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

Go to the source code of this file.

Functions

void plasma_pcgerbb (plasma_context_t *plasma)
void plasma_pcgerbb_quark (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 c Tue Nov 22 14:35:43 2011

Definition in file pcgerbb.c.

Function Documentation

void plasma_pcgerbb ( plasma_context_t *  plasma)

Parallel tile BAND Bidiagonal Reduction - dynamic scheduler Could be optimized by using the algorithms from Trefethen book

WARNING: do never call this function because unmqr and unmlq are not implementing all the cases required in static.

Definition at line 26 of file pcgerbb.c.

References A, 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_desc_submatrix(), plasma_pcgelqf(), plasma_pcgeqrf(), plasma_pcunmlq(), plasma_pcunmqr(), plasma_static_call_4, plasma_static_call_7, PLASMA_SUCCESS, plasma_unpack_args_4, PlasmaConjTrans, PlasmaLeft, PlasmaRight, plasma_sequence_t::status, and T.

{
PLASMA_desc A;
PLASMA_desc T;
PLASMA_sequence *sequence;
PLASMA_request *request;
int k;
int tempkm, tempkn;
plasma_unpack_args_4(A, T, sequence, request);
if (sequence->status != PLASMA_SUCCESS)
return;
if (A.m >= A.n){
for (k = 0; k < A.nt; k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
plasma_static_call_4(plasma_pcgeqrf,
PLASMA_desc, plasma_desc_submatrix(A, k*A.mb, k*A.nb, A.m-k*A.mb, tempkn),
PLASMA_desc, plasma_desc_submatrix(T, k*T.mb, k*T.nb, T.m-k*T.mb, tempkn),
PLASMA_sequence*, sequence,
PLASMA_request*, request);
plasma_static_call_7(plasma_pcunmqr,
PLASMA_enum, PlasmaLeft,
PLASMA_enum, PlasmaConjTrans,
PLASMA_desc, plasma_desc_submatrix(A, k*A.mb, k*A.nb, A.m-k*A.mb, tempkn),
PLASMA_desc, plasma_desc_submatrix(A, k*A.mb, (k+1)*A.nb, A.m-k*A.mb, A.n-(k+1)*A.nb),
PLASMA_desc, plasma_desc_submatrix(T, k*T.mb, k*T.nb, T.m-k*T.mb, tempkn),
PLASMA_sequence*, sequence,
PLASMA_request*, request);
if (k+1 < A.nt){
tempkn = k+1 == A.nt-1 ? A.n-(k+1)*A.nb : A.nb;
plasma_static_call_4(plasma_pcgelqf,
PLASMA_desc, plasma_desc_submatrix(A, k*A.mb, (k+1)*A.nb, tempkm, A.n-(k+1)*A.nb),
PLASMA_desc, plasma_desc_submatrix(T, k*T.mb, (k+1)*T.nb, tempkm, T.n-(k+1)*T.nb),
PLASMA_sequence*, sequence,
PLASMA_request*, request);
plasma_static_call_7(plasma_pcunmlq,
PLASMA_enum, PlasmaRight,
PLASMA_enum, PlasmaConjTrans,
PLASMA_desc, plasma_desc_submatrix(A, k*A.mb, (k+1)*A.nb, tempkm, A.n-(k+1)*A.nb),
PLASMA_desc, plasma_desc_submatrix(A, (k+1)*A.mb, (k+1)*A.nb, A.m-(k+1)*A.mb, A.n-(k+1)*A.nb),
PLASMA_desc, plasma_desc_submatrix(T, k*T.mb, (k+1)*T.nb, tempkm, T.n-(k+1)*T.nb),
PLASMA_sequence*, sequence,
PLASMA_request*, request);
}
}
}
else{
for (k = 0; k < A.mt; k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
plasma_static_call_4(plasma_pcgelqf,
PLASMA_desc, plasma_desc_submatrix(A, k*A.mb, k*A.nb, tempkm, A.n-k*A.nb),
PLASMA_desc, plasma_desc_submatrix(T, k*T.mb, k*T.nb, tempkm, T.n-k*T.nb),
PLASMA_sequence*, sequence,
PLASMA_request*, request);
plasma_static_call_7(plasma_pcunmlq,
PLASMA_enum, PlasmaRight,
PLASMA_enum, PlasmaConjTrans,
PLASMA_desc, plasma_desc_submatrix(A, k*A.mb, k*A.nb, tempkm, A.n-k*A.nb),
PLASMA_desc, plasma_desc_submatrix(A, (k+1)*A.mb, k*A.nb, A.m-(k+1)*A.mb, A.n-k*A.nb),
PLASMA_desc, plasma_desc_submatrix(T, k*T.mb, k*T.nb, tempkm, T.n-k*T.nb),
PLASMA_sequence*, sequence,
PLASMA_request*, request);
if (k+1 < A.mt){
tempkm = k+1 == A.mt-1 ? A.m-(k+1)*A.mb : A.mb;
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
plasma_static_call_4(plasma_pcgeqrf,
PLASMA_desc, plasma_desc_submatrix(A, (k+1)*A.mb, k*A.nb, A.m-(k+1)*A.mb, tempkn),
PLASMA_desc, plasma_desc_submatrix(T, (k+1)*T.mb, k*T.nb, T.m-(k+1)*T.mb, tempkn),
PLASMA_sequence*, sequence,
PLASMA_request*, request);
plasma_static_call_7(plasma_pcunmqr,
PLASMA_enum, PlasmaLeft,
PLASMA_enum, PlasmaConjTrans,
PLASMA_desc, plasma_desc_submatrix(A, (k+1)*A.mb, k*A.nb, A.m-(k+1)*A.mb, tempkn),
PLASMA_desc, plasma_desc_submatrix(A, (k+1)*A.mb, (k+1)*A.nb, A.m-(k+1)*A.mb, A.n-(k+1)*A.nb),
PLASMA_desc, plasma_desc_submatrix(T, (k+1)*T.mb, k*T.nb, T.m-(k+1)*T.mb, tempkn),
PLASMA_sequence*, sequence,
PLASMA_request*, request);
}
}
}
}

Here is the call graph for this function:

Here is the caller graph for this function:

void plasma_pcgerbb_quark ( PLASMA_desc  A,
PLASMA_desc  T,
PLASMA_sequence *  sequence,
PLASMA_request *  request 
)

Parallel tile BAND Bidiagonal Reduction - dynamic scheduler Could be optimized by using the algorithms from Trefethen book

Definition at line 127 of file pcgerbb.c.

References 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_desc_submatrix(), plasma_pcgelqf_quark(), plasma_pcgeqrf_quark(), plasma_pcunmlq_quark(), plasma_pcunmqr_quark(), PlasmaConjTrans, PlasmaLeft, and PlasmaRight.

{
int k;
int tempkm, tempkn;
if (A.m >= A.n){
for (k = 0; k < A.nt; k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
plasma_pcgeqrf_quark(
plasma_desc_submatrix(A, k*A.mb, k*A.nb, A.m-k*A.mb, tempkn),
plasma_desc_submatrix(T, k*T.mb, k*T.nb, T.m-k*T.mb, tempkn),
sequence, request);
plasma_pcunmqr_quark(
PlasmaLeft,
PlasmaConjTrans,
plasma_desc_submatrix(A, k*A.mb, k*A.nb, A.m-k*A.mb, tempkn),
plasma_desc_submatrix(A, k*A.mb, (k+1)*A.nb, A.m-k*A.mb, A.n-(k+1)*A.nb),
plasma_desc_submatrix(T, k*T.mb, k*T.nb, T.m-k*T.mb, tempkn),
sequence, request);
if (k+1 < A.nt){
tempkn = k+1 == A.nt-1 ? A.n-(k+1)*A.nb : A.nb;
plasma_pcgelqf_quark(
plasma_desc_submatrix(A, k*A.mb, (k+1)*A.nb, tempkm, A.n-(k+1)*A.nb),
plasma_desc_submatrix(T, k*T.mb, (k+1)*T.nb, tempkm, T.n-(k+1)*T.nb),
sequence, request);
plasma_pcunmlq_quark(
PlasmaRight, PlasmaConjTrans,
plasma_desc_submatrix(A, k*A.mb, (k+1)*A.nb, tempkm, A.n-(k+1)*A.nb),
plasma_desc_submatrix(A, (k+1)*A.mb, (k+1)*A.nb, A.m-(k+1)*A.mb, A.n-(k+1)*A.nb),
plasma_desc_submatrix(T, k*T.mb, (k+1)*T.nb, tempkm, T.n-(k+1)*T.nb),
sequence, request);
}
}
}
else{
for (k = 0; k < A.mt; k++) {
tempkm = k == A.mt-1 ? A.m-k*A.mb : A.mb;
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
plasma_pcgelqf_quark(
plasma_desc_submatrix(A, k*A.mb, k*A.nb, tempkm, A.n-k*A.nb),
plasma_desc_submatrix(T, k*T.mb, k*T.nb, tempkm, T.n-k*T.nb),
sequence, request);
plasma_pcunmlq_quark(
PlasmaRight, PlasmaConjTrans,
plasma_desc_submatrix(A, k*A.mb, k*A.nb, tempkm, A.n-k*A.nb),
plasma_desc_submatrix(A, (k+1)*A.mb, k*A.nb, A.m-(k+1)*A.mb, A.n-k*A.nb),
plasma_desc_submatrix(T, k*T.mb, k*T.nb, tempkm, T.n-k*T.nb),
sequence, request);
if (k+1 < A.mt){
tempkm = k+1 == A.mt-1 ? A.m-(k+1)*A.mb : A.mb;
tempkn = k == A.nt-1 ? A.n-k*A.nb : A.nb;
plasma_pcgeqrf_quark(
plasma_desc_submatrix(A, (k+1)*A.mb, k*A.nb, A.m-(k+1)*A.mb, tempkn),
plasma_desc_submatrix(T, (k+1)*T.mb, k*T.nb, T.m-(k+1)*T.mb, tempkn),
sequence, request);
plasma_pcunmqr_quark(
PlasmaLeft, PlasmaConjTrans,
plasma_desc_submatrix(A, (k+1)*A.mb, k*A.nb, A.m-(k+1)*A.mb, tempkn),
plasma_desc_submatrix(A, (k+1)*A.mb, (k+1)*A.nb, A.m-(k+1)*A.mb, A.n-(k+1)*A.nb),
plasma_desc_submatrix(T, (k+1)*T.mb, k*T.nb, T.m-(k+1)*T.mb, tempkn),
sequence, request);
}
}
}
}

Here is the call graph for this function: