_Flag_Set(&task_flags, TASK_SEQUENCE, (intptr_t)sequence->quark_sequence);
ib = PLASMA_IB;
if (uplo == PlasmaLower) {
for (k = 0; k < A.nt-1; k++){
tempkn = k+1 == A.nt-1 ? A.n-(k+1)*A.nb : A.nb;
ldak = BLKLDD(A, k+1);
QUARK_CORE_dgeqrt(
plasma->quark, &task_flags,
tempkn, A.nb, ib, T.nb,
A(k+1, k), ldak,
T(k+1, k), T.mb);
/* LEFT and RIGHT on the symmetric diagonal block */
QUARK_CORE_dsyrfb(
plasma->quark, &task_flags,
PlasmaLower,
tempkn, tempkn, ib, T.nb,
A(k+1, k), ldak,
T(k+1, k), T.mb,
A(k+1, k+1), ldak);
/* RIGHT on the remaining tiles until the bottom */
for (m = k+2; m < A.mt ; m++) {
tempmm = m == A.mt-1 ? A.m-m*A.mb : A.mb;
ldam = BLKLDD(A, m);
QUARK_CORE_dormqr(
plasma->quark, &task_flags,
PlasmaRight, PlasmaNoTrans,
tempmm, A.nb, tempkn, ib, T.nb,
A(k+1, k), ldak,
T(k+1, k), T.mb,
A(m , k+1), ldam);
}
for (m = k+2; m < A.mt; m++) {
tempmm = m == A.mt-1 ? A.m-m*A.mb : A.mb;
ldam = BLKLDD(A, m);
QUARK_CORE_dtsqrt(
plasma->quark, &task_flags,
tempmm, A.nb, ib, T.nb,
A(k+1, k), ldak,
A(m , k), ldam,
T(m , k), T.mb);
/* LEFT */
for (i = k+2; i < m; i++) {
ldai = BLKLDD(A, i);
QUARK_CORE_dtsmqr_sytra1(
plasma->quark, &task_flags,
PlasmaLeft, PlasmaTrans,
A.mb, A.nb, tempmm, A.nb, A.nb, ib, T.nb,
A(i, k+1), ldai,
A(m, i), ldam,
A(m, k), ldam,
T(m, k), T.mb);
}
/* RIGHT */
for (j = m+1; j < A.mt ; j++) {
tempjj = j == A.mt-1 ? A.m-j*A.mb : A.mb;
ldaj = BLKLDD(A, j);
QUARK_CORE_dtsmqr(
plasma->quark, &task_flags,
PlasmaRight, PlasmaNoTrans,
tempjj, A.nb, tempjj, tempmm, A.nb, ib, T.nb,
A(j, k+1), ldaj,
A(j, m), ldaj,
A(m, k), ldam,
T(m, k), T.mb);
}
/* LEFT->RIGHT */
QUARK_CORE_dtsmqr_corner(
plasma->quark, &task_flags,
A.nb, A.nb, tempmm, A.nb, tempmm, tempmm, A.nb, ib, T.nb,
A(k+1, k+1), ldak,
A(m , k+1), ldam,
A(m , m), ldam,
A(m , k), ldam,
T(m , k), T.mb);
}
}
}
else {
for (k = 0; k < A.nt-1; k++){
tempkn = k+1 == A.nt-1 ? A.n-(k+1)*A.nb : A.nb;
ldak = BLKLDD(A, k+1);
QUARK_CORE_dgelqt(
plasma->quark, &task_flags,
A.nb, tempkn, ib, T.nb,
A(k, k+1), A.nb,
T(k, k+1), T.mb);
/* RIGHT and LEFT on the symmetric diagonal block */
QUARK_CORE_dsyrfb(
plasma->quark, &task_flags,
PlasmaUpper,
tempkn, tempkn, ib, T.nb,
A(k, k+1), A.nb,
T(k, k+1), T.mb,
A(k+1, k+1), ldak);
/* LEFT on the remaining tiles until the left side */
for (n = k+2; n < A.nt ; n++) {
tempnn = n == A.nt-1 ? A.n-n*A.nb : A.nb;
QUARK_CORE_dormlq(
plasma->quark, &task_flags,
PlasmaLeft, PlasmaNoTrans,
A.nb, tempnn, tempkn, ib, T.nb,
A(k, k+1), A.nb,
T(k, k+1), T.mb,
A(k+1, n), ldak);
}
for (n = k+2; n < A.nt; n++) {
tempnn = n == A.nt-1 ? A.n-n*A.nb : A.nb;
ldan = BLKLDD(A, n);
QUARK_CORE_dtslqt(
plasma->quark, &task_flags,
A.nb, tempnn, ib, T.nb,
A(k, k+1), A.nb,
A(k, n), A.nb,
T(k, n), T.mb);
/* RIGHT */
for (i = k+2; i < n; i++) {
ldai = BLKLDD(A, i);
QUARK_CORE_dtsmlq_sytra1(
plasma->quark, &task_flags,
PlasmaRight, PlasmaTrans,
A.mb, A.nb, A.nb, tempnn, A.nb, ib, T.nb,
A(k+1, i), ldak,
A(i, n), ldai,
A(k, n), A.nb,
T(k, n), T.mb);
}
/* LEFT */
for (j = n+1; j < A.nt ; j++) {
tempjj = j == A.nt-1 ? A.n-j*A.nb : A.nb;
ldaj = BLKLDD(A, j);
QUARK_CORE_dtsmlq(
plasma->quark, &task_flags,
PlasmaLeft, PlasmaNoTrans,
A.nb, tempjj, tempnn, tempjj, A.nb, ib, T.nb,
A(k+1, j), ldak,
A(n, j), ldan,
A(k, n), A.nb,
T(k, n), T.mb);
}
/* RIGHT->LEFT */
QUARK_CORE_dtsmlq_corner(
plasma->quark, &task_flags,
A.nb, A.nb, A.nb, tempnn, tempnn, tempnn, A.nb, ib, T.nb,
A(k+1, k+1), ldak,
A(k+1, n), ldak,
A(n , n), ldan,
A(k , n), A.nb,
T(k , n), T.mb);
}
}
}
}

Here is the call graph for this function: