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
core_cttqrt.c File Reference
#include <lapacke.h>
#include "common.h"
Include dependency graph for core_cttqrt.c:

Go to the source code of this file.

Macros

#define COMPLEX

Functions

int CORE_cttqrt (int M, int N, int IB, PLASMA_Complex32_t *A1, int LDA1, PLASMA_Complex32_t *A2, int LDA2, PLASMA_Complex32_t *T, int LDT, PLASMA_Complex32_t *TAU, PLASMA_Complex32_t *WORK)
void QUARK_CORE_cttqrt (Quark *quark, Quark_Task_Flags *task_flags, int m, int n, int ib, int nb, PLASMA_Complex32_t *A1, int lda1, PLASMA_Complex32_t *A2, int lda2, PLASMA_Complex32_t *T, int ldt)
void CORE_cttqrt_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:
Hatem Ltaief
Mathieu Faverge
Dulceneia Becker
Date:
2010-11-15 c Tue Nov 22 14:35:17 2011

Definition in file core_cttqrt.c.

Macro Definition Documentation

#define COMPLEX

Definition at line 20 of file core_cttqrt.c.

Function Documentation

int CORE_cttqrt ( int  M,
int  N,
int  IB,
PLASMA_Complex32_t *  A1,
int  LDA1,
PLASMA_Complex32_t *  A2,
int  LDA2,
PLASMA_Complex32_t *  T,
int  LDT,
PLASMA_Complex32_t *  TAU,
PLASMA_Complex32_t *  WORK 
)

CORE_cttqrt computes a QR factorization of a rectangular matrix formed by coupling a complex N-by-N upper triangular tile A1 on top of a complex M-by-N upper trapezoidal tile A2:

| A1 | = Q * R | A2 |

The tile Q is represented as a product of elementary reflectors

Q = H(1) H(2) . . . H(k), where k = min(M,N).

Each H(i) has the form

H(i) = I - tau * v * v'

where tau is a complex scalar, and v is a complex vector with v(1:i-1) = 0 and v(i) = 1; v(i+1:m) is stored on exit in A2(1:m,i), and tau in TAU(i).

Parameters:
[in]MThe number of rows of the tile A2. M >= 0.
[in]NThe number of columns of the tile A1 and A2. N >= 0.
[in]IBThe inner-blocking size. IB >= 0.
[in,out]A1On entry, the N-by-N tile A1. On exit, the elements on and above the diagonal of the array contain the N-by-N upper trapezoidal tile R; the elements below the diagonal are not referenced.
[in]LDA1The leading dimension of the array A1. LDA1 >= max(1,N).
[in,out]A2On entry, the M-by-N upper triangular tile A2. On exit, the elements on and above the diagonal of the array with the array TAU, represent the unitary tile Q as a product of elementary reflectors (see Further Details).
[in]LDA2The leading dimension of the array A2. LDA2 >= max(1,M).
[out]TThe IB-by-N triangular factor T of the block reflector. T is upper triangular by block (economic storage); The rest of the array is not referenced.
[in]LDTThe leading dimension of the array T. LDT >= IB.
[out]TAUThe scalar factors of the elementary reflectors (see Further Details).
[in,out]WORK
Returns:
Return values:
PLASMA_SUCCESSsuccessful exit
<0if -i, the i-th argument had an illegal value

Definition at line 100 of file core_cttqrt.c.

References cblas_caxpy(), cblas_ccopy(), cblas_cgemv(), cblas_cgerc(), cblas_ctrmv(), CBLAS_SADDR, CblasColMajor, CORE_claset(), CORE_cparfb(), CORE_cpemv(), coreblas_error, max, min, PLASMA_SUCCESS, PlasmaColumnwise, PlasmaConjTrans, PlasmaForward, PlasmaLeft, PlasmaNonUnit, PlasmaNoTrans, PlasmaUpper, and PlasmaUpperLower.

{
static int ione = 1;
static PLASMA_Complex32_t zone = 1.0;
static PLASMA_Complex32_t zzero = 0.0;
PLASMA_Complex32_t alpha;
int i, j, l, ii, sb, mi, ni;
/* Check input arguments */
if (M < 0) {
coreblas_error(1, "Illegal value of M");
return -1;
}
if (N < 0) {
coreblas_error(2, "Illegal value of N");
return -2;
}
if (IB < 0) {
coreblas_error(3, "Illegal value of IB");
return -3;
}
if ((LDA2 < max(1,M)) && (M > 0)) {
coreblas_error(7, "Illegal value of LDA2");
return -7;
}
/* Quick return */
if ((M == 0) || (N == 0) || (IB == 0))
return PLASMA_SUCCESS;
/* TODO: Need to check why some cases require
* this to not have uninitialized values */
CORE_claset( PlasmaUpperLower, IB, N,
0., 0., T, LDT);
for (ii = 0; ii < N; ii += IB) {
sb = min(N-ii, IB);
for (i = 0; i < sb; i++) {
j = ii + i;
mi = min( j + 1, M );
ni = sb-i-1;
/*
* Generate elementary reflector H( II*IB+I ) to annihilate
* A( II*IB+I:mi, II*IB+I ).
*/
LAPACKE_clarfg_work(
mi+1, &A1[LDA1*j+j], &A2[LDA2*j], ione, &TAU[j]);
if (ni > 0) {
/*
* Apply H( II*IB+I ) to A( II*IB+I:M, II*IB+I+1:II*IB+IB ) from the left.
*/
cblas_ccopy(
ni,
&A1[LDA1*(j+1)+j], LDA1,
WORK, 1);
#ifdef COMPLEX
LAPACKE_clacgv_work(ni, WORK, 1);
#endif
cblas_cgemv(
CblasColMajor, (CBLAS_TRANSPOSE)PlasmaConjTrans,
mi, ni,
CBLAS_SADDR(zone), &A2[LDA2*(j+1)], LDA2,
&A2[LDA2*j], 1,
CBLAS_SADDR(zone), WORK, 1);
#ifdef COMPLEX
LAPACKE_clacgv_work(ni, WORK, 1);
#endif
alpha = -conjf(TAU[j]);
cblas_caxpy(
ni, CBLAS_SADDR(alpha),
WORK, 1,
&A1[LDA1*(j+1)+j], LDA1);
#ifdef COMPLEX
LAPACKE_clacgv_work(ni, WORK, 1);
#endif
cblas_cgerc(
CblasColMajor, mi, ni,
CBLAS_SADDR(alpha), &A2[LDA2*j], 1,
WORK, 1,
&A2[LDA2*(j+1)], LDA2);
}
/*
* Calculate T
*
* T(0:i-1, j) = alpha * A2(0:M-1, ii:j-1)' * A2(0:M-1, j)
*/
if ( i > 0 ) {
l = min(i, max(0, M-ii));
alpha = -(TAU[j]);
CORE_cpemv(
PlasmaConjTrans, PlasmaColumnwise,
min(j, M), i, l,
alpha, &A2[LDA2*ii], LDA2,
&A2[LDA2*j], 1,
zzero, &T[LDT*j], 1,
WORK);
/* T(0:i-1, j) = T(0:i-1, ii:j-1) * T(0:i-1, j) */
cblas_ctrmv(
CblasColMajor, (CBLAS_UPLO)PlasmaUpper,
(CBLAS_TRANSPOSE)PlasmaNoTrans,
(CBLAS_DIAG)PlasmaNonUnit,
i, &T[LDT*ii], LDT,
&T[LDT*j], 1);
}
T[LDT*j+i] = TAU[j];
}
/* Apply Q' to the rest of the matrix to the left */
if (N > ii+sb) {
mi = min(ii+sb, M);
ni = N-(ii+sb);
l = min(sb, max(0, mi-ii));
CORE_cparfb(
PlasmaLeft, PlasmaConjTrans,
PlasmaForward, PlasmaColumnwise,
IB, ni, mi, ni, sb, l, //replaced sb by IB
&A1[LDA1*(ii+sb)+ii], LDA1,
&A2[LDA2*(ii+sb)], LDA2,
&A2[LDA2*ii], LDA2,
&T[LDT*ii], LDT,
WORK, sb);
}
}
return PLASMA_SUCCESS;
}

Here is the call graph for this function:

Here is the caller graph for this function:

void CORE_cttqrt_quark ( Quark *  quark)

Definition at line 273 of file core_cttqrt.c.

References CORE_cttqrt(), quark_unpack_args_11, T, and TAU.

{
int m;
int n;
int ib;
PLASMA_Complex32_t *A1;
int lda1;
PLASMA_Complex32_t *A2;
int lda2;
PLASMA_Complex32_t *T;
int ldt;
PLASMA_Complex32_t *TAU;
PLASMA_Complex32_t *WORK;
quark_unpack_args_11(quark, m, n, ib, A1, lda1, A2, lda2, T, ldt, TAU, WORK);
CORE_cttqrt(m, n, ib, A1, lda1, A2, lda2, T, ldt, TAU, WORK);
}

Here is the call graph for this function:

Here is the caller graph for this function:

void QUARK_CORE_cttqrt ( Quark *  quark,
Quark_Task_Flags *  task_flags,
int  m,
int  n,
int  ib,
int  nb,
PLASMA_Complex32_t *  A1,
int  lda1,
PLASMA_Complex32_t *  A2,
int  lda2,
PLASMA_Complex32_t *  T,
int  ldt 
)

Definition at line 244 of file core_cttqrt.c.

References CORE_cttqrt_quark(), DAG_CORE_TTQRT, INOUT, LOCALITY, OUTPUT, QUARK_Insert_Task(), QUARK_REGION_D, QUARK_REGION_U, SCRATCH, and VALUE.

{
DAG_CORE_TTQRT;
QUARK_Insert_Task(quark, CORE_cttqrt_quark, task_flags,
sizeof(int), &m, VALUE,
sizeof(int), &n, VALUE,
sizeof(int), &ib, VALUE,
sizeof(PLASMA_Complex32_t)*nb*nb, A1, INOUT|QUARK_REGION_D|QUARK_REGION_U,
sizeof(int), &lda1, VALUE,
sizeof(PLASMA_Complex32_t)*nb*nb, A2, INOUT|QUARK_REGION_D|QUARK_REGION_U|LOCALITY,
sizeof(int), &lda2, VALUE,
sizeof(PLASMA_Complex32_t)*ib*nb, T, OUTPUT,
sizeof(int), &ldt, VALUE,
sizeof(PLASMA_Complex32_t)*nb, NULL, SCRATCH,
sizeof(PLASMA_Complex32_t)*ib*nb, NULL, SCRATCH,
0);
}

Here is the call graph for this function:

Here is the caller graph for this function: