/* Subroutine */ int cpbcon_(char *uplo, integer *n, integer *kd, complex *ab,
integer *ldab, real *anorm, real *rcond, complex *work, real *rwork,
integer *info)
{
/* -- LAPACK routine (version 2.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
September 30, 1994
Purpose
=======
CPBCON estimates the reciprocal of the condition number (in the
1-norm) of a complex Hermitian positive definite band matrix using
the Cholesky factorization A = U**H*U or A = L*L**H computed by
CPBTRF.
An estimate is obtained for norm(inv(A)), and the reciprocal of the
condition number is computed as RCOND = 1 / (ANORM * norm(inv(A))).
Arguments
=========
UPLO (input) CHARACTER*1
= 'U': Upper triangular factor stored in AB;
= 'L': Lower triangular factor stored in AB.
N (input) INTEGER
The order of the matrix A. N >= 0.
KD (input) INTEGER
The number of superdiagonals of the matrix A if UPLO = 'U',
or the number of sub-diagonals if UPLO = 'L'. KD >= 0.
AB (input) COMPLEX array, dimension (LDAB,N)
The triangular factor U or L from the Cholesky factorization
A = U**H*U or A = L*L**H of the band matrix A, stored in the
first KD+1 rows of the array. The j-th column of U or L is
stored in the j-th column of the array AB as follows:
if UPLO ='U', AB(kd+1+i-j,j) = U(i,j) for max(1,j-kd)<=i<=j;
if UPLO ='L', AB(1+i-j,j) = L(i,j) for j<=i<=min(n,j+kd).
LDAB (input) INTEGER
The leading dimension of the array AB. LDAB >= KD+1.
ANORM (input) REAL
The 1-norm (or infinity-norm) of the Hermitian band matrix A.
RCOND (output) REAL
The reciprocal of the condition number of the matrix A,
computed as RCOND = 1/(ANORM * AINVNM), where AINVNM is an
estimate of the 1-norm of inv(A) computed in this routine.
WORK (workspace) COMPLEX array, dimension (2*N)
RWORK (workspace) REAL array, dimension (N)
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
=====================================================================
Test the input parameters.
Parameter adjustments
Function Body */
/* Table of constant values */
static integer c__1 = 1;
/* System generated locals */
integer ab_dim1, ab_offset, i__1;
real r__1, r__2;
/* Builtin functions */
double r_imag(complex *);
/* Local variables */
static integer kase;
static real scale;
extern logical lsame_(char *, char *);
static logical upper;
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *);
static integer ix;
extern integer icamax_(integer *, complex *, integer *);
static real scalel;
extern doublereal slamch_(char *);
extern /* Subroutine */ int clatbs_(char *, char *, char *, char *,
integer *, integer *, complex *, integer *, complex *, real *,
real *, integer *);
static real scaleu;
extern /* Subroutine */ int xerbla_(char *, integer *);
static real ainvnm;
extern /* Subroutine */ int csrscl_(integer *, real *, complex *, integer
*);
static char normin[1];
static real smlnum;
#define WORK(I) work[(I)-1]
#define RWORK(I) rwork[(I)-1]
#define AB(I,J) ab[(I)-1 + ((J)-1)* ( *ldab)]
*info = 0;
upper = lsame_(uplo, "U");
if (! upper && ! lsame_(uplo, "L")) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*kd < 0) {
*info = -3;
} else if (*ldab < *kd + 1) {
*info = -5;
} else if (*anorm < 0.f) {
*info = -6;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CPBCON", &i__1);
return 0;
}
/* Quick return if possible */
*rcond = 0.f;
if (*n == 0) {
*rcond = 1.f;
return 0;
} else if (*anorm == 0.f) {
return 0;
}
smlnum = slamch_("Safe minimum");
/* Estimate the 1-norm of the inverse. */
kase = 0;
*(unsigned char *)normin = 'N';
L10:
clacon_(n, &WORK(*n + 1), &WORK(1), &ainvnm, &kase);
if (kase != 0) {
if (upper) {
/* Multiply by inv(U'). */
clatbs_("Upper", "Conjugate transpose", "Non-unit", normin, n, kd,
&AB(1,1), ldab, &WORK(1), &scalel, &RWORK(1), info);
*(unsigned char *)normin = 'Y';
/* Multiply by inv(U). */
clatbs_("Upper", "No transpose", "Non-unit", normin, n, kd, &AB(1,1), ldab, &WORK(1), &scaleu, &RWORK(1), info);
} else {
/* Multiply by inv(L). */
clatbs_("Lower", "No transpose", "Non-unit", normin, n, kd, &AB(1,1), ldab, &WORK(1), &scalel, &RWORK(1), info);
*(unsigned char *)normin = 'Y';
/* Multiply by inv(L'). */
clatbs_("Lower", "Conjugate transpose", "Non-unit", normin, n, kd,
&AB(1,1), ldab, &WORK(1), &scaleu, &RWORK(1), info);
}
/* Multiply by 1/SCALE if doing so will not cause overflow. */
scale = scalel * scaleu;
if (scale != 1.f) {
ix = icamax_(n, &WORK(1), &c__1);
i__1 = ix;
if (scale < ((r__1 = WORK(ix).r, dabs(r__1)) + (r__2 = r_imag(&
WORK(ix)), dabs(r__2))) * smlnum || scale == 0.f) {
goto L20;
}
csrscl_(n, &scale, &WORK(1), &c__1);
}
goto L10;
}
/* Compute the estimate of the reciprocal condition number. */
if (ainvnm != 0.f) {
*rcond = 1.f / ainvnm / *anorm;
}
L20:
return 0;
/* End of CPBCON */
} /* cpbcon_ */
/* Subroutine */ int cgecon_(char *norm, integer *n, complex *a, integer *lda,
real *anorm, real *rcond, complex *work, real *rwork, integer *info,
ftnlen norm_len)
{
/* System generated locals */
integer a_dim1, a_offset, i__1;
real r__1, r__2;
/* Builtin functions */
double r_imag(complex *);
/* Local variables */
static real sl;
static integer ix;
static real su;
static integer kase, kase1;
static real scale;
extern logical lsame_(char *, char *, ftnlen, ftnlen);
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *);
extern integer icamax_(integer *, complex *, integer *);
extern doublereal slamch_(char *, ftnlen);
extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
static real ainvnm;
extern /* Subroutine */ int clatrs_(char *, char *, char *, char *,
integer *, complex *, integer *, complex *, real *, real *,
integer *, ftnlen, ftnlen, ftnlen, ftnlen), csrscl_(integer *,
real *, complex *, integer *);
static logical onenrm;
static char normin[1];
static real smlnum;
/* -- LAPACK routine (version 3.0) -- */
/* Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., */
/* Courant Institute, Argonne National Lab, and Rice University */
/* March 31, 1993 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* CGECON estimates the reciprocal of the condition number of a general */
/* complex matrix A, in either the 1-norm or the infinity-norm, using */
/* the LU factorization computed by CGETRF. */
/* An estimate is obtained for norm(inv(A)), and the reciprocal of the */
/* condition number is computed as */
/* RCOND = 1 / ( norm(A) * norm(inv(A)) ). */
/* Arguments */
/* ========= */
/* NORM (input) CHARACTER*1 */
/* Specifies whether the 1-norm condition number or the */
/* infinity-norm condition number is required: */
/* = '1' or 'O': 1-norm; */
/* = 'I': Infinity-norm. */
/* N (input) INTEGER */
/* The order of the matrix A. N >= 0. */
/* A (input) COMPLEX array, dimension (LDA,N) */
/* The factors L and U from the factorization A = P*L*U */
/* as computed by CGETRF. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1,N). */
/* ANORM (input) REAL */
/* If NORM = '1' or 'O', the 1-norm of the original matrix A. */
/* If NORM = 'I', the infinity-norm of the original matrix A. */
/* RCOND (output) REAL */
/* The reciprocal of the condition number of the matrix A, */
/* computed as RCOND = 1/(norm(A) * norm(inv(A))). */
/* WORK (workspace) COMPLEX array, dimension (2*N) */
/* RWORK (workspace) REAL array, dimension (2*N) */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Statement Functions .. */
/* .. */
/* .. Statement Function definitions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
--work;
--rwork;
/* Function Body */
*info = 0;
onenrm = *(unsigned char *)norm == '1' || lsame_(norm, "O", (ftnlen)1, (
ftnlen)1);
if (! onenrm && ! lsame_(norm, "I", (ftnlen)1, (ftnlen)1)) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*lda < max(1,*n)) {
*info = -4;
} else if (*anorm < 0.f) {
*info = -5;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CGECON", &i__1, (ftnlen)6);
return 0;
}
/* Quick return if possible */
*rcond = 0.f;
if (*n == 0) {
*rcond = 1.f;
return 0;
} else if (*anorm == 0.f) {
return 0;
}
smlnum = slamch_("Safe minimum", (ftnlen)12);
/* Estimate the norm of inv(A). */
ainvnm = 0.f;
*(unsigned char *)normin = 'N';
if (onenrm) {
kase1 = 1;
} else {
kase1 = 2;
}
kase = 0;
L10:
clacon_(n, &work[*n + 1], &work[1], &ainvnm, &kase);
if (kase != 0) {
if (kase == kase1) {
/* Multiply by inv(L). */
clatrs_("Lower", "No transpose", "Unit", normin, n, &a[a_offset],
lda, &work[1], &sl, &rwork[1], info, (ftnlen)5, (ftnlen)
12, (ftnlen)4, (ftnlen)1);
/* Multiply by inv(U). */
clatrs_("Upper", "No transpose", "Non-unit", normin, n, &a[
a_offset], lda, &work[1], &su, &rwork[*n + 1], info, (
ftnlen)5, (ftnlen)12, (ftnlen)8, (ftnlen)1);
} else {
/* Multiply by inv(U'). */
clatrs_("Upper", "Conjugate transpose", "Non-unit", normin, n, &a[
a_offset], lda, &work[1], &su, &rwork[*n + 1], info, (
ftnlen)5, (ftnlen)19, (ftnlen)8, (ftnlen)1);
/* Multiply by inv(L'). */
clatrs_("Lower", "Conjugate transpose", "Unit", normin, n, &a[
a_offset], lda, &work[1], &sl, &rwork[1], info, (ftnlen)5,
(ftnlen)19, (ftnlen)4, (ftnlen)1);
}
/* Divide X by 1/(SL*SU) if doing so will not cause overflow. */
scale = sl * su;
*(unsigned char *)normin = 'Y';
if (scale != 1.f) {
ix = icamax_(n, &work[1], &c__1);
i__1 = ix;
if (scale < ((r__1 = work[i__1].r, dabs(r__1)) + (r__2 = r_imag(&
work[ix]), dabs(r__2))) * smlnum || scale == 0.f) {
goto L20;
}
csrscl_(n, &scale, &work[1], &c__1);
}
goto L10;
}
/* Compute the estimate of the reciprocal condition number. */
if (ainvnm != 0.f) {
*rcond = 1.f / ainvnm / *anorm;
}
L20:
return 0;
/* End of CGECON */
} /* cgecon_ */
void
cgscon(char *norm, SuperMatrix *L, SuperMatrix *U,
float anorm, float *rcond, SuperLUStat_t *stat, int *info)
{
/* Local variables */
int kase, kase1, onenrm, i;
float ainvnm;
complex *work;
extern int crscl_(int *, complex *, complex *, int *);
extern int clacon_(int *, complex *, complex *, float *, int *);
/* Test the input parameters. */
*info = 0;
onenrm = *(unsigned char *)norm == '1' || lsame_(norm, "O");
if (! onenrm && ! lsame_(norm, "I")) *info = -1;
else if (L->nrow < 0 || L->nrow != L->ncol ||
L->Stype != SLU_SC || L->Dtype != SLU_C || L->Mtype != SLU_TRLU)
*info = -2;
else if (U->nrow < 0 || U->nrow != U->ncol ||
U->Stype != SLU_NC || U->Dtype != SLU_C || U->Mtype != SLU_TRU)
*info = -3;
if (*info != 0) {
i = -(*info);
xerbla_("cgscon", &i);
return;
}
/* Quick return if possible */
*rcond = 0.;
if ( L->nrow == 0 || U->nrow == 0) {
*rcond = 1.;
return;
}
work = complexCalloc( 3*L->nrow );
if ( !work )
ABORT("Malloc fails for work arrays in cgscon.");
/* Estimate the norm of inv(A). */
ainvnm = 0.;
if ( onenrm ) kase1 = 1;
else kase1 = 2;
kase = 0;
do {
clacon_(&L->nrow, &work[L->nrow], &work[0], &ainvnm, &kase);
if (kase == 0) break;
if (kase == kase1) {
/* Multiply by inv(L). */
sp_ctrsv("L", "No trans", "Unit", L, U, &work[0], stat, info);
/* Multiply by inv(U). */
sp_ctrsv("U", "No trans", "Non-unit", L, U, &work[0], stat, info);
} else {
/* Multiply by inv(U'). */
sp_ctrsv("U", "Transpose", "Non-unit", L, U, &work[0], stat, info);
/* Multiply by inv(L'). */
sp_ctrsv("L", "Transpose", "Unit", L, U, &work[0], stat, info);
}
} while ( kase != 0 );
/* Compute the estimate of the reciprocal condition number. */
if (ainvnm != 0.) *rcond = (1. / ainvnm) / anorm;
SUPERLU_FREE (work);
return;
} /* cgscon */
/* Subroutine */ int ctpcon_(char *norm, char *uplo, char *diag, integer *n,
complex *ap, real *rcond, complex *work, real *rwork, integer *info,
ftnlen norm_len, ftnlen uplo_len, ftnlen diag_len)
{
/* System generated locals */
integer i__1;
real r__1, r__2;
/* Builtin functions */
double r_imag(complex *);
/* Local variables */
static integer ix, kase, kase1;
static real scale;
extern logical lsame_(char *, char *, ftnlen, ftnlen);
static real anorm;
static logical upper;
static real xnorm;
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *);
extern integer icamax_(integer *, complex *, integer *);
extern doublereal slamch_(char *, ftnlen);
extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
extern doublereal clantp_(char *, char *, char *, integer *, complex *,
real *, ftnlen, ftnlen, ftnlen);
extern /* Subroutine */ int clatps_(char *, char *, char *, char *,
integer *, complex *, complex *, real *, real *, integer *,
ftnlen, ftnlen, ftnlen, ftnlen);
static real ainvnm;
extern /* Subroutine */ int csrscl_(integer *, real *, complex *, integer
*);
static logical onenrm;
static char normin[1];
static real smlnum;
static logical nounit;
/* -- LAPACK routine (version 3.0) -- */
/* Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., */
/* Courant Institute, Argonne National Lab, and Rice University */
/* March 31, 1993 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* CTPCON estimates the reciprocal of the condition number of a packed */
/* triangular matrix A, in either the 1-norm or the infinity-norm. */
/* The norm of A is computed and an estimate is obtained for */
/* norm(inv(A)), then the reciprocal of the condition number is */
/* computed as */
/* RCOND = 1 / ( norm(A) * norm(inv(A)) ). */
/* Arguments */
/* ========= */
/* NORM (input) CHARACTER*1 */
/* Specifies whether the 1-norm condition number or the */
/* infinity-norm condition number is required: */
/* = '1' or 'O': 1-norm; */
/* = 'I': Infinity-norm. */
/* UPLO (input) CHARACTER*1 */
/* = 'U': A is upper triangular; */
/* = 'L': A is lower triangular. */
/* DIAG (input) CHARACTER*1 */
/* = 'N': A is non-unit triangular; */
/* = 'U': A is unit triangular. */
/* N (input) INTEGER */
/* The order of the matrix A. N >= 0. */
/* AP (input) COMPLEX array, dimension (N*(N+1)/2) */
/* The upper or lower triangular matrix A, packed columnwise in */
/* a linear array. The j-th column of A is stored in the array */
/* AP as follows: */
/* if UPLO = 'U', AP(i + (j-1)*j/2) = A(i,j) for 1<=i<=j; */
/* if UPLO = 'L', AP(i + (j-1)*(2n-j)/2) = A(i,j) for j<=i<=n. */
/* If DIAG = 'U', the diagonal elements of A are not referenced */
/* and are assumed to be 1. */
/* RCOND (output) REAL */
/* The reciprocal of the condition number of the matrix A, */
/* computed as RCOND = 1/(norm(A) * norm(inv(A))). */
/* WORK (workspace) COMPLEX array, dimension (2*N) */
/* RWORK (workspace) REAL array, dimension (N) */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Statement Functions .. */
/* .. */
/* .. Statement Function definitions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
--rwork;
--work;
--ap;
/* Function Body */
*info = 0;
upper = lsame_(uplo, "U", (ftnlen)1, (ftnlen)1);
onenrm = *(unsigned char *)norm == '1' || lsame_(norm, "O", (ftnlen)1, (
ftnlen)1);
nounit = lsame_(diag, "N", (ftnlen)1, (ftnlen)1);
if (! onenrm && ! lsame_(norm, "I", (ftnlen)1, (ftnlen)1)) {
*info = -1;
} else if (! upper && ! lsame_(uplo, "L", (ftnlen)1, (ftnlen)1)) {
*info = -2;
} else if (! nounit && ! lsame_(diag, "U", (ftnlen)1, (ftnlen)1)) {
*info = -3;
} else if (*n < 0) {
*info = -4;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CTPCON", &i__1, (ftnlen)6);
return 0;
}
/* Quick return if possible */
if (*n == 0) {
*rcond = 1.f;
return 0;
}
*rcond = 0.f;
smlnum = slamch_("Safe minimum", (ftnlen)12) * (real) max(1,*n);
/* Compute the norm of the triangular matrix A. */
anorm = clantp_(norm, uplo, diag, n, &ap[1], &rwork[1], (ftnlen)1, (
ftnlen)1, (ftnlen)1);
/* Continue only if ANORM > 0. */
if (anorm > 0.f) {
/* Estimate the norm of the inverse of A. */
ainvnm = 0.f;
*(unsigned char *)normin = 'N';
if (onenrm) {
kase1 = 1;
} else {
kase1 = 2;
}
kase = 0;
L10:
clacon_(n, &work[*n + 1], &work[1], &ainvnm, &kase);
if (kase != 0) {
if (kase == kase1) {
/* Multiply by inv(A). */
clatps_(uplo, "No transpose", diag, normin, n, &ap[1], &work[
1], &scale, &rwork[1], info, (ftnlen)1, (ftnlen)12, (
ftnlen)1, (ftnlen)1);
} else {
/* Multiply by inv(A'). */
clatps_(uplo, "Conjugate transpose", diag, normin, n, &ap[1],
&work[1], &scale, &rwork[1], info, (ftnlen)1, (ftnlen)
19, (ftnlen)1, (ftnlen)1);
}
*(unsigned char *)normin = 'Y';
/* Multiply by 1/SCALE if doing so will not cause overflow. */
if (scale != 1.f) {
ix = icamax_(n, &work[1], &c__1);
i__1 = ix;
xnorm = (r__1 = work[i__1].r, dabs(r__1)) + (r__2 = r_imag(&
work[ix]), dabs(r__2));
if (scale < xnorm * smlnum || scale == 0.f) {
goto L20;
}
csrscl_(n, &scale, &work[1], &c__1);
}
goto L10;
}
/* Compute the estimate of the reciprocal condition number. */
if (ainvnm != 0.f) {
*rcond = 1.f / anorm / ainvnm;
}
}
L20:
return 0;
/* End of CTPCON */
} /* ctpcon_ */
/* Subroutine */ int cgbrfs_(char *trans, integer *n, integer *kl, integer *
ku, integer *nrhs, complex *ab, integer *ldab, complex *afb, integer *
ldafb, integer *ipiv, complex *b, integer *ldb, complex *x, integer *
ldx, real *ferr, real *berr, complex *work, real *rwork, integer *
info, ftnlen trans_len)
{
/* System generated locals */
integer ab_dim1, ab_offset, afb_dim1, afb_offset, b_dim1, b_offset,
x_dim1, x_offset, i__1, i__2, i__3, i__4, i__5, i__6, i__7;
real r__1, r__2, r__3, r__4;
complex q__1;
/* Builtin functions */
double r_imag(complex *);
/* Local variables */
static integer i__, j, k;
static real s;
static integer kk;
static real xk;
static integer nz;
static real eps;
static integer kase;
static real safe1, safe2;
extern /* Subroutine */ int cgbmv_(char *, integer *, integer *, integer *
, integer *, complex *, complex *, integer *, complex *, integer *
, complex *, complex *, integer *, ftnlen);
extern logical lsame_(char *, char *, ftnlen, ftnlen);
extern /* Subroutine */ int ccopy_(integer *, complex *, integer *,
complex *, integer *), caxpy_(integer *, complex *, complex *,
integer *, complex *, integer *);
static integer count;
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *);
extern doublereal slamch_(char *, ftnlen);
static real safmin;
extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen), cgbtrs_(
char *, integer *, integer *, integer *, integer *, complex *,
integer *, integer *, complex *, integer *, integer *, ftnlen);
static logical notran;
static char transn[1], transt[1];
static real lstres;
/* -- LAPACK routine (version 3.0) -- */
/* Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., */
/* Courant Institute, Argonne National Lab, and Rice University */
/* September 30, 1994 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* CGBRFS improves the computed solution to a system of linear */
/* equations when the coefficient matrix is banded, and provides */
/* error bounds and backward error estimates for the solution. */
/* Arguments */
/* ========= */
/* TRANS (input) CHARACTER*1 */
/* Specifies the form of the system of equations: */
/* = 'N': A * X = B (No transpose) */
/* = 'T': A**T * X = B (Transpose) */
/* = 'C': A**H * X = B (Conjugate transpose) */
/* N (input) INTEGER */
/* The order of the matrix A. N >= 0. */
/* KL (input) INTEGER */
/* The number of subdiagonals within the band of A. KL >= 0. */
/* KU (input) INTEGER */
/* The number of superdiagonals within the band of A. KU >= 0. */
/* NRHS (input) INTEGER */
/* The number of right hand sides, i.e., the number of columns */
/* of the matrices B and X. NRHS >= 0. */
/* AB (input) COMPLEX array, dimension (LDAB,N) */
/* The original band matrix A, stored in rows 1 to KL+KU+1. */
/* The j-th column of A is stored in the j-th column of the */
/* array AB as follows: */
/* AB(ku+1+i-j,j) = A(i,j) for max(1,j-ku)<=i<=min(n,j+kl). */
/* LDAB (input) INTEGER */
/* The leading dimension of the array AB. LDAB >= KL+KU+1. */
/* AFB (input) COMPLEX array, dimension (LDAFB,N) */
/* Details of the LU factorization of the band matrix A, as */
/* computed by CGBTRF. U is stored as an upper triangular band */
/* matrix with KL+KU superdiagonals in rows 1 to KL+KU+1, and */
/* the multipliers used during the factorization are stored in */
/* rows KL+KU+2 to 2*KL+KU+1. */
/* LDAFB (input) INTEGER */
/* The leading dimension of the array AFB. LDAFB >= 2*KL*KU+1. */
/* IPIV (input) INTEGER array, dimension (N) */
/* The pivot indices from CGBTRF; for 1<=i<=N, row i of the */
/* matrix was interchanged with row IPIV(i). */
/* B (input) COMPLEX array, dimension (LDB,NRHS) */
/* The right hand side matrix B. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= max(1,N). */
/* X (input/output) COMPLEX array, dimension (LDX,NRHS) */
/* On entry, the solution matrix X, as computed by CGBTRS. */
/* On exit, the improved solution matrix X. */
/* LDX (input) INTEGER */
/* The leading dimension of the array X. LDX >= max(1,N). */
/* FERR (output) REAL array, dimension (NRHS) */
/* The estimated forward error bound for each solution vector */
/* X(j) (the j-th column of the solution matrix X). */
/* If XTRUE is the true solution corresponding to X(j), FERR(j) */
/* is an estimated upper bound for the magnitude of the largest */
/* element in (X(j) - XTRUE) divided by the magnitude of the */
/* largest element in X(j). The estimate is as reliable as */
/* the estimate for RCOND, and is almost always a slight */
/* overestimate of the true error. */
/* BERR (output) REAL array, dimension (NRHS) */
/* The componentwise relative backward error of each solution */
/* vector X(j) (i.e., the smallest relative change in */
/* any element of A or B that makes X(j) an exact solution). */
/* WORK (workspace) COMPLEX array, dimension (2*N) */
/* RWORK (workspace) REAL array, dimension (N) */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* Internal Parameters */
/* =================== */
/* ITMAX is the maximum number of steps of iterative refinement. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Statement Functions .. */
/* .. */
/* .. Statement Function definitions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
ab_dim1 = *ldab;
ab_offset = 1 + ab_dim1;
ab -= ab_offset;
afb_dim1 = *ldafb;
afb_offset = 1 + afb_dim1;
afb -= afb_offset;
--ipiv;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
x_dim1 = *ldx;
x_offset = 1 + x_dim1;
x -= x_offset;
--ferr;
--berr;
--work;
--rwork;
/* Function Body */
*info = 0;
notran = lsame_(trans, "N", (ftnlen)1, (ftnlen)1);
if (! notran && ! lsame_(trans, "T", (ftnlen)1, (ftnlen)1) && ! lsame_(
trans, "C", (ftnlen)1, (ftnlen)1)) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*kl < 0) {
*info = -3;
} else if (*ku < 0) {
*info = -4;
} else if (*nrhs < 0) {
*info = -5;
} else if (*ldab < *kl + *ku + 1) {
*info = -7;
} else if (*ldafb < (*kl << 1) + *ku + 1) {
*info = -9;
} else if (*ldb < max(1,*n)) {
*info = -12;
} else if (*ldx < max(1,*n)) {
*info = -14;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CGBRFS", &i__1, (ftnlen)6);
return 0;
}
/* Quick return if possible */
if (*n == 0 || *nrhs == 0) {
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
ferr[j] = 0.f;
berr[j] = 0.f;
/* L10: */
}
return 0;
}
if (notran) {
*(unsigned char *)transn = 'N';
*(unsigned char *)transt = 'C';
} else {
*(unsigned char *)transn = 'C';
*(unsigned char *)transt = 'N';
}
/* NZ = maximum number of nonzero elements in each row of A, plus 1 */
/* Computing MIN */
i__1 = *kl + *ku + 2, i__2 = *n + 1;
nz = min(i__1,i__2);
eps = slamch_("Epsilon", (ftnlen)7);
safmin = slamch_("Safe minimum", (ftnlen)12);
safe1 = nz * safmin;
safe2 = safe1 / eps;
/* Do for each right hand side */
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
count = 1;
lstres = 3.f;
L20:
/* Loop until stopping criterion is satisfied. */
/* Compute residual R = B - op(A) * X, */
/* where op(A) = A, A**T, or A**H, depending on TRANS. */
ccopy_(n, &b[j * b_dim1 + 1], &c__1, &work[1], &c__1);
q__1.r = -1.f, q__1.i = -0.f;
cgbmv_(trans, n, n, kl, ku, &q__1, &ab[ab_offset], ldab, &x[j *
x_dim1 + 1], &c__1, &c_b1, &work[1], &c__1, (ftnlen)1);
/* Compute componentwise relative backward error from formula */
/* max(i) ( abs(R(i)) / ( abs(op(A))*abs(X) + abs(B) )(i) ) */
/* where abs(Z) is the componentwise absolute value of the matrix */
/* or vector Z. If the i-th component of the denominator is less */
/* than SAFE2, then SAFE1 is added to the i-th components of the */
/* numerator and denominator before dividing. */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__ + j * b_dim1;
rwork[i__] = (r__1 = b[i__3].r, dabs(r__1)) + (r__2 = r_imag(&b[
i__ + j * b_dim1]), dabs(r__2));
/* L30: */
}
/* Compute abs(op(A))*abs(X) + abs(B). */
if (notran) {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
kk = *ku + 1 - k;
i__3 = k + j * x_dim1;
xk = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&x[k + j
* x_dim1]), dabs(r__2));
/* Computing MAX */
i__3 = 1, i__4 = k - *ku;
/* Computing MIN */
i__6 = *n, i__7 = k + *kl;
i__5 = min(i__6,i__7);
for (i__ = max(i__3,i__4); i__ <= i__5; ++i__) {
i__3 = kk + i__ + k * ab_dim1;
rwork[i__] += ((r__1 = ab[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&ab[kk + i__ + k * ab_dim1]), dabs(r__2)))
* xk;
/* L40: */
}
/* L50: */
}
} else {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
s = 0.f;
kk = *ku + 1 - k;
/* Computing MAX */
i__5 = 1, i__3 = k - *ku;
/* Computing MIN */
i__6 = *n, i__7 = k + *kl;
i__4 = min(i__6,i__7);
for (i__ = max(i__5,i__3); i__ <= i__4; ++i__) {
i__5 = kk + i__ + k * ab_dim1;
i__3 = i__ + j * x_dim1;
s += ((r__1 = ab[i__5].r, dabs(r__1)) + (r__2 = r_imag(&
ab[kk + i__ + k * ab_dim1]), dabs(r__2))) * ((
r__3 = x[i__3].r, dabs(r__3)) + (r__4 = r_imag(&x[
i__ + j * x_dim1]), dabs(r__4)));
/* L60: */
}
rwork[k] += s;
/* L70: */
}
}
s = 0.f;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
/* Computing MAX */
i__4 = i__;
r__3 = s, r__4 = ((r__1 = work[i__4].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2))) / rwork[i__];
s = dmax(r__3,r__4);
} else {
/* Computing MAX */
i__4 = i__;
r__3 = s, r__4 = ((r__1 = work[i__4].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + safe1) / (rwork[i__]
+ safe1);
s = dmax(r__3,r__4);
}
/* L80: */
}
berr[j] = s;
/* Test stopping criterion. Continue iterating if */
/* 1) The residual BERR(J) is larger than machine epsilon, and */
/* 2) BERR(J) decreased by at least a factor of 2 during the */
/* last iteration, and */
/* 3) At most ITMAX iterations tried. */
if (berr[j] > eps && berr[j] * 2.f <= lstres && count <= 5) {
/* Update solution and try again. */
cgbtrs_(trans, n, kl, ku, &c__1, &afb[afb_offset], ldafb, &ipiv[1]
, &work[1], n, info, (ftnlen)1);
caxpy_(n, &c_b1, &work[1], &c__1, &x[j * x_dim1 + 1], &c__1);
lstres = berr[j];
++count;
goto L20;
}
/* Bound error from formula */
/* norm(X - XTRUE) / norm(X) .le. FERR = */
/* norm( abs(inv(op(A)))* */
/* ( abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) / norm(X) */
/* where */
/* norm(Z) is the magnitude of the largest component of Z */
/* inv(op(A)) is the inverse of op(A) */
/* abs(Z) is the componentwise absolute value of the matrix or */
/* vector Z */
/* NZ is the maximum number of nonzeros in any row of A, plus 1 */
/* EPS is machine epsilon */
/* The i-th component of abs(R)+NZ*EPS*(abs(op(A))*abs(X)+abs(B)) */
/* is incremented by SAFE1 if the i-th component of */
/* abs(op(A))*abs(X) + abs(B) is less than SAFE2. */
/* Use CLACON to estimate the infinity-norm of the matrix */
/* inv(op(A)) * diag(W), */
/* where W = abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
i__4 = i__;
rwork[i__] = (r__1 = work[i__4].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + nz * eps * rwork[
i__];
} else {
i__4 = i__;
rwork[i__] = (r__1 = work[i__4].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + nz * eps * rwork[
i__] + safe1;
}
/* L90: */
}
kase = 0;
L100:
clacon_(n, &work[*n + 1], &work[1], &ferr[j], &kase);
if (kase != 0) {
if (kase == 1) {
/* Multiply by diag(W)*inv(op(A)**H). */
cgbtrs_(transt, n, kl, ku, &c__1, &afb[afb_offset], ldafb, &
ipiv[1], &work[1], n, info, (ftnlen)1);
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__4 = i__;
i__5 = i__;
i__3 = i__;
q__1.r = rwork[i__5] * work[i__3].r, q__1.i = rwork[i__5]
* work[i__3].i;
work[i__4].r = q__1.r, work[i__4].i = q__1.i;
/* L110: */
}
} else {
/* Multiply by inv(op(A))*diag(W). */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__4 = i__;
i__5 = i__;
i__3 = i__;
q__1.r = rwork[i__5] * work[i__3].r, q__1.i = rwork[i__5]
* work[i__3].i;
work[i__4].r = q__1.r, work[i__4].i = q__1.i;
/* L120: */
}
cgbtrs_(transn, n, kl, ku, &c__1, &afb[afb_offset], ldafb, &
ipiv[1], &work[1], n, info, (ftnlen)1);
}
goto L100;
}
/* Normalize error. */
lstres = 0.f;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
/* Computing MAX */
i__4 = i__ + j * x_dim1;
r__3 = lstres, r__4 = (r__1 = x[i__4].r, dabs(r__1)) + (r__2 =
r_imag(&x[i__ + j * x_dim1]), dabs(r__2));
lstres = dmax(r__3,r__4);
/* L130: */
}
if (lstres != 0.f) {
ferr[j] /= lstres;
}
/* L140: */
}
return 0;
/* End of CGBRFS */
} /* cgbrfs_ */
/* Subroutine */ int ctrsen_(char *job, char *compq, logical *select, integer
*n, complex *t, integer *ldt, complex *q, integer *ldq, complex *w,
integer *m, real *s, real *sep, complex *work, integer *lwork,
integer *info)
{
/* -- LAPACK routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
June 30, 1999
Purpose
=======
CTRSEN reorders the Schur factorization of a complex matrix
A = Q*T*Q**H, so that a selected cluster of eigenvalues appears in
the leading positions on the diagonal of the upper triangular matrix
T, and the leading columns of Q form an orthonormal basis of the
corresponding right invariant subspace.
Optionally the routine computes the reciprocal condition numbers of
the cluster of eigenvalues and/or the invariant subspace.
Arguments
=========
JOB (input) CHARACTER*1
Specifies whether condition numbers are required for the
cluster of eigenvalues (S) or the invariant subspace (SEP):
= 'N': none;
= 'E': for eigenvalues only (S);
= 'V': for invariant subspace only (SEP);
= 'B': for both eigenvalues and invariant subspace (S and
SEP).
COMPQ (input) CHARACTER*1
= 'V': update the matrix Q of Schur vectors;
= 'N': do not update Q.
SELECT (input) LOGICAL array, dimension (N)
SELECT specifies the eigenvalues in the selected cluster. To
select the j-th eigenvalue, SELECT(j) must be set to .TRUE..
N (input) INTEGER
The order of the matrix T. N >= 0.
T (input/output) COMPLEX array, dimension (LDT,N)
On entry, the upper triangular matrix T.
On exit, T is overwritten by the reordered matrix T, with the
selected eigenvalues as the leading diagonal elements.
LDT (input) INTEGER
The leading dimension of the array T. LDT >= max(1,N).
Q (input/output) COMPLEX array, dimension (LDQ,N)
On entry, if COMPQ = 'V', the matrix Q of Schur vectors.
On exit, if COMPQ = 'V', Q has been postmultiplied by the
unitary transformation matrix which reorders T; the leading M
columns of Q form an orthonormal basis for the specified
invariant subspace.
If COMPQ = 'N', Q is not referenced.
LDQ (input) INTEGER
The leading dimension of the array Q.
LDQ >= 1; and if COMPQ = 'V', LDQ >= N.
W (output) COMPLEX array, dimension (N)
The reordered eigenvalues of T, in the same order as they
appear on the diagonal of T.
M (output) INTEGER
The dimension of the specified invariant subspace.
0 <= M <= N.
S (output) REAL
If JOB = 'E' or 'B', S is a lower bound on the reciprocal
condition number for the selected cluster of eigenvalues.
S cannot underestimate the true reciprocal condition number
by more than a factor of sqrt(N). If M = 0 or N, S = 1.
If JOB = 'N' or 'V', S is not referenced.
SEP (output) REAL
If JOB = 'V' or 'B', SEP is the estimated reciprocal
condition number of the specified invariant subspace. If
M = 0 or N, SEP = norm(T).
If JOB = 'N' or 'E', SEP is not referenced.
WORK (workspace/output) COMPLEX array, dimension (LWORK)
If JOB = 'N', WORK is not referenced. Otherwise,
on exit, if INFO = 0, WORK(1) returns the optimal LWORK.
LWORK (input) INTEGER
The dimension of the array WORK.
If JOB = 'N', LWORK >= 1;
if JOB = 'E', LWORK = M*(N-M);
if JOB = 'V' or 'B', LWORK >= 2*M*(N-M).
If LWORK = -1, then a workspace query is assumed; the routine
only calculates the optimal size of the WORK array, returns
this value as the first entry of the WORK array, and no error
message related to LWORK is issued by XERBLA.
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
Further Details
===============
CTRSEN first collects the selected eigenvalues by computing a unitary
transformation Z to move them to the top left corner of T. In other
words, the selected eigenvalues are the eigenvalues of T11 in:
Z'*T*Z = ( T11 T12 ) n1
( 0 T22 ) n2
n1 n2
where N = n1+n2 and Z' means the conjugate transpose of Z. The first
n1 columns of Z span the specified invariant subspace of T.
If T has been obtained from the Schur factorization of a matrix
A = Q*T*Q', then the reordered Schur factorization of A is given by
A = (Q*Z)*(Z'*T*Z)*(Q*Z)', and the first n1 columns of Q*Z span the
corresponding invariant subspace of A.
The reciprocal condition number of the average of the eigenvalues of
T11 may be returned in S. S lies between 0 (very badly conditioned)
and 1 (very well conditioned). It is computed as follows. First we
compute R so that
P = ( I R ) n1
( 0 0 ) n2
n1 n2
is the projector on the invariant subspace associated with T11.
R is the solution of the Sylvester equation:
T11*R - R*T22 = T12.
Let F-norm(M) denote the Frobenius-norm of M and 2-norm(M) denote
the two-norm of M. Then S is computed as the lower bound
(1 + F-norm(R)**2)**(-1/2)
on the reciprocal of 2-norm(P), the true reciprocal condition number.
S cannot underestimate 1 / 2-norm(P) by more than a factor of
sqrt(N).
An approximate error bound for the computed average of the
eigenvalues of T11 is
EPS * norm(T) / S
where EPS is the machine precision.
The reciprocal condition number of the right invariant subspace
spanned by the first n1 columns of Z (or of Q*Z) is returned in SEP.
SEP is defined as the separation of T11 and T22:
sep( T11, T22 ) = sigma-min( C )
where sigma-min(C) is the smallest singular value of the
n1*n2-by-n1*n2 matrix
C = kprod( I(n2), T11 ) - kprod( transpose(T22), I(n1) )
I(m) is an m by m identity matrix, and kprod denotes the Kronecker
product. We estimate sigma-min(C) by the reciprocal of an estimate of
the 1-norm of inverse(C). The true reciprocal 1-norm of inverse(C)
cannot differ from sigma-min(C) by more than a factor of sqrt(n1*n2).
When SEP is small, small changes in T can cause large changes in
the invariant subspace. An approximate bound on the maximum angular
error in the computed right invariant subspace is
EPS * norm(T) / SEP
=====================================================================
Decode and test the input parameters.
Parameter adjustments */
/* Table of constant values */
static integer c_n1 = -1;
/* System generated locals */
integer q_dim1, q_offset, t_dim1, t_offset, i__1, i__2, i__3;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
static integer kase, ierr, k;
static real scale;
extern logical lsame_(char *, char *);
static integer lwmin;
static logical wantq, wants;
static real rnorm;
static integer n1, n2;
static real rwork[1];
extern doublereal clange_(char *, integer *, integer *, complex *,
integer *, real *);
static integer nn, ks;
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *), clacpy_(char *, integer *, integer *, complex *,
integer *, complex *, integer *), xerbla_(char *, integer
*);
static logical wantbh;
extern /* Subroutine */ int ctrexc_(char *, integer *, complex *, integer
*, complex *, integer *, integer *, integer *, integer *);
static logical wantsp;
extern /* Subroutine */ int ctrsyl_(char *, char *, integer *, integer *,
integer *, complex *, integer *, complex *, integer *, complex *,
integer *, real *, integer *);
static logical lquery;
static real est;
#define t_subscr(a_1,a_2) (a_2)*t_dim1 + a_1
#define t_ref(a_1,a_2) t[t_subscr(a_1,a_2)]
--select;
t_dim1 = *ldt;
t_offset = 1 + t_dim1 * 1;
t -= t_offset;
q_dim1 = *ldq;
q_offset = 1 + q_dim1 * 1;
q -= q_offset;
--w;
--work;
/* Function Body */
wantbh = lsame_(job, "B");
wants = lsame_(job, "E") || wantbh;
wantsp = lsame_(job, "V") || wantbh;
wantq = lsame_(compq, "V");
/* Set M to the number of selected eigenvalues. */
*m = 0;
i__1 = *n;
for (k = 1; k <= i__1; ++k) {
if (select[k]) {
++(*m);
}
/* L10: */
}
n1 = *m;
n2 = *n - *m;
nn = n1 * n2;
*info = 0;
lquery = *lwork == -1;
if (wantsp) {
/* Computing MAX */
i__1 = 1, i__2 = nn << 1;
lwmin = max(i__1,i__2);
} else if (lsame_(job, "N")) {
lwmin = 1;
} else if (lsame_(job, "E")) {
lwmin = max(1,nn);
}
if (! lsame_(job, "N") && ! wants && ! wantsp) {
*info = -1;
} else if (! lsame_(compq, "N") && ! wantq) {
*info = -2;
} else if (*n < 0) {
*info = -4;
} else if (*ldt < max(1,*n)) {
*info = -6;
} else if (*ldq < 1 || wantq && *ldq < *n) {
*info = -8;
} else if (*lwork < lwmin && ! lquery) {
*info = -14;
}
if (*info == 0) {
work[1].r = (real) lwmin, work[1].i = 0.f;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CTRSEN", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (*m == *n || *m == 0) {
if (wants) {
*s = 1.f;
}
if (wantsp) {
*sep = clange_("1", n, n, &t[t_offset], ldt, rwork);
}
goto L40;
}
/* Collect the selected eigenvalues at the top left corner of T. */
ks = 0;
i__1 = *n;
for (k = 1; k <= i__1; ++k) {
if (select[k]) {
++ks;
/* Swap the K-th eigenvalue to position KS. */
if (k != ks) {
ctrexc_(compq, n, &t[t_offset], ldt, &q[q_offset], ldq, &k, &
ks, &ierr);
}
}
/* L20: */
}
if (wants) {
/* Solve the Sylvester equation for R:
T11*R - R*T22 = scale*T12 */
clacpy_("F", &n1, &n2, &t_ref(1, n1 + 1), ldt, &work[1], &n1);
ctrsyl_("N", "N", &c_n1, &n1, &n2, &t[t_offset], ldt, &t_ref(n1 + 1,
n1 + 1), ldt, &work[1], &n1, &scale, &ierr);
/* Estimate the reciprocal of the condition number of the cluster
of eigenvalues. */
rnorm = clange_("F", &n1, &n2, &work[1], &n1, rwork);
if (rnorm == 0.f) {
*s = 1.f;
} else {
*s = scale / (sqrt(scale * scale / rnorm + rnorm) * sqrt(rnorm));
}
}
if (wantsp) {
/* Estimate sep(T11,T22). */
est = 0.f;
kase = 0;
L30:
clacon_(&nn, &work[nn + 1], &work[1], &est, &kase);
if (kase != 0) {
if (kase == 1) {
/* Solve T11*R - R*T22 = scale*X. */
ctrsyl_("N", "N", &c_n1, &n1, &n2, &t[t_offset], ldt, &t_ref(
n1 + 1, n1 + 1), ldt, &work[1], &n1, &scale, &ierr);
} else {
/* Solve T11'*R - R*T22' = scale*X. */
ctrsyl_("C", "C", &c_n1, &n1, &n2, &t[t_offset], ldt, &t_ref(
n1 + 1, n1 + 1), ldt, &work[1], &n1, &scale, &ierr);
}
goto L30;
}
*sep = scale / est;
}
L40:
/* Copy reordered eigenvalues to W. */
i__1 = *n;
for (k = 1; k <= i__1; ++k) {
i__2 = k;
i__3 = t_subscr(k, k);
w[i__2].r = t[i__3].r, w[i__2].i = t[i__3].i;
/* L50: */
}
work[1].r = (real) lwmin, work[1].i = 0.f;
return 0;
/* End of CTRSEN */
} /* ctrsen_ */
/* Subroutine */ int csycon_(char *uplo, integer *n, complex *a, integer *lda,
integer *ipiv, real *anorm, real *rcond, complex *work, integer *
info)
{
/* -- LAPACK routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
March 31, 1993
Purpose
=======
CSYCON estimates the reciprocal of the condition number (in the
1-norm) of a complex symmetric matrix A using the factorization
A = U*D*U**T or A = L*D*L**T computed by CSYTRF.
An estimate is obtained for norm(inv(A)), and the reciprocal of the
condition number is computed as RCOND = 1 / (ANORM * norm(inv(A))).
Arguments
=========
UPLO (input) CHARACTER*1
Specifies whether the details of the factorization are stored
as an upper or lower triangular matrix.
= 'U': Upper triangular, form is A = U*D*U**T;
= 'L': Lower triangular, form is A = L*D*L**T.
N (input) INTEGER
The order of the matrix A. N >= 0.
A (input) COMPLEX array, dimension (LDA,N)
The block diagonal matrix D and the multipliers used to
obtain the factor U or L as computed by CSYTRF.
LDA (input) INTEGER
The leading dimension of the array A. LDA >= max(1,N).
IPIV (input) INTEGER array, dimension (N)
Details of the interchanges and the block structure of D
as determined by CSYTRF.
ANORM (input) REAL
The 1-norm of the original matrix A.
RCOND (output) REAL
The reciprocal of the condition number of the matrix A,
computed as RCOND = 1/(ANORM * AINVNM), where AINVNM is an
estimate of the 1-norm of inv(A) computed in this routine.
WORK (workspace) COMPLEX array, dimension (2*N)
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
=====================================================================
Test the input parameters.
Parameter adjustments */
/* Table of constant values */
static integer c__1 = 1;
/* System generated locals */
integer a_dim1, a_offset, i__1, i__2;
/* Local variables */
static integer kase, i__;
extern logical lsame_(char *, char *);
static logical upper;
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *), xerbla_(char *, integer *);
static real ainvnm;
extern /* Subroutine */ int csytrs_(char *, integer *, integer *, complex
*, integer *, integer *, complex *, integer *, integer *);
#define a_subscr(a_1,a_2) (a_2)*a_dim1 + a_1
#define a_ref(a_1,a_2) a[a_subscr(a_1,a_2)]
a_dim1 = *lda;
a_offset = 1 + a_dim1 * 1;
a -= a_offset;
--ipiv;
--work;
/* Function Body */
*info = 0;
upper = lsame_(uplo, "U");
if (! upper && ! lsame_(uplo, "L")) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*lda < max(1,*n)) {
*info = -4;
} else if (*anorm < 0.f) {
*info = -6;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CSYCON", &i__1);
return 0;
}
/* Quick return if possible */
*rcond = 0.f;
if (*n == 0) {
*rcond = 1.f;
return 0;
} else if (*anorm <= 0.f) {
return 0;
}
/* Check that the diagonal matrix D is nonsingular. */
if (upper) {
/* Upper triangular storage: examine D from bottom to top */
for (i__ = *n; i__ >= 1; --i__) {
i__1 = a_subscr(i__, i__);
if (ipiv[i__] > 0 && (a[i__1].r == 0.f && a[i__1].i == 0.f)) {
return 0;
}
/* L10: */
}
} else {
/* Lower triangular storage: examine D from top to bottom. */
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = a_subscr(i__, i__);
if (ipiv[i__] > 0 && (a[i__2].r == 0.f && a[i__2].i == 0.f)) {
return 0;
}
/* L20: */
}
}
/* Estimate the 1-norm of the inverse. */
kase = 0;
L30:
clacon_(n, &work[*n + 1], &work[1], &ainvnm, &kase);
if (kase != 0) {
/* Multiply by inv(L*D*L') or inv(U*D*U'). */
csytrs_(uplo, n, &c__1, &a[a_offset], lda, &ipiv[1], &work[1], n,
info);
goto L30;
}
/* Compute the estimate of the reciprocal condition number. */
if (ainvnm != 0.f) {
*rcond = 1.f / ainvnm / *anorm;
}
return 0;
/* End of CSYCON */
} /* csycon_ */
/* Subroutine */ int ctgsen_(integer *ijob, logical *wantq, logical *wantz,
logical *select, integer *n, complex *a, integer *lda, complex *b,
integer *ldb, complex *alpha, complex *beta, complex *q, integer *ldq,
complex *z__, integer *ldz, integer *m, real *pl, real *pr, real *
dif, complex *work, integer *lwork, integer *iwork, integer *liwork,
integer *info)
{
/* -- LAPACK routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
June 30, 1999
Purpose
=======
CTGSEN reorders the generalized Schur decomposition of a complex
matrix pair (A, B) (in terms of an unitary equivalence trans-
formation Q' * (A, B) * Z), so that a selected cluster of eigenvalues
appears in the leading diagonal blocks of the pair (A,B). The leading
columns of Q and Z form unitary bases of the corresponding left and
right eigenspaces (deflating subspaces). (A, B) must be in
generalized Schur canonical form, that is, A and B are both upper
triangular.
CTGSEN also computes the generalized eigenvalues
w(j)= ALPHA(j) / BETA(j)
of the reordered matrix pair (A, B).
Optionally, the routine computes estimates of reciprocal condition
numbers for eigenvalues and eigenspaces. These are Difu[(A11,B11),
(A22,B22)] and Difl[(A11,B11), (A22,B22)], i.e. the separation(s)
between the matrix pairs (A11, B11) and (A22,B22) that correspond to
the selected cluster and the eigenvalues outside the cluster, resp.,
and norms of "projections" onto left and right eigenspaces w.r.t.
the selected cluster in the (1,1)-block.
Arguments
=========
IJOB (input) integer
Specifies whether condition numbers are required for the
cluster of eigenvalues (PL and PR) or the deflating subspaces
(Difu and Difl):
=0: Only reorder w.r.t. SELECT. No extras.
=1: Reciprocal of norms of "projections" onto left and right
eigenspaces w.r.t. the selected cluster (PL and PR).
=2: Upper bounds on Difu and Difl. F-norm-based estimate
(DIF(1:2)).
=3: Estimate of Difu and Difl. 1-norm-based estimate
(DIF(1:2)).
About 5 times as expensive as IJOB = 2.
=4: Compute PL, PR and DIF (i.e. 0, 1 and 2 above): Economic
version to get it all.
=5: Compute PL, PR and DIF (i.e. 0, 1 and 3 above)
WANTQ (input) LOGICAL
.TRUE. : update the left transformation matrix Q;
.FALSE.: do not update Q.
WANTZ (input) LOGICAL
.TRUE. : update the right transformation matrix Z;
.FALSE.: do not update Z.
SELECT (input) LOGICAL array, dimension (N)
SELECT specifies the eigenvalues in the selected cluster. To
select an eigenvalue w(j), SELECT(j) must be set to
.TRUE..
N (input) INTEGER
The order of the matrices A and B. N >= 0.
A (input/output) COMPLEX array, dimension(LDA,N)
On entry, the upper triangular matrix A, in generalized
Schur canonical form.
On exit, A is overwritten by the reordered matrix A.
LDA (input) INTEGER
The leading dimension of the array A. LDA >= max(1,N).
B (input/output) COMPLEX array, dimension(LDB,N)
On entry, the upper triangular matrix B, in generalized
Schur canonical form.
On exit, B is overwritten by the reordered matrix B.
LDB (input) INTEGER
The leading dimension of the array B. LDB >= max(1,N).
ALPHA (output) COMPLEX array, dimension (N)
BETA (output) COMPLEX array, dimension (N)
The diagonal elements of A and B, respectively,
when the pair (A,B) has been reduced to generalized Schur
form. ALPHA(i)/BETA(i) i=1,...,N are the generalized
eigenvalues.
Q (input/output) COMPLEX array, dimension (LDQ,N)
On entry, if WANTQ = .TRUE., Q is an N-by-N matrix.
On exit, Q has been postmultiplied by the left unitary
transformation matrix which reorder (A, B); The leading M
columns of Q form orthonormal bases for the specified pair of
left eigenspaces (deflating subspaces).
If WANTQ = .FALSE., Q is not referenced.
LDQ (input) INTEGER
The leading dimension of the array Q. LDQ >= 1.
If WANTQ = .TRUE., LDQ >= N.
Z (input/output) COMPLEX array, dimension (LDZ,N)
On entry, if WANTZ = .TRUE., Z is an N-by-N matrix.
On exit, Z has been postmultiplied by the left unitary
transformation matrix which reorder (A, B); The leading M
columns of Z form orthonormal bases for the specified pair of
left eigenspaces (deflating subspaces).
If WANTZ = .FALSE., Z is not referenced.
LDZ (input) INTEGER
The leading dimension of the array Z. LDZ >= 1.
If WANTZ = .TRUE., LDZ >= N.
M (output) INTEGER
The dimension of the specified pair of left and right
eigenspaces, (deflating subspaces) 0 <= M <= N.
PL, PR (output) REAL
If IJOB = 1, 4 or 5, PL, PR are lower bounds on the
reciprocal of the norm of "projections" onto left and right
eigenspace with respect to the selected cluster.
0 < PL, PR <= 1.
If M = 0 or M = N, PL = PR = 1.
If IJOB = 0, 2 or 3 PL, PR are not referenced.
DIF (output) REAL array, dimension (2).
If IJOB >= 2, DIF(1:2) store the estimates of Difu and Difl.
If IJOB = 2 or 4, DIF(1:2) are F-norm-based upper bounds on
Difu and Difl. If IJOB = 3 or 5, DIF(1:2) are 1-norm-based
estimates of Difu and Difl, computed using reversed
communication with CLACON.
If M = 0 or N, DIF(1:2) = F-norm([A, B]).
If IJOB = 0 or 1, DIF is not referenced.
WORK (workspace/output) COMPLEX array, dimension (LWORK)
IF IJOB = 0, WORK is not referenced. Otherwise,
on exit, if INFO = 0, WORK(1) returns the optimal LWORK.
LWORK (input) INTEGER
The dimension of the array WORK. LWORK >= 1
If IJOB = 1, 2 or 4, LWORK >= 2*M*(N-M)
If IJOB = 3 or 5, LWORK >= 4*M*(N-M)
If LWORK = -1, then a workspace query is assumed; the routine
only calculates the optimal size of the WORK array, returns
this value as the first entry of the WORK array, and no error
message related to LWORK is issued by XERBLA.
IWORK (workspace/output) INTEGER, dimension (LIWORK)
IF IJOB = 0, IWORK is not referenced. Otherwise,
on exit, if INFO = 0, IWORK(1) returns the optimal LIWORK.
LIWORK (input) INTEGER
The dimension of the array IWORK. LIWORK >= 1.
If IJOB = 1, 2 or 4, LIWORK >= N+2;
If IJOB = 3 or 5, LIWORK >= MAX(N+2, 2*M*(N-M));
If LIWORK = -1, then a workspace query is assumed; the
routine only calculates the optimal size of the IWORK array,
returns this value as the first entry of the IWORK array, and
no error message related to LIWORK is issued by XERBLA.
INFO (output) INTEGER
=0: Successful exit.
<0: If INFO = -i, the i-th argument had an illegal value.
=1: Reordering of (A, B) failed because the transformed
matrix pair (A, B) would be too far from generalized
Schur form; the problem is very ill-conditioned.
(A, B) may have been partially reordered.
If requested, 0 is returned in DIF(*), PL and PR.
Further Details
===============
CTGSEN first collects the selected eigenvalues by computing unitary
U and W that move them to the top left corner of (A, B). In other
words, the selected eigenvalues are the eigenvalues of (A11, B11) in
U'*(A, B)*W = (A11 A12) (B11 B12) n1
( 0 A22),( 0 B22) n2
n1 n2 n1 n2
where N = n1+n2 and U' means the conjugate transpose of U. The first
n1 columns of U and W span the specified pair of left and right
eigenspaces (deflating subspaces) of (A, B).
If (A, B) has been obtained from the generalized real Schur
decomposition of a matrix pair (C, D) = Q*(A, B)*Z', then the
reordered generalized Schur form of (C, D) is given by
(C, D) = (Q*U)*(U'*(A, B)*W)*(Z*W)',
and the first n1 columns of Q*U and Z*W span the corresponding
deflating subspaces of (C, D) (Q and Z store Q*U and Z*W, resp.).
Note that if the selected eigenvalue is sufficiently ill-conditioned,
then its value may differ significantly from its value before
reordering.
The reciprocal condition numbers of the left and right eigenspaces
spanned by the first n1 columns of U and W (or Q*U and Z*W) may
be returned in DIF(1:2), corresponding to Difu and Difl, resp.
The Difu and Difl are defined as:
Difu[(A11, B11), (A22, B22)] = sigma-min( Zu )
and
Difl[(A11, B11), (A22, B22)] = Difu[(A22, B22), (A11, B11)],
where sigma-min(Zu) is the smallest singular value of the
(2*n1*n2)-by-(2*n1*n2) matrix
Zu = [ kron(In2, A11) -kron(A22', In1) ]
[ kron(In2, B11) -kron(B22', In1) ].
Here, Inx is the identity matrix of size nx and A22' is the
transpose of A22. kron(X, Y) is the Kronecker product between
the matrices X and Y.
When DIF(2) is small, small changes in (A, B) can cause large changes
in the deflating subspace. An approximate (asymptotic) bound on the
maximum angular error in the computed deflating subspaces is
EPS * norm((A, B)) / DIF(2),
where EPS is the machine precision.
The reciprocal norm of the projectors on the left and right
eigenspaces associated with (A11, B11) may be returned in PL and PR.
They are computed as follows. First we compute L and R so that
P*(A, B)*Q is block diagonal, where
P = ( I -L ) n1 Q = ( I R ) n1
( 0 I ) n2 and ( 0 I ) n2
n1 n2 n1 n2
and (L, R) is the solution to the generalized Sylvester equation
A11*R - L*A22 = -A12
B11*R - L*B22 = -B12
Then PL = (F-norm(L)**2+1)**(-1/2) and PR = (F-norm(R)**2+1)**(-1/2).
An approximate (asymptotic) bound on the average absolute error of
the selected eigenvalues is
EPS * norm((A, B)) / PL.
There are also global error bounds which valid for perturbations up
to a certain restriction: A lower bound (x) on the smallest
F-norm(E,F) for which an eigenvalue of (A11, B11) may move and
coalesce with an eigenvalue of (A22, B22) under perturbation (E,F),
(i.e. (A + E, B + F), is
x = min(Difu,Difl)/((1/(PL*PL)+1/(PR*PR))**(1/2)+2*max(1/PL,1/PR)).
An approximate bound on x can be computed from DIF(1:2), PL and PR.
If y = ( F-norm(E,F) / x) <= 1, the angles between the perturbed
(L', R') and unperturbed (L, R) left and right deflating subspaces
associated with the selected cluster in the (1,1)-blocks can be
bounded as
max-angle(L, L') <= arctan( y * PL / (1 - y * (1 - PL * PL)**(1/2))
max-angle(R, R') <= arctan( y * PR / (1 - y * (1 - PR * PR)**(1/2))
See LAPACK User's Guide section 4.11 or the following references
for more information.
Note that if the default method for computing the Frobenius-norm-
based estimate DIF is not wanted (see CLATDF), then the parameter
IDIFJB (see below) should be changed from 3 to 4 (routine CLATDF
(IJOB = 2 will be used)). See CTGSYL for more details.
Based on contributions by
Bo Kagstrom and Peter Poromaa, Department of Computing Science,
Umea University, S-901 87 Umea, Sweden.
References
==========
[1] B. Kagstrom; A Direct Method for Reordering Eigenvalues in the
Generalized Real Schur Form of a Regular Matrix Pair (A, B), in
M.S. Moonen et al (eds), Linear Algebra for Large Scale and
Real-Time Applications, Kluwer Academic Publ. 1993, pp 195-218.
[2] B. Kagstrom and P. Poromaa; Computing Eigenspaces with Specified
Eigenvalues of a Regular Matrix Pair (A, B) and Condition
Estimation: Theory, Algorithms and Software, Report
UMINF - 94.04, Department of Computing Science, Umea University,
S-901 87 Umea, Sweden, 1994. Also as LAPACK Working Note 87.
To appear in Numerical Algorithms, 1996.
[3] B. Kagstrom and P. Poromaa, LAPACK-Style Algorithms and Software
for Solving the Generalized Sylvester Equation and Estimating the
Separation between Regular Matrix Pairs, Report UMINF - 93.23,
Department of Computing Science, Umea University, S-901 87 Umea,
Sweden, December 1993, Revised April 1994, Also as LAPACK working
Note 75. To appear in ACM Trans. on Math. Software, Vol 22, No 1,
1996.
=====================================================================
Decode and test the input parameters
Parameter adjustments */
/* Table of constant values */
static integer c__1 = 1;
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, q_dim1, q_offset, z_dim1,
z_offset, i__1, i__2, i__3;
complex q__1, q__2;
/* Builtin functions */
double sqrt(doublereal), c_abs(complex *);
void r_cnjg(complex *, complex *);
/* Local variables */
static integer kase, ierr;
static real dsum;
static logical swap;
static integer i__, k;
extern /* Subroutine */ int cscal_(integer *, complex *, complex *,
integer *);
static logical wantd;
static integer lwmin;
static logical wantp;
static integer n1, n2;
static logical wantd1, wantd2;
static real dscale;
static integer ks;
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *);
extern doublereal slamch_(char *);
static real rdscal;
extern /* Subroutine */ int clacpy_(char *, integer *, integer *, complex
*, integer *, complex *, integer *);
static real safmin;
extern /* Subroutine */ int ctgexc_(logical *, logical *, integer *,
complex *, integer *, complex *, integer *, complex *, integer *,
complex *, integer *, integer *, integer *, integer *), xerbla_(
char *, integer *), classq_(integer *, complex *, integer
*, real *, real *);
static integer liwmin;
extern /* Subroutine */ int ctgsyl_(char *, integer *, integer *, integer
*, complex *, integer *, complex *, integer *, complex *, integer
*, complex *, integer *, complex *, integer *, complex *, integer
*, real *, real *, complex *, integer *, integer *, integer *);
static integer mn2;
static logical lquery;
static integer ijb;
#define a_subscr(a_1,a_2) (a_2)*a_dim1 + a_1
#define a_ref(a_1,a_2) a[a_subscr(a_1,a_2)]
#define b_subscr(a_1,a_2) (a_2)*b_dim1 + a_1
#define b_ref(a_1,a_2) b[b_subscr(a_1,a_2)]
#define q_subscr(a_1,a_2) (a_2)*q_dim1 + a_1
#define q_ref(a_1,a_2) q[q_subscr(a_1,a_2)]
--select;
a_dim1 = *lda;
a_offset = 1 + a_dim1 * 1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1 * 1;
b -= b_offset;
--alpha;
--beta;
q_dim1 = *ldq;
q_offset = 1 + q_dim1 * 1;
q -= q_offset;
z_dim1 = *ldz;
z_offset = 1 + z_dim1 * 1;
z__ -= z_offset;
--dif;
--work;
--iwork;
/* Function Body */
*info = 0;
lquery = *lwork == -1 || *liwork == -1;
if (*ijob < 0 || *ijob > 5) {
*info = -1;
} else if (*n < 0) {
*info = -5;
} else if (*lda < max(1,*n)) {
*info = -7;
} else if (*ldb < max(1,*n)) {
*info = -9;
} else if (*ldq < 1 || *wantq && *ldq < *n) {
*info = -13;
} else if (*ldz < 1 || *wantz && *ldz < *n) {
*info = -15;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CTGSEN", &i__1);
return 0;
}
ierr = 0;
wantp = *ijob == 1 || *ijob >= 4;
wantd1 = *ijob == 2 || *ijob == 4;
wantd2 = *ijob == 3 || *ijob == 5;
wantd = wantd1 || wantd2;
/* Set M to the dimension of the specified pair of deflating
subspaces. */
*m = 0;
i__1 = *n;
for (k = 1; k <= i__1; ++k) {
i__2 = k;
i__3 = a_subscr(k, k);
alpha[i__2].r = a[i__3].r, alpha[i__2].i = a[i__3].i;
i__2 = k;
i__3 = b_subscr(k, k);
beta[i__2].r = b[i__3].r, beta[i__2].i = b[i__3].i;
if (k < *n) {
if (select[k]) {
++(*m);
}
} else {
if (select[*n]) {
++(*m);
}
}
/* L10: */
}
if (*ijob == 1 || *ijob == 2 || *ijob == 4) {
/* Computing MAX */
i__1 = 1, i__2 = (*m << 1) * (*n - *m);
lwmin = max(i__1,i__2);
/* Computing MAX */
i__1 = 1, i__2 = *n + 2;
liwmin = max(i__1,i__2);
} else if (*ijob == 3 || *ijob == 5) {
/* Computing MAX */
i__1 = 1, i__2 = (*m << 2) * (*n - *m);
lwmin = max(i__1,i__2);
/* Computing MAX */
i__1 = 1, i__2 = (*m << 1) * (*n - *m), i__1 = max(i__1,i__2), i__2 =
*n + 2;
liwmin = max(i__1,i__2);
} else {
lwmin = 1;
liwmin = 1;
}
work[1].r = (real) lwmin, work[1].i = 0.f;
iwork[1] = liwmin;
if (*lwork < lwmin && ! lquery) {
*info = -21;
} else if (*liwork < liwmin && ! lquery) {
*info = -23;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CTGSEN", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible. */
if (*m == *n || *m == 0) {
if (wantp) {
*pl = 1.f;
*pr = 1.f;
}
if (wantd) {
dscale = 0.f;
dsum = 1.f;
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
classq_(n, &a_ref(1, i__), &c__1, &dscale, &dsum);
classq_(n, &b_ref(1, i__), &c__1, &dscale, &dsum);
/* L20: */
}
dif[1] = dscale * sqrt(dsum);
dif[2] = dif[1];
}
goto L70;
}
/* Get machine constant */
safmin = slamch_("S");
/* Collect the selected blocks at the top-left corner of (A, B). */
ks = 0;
i__1 = *n;
for (k = 1; k <= i__1; ++k) {
swap = select[k];
if (swap) {
++ks;
/* Swap the K-th block to position KS. Compute unitary Q
and Z that will swap adjacent diagonal blocks in (A, B). */
if (k != ks) {
ctgexc_(wantq, wantz, n, &a[a_offset], lda, &b[b_offset], ldb,
&q[q_offset], ldq, &z__[z_offset], ldz, &k, &ks, &
ierr);
}
if (ierr > 0) {
/* Swap is rejected: exit. */
*info = 1;
if (wantp) {
*pl = 0.f;
*pr = 0.f;
}
if (wantd) {
dif[1] = 0.f;
dif[2] = 0.f;
}
goto L70;
}
}
/* L30: */
}
if (wantp) {
/* Solve generalized Sylvester equation for R and L:
A11 * R - L * A22 = A12
B11 * R - L * B22 = B12 */
n1 = *m;
n2 = *n - *m;
i__ = n1 + 1;
clacpy_("Full", &n1, &n2, &a_ref(1, i__), lda, &work[1], &n1);
clacpy_("Full", &n1, &n2, &b_ref(1, i__), ldb, &work[n1 * n2 + 1], &
n1);
ijb = 0;
i__1 = *lwork - (n1 << 1) * n2;
ctgsyl_("N", &ijb, &n1, &n2, &a[a_offset], lda, &a_ref(i__, i__), lda,
&work[1], &n1, &b[b_offset], ldb, &b_ref(i__, i__), ldb, &
work[n1 * n2 + 1], &n1, &dscale, &dif[1], &work[(n1 * n2 << 1)
+ 1], &i__1, &iwork[1], &ierr);
/* Estimate the reciprocal of norms of "projections" onto
left and right eigenspaces */
rdscal = 0.f;
dsum = 1.f;
i__1 = n1 * n2;
classq_(&i__1, &work[1], &c__1, &rdscal, &dsum);
*pl = rdscal * sqrt(dsum);
if (*pl == 0.f) {
*pl = 1.f;
} else {
*pl = dscale / (sqrt(dscale * dscale / *pl + *pl) * sqrt(*pl));
}
rdscal = 0.f;
dsum = 1.f;
i__1 = n1 * n2;
classq_(&i__1, &work[n1 * n2 + 1], &c__1, &rdscal, &dsum);
*pr = rdscal * sqrt(dsum);
if (*pr == 0.f) {
*pr = 1.f;
} else {
*pr = dscale / (sqrt(dscale * dscale / *pr + *pr) * sqrt(*pr));
}
}
if (wantd) {
/* Compute estimates Difu and Difl. */
if (wantd1) {
n1 = *m;
n2 = *n - *m;
i__ = n1 + 1;
ijb = 3;
/* Frobenius norm-based Difu estimate. */
i__1 = *lwork - (n1 << 1) * n2;
ctgsyl_("N", &ijb, &n1, &n2, &a[a_offset], lda, &a_ref(i__, i__),
lda, &work[1], &n1, &b[b_offset], ldb, &b_ref(i__, i__),
ldb, &work[n1 * n2 + 1], &n1, &dscale, &dif[1], &work[(n1
* n2 << 1) + 1], &i__1, &iwork[1], &ierr);
/* Frobenius norm-based Difl estimate. */
i__1 = *lwork - (n1 << 1) * n2;
ctgsyl_("N", &ijb, &n2, &n1, &a_ref(i__, i__), lda, &a[a_offset],
lda, &work[1], &n2, &b_ref(i__, i__), ldb, &b[b_offset],
ldb, &work[n1 * n2 + 1], &n2, &dscale, &dif[2], &work[(n1
* n2 << 1) + 1], &i__1, &iwork[1], &ierr);
} else {
/* Compute 1-norm-based estimates of Difu and Difl using
reversed communication with CLACON. In each step a
generalized Sylvester equation or a transposed variant
is solved. */
kase = 0;
n1 = *m;
n2 = *n - *m;
i__ = n1 + 1;
ijb = 0;
mn2 = (n1 << 1) * n2;
/* 1-norm-based estimate of Difu. */
L40:
clacon_(&mn2, &work[mn2 + 1], &work[1], &dif[1], &kase);
if (kase != 0) {
if (kase == 1) {
/* Solve generalized Sylvester equation */
i__1 = *lwork - (n1 << 1) * n2;
ctgsyl_("N", &ijb, &n1, &n2, &a[a_offset], lda, &a_ref(
i__, i__), lda, &work[1], &n1, &b[b_offset], ldb,
&b_ref(i__, i__), ldb, &work[n1 * n2 + 1], &n1, &
dscale, &dif[1], &work[(n1 * n2 << 1) + 1], &i__1,
&iwork[1], &ierr);
} else {
/* Solve the transposed variant. */
i__1 = *lwork - (n1 << 1) * n2;
ctgsyl_("C", &ijb, &n1, &n2, &a[a_offset], lda, &a_ref(
i__, i__), lda, &work[1], &n1, &b[b_offset], ldb,
&b_ref(i__, i__), ldb, &work[n1 * n2 + 1], &n1, &
dscale, &dif[1], &work[(n1 * n2 << 1) + 1], &i__1,
&iwork[1], &ierr);
}
goto L40;
}
dif[1] = dscale / dif[1];
/* 1-norm-based estimate of Difl. */
L50:
clacon_(&mn2, &work[mn2 + 1], &work[1], &dif[2], &kase);
if (kase != 0) {
if (kase == 1) {
/* Solve generalized Sylvester equation */
i__1 = *lwork - (n1 << 1) * n2;
ctgsyl_("N", &ijb, &n2, &n1, &a_ref(i__, i__), lda, &a[
a_offset], lda, &work[1], &n2, &b_ref(i__, i__),
ldb, &b[b_offset], ldb, &work[n1 * n2 + 1], &n2, &
dscale, &dif[2], &work[(n1 * n2 << 1) + 1], &i__1,
&iwork[1], &ierr);
} else {
/* Solve the transposed variant. */
i__1 = *lwork - (n1 << 1) * n2;
ctgsyl_("C", &ijb, &n2, &n1, &a_ref(i__, i__), lda, &a[
a_offset], lda, &work[1], &n2, &b[b_offset], ldb,
&b_ref(i__, i__), ldb, &work[n1 * n2 + 1], &n2, &
dscale, &dif[2], &work[(n1 * n2 << 1) + 1], &i__1,
&iwork[1], &ierr);
}
goto L50;
}
dif[2] = dscale / dif[2];
}
}
/* If B(K,K) is complex, make it real and positive (normalization
of the generalized Schur form) and Store the generalized
eigenvalues of reordered pair (A, B) */
i__1 = *n;
for (k = 1; k <= i__1; ++k) {
dscale = c_abs(&b_ref(k, k));
if (dscale > safmin) {
i__2 = b_subscr(k, k);
q__2.r = b[i__2].r / dscale, q__2.i = b[i__2].i / dscale;
r_cnjg(&q__1, &q__2);
work[1].r = q__1.r, work[1].i = q__1.i;
i__2 = b_subscr(k, k);
q__1.r = b[i__2].r / dscale, q__1.i = b[i__2].i / dscale;
work[2].r = q__1.r, work[2].i = q__1.i;
i__2 = b_subscr(k, k);
b[i__2].r = dscale, b[i__2].i = 0.f;
i__2 = *n - k;
cscal_(&i__2, &work[1], &b_ref(k, k + 1), ldb);
i__2 = *n - k + 1;
cscal_(&i__2, &work[1], &a_ref(k, k), lda);
if (*wantq) {
cscal_(n, &work[2], &q_ref(1, k), &c__1);
}
} else {
i__2 = b_subscr(k, k);
b[i__2].r = 0.f, b[i__2].i = 0.f;
}
i__2 = k;
i__3 = a_subscr(k, k);
alpha[i__2].r = a[i__3].r, alpha[i__2].i = a[i__3].i;
i__2 = k;
i__3 = b_subscr(k, k);
beta[i__2].r = b[i__3].r, beta[i__2].i = b[i__3].i;
/* L60: */
}
L70:
work[1].r = (real) lwmin, work[1].i = 0.f;
iwork[1] = liwmin;
return 0;
/* End of CTGSEN */
} /* ctgsen_ */
/* Subroutine */ int cgtcon_(char *norm, integer *n, complex *dl, complex *d,
complex *du, complex *du2, integer *ipiv, real *anorm, real *rcond,
complex *work, integer *info)
{
/* -- LAPACK routine (version 2.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
September 30, 1994
Purpose
=======
CGTCON estimates the reciprocal of the condition number of a complex
tridiagonal matrix A using the LU factorization as computed by
CGTTRF.
An estimate is obtained for norm(inv(A)), and the reciprocal of the
condition number is computed as RCOND = 1 / (ANORM * norm(inv(A))).
Arguments
=========
NORM (input) CHARACTER*1
Specifies whether the 1-norm condition number or the
infinity-norm condition number is required:
= '1' or 'O': 1-norm;
= 'I': Infinity-norm.
N (input) INTEGER
The order of the matrix A. N >= 0.
DL (input) COMPLEX array, dimension (N-1)
The (n-1) multipliers that define the matrix L from the
LU factorization of A as computed by CGTTRF.
D (input) COMPLEX array, dimension (N)
The n diagonal elements of the upper triangular matrix U from
the LU factorization of A.
DU (input) COMPLEX array, dimension (N-1)
The (n-1) elements of the first superdiagonal of U.
DU2 (input) COMPLEX array, dimension (N-2)
The (n-2) elements of the second superdiagonal of U.
IPIV (input) INTEGER array, dimension (N)
The pivot indices; for 1 <= i <= n, row i of the matrix was
interchanged with row IPIV(i). IPIV(i) will always be either
i or i+1; IPIV(i) = i indicates a row interchange was not
required.
ANORM (input) REAL
If NORM = '1' or 'O', the 1-norm of the original matrix A.
If NORM = 'I', the infinity-norm of the original matrix A.
RCOND (output) REAL
The reciprocal of the condition number of the matrix A,
computed as RCOND = 1/(ANORM * AINVNM), where AINVNM is an
estimate of the 1-norm of inv(A) computed in this routine.
WORK (workspace) COMPLEX array, dimension (2*N)
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
=====================================================================
Test the input arguments.
Parameter adjustments
Function Body */
/* Table of constant values */
static integer c__1 = 1;
/* System generated locals */
integer i__1, i__2;
/* Local variables */
static integer kase, kase1, i;
extern logical lsame_(char *, char *);
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *), xerbla_(char *, integer *);
static real ainvnm;
static logical onenrm;
extern /* Subroutine */ int cgttrs_(char *, integer *, integer *, complex
*, complex *, complex *, complex *, integer *, complex *, integer
*, integer *);
#define WORK(I) work[(I)-1]
#define IPIV(I) ipiv[(I)-1]
#define DU2(I) du2[(I)-1]
#define DU(I) du[(I)-1]
#define D(I) d[(I)-1]
#define DL(I) dl[(I)-1]
*info = 0;
onenrm = *(unsigned char *)norm == '1' || lsame_(norm, "O");
if (! onenrm && ! lsame_(norm, "I")) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*anorm < 0.f) {
*info = -8;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CGTCON", &i__1);
return 0;
}
/* Quick return if possible */
*rcond = 0.f;
if (*n == 0) {
*rcond = 1.f;
return 0;
} else if (*anorm == 0.f) {
return 0;
}
/* Check that D(1:N) is non-zero. */
i__1 = *n;
for (i = 1; i <= *n; ++i) {
i__2 = i;
if (D(i).r == 0.f && D(i).i == 0.f) {
return 0;
}
/* L10: */
}
ainvnm = 0.f;
if (onenrm) {
kase1 = 1;
} else {
kase1 = 2;
}
kase = 0;
L20:
clacon_(n, &WORK(*n + 1), &WORK(1), &ainvnm, &kase);
if (kase != 0) {
if (kase == kase1) {
/* Multiply by inv(U)*inv(L). */
cgttrs_("No transpose", n, &c__1, &DL(1), &D(1), &DU(1), &DU2(1),
&IPIV(1), &WORK(1), n, info);
} else {
/* Multiply by inv(L')*inv(U'). */
cgttrs_("Conjugate transpose", n, &c__1, &DL(1), &D(1), &DU(1), &
DU2(1), &IPIV(1), &WORK(1), n, info);
}
goto L20;
}
/* Compute the estimate of the reciprocal condition number. */
if (ainvnm != 0.f) {
*rcond = 1.f / ainvnm / *anorm;
}
return 0;
/* End of CGTCON */
} /* cgtcon_ */
/* Subroutine */ int cpbrfs_(char *uplo, integer *n, integer *kd, integer *
nrhs, complex *ab, integer *ldab, complex *afb, integer *ldafb,
complex *b, integer *ldb, complex *x, integer *ldx, real *ferr, real *
berr, complex *work, real *rwork, integer *info)
{
/* -- LAPACK routine (version 2.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
September 30, 1994
Purpose
=======
CPBRFS improves the computed solution to a system of linear
equations when the coefficient matrix is Hermitian positive definite
and banded, and provides error bounds and backward error estimates
for the solution.
Arguments
=========
UPLO (input) CHARACTER*1
= 'U': Upper triangle of A is stored;
= 'L': Lower triangle of A is stored.
N (input) INTEGER
The order of the matrix A. N >= 0.
KD (input) INTEGER
The number of superdiagonals of the matrix A if UPLO = 'U',
or the number of subdiagonals if UPLO = 'L'. KD >= 0.
NRHS (input) INTEGER
The number of right hand sides, i.e., the number of columns
of the matrices B and X. NRHS >= 0.
AB (input) REAL array, dimension (LDAB,N)
The upper or lower triangle of the Hermitian band matrix A,
stored in the first KD+1 rows of the array. The j-th column
of A is stored in the j-th column of the array AB as follows:
if UPLO = 'U', AB(kd+1+i-j,j) = A(i,j) for max(1,j-kd)<=i<=j;
if UPLO = 'L', AB(1+i-j,j) = A(i,j) for j<=i<=min(n,j+kd).
LDAB (input) INTEGER
The leading dimension of the array AB. LDAB >= KD+1.
AFB (input) COMPLEX array, dimension (LDAFB,N)
The triangular factor U or L from the Cholesky factorization
A = U**H*U or A = L*L**H of the band matrix A as computed by
CPBTRF, in the same storage format as A (see AB).
LDAFB (input) INTEGER
The leading dimension of the array AFB. LDAFB >= KD+1.
B (input) COMPLEX array, dimension (LDB,NRHS)
The right hand side matrix B.
LDB (input) INTEGER
The leading dimension of the array B. LDB >= max(1,N).
X (input/output) COMPLEX array, dimension (LDX,NRHS)
On entry, the solution matrix X, as computed by CPBTRS.
On exit, the improved solution matrix X.
LDX (input) INTEGER
The leading dimension of the array X. LDX >= max(1,N).
FERR (output) REAL array, dimension (NRHS)
The estimated forward error bound for each solution vector
X(j) (the j-th column of the solution matrix X).
If XTRUE is the true solution corresponding to X(j), FERR(j)
is an estimated upper bound for the magnitude of the largest
element in (X(j) - XTRUE) divided by the magnitude of the
largest element in X(j). The estimate is as reliable as
the estimate for RCOND, and is almost always a slight
overestimate of the true error.
BERR (output) REAL array, dimension (NRHS)
The componentwise relative backward error of each solution
vector X(j) (i.e., the smallest relative change in
any element of A or B that makes X(j) an exact solution).
WORK (workspace) COMPLEX array, dimension (2*N)
RWORK (workspace) REAL array, dimension (N)
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
Internal Parameters
===================
ITMAX is the maximum number of steps of iterative refinement.
=====================================================================
Test the input parameters.
Parameter adjustments
Function Body */
/* Table of constant values */
static complex c_b1 = {1.f,0.f};
static integer c__1 = 1;
/* System generated locals */
integer ab_dim1, ab_offset, afb_dim1, afb_offset, b_dim1, b_offset,
x_dim1, x_offset, i__1, i__2, i__3, i__4, i__5;
real r__1, r__2, r__3, r__4;
complex q__1;
/* Builtin functions */
double r_imag(complex *);
/* Local variables */
static integer kase;
static real safe1, safe2;
static integer i, j, k, l;
static real s;
extern /* Subroutine */ int chbmv_(char *, integer *, integer *, complex *
, complex *, integer *, complex *, integer *, complex *, complex *
, integer *);
extern logical lsame_(char *, char *);
extern /* Subroutine */ int ccopy_(integer *, complex *, integer *,
complex *, integer *), caxpy_(integer *, complex *, complex *,
integer *, complex *, integer *);
static integer count;
static logical upper;
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *);
static real xk;
extern doublereal slamch_(char *);
static integer nz;
static real safmin;
extern /* Subroutine */ int xerbla_(char *, integer *), cpbtrs_(
char *, integer *, integer *, integer *, complex *, integer *,
complex *, integer *, integer *);
static real lstres, eps;
#define FERR(I) ferr[(I)-1]
#define BERR(I) berr[(I)-1]
#define WORK(I) work[(I)-1]
#define RWORK(I) rwork[(I)-1]
#define AB(I,J) ab[(I)-1 + ((J)-1)* ( *ldab)]
#define AFB(I,J) afb[(I)-1 + ((J)-1)* ( *ldafb)]
#define B(I,J) b[(I)-1 + ((J)-1)* ( *ldb)]
#define X(I,J) x[(I)-1 + ((J)-1)* ( *ldx)]
*info = 0;
upper = lsame_(uplo, "U");
if (! upper && ! lsame_(uplo, "L")) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*kd < 0) {
*info = -3;
} else if (*nrhs < 0) {
*info = -4;
} else if (*ldab < *kd + 1) {
*info = -6;
} else if (*ldafb < *kd + 1) {
*info = -8;
} else if (*ldb < max(1,*n)) {
*info = -10;
} else if (*ldx < max(1,*n)) {
*info = -12;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CPBRFS", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0 || *nrhs == 0) {
i__1 = *nrhs;
for (j = 1; j <= *nrhs; ++j) {
FERR(j) = 0.f;
BERR(j) = 0.f;
/* L10: */
}
return 0;
}
/* NZ = maximum number of nonzero elements in each row of A, plus 1
Computing MIN */
i__1 = *n + 1, i__2 = (*kd << 1) + 2;
nz = min(i__1,i__2);
eps = slamch_("Epsilon");
safmin = slamch_("Safe minimum");
safe1 = nz * safmin;
safe2 = safe1 / eps;
/* Do for each right hand side */
i__1 = *nrhs;
for (j = 1; j <= *nrhs; ++j) {
count = 1;
lstres = 3.f;
L20:
/* Loop until stopping criterion is satisfied.
Compute residual R = B - A * X */
ccopy_(n, &B(1,j), &c__1, &WORK(1), &c__1);
q__1.r = -1.f, q__1.i = 0.f;
chbmv_(uplo, n, kd, &q__1, &AB(1,1), ldab, &X(1,j), &
c__1, &c_b1, &WORK(1), &c__1);
/* Compute componentwise relative backward error from formula
max(i) ( abs(R(i)) / ( abs(A)*abs(X) + abs(B) )(i) )
where abs(Z) is the componentwise absolute value of the matr
ix
or vector Z. If the i-th component of the denominator is le
ss
than SAFE2, then SAFE1 is added to the i-th components of th
e
numerator and denominator before dividing. */
i__2 = *n;
for (i = 1; i <= *n; ++i) {
i__3 = i + j * b_dim1;
RWORK(i) = (r__1 = B(i,j).r, dabs(r__1)) + (r__2 = r_imag(&B(i,j)), dabs(r__2));
/* L30: */
}
/* Compute abs(A)*abs(X) + abs(B). */
if (upper) {
i__2 = *n;
for (k = 1; k <= *n; ++k) {
s = 0.f;
i__3 = k + j * x_dim1;
xk = (r__1 = X(k,j).r, dabs(r__1)) + (r__2 = r_imag(&X(k,j)), dabs(r__2));
l = *kd + 1 - k;
/* Computing MAX */
i__3 = 1, i__4 = k - *kd;
i__5 = k - 1;
for (i = max(1,k-*kd); i <= k-1; ++i) {
i__3 = l + i + k * ab_dim1;
RWORK(i) += ((r__1 = AB(l+i,k).r, dabs(r__1)) + (r__2 =
r_imag(&AB(l+i,k)), dabs(r__2))) *
xk;
i__3 = l + i + k * ab_dim1;
i__4 = i + j * x_dim1;
s += ((r__1 = AB(l+i,k).r, dabs(r__1)) + (r__2 = r_imag(&
AB(l+i,k)), dabs(r__2))) * ((r__3 =
X(i,j).r, dabs(r__3)) + (r__4 = r_imag(&X(i,j)), dabs(r__4)));
/* L40: */
}
i__5 = *kd + 1 + k * ab_dim1;
RWORK(k) = RWORK(k) + (r__1 = AB(*kd+1,k).r, dabs(r__1)) * xk +
s;
/* L50: */
}
} else {
i__2 = *n;
for (k = 1; k <= *n; ++k) {
s = 0.f;
i__5 = k + j * x_dim1;
xk = (r__1 = X(k,j).r, dabs(r__1)) + (r__2 = r_imag(&X(k,j)), dabs(r__2));
i__5 = k * ab_dim1 + 1;
RWORK(k) += (r__1 = AB(1,k).r, dabs(r__1)) * xk;
l = 1 - k;
/* Computing MIN */
i__3 = *n, i__4 = k + *kd;
i__5 = min(i__3,i__4);
for (i = k + 1; i <= min(*n,k+*kd); ++i) {
i__3 = l + i + k * ab_dim1;
RWORK(i) += ((r__1 = AB(l+i,k).r, dabs(r__1)) + (r__2 =
r_imag(&AB(l+i,k)), dabs(r__2))) *
xk;
i__3 = l + i + k * ab_dim1;
i__4 = i + j * x_dim1;
s += ((r__1 = AB(l+i,k).r, dabs(r__1)) + (r__2 = r_imag(&
AB(l+i,k)), dabs(r__2))) * ((r__3 =
X(i,j).r, dabs(r__3)) + (r__4 = r_imag(&X(i,j)), dabs(r__4)));
/* L60: */
}
RWORK(k) += s;
/* L70: */
}
}
s = 0.f;
i__2 = *n;
for (i = 1; i <= *n; ++i) {
if (RWORK(i) > safe2) {
/* Computing MAX */
i__5 = i;
r__3 = s, r__4 = ((r__1 = WORK(i).r, dabs(r__1)) + (r__2 =
r_imag(&WORK(i)), dabs(r__2))) / RWORK(i);
s = dmax(r__3,r__4);
} else {
/* Computing MAX */
i__5 = i;
r__3 = s, r__4 = ((r__1 = WORK(i).r, dabs(r__1)) + (r__2 =
r_imag(&WORK(i)), dabs(r__2)) + safe1) / (RWORK(i) +
safe1);
s = dmax(r__3,r__4);
}
/* L80: */
}
BERR(j) = s;
/* Test stopping criterion. Continue iterating if
1) The residual BERR(J) is larger than machine epsilon, a
nd
2) BERR(J) decreased by at least a factor of 2 during the
last iteration, and
3) At most ITMAX iterations tried. */
if (BERR(j) > eps && BERR(j) * 2.f <= lstres && count <= 5) {
/* Update solution and try again. */
cpbtrs_(uplo, n, kd, &c__1, &AFB(1,1), ldafb, &WORK(1), n,
info);
caxpy_(n, &c_b1, &WORK(1), &c__1, &X(1,j), &c__1);
lstres = BERR(j);
++count;
goto L20;
}
/* Bound error from formula
norm(X - XTRUE) / norm(X) .le. FERR =
norm( abs(inv(A))*
( abs(R) + NZ*EPS*( abs(A)*abs(X)+abs(B) ))) / norm(X)
where
norm(Z) is the magnitude of the largest component of Z
inv(A) is the inverse of A
abs(Z) is the componentwise absolute value of the matrix o
r
vector Z
NZ is the maximum number of nonzeros in any row of A, plus
1
EPS is machine epsilon
The i-th component of abs(R)+NZ*EPS*(abs(A)*abs(X)+abs(B))
is incremented by SAFE1 if the i-th component of
abs(A)*abs(X) + abs(B) is less than SAFE2.
Use CLACON to estimate the infinity-norm of the matrix
inv(A) * diag(W),
where W = abs(R) + NZ*EPS*( abs(A)*abs(X)+abs(B) ))) */
i__2 = *n;
for (i = 1; i <= *n; ++i) {
if (RWORK(i) > safe2) {
i__5 = i;
RWORK(i) = (r__1 = WORK(i).r, dabs(r__1)) + (r__2 = r_imag(
&WORK(i)), dabs(r__2)) + nz * eps * RWORK(i);
} else {
i__5 = i;
RWORK(i) = (r__1 = WORK(i).r, dabs(r__1)) + (r__2 = r_imag(
&WORK(i)), dabs(r__2)) + nz * eps * RWORK(i) + safe1;
}
/* L90: */
}
kase = 0;
L100:
clacon_(n, &WORK(*n + 1), &WORK(1), &FERR(j), &kase);
if (kase != 0) {
if (kase == 1) {
/* Multiply by diag(W)*inv(A'). */
cpbtrs_(uplo, n, kd, &c__1, &AFB(1,1), ldafb, &WORK(1),
n, info);
i__2 = *n;
for (i = 1; i <= *n; ++i) {
i__5 = i;
i__3 = i;
i__4 = i;
q__1.r = RWORK(i) * WORK(i).r, q__1.i = RWORK(i)
* WORK(i).i;
WORK(i).r = q__1.r, WORK(i).i = q__1.i;
/* L110: */
}
} else if (kase == 2) {
/* Multiply by inv(A)*diag(W). */
i__2 = *n;
for (i = 1; i <= *n; ++i) {
i__5 = i;
i__3 = i;
i__4 = i;
q__1.r = RWORK(i) * WORK(i).r, q__1.i = RWORK(i)
* WORK(i).i;
WORK(i).r = q__1.r, WORK(i).i = q__1.i;
/* L120: */
}
cpbtrs_(uplo, n, kd, &c__1, &AFB(1,1), ldafb, &WORK(1),
n, info);
}
goto L100;
}
/* Normalize error. */
lstres = 0.f;
i__2 = *n;
for (i = 1; i <= *n; ++i) {
/* Computing MAX */
i__5 = i + j * x_dim1;
r__3 = lstres, r__4 = (r__1 = X(i,j).r, dabs(r__1)) + (r__2 =
r_imag(&X(i,j)), dabs(r__2));
lstres = dmax(r__3,r__4);
/* L130: */
}
if (lstres != 0.f) {
FERR(j) /= lstres;
}
/* L140: */
}
return 0;
/* End of CPBRFS */
} /* cpbrfs_ */
/* Subroutine */ int ctrsna_(char *job, char *howmny, logical *select,
integer *n, complex *t, integer *ldt, complex *vl, integer *ldvl,
complex *vr, integer *ldvr, real *s, real *sep, integer *mm, integer *
m, complex *work, integer *ldwork, real *rwork, integer *info)
{
/* -- LAPACK routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
September 30, 1994
Purpose
=======
CTRSNA estimates reciprocal condition numbers for specified
eigenvalues and/or right eigenvectors of a complex upper triangular
matrix T (or of any matrix Q*T*Q**H with Q unitary).
Arguments
=========
JOB (input) CHARACTER*1
Specifies whether condition numbers are required for
eigenvalues (S) or eigenvectors (SEP):
= 'E': for eigenvalues only (S);
= 'V': for eigenvectors only (SEP);
= 'B': for both eigenvalues and eigenvectors (S and SEP).
HOWMNY (input) CHARACTER*1
= 'A': compute condition numbers for all eigenpairs;
= 'S': compute condition numbers for selected eigenpairs
specified by the array SELECT.
SELECT (input) LOGICAL array, dimension (N)
If HOWMNY = 'S', SELECT specifies the eigenpairs for which
condition numbers are required. To select condition numbers
for the j-th eigenpair, SELECT(j) must be set to .TRUE..
If HOWMNY = 'A', SELECT is not referenced.
N (input) INTEGER
The order of the matrix T. N >= 0.
T (input) COMPLEX array, dimension (LDT,N)
The upper triangular matrix T.
LDT (input) INTEGER
The leading dimension of the array T. LDT >= max(1,N).
VL (input) COMPLEX array, dimension (LDVL,M)
If JOB = 'E' or 'B', VL must contain left eigenvectors of T
(or of any Q*T*Q**H with Q unitary), corresponding to the
eigenpairs specified by HOWMNY and SELECT. The eigenvectors
must be stored in consecutive columns of VL, as returned by
CHSEIN or CTREVC.
If JOB = 'V', VL is not referenced.
LDVL (input) INTEGER
The leading dimension of the array VL.
LDVL >= 1; and if JOB = 'E' or 'B', LDVL >= N.
VR (input) COMPLEX array, dimension (LDVR,M)
If JOB = 'E' or 'B', VR must contain right eigenvectors of T
(or of any Q*T*Q**H with Q unitary), corresponding to the
eigenpairs specified by HOWMNY and SELECT. The eigenvectors
must be stored in consecutive columns of VR, as returned by
CHSEIN or CTREVC.
If JOB = 'V', VR is not referenced.
LDVR (input) INTEGER
The leading dimension of the array VR.
LDVR >= 1; and if JOB = 'E' or 'B', LDVR >= N.
S (output) REAL array, dimension (MM)
If JOB = 'E' or 'B', the reciprocal condition numbers of the
selected eigenvalues, stored in consecutive elements of the
array. Thus S(j), SEP(j), and the j-th columns of VL and VR
all correspond to the same eigenpair (but not in general the
j-th eigenpair, unless all eigenpairs are selected).
If JOB = 'V', S is not referenced.
SEP (output) REAL array, dimension (MM)
If JOB = 'V' or 'B', the estimated reciprocal condition
numbers of the selected eigenvectors, stored in consecutive
elements of the array.
If JOB = 'E', SEP is not referenced.
MM (input) INTEGER
The number of elements in the arrays S (if JOB = 'E' or 'B')
and/or SEP (if JOB = 'V' or 'B'). MM >= M.
M (output) INTEGER
The number of elements of the arrays S and/or SEP actually
used to store the estimated condition numbers.
If HOWMNY = 'A', M is set to N.
WORK (workspace) COMPLEX array, dimension (LDWORK,N+1)
If JOB = 'E', WORK is not referenced.
LDWORK (input) INTEGER
The leading dimension of the array WORK.
LDWORK >= 1; and if JOB = 'V' or 'B', LDWORK >= N.
RWORK (workspace) REAL array, dimension (N)
If JOB = 'E', RWORK is not referenced.
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
Further Details
===============
The reciprocal of the condition number of an eigenvalue lambda is
defined as
S(lambda) = |v'*u| / (norm(u)*norm(v))
where u and v are the right and left eigenvectors of T corresponding
to lambda; v' denotes the conjugate transpose of v, and norm(u)
denotes the Euclidean norm. These reciprocal condition numbers always
lie between zero (very badly conditioned) and one (very well
conditioned). If n = 1, S(lambda) is defined to be 1.
An approximate error bound for a computed eigenvalue W(i) is given by
EPS * norm(T) / S(i)
where EPS is the machine precision.
The reciprocal of the condition number of the right eigenvector u
corresponding to lambda is defined as follows. Suppose
T = ( lambda c )
( 0 T22 )
Then the reciprocal condition number is
SEP( lambda, T22 ) = sigma-min( T22 - lambda*I )
where sigma-min denotes the smallest singular value. We approximate
the smallest singular value by the reciprocal of an estimate of the
one-norm of the inverse of T22 - lambda*I. If n = 1, SEP(1) is
defined to be abs(T(1,1)).
An approximate error bound for a computed right eigenvector VR(i)
is given by
EPS * norm(T) / SEP(i)
=====================================================================
Decode and test the input parameters
Parameter adjustments */
/* Table of constant values */
static integer c__1 = 1;
/* System generated locals */
integer t_dim1, t_offset, vl_dim1, vl_offset, vr_dim1, vr_offset,
work_dim1, work_offset, i__1, i__2, i__3, i__4, i__5;
real r__1, r__2;
complex q__1;
/* Builtin functions */
double c_abs(complex *), r_imag(complex *);
/* Local variables */
static integer kase, ierr;
static complex prod;
static real lnrm, rnrm;
static integer i__, j, k;
static real scale;
extern /* Complex */ VOID cdotc_(complex *, integer *, complex *, integer
*, complex *, integer *);
extern logical lsame_(char *, char *);
static complex dummy[1];
static logical wants;
static real xnorm;
extern doublereal scnrm2_(integer *, complex *, integer *);
extern /* Subroutine */ int slabad_(real *, real *);
static integer ks;
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *);
static integer ix;
extern integer icamax_(integer *, complex *, integer *);
extern doublereal slamch_(char *);
extern /* Subroutine */ int clacpy_(char *, integer *, integer *, complex
*, integer *, complex *, integer *), xerbla_(char *,
integer *);
static real bignum;
static logical wantbh;
extern /* Subroutine */ int clatrs_(char *, char *, char *, char *,
integer *, complex *, integer *, complex *, real *, real *,
integer *), csrscl_(integer *,
real *, complex *, integer *), ctrexc_(char *, integer *, complex
*, integer *, complex *, integer *, integer *, integer *, integer
*);
static logical somcon;
static char normin[1];
static real smlnum;
static logical wantsp;
static real eps, est;
#define work_subscr(a_1,a_2) (a_2)*work_dim1 + a_1
#define work_ref(a_1,a_2) work[work_subscr(a_1,a_2)]
#define t_subscr(a_1,a_2) (a_2)*t_dim1 + a_1
#define t_ref(a_1,a_2) t[t_subscr(a_1,a_2)]
#define vl_subscr(a_1,a_2) (a_2)*vl_dim1 + a_1
#define vl_ref(a_1,a_2) vl[vl_subscr(a_1,a_2)]
#define vr_subscr(a_1,a_2) (a_2)*vr_dim1 + a_1
#define vr_ref(a_1,a_2) vr[vr_subscr(a_1,a_2)]
--select;
t_dim1 = *ldt;
t_offset = 1 + t_dim1 * 1;
t -= t_offset;
vl_dim1 = *ldvl;
vl_offset = 1 + vl_dim1 * 1;
vl -= vl_offset;
vr_dim1 = *ldvr;
vr_offset = 1 + vr_dim1 * 1;
vr -= vr_offset;
--s;
--sep;
work_dim1 = *ldwork;
work_offset = 1 + work_dim1 * 1;
work -= work_offset;
--rwork;
/* Function Body */
wantbh = lsame_(job, "B");
wants = lsame_(job, "E") || wantbh;
wantsp = lsame_(job, "V") || wantbh;
somcon = lsame_(howmny, "S");
/* Set M to the number of eigenpairs for which condition numbers are
to be computed. */
if (somcon) {
*m = 0;
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
if (select[j]) {
++(*m);
}
/* L10: */
}
} else {
*m = *n;
}
*info = 0;
if (! wants && ! wantsp) {
*info = -1;
} else if (! lsame_(howmny, "A") && ! somcon) {
*info = -2;
} else if (*n < 0) {
*info = -4;
} else if (*ldt < max(1,*n)) {
*info = -6;
} else if (*ldvl < 1 || wants && *ldvl < *n) {
*info = -8;
} else if (*ldvr < 1 || wants && *ldvr < *n) {
*info = -10;
} else if (*mm < *m) {
*info = -13;
} else if (*ldwork < 1 || wantsp && *ldwork < *n) {
*info = -16;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CTRSNA", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
if (*n == 1) {
if (somcon) {
if (! select[1]) {
return 0;
}
}
if (wants) {
s[1] = 1.f;
}
if (wantsp) {
sep[1] = c_abs(&t_ref(1, 1));
}
return 0;
}
/* Get machine constants */
eps = slamch_("P");
smlnum = slamch_("S") / eps;
bignum = 1.f / smlnum;
slabad_(&smlnum, &bignum);
ks = 1;
i__1 = *n;
for (k = 1; k <= i__1; ++k) {
if (somcon) {
if (! select[k]) {
goto L50;
}
}
if (wants) {
/* Compute the reciprocal condition number of the k-th
eigenvalue. */
cdotc_(&q__1, n, &vr_ref(1, ks), &c__1, &vl_ref(1, ks), &c__1);
prod.r = q__1.r, prod.i = q__1.i;
rnrm = scnrm2_(n, &vr_ref(1, ks), &c__1);
lnrm = scnrm2_(n, &vl_ref(1, ks), &c__1);
s[ks] = c_abs(&prod) / (rnrm * lnrm);
}
if (wantsp) {
/* Estimate the reciprocal condition number of the k-th
eigenvector.
Copy the matrix T to the array WORK and swap the k-th
diagonal element to the (1,1) position. */
clacpy_("Full", n, n, &t[t_offset], ldt, &work[work_offset],
ldwork);
ctrexc_("No Q", n, &work[work_offset], ldwork, dummy, &c__1, &k, &
c__1, &ierr);
/* Form C = T22 - lambda*I in WORK(2:N,2:N). */
i__2 = *n;
for (i__ = 2; i__ <= i__2; ++i__) {
i__3 = work_subscr(i__, i__);
i__4 = work_subscr(i__, i__);
i__5 = work_subscr(1, 1);
q__1.r = work[i__4].r - work[i__5].r, q__1.i = work[i__4].i -
work[i__5].i;
work[i__3].r = q__1.r, work[i__3].i = q__1.i;
/* L20: */
}
/* Estimate a lower bound for the 1-norm of inv(C'). The 1st
and (N+1)th columns of WORK are used to store work vectors. */
sep[ks] = 0.f;
est = 0.f;
kase = 0;
*(unsigned char *)normin = 'N';
L30:
i__2 = *n - 1;
clacon_(&i__2, &work_ref(1, *n + 1), &work[work_offset], &est, &
kase);
if (kase != 0) {
if (kase == 1) {
/* Solve C'*x = scale*b */
i__2 = *n - 1;
clatrs_("Upper", "Conjugate transpose", "Nonunit", normin,
&i__2, &work_ref(2, 2), ldwork, &work[
work_offset], &scale, &rwork[1], &ierr);
} else {
/* Solve C*x = scale*b */
i__2 = *n - 1;
clatrs_("Upper", "No transpose", "Nonunit", normin, &i__2,
&work_ref(2, 2), ldwork, &work[work_offset], &
scale, &rwork[1], &ierr);
}
*(unsigned char *)normin = 'Y';
if (scale != 1.f) {
/* Multiply by 1/SCALE if doing so will not cause
overflow. */
i__2 = *n - 1;
ix = icamax_(&i__2, &work[work_offset], &c__1);
i__2 = work_subscr(ix, 1);
xnorm = (r__1 = work[i__2].r, dabs(r__1)) + (r__2 =
r_imag(&work_ref(ix, 1)), dabs(r__2));
if (scale < xnorm * smlnum || scale == 0.f) {
goto L40;
}
csrscl_(n, &scale, &work[work_offset], &c__1);
}
goto L30;
}
sep[ks] = 1.f / dmax(est,smlnum);
}
L40:
++ks;
L50:
;
}
return 0;
/* End of CTRSNA */
} /* ctrsna_ */
/* Subroutine */ int ctprfs_(char *uplo, char *trans, char *diag, integer *n,
integer *nrhs, complex *ap, complex *b, integer *ldb, complex *x,
integer *ldx, real *ferr, real *berr, complex *work, real *rwork,
integer *info)
{
/* -- LAPACK routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
September 30, 1994
Purpose
=======
CTPRFS provides error bounds and backward error estimates for the
solution to a system of linear equations with a triangular packed
coefficient matrix.
The solution matrix X must be computed by CTPTRS or some other
means before entering this routine. CTPRFS does not do iterative
refinement because doing so cannot improve the backward error.
Arguments
=========
UPLO (input) CHARACTER*1
= 'U': A is upper triangular;
= 'L': A is lower triangular.
TRANS (input) CHARACTER*1
Specifies the form of the system of equations:
= 'N': A * X = B (No transpose)
= 'T': A**T * X = B (Transpose)
= 'C': A**H * X = B (Conjugate transpose)
DIAG (input) CHARACTER*1
= 'N': A is non-unit triangular;
= 'U': A is unit triangular.
N (input) INTEGER
The order of the matrix A. N >= 0.
NRHS (input) INTEGER
The number of right hand sides, i.e., the number of columns
of the matrices B and X. NRHS >= 0.
AP (input) COMPLEX array, dimension (N*(N+1)/2)
The upper or lower triangular matrix A, packed columnwise in
a linear array. The j-th column of A is stored in the array
AP as follows:
if UPLO = 'U', AP(i + (j-1)*j/2) = A(i,j) for 1<=i<=j;
if UPLO = 'L', AP(i + (j-1)*(2n-j)/2) = A(i,j) for j<=i<=n.
If DIAG = 'U', the diagonal elements of A are not referenced
and are assumed to be 1.
B (input) COMPLEX array, dimension (LDB,NRHS)
The right hand side matrix B.
LDB (input) INTEGER
The leading dimension of the array B. LDB >= max(1,N).
X (input) COMPLEX array, dimension (LDX,NRHS)
The solution matrix X.
LDX (input) INTEGER
The leading dimension of the array X. LDX >= max(1,N).
FERR (output) REAL array, dimension (NRHS)
The estimated forward error bound for each solution vector
X(j) (the j-th column of the solution matrix X).
If XTRUE is the true solution corresponding to X(j), FERR(j)
is an estimated upper bound for the magnitude of the largest
element in (X(j) - XTRUE) divided by the magnitude of the
largest element in X(j). The estimate is as reliable as
the estimate for RCOND, and is almost always a slight
overestimate of the true error.
BERR (output) REAL array, dimension (NRHS)
The componentwise relative backward error of each solution
vector X(j) (i.e., the smallest relative change in
any element of A or B that makes X(j) an exact solution).
WORK (workspace) COMPLEX array, dimension (2*N)
RWORK (workspace) REAL array, dimension (N)
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
=====================================================================
Test the input parameters.
Parameter adjustments */
/* Table of constant values */
static integer c__1 = 1;
/* System generated locals */
integer b_dim1, b_offset, x_dim1, x_offset, i__1, i__2, i__3, i__4, i__5;
real r__1, r__2, r__3, r__4;
complex q__1;
/* Builtin functions */
double r_imag(complex *);
/* Local variables */
static integer kase;
static real safe1, safe2;
static integer i__, j, k;
static real s;
extern logical lsame_(char *, char *);
extern /* Subroutine */ int ccopy_(integer *, complex *, integer *,
complex *, integer *), caxpy_(integer *, complex *, complex *,
integer *, complex *, integer *), ctpmv_(char *, char *, char *,
integer *, complex *, complex *, integer *);
static logical upper;
extern /* Subroutine */ int ctpsv_(char *, char *, char *, integer *,
complex *, complex *, integer *);
static integer kc;
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *);
static real xk;
extern doublereal slamch_(char *);
static integer nz;
static real safmin;
extern /* Subroutine */ int xerbla_(char *, integer *);
static logical notran;
static char transn[1], transt[1];
static logical nounit;
static real lstres, eps;
#define b_subscr(a_1,a_2) (a_2)*b_dim1 + a_1
#define b_ref(a_1,a_2) b[b_subscr(a_1,a_2)]
#define x_subscr(a_1,a_2) (a_2)*x_dim1 + a_1
#define x_ref(a_1,a_2) x[x_subscr(a_1,a_2)]
--ap;
b_dim1 = *ldb;
b_offset = 1 + b_dim1 * 1;
b -= b_offset;
x_dim1 = *ldx;
x_offset = 1 + x_dim1 * 1;
x -= x_offset;
--ferr;
--berr;
--work;
--rwork;
/* Function Body */
*info = 0;
upper = lsame_(uplo, "U");
notran = lsame_(trans, "N");
nounit = lsame_(diag, "N");
if (! upper && ! lsame_(uplo, "L")) {
*info = -1;
} else if (! notran && ! lsame_(trans, "T") && !
lsame_(trans, "C")) {
*info = -2;
} else if (! nounit && ! lsame_(diag, "U")) {
*info = -3;
} else if (*n < 0) {
*info = -4;
} else if (*nrhs < 0) {
*info = -5;
} else if (*ldb < max(1,*n)) {
*info = -8;
} else if (*ldx < max(1,*n)) {
*info = -10;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CTPRFS", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0 || *nrhs == 0) {
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
ferr[j] = 0.f;
berr[j] = 0.f;
/* L10: */
}
return 0;
}
if (notran) {
*(unsigned char *)transn = 'N';
*(unsigned char *)transt = 'C';
} else {
*(unsigned char *)transn = 'C';
*(unsigned char *)transt = 'N';
}
/* NZ = maximum number of nonzero elements in each row of A, plus 1 */
nz = *n + 1;
eps = slamch_("Epsilon");
safmin = slamch_("Safe minimum");
safe1 = nz * safmin;
safe2 = safe1 / eps;
/* Do for each right hand side */
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
/* Compute residual R = B - op(A) * X,
where op(A) = A, A**T, or A**H, depending on TRANS. */
ccopy_(n, &x_ref(1, j), &c__1, &work[1], &c__1);
ctpmv_(uplo, trans, diag, n, &ap[1], &work[1], &c__1);
q__1.r = -1.f, q__1.i = 0.f;
caxpy_(n, &q__1, &b_ref(1, j), &c__1, &work[1], &c__1);
/* Compute componentwise relative backward error from formula
max(i) ( abs(R(i)) / ( abs(op(A))*abs(X) + abs(B) )(i) )
where abs(Z) is the componentwise absolute value of the matrix
or vector Z. If the i-th component of the denominator is less
than SAFE2, then SAFE1 is added to the i-th components of the
numerator and denominator before dividing. */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = b_subscr(i__, j);
rwork[i__] = (r__1 = b[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
b_ref(i__, j)), dabs(r__2));
/* L20: */
}
if (notran) {
/* Compute abs(A)*abs(X) + abs(B). */
if (upper) {
kc = 1;
if (nounit) {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
i__3 = x_subscr(k, j);
xk = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
x_ref(k, j)), dabs(r__2));
i__3 = k;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = kc + i__ - 1;
rwork[i__] += ((r__1 = ap[i__4].r, dabs(r__1)) + (
r__2 = r_imag(&ap[kc + i__ - 1]), dabs(
r__2))) * xk;
/* L30: */
}
kc += k;
/* L40: */
}
} else {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
i__3 = x_subscr(k, j);
xk = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
x_ref(k, j)), dabs(r__2));
i__3 = k - 1;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = kc + i__ - 1;
rwork[i__] += ((r__1 = ap[i__4].r, dabs(r__1)) + (
r__2 = r_imag(&ap[kc + i__ - 1]), dabs(
r__2))) * xk;
/* L50: */
}
rwork[k] += xk;
kc += k;
/* L60: */
}
}
} else {
kc = 1;
if (nounit) {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
i__3 = x_subscr(k, j);
xk = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
x_ref(k, j)), dabs(r__2));
i__3 = *n;
for (i__ = k; i__ <= i__3; ++i__) {
i__4 = kc + i__ - k;
rwork[i__] += ((r__1 = ap[i__4].r, dabs(r__1)) + (
r__2 = r_imag(&ap[kc + i__ - k]), dabs(
r__2))) * xk;
/* L70: */
}
kc = kc + *n - k + 1;
/* L80: */
}
} else {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
i__3 = x_subscr(k, j);
xk = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
x_ref(k, j)), dabs(r__2));
i__3 = *n;
for (i__ = k + 1; i__ <= i__3; ++i__) {
i__4 = kc + i__ - k;
rwork[i__] += ((r__1 = ap[i__4].r, dabs(r__1)) + (
r__2 = r_imag(&ap[kc + i__ - k]), dabs(
r__2))) * xk;
/* L90: */
}
rwork[k] += xk;
kc = kc + *n - k + 1;
/* L100: */
}
}
}
} else {
/* Compute abs(A**H)*abs(X) + abs(B). */
if (upper) {
kc = 1;
if (nounit) {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
s = 0.f;
i__3 = k;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = kc + i__ - 1;
i__5 = x_subscr(i__, j);
s += ((r__1 = ap[i__4].r, dabs(r__1)) + (r__2 =
r_imag(&ap[kc + i__ - 1]), dabs(r__2))) *
((r__3 = x[i__5].r, dabs(r__3)) + (r__4 =
r_imag(&x_ref(i__, j)), dabs(r__4)));
/* L110: */
}
rwork[k] += s;
kc += k;
/* L120: */
}
} else {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
i__3 = x_subscr(k, j);
s = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
x_ref(k, j)), dabs(r__2));
i__3 = k - 1;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = kc + i__ - 1;
i__5 = x_subscr(i__, j);
s += ((r__1 = ap[i__4].r, dabs(r__1)) + (r__2 =
r_imag(&ap[kc + i__ - 1]), dabs(r__2))) *
((r__3 = x[i__5].r, dabs(r__3)) + (r__4 =
r_imag(&x_ref(i__, j)), dabs(r__4)));
/* L130: */
}
rwork[k] += s;
kc += k;
/* L140: */
}
}
} else {
kc = 1;
if (nounit) {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
s = 0.f;
i__3 = *n;
for (i__ = k; i__ <= i__3; ++i__) {
i__4 = kc + i__ - k;
i__5 = x_subscr(i__, j);
s += ((r__1 = ap[i__4].r, dabs(r__1)) + (r__2 =
r_imag(&ap[kc + i__ - k]), dabs(r__2))) *
((r__3 = x[i__5].r, dabs(r__3)) + (r__4 =
r_imag(&x_ref(i__, j)), dabs(r__4)));
/* L150: */
}
rwork[k] += s;
kc = kc + *n - k + 1;
/* L160: */
}
} else {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
i__3 = x_subscr(k, j);
s = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
x_ref(k, j)), dabs(r__2));
i__3 = *n;
for (i__ = k + 1; i__ <= i__3; ++i__) {
i__4 = kc + i__ - k;
i__5 = x_subscr(i__, j);
s += ((r__1 = ap[i__4].r, dabs(r__1)) + (r__2 =
r_imag(&ap[kc + i__ - k]), dabs(r__2))) *
((r__3 = x[i__5].r, dabs(r__3)) + (r__4 =
r_imag(&x_ref(i__, j)), dabs(r__4)));
/* L170: */
}
rwork[k] += s;
kc = kc + *n - k + 1;
/* L180: */
}
}
}
}
s = 0.f;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
/* Computing MAX */
i__3 = i__;
r__3 = s, r__4 = ((r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2))) / rwork[i__];
s = dmax(r__3,r__4);
} else {
/* Computing MAX */
i__3 = i__;
r__3 = s, r__4 = ((r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + safe1) / (rwork[i__]
+ safe1);
s = dmax(r__3,r__4);
}
/* L190: */
}
berr[j] = s;
/* Bound error from formula
norm(X - XTRUE) / norm(X) .le. FERR =
norm( abs(inv(op(A)))*
( abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) / norm(X)
where
norm(Z) is the magnitude of the largest component of Z
inv(op(A)) is the inverse of op(A)
abs(Z) is the componentwise absolute value of the matrix or
vector Z
NZ is the maximum number of nonzeros in any row of A, plus 1
EPS is machine epsilon
The i-th component of abs(R)+NZ*EPS*(abs(op(A))*abs(X)+abs(B))
is incremented by SAFE1 if the i-th component of
abs(op(A))*abs(X) + abs(B) is less than SAFE2.
Use CLACON to estimate the infinity-norm of the matrix
inv(op(A)) * diag(W),
where W = abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
i__3 = i__;
rwork[i__] = (r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + nz * eps * rwork[
i__];
} else {
i__3 = i__;
rwork[i__] = (r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + nz * eps * rwork[
i__] + safe1;
}
/* L200: */
}
kase = 0;
L210:
clacon_(n, &work[*n + 1], &work[1], &ferr[j], &kase);
if (kase != 0) {
if (kase == 1) {
/* Multiply by diag(W)*inv(op(A)**H). */
ctpsv_(uplo, transt, diag, n, &ap[1], &work[1], &c__1);
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__;
i__4 = i__;
i__5 = i__;
q__1.r = rwork[i__4] * work[i__5].r, q__1.i = rwork[i__4]
* work[i__5].i;
work[i__3].r = q__1.r, work[i__3].i = q__1.i;
/* L220: */
}
} else {
/* Multiply by inv(op(A))*diag(W). */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__;
i__4 = i__;
i__5 = i__;
q__1.r = rwork[i__4] * work[i__5].r, q__1.i = rwork[i__4]
* work[i__5].i;
work[i__3].r = q__1.r, work[i__3].i = q__1.i;
/* L230: */
}
ctpsv_(uplo, transn, diag, n, &ap[1], &work[1], &c__1);
}
goto L210;
}
/* Normalize error. */
lstres = 0.f;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
/* Computing MAX */
i__3 = x_subscr(i__, j);
r__3 = lstres, r__4 = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&x_ref(i__, j)), dabs(r__2));
lstres = dmax(r__3,r__4);
/* L240: */
}
if (lstres != 0.f) {
ferr[j] /= lstres;
}
/* L250: */
}
return 0;
/* End of CTPRFS */
} /* ctprfs_ */
/* Subroutine */ int cpbrfs_(char *uplo, integer *n, integer *kd, integer *
nrhs, complex *ab, integer *ldab, complex *afb, integer *ldafb,
complex *b, integer *ldb, complex *x, integer *ldx, real *ferr, real *
berr, complex *work, real *rwork, integer *info, ftnlen uplo_len)
{
/* System generated locals */
integer ab_dim1, ab_offset, afb_dim1, afb_offset, b_dim1, b_offset,
x_dim1, x_offset, i__1, i__2, i__3, i__4, i__5;
real r__1, r__2, r__3, r__4;
complex q__1;
/* Builtin functions */
double r_imag(complex *);
/* Local variables */
static integer i__, j, k, l;
static real s, xk;
static integer nz;
static real eps;
static integer kase;
static real safe1, safe2;
extern /* Subroutine */ int chbmv_(char *, integer *, integer *, complex *
, complex *, integer *, complex *, integer *, complex *, complex *
, integer *, ftnlen);
extern logical lsame_(char *, char *, ftnlen, ftnlen);
extern /* Subroutine */ int ccopy_(integer *, complex *, integer *,
complex *, integer *), caxpy_(integer *, complex *, complex *,
integer *, complex *, integer *);
static integer count;
static logical upper;
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *);
extern doublereal slamch_(char *, ftnlen);
static real safmin;
extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen), cpbtrs_(
char *, integer *, integer *, integer *, complex *, integer *,
complex *, integer *, integer *, ftnlen);
static real lstres;
/* -- LAPACK routine (version 3.0) -- */
/* Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., */
/* Courant Institute, Argonne National Lab, and Rice University */
/* June 30, 1999 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* CPBRFS improves the computed solution to a system of linear */
/* equations when the coefficient matrix is Hermitian positive definite */
/* and banded, and provides error bounds and backward error estimates */
/* for the solution. */
/* Arguments */
/* ========= */
/* UPLO (input) CHARACTER*1 */
/* = 'U': Upper triangle of A is stored; */
/* = 'L': Lower triangle of A is stored. */
/* N (input) INTEGER */
/* The order of the matrix A. N >= 0. */
/* KD (input) INTEGER */
/* The number of superdiagonals of the matrix A if UPLO = 'U', */
/* or the number of subdiagonals if UPLO = 'L'. KD >= 0. */
/* NRHS (input) INTEGER */
/* The number of right hand sides, i.e., the number of columns */
/* of the matrices B and X. NRHS >= 0. */
/* AB (input) COMPLEX array, dimension (LDAB,N) */
/* The upper or lower triangle of the Hermitian band matrix A, */
/* stored in the first KD+1 rows of the array. The j-th column */
/* of A is stored in the j-th column of the array AB as follows: */
/* if UPLO = 'U', AB(kd+1+i-j,j) = A(i,j) for max(1,j-kd)<=i<=j; */
/* if UPLO = 'L', AB(1+i-j,j) = A(i,j) for j<=i<=min(n,j+kd). */
/* LDAB (input) INTEGER */
/* The leading dimension of the array AB. LDAB >= KD+1. */
/* AFB (input) COMPLEX array, dimension (LDAFB,N) */
/* The triangular factor U or L from the Cholesky factorization */
/* A = U**H*U or A = L*L**H of the band matrix A as computed by */
/* CPBTRF, in the same storage format as A (see AB). */
/* LDAFB (input) INTEGER */
/* The leading dimension of the array AFB. LDAFB >= KD+1. */
/* B (input) COMPLEX array, dimension (LDB,NRHS) */
/* The right hand side matrix B. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= max(1,N). */
/* X (input/output) COMPLEX array, dimension (LDX,NRHS) */
/* On entry, the solution matrix X, as computed by CPBTRS. */
/* On exit, the improved solution matrix X. */
/* LDX (input) INTEGER */
/* The leading dimension of the array X. LDX >= max(1,N). */
/* FERR (output) REAL array, dimension (NRHS) */
/* The estimated forward error bound for each solution vector */
/* X(j) (the j-th column of the solution matrix X). */
/* If XTRUE is the true solution corresponding to X(j), FERR(j) */
/* is an estimated upper bound for the magnitude of the largest */
/* element in (X(j) - XTRUE) divided by the magnitude of the */
/* largest element in X(j). The estimate is as reliable as */
/* the estimate for RCOND, and is almost always a slight */
/* overestimate of the true error. */
/* BERR (output) REAL array, dimension (NRHS) */
/* The componentwise relative backward error of each solution */
/* vector X(j) (i.e., the smallest relative change in */
/* any element of A or B that makes X(j) an exact solution). */
/* WORK (workspace) COMPLEX array, dimension (2*N) */
/* RWORK (workspace) REAL array, dimension (N) */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* Internal Parameters */
/* =================== */
/* ITMAX is the maximum number of steps of iterative refinement. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Statement Functions .. */
/* .. */
/* .. Statement Function definitions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
ab_dim1 = *ldab;
ab_offset = 1 + ab_dim1;
ab -= ab_offset;
afb_dim1 = *ldafb;
afb_offset = 1 + afb_dim1;
afb -= afb_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
x_dim1 = *ldx;
x_offset = 1 + x_dim1;
x -= x_offset;
--ferr;
--berr;
--work;
--rwork;
/* Function Body */
*info = 0;
upper = lsame_(uplo, "U", (ftnlen)1, (ftnlen)1);
if (! upper && ! lsame_(uplo, "L", (ftnlen)1, (ftnlen)1)) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*kd < 0) {
*info = -3;
} else if (*nrhs < 0) {
*info = -4;
} else if (*ldab < *kd + 1) {
*info = -6;
} else if (*ldafb < *kd + 1) {
*info = -8;
} else if (*ldb < max(1,*n)) {
*info = -10;
} else if (*ldx < max(1,*n)) {
*info = -12;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CPBRFS", &i__1, (ftnlen)6);
return 0;
}
/* Quick return if possible */
if (*n == 0 || *nrhs == 0) {
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
ferr[j] = 0.f;
berr[j] = 0.f;
/* L10: */
}
return 0;
}
/* NZ = maximum number of nonzero elements in each row of A, plus 1 */
/* Computing MIN */
i__1 = *n + 1, i__2 = (*kd << 1) + 2;
nz = min(i__1,i__2);
eps = slamch_("Epsilon", (ftnlen)7);
safmin = slamch_("Safe minimum", (ftnlen)12);
safe1 = nz * safmin;
safe2 = safe1 / eps;
/* Do for each right hand side */
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
count = 1;
lstres = 3.f;
L20:
/* Loop until stopping criterion is satisfied. */
/* Compute residual R = B - A * X */
ccopy_(n, &b[j * b_dim1 + 1], &c__1, &work[1], &c__1);
q__1.r = -1.f, q__1.i = -0.f;
chbmv_(uplo, n, kd, &q__1, &ab[ab_offset], ldab, &x[j * x_dim1 + 1], &
c__1, &c_b1, &work[1], &c__1, (ftnlen)1);
/* Compute componentwise relative backward error from formula */
/* max(i) ( abs(R(i)) / ( abs(A)*abs(X) + abs(B) )(i) ) */
/* where abs(Z) is the componentwise absolute value of the matrix */
/* or vector Z. If the i-th component of the denominator is less */
/* than SAFE2, then SAFE1 is added to the i-th components of the */
/* numerator and denominator before dividing. */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__ + j * b_dim1;
rwork[i__] = (r__1 = b[i__3].r, dabs(r__1)) + (r__2 = r_imag(&b[
i__ + j * b_dim1]), dabs(r__2));
/* L30: */
}
/* Compute abs(A)*abs(X) + abs(B). */
if (upper) {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
s = 0.f;
i__3 = k + j * x_dim1;
xk = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&x[k + j
* x_dim1]), dabs(r__2));
l = *kd + 1 - k;
/* Computing MAX */
i__3 = 1, i__4 = k - *kd;
i__5 = k - 1;
for (i__ = max(i__3,i__4); i__ <= i__5; ++i__) {
i__3 = l + i__ + k * ab_dim1;
rwork[i__] += ((r__1 = ab[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&ab[l + i__ + k * ab_dim1]), dabs(r__2))) *
xk;
i__3 = l + i__ + k * ab_dim1;
i__4 = i__ + j * x_dim1;
s += ((r__1 = ab[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
ab[l + i__ + k * ab_dim1]), dabs(r__2))) * ((r__3
= x[i__4].r, dabs(r__3)) + (r__4 = r_imag(&x[i__
+ j * x_dim1]), dabs(r__4)));
/* L40: */
}
i__5 = *kd + 1 + k * ab_dim1;
rwork[k] = rwork[k] + (r__1 = ab[i__5].r, dabs(r__1)) * xk +
s;
/* L50: */
}
} else {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
s = 0.f;
i__5 = k + j * x_dim1;
xk = (r__1 = x[i__5].r, dabs(r__1)) + (r__2 = r_imag(&x[k + j
* x_dim1]), dabs(r__2));
i__5 = k * ab_dim1 + 1;
rwork[k] += (r__1 = ab[i__5].r, dabs(r__1)) * xk;
l = 1 - k;
/* Computing MIN */
i__3 = *n, i__4 = k + *kd;
i__5 = min(i__3,i__4);
for (i__ = k + 1; i__ <= i__5; ++i__) {
i__3 = l + i__ + k * ab_dim1;
rwork[i__] += ((r__1 = ab[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&ab[l + i__ + k * ab_dim1]), dabs(r__2))) *
xk;
i__3 = l + i__ + k * ab_dim1;
i__4 = i__ + j * x_dim1;
s += ((r__1 = ab[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
ab[l + i__ + k * ab_dim1]), dabs(r__2))) * ((r__3
= x[i__4].r, dabs(r__3)) + (r__4 = r_imag(&x[i__
+ j * x_dim1]), dabs(r__4)));
/* L60: */
}
rwork[k] += s;
/* L70: */
}
}
s = 0.f;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
/* Computing MAX */
i__5 = i__;
r__3 = s, r__4 = ((r__1 = work[i__5].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2))) / rwork[i__];
s = dmax(r__3,r__4);
} else {
/* Computing MAX */
i__5 = i__;
r__3 = s, r__4 = ((r__1 = work[i__5].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + safe1) / (rwork[i__]
+ safe1);
s = dmax(r__3,r__4);
}
/* L80: */
}
berr[j] = s;
/* Test stopping criterion. Continue iterating if */
/* 1) The residual BERR(J) is larger than machine epsilon, and */
/* 2) BERR(J) decreased by at least a factor of 2 during the */
/* last iteration, and */
/* 3) At most ITMAX iterations tried. */
if (berr[j] > eps && berr[j] * 2.f <= lstres && count <= 5) {
/* Update solution and try again. */
cpbtrs_(uplo, n, kd, &c__1, &afb[afb_offset], ldafb, &work[1], n,
info, (ftnlen)1);
caxpy_(n, &c_b1, &work[1], &c__1, &x[j * x_dim1 + 1], &c__1);
lstres = berr[j];
++count;
goto L20;
}
/* Bound error from formula */
/* norm(X - XTRUE) / norm(X) .le. FERR = */
/* norm( abs(inv(A))* */
/* ( abs(R) + NZ*EPS*( abs(A)*abs(X)+abs(B) ))) / norm(X) */
/* where */
/* norm(Z) is the magnitude of the largest component of Z */
/* inv(A) is the inverse of A */
/* abs(Z) is the componentwise absolute value of the matrix or */
/* vector Z */
/* NZ is the maximum number of nonzeros in any row of A, plus 1 */
/* EPS is machine epsilon */
/* The i-th component of abs(R)+NZ*EPS*(abs(A)*abs(X)+abs(B)) */
/* is incremented by SAFE1 if the i-th component of */
/* abs(A)*abs(X) + abs(B) is less than SAFE2. */
/* Use CLACON to estimate the infinity-norm of the matrix */
/* inv(A) * diag(W), */
/* where W = abs(R) + NZ*EPS*( abs(A)*abs(X)+abs(B) ))) */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
i__5 = i__;
rwork[i__] = (r__1 = work[i__5].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + nz * eps * rwork[
i__];
} else {
i__5 = i__;
rwork[i__] = (r__1 = work[i__5].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + nz * eps * rwork[
i__] + safe1;
}
/* L90: */
}
kase = 0;
L100:
clacon_(n, &work[*n + 1], &work[1], &ferr[j], &kase);
if (kase != 0) {
if (kase == 1) {
/* Multiply by diag(W)*inv(A'). */
cpbtrs_(uplo, n, kd, &c__1, &afb[afb_offset], ldafb, &work[1],
n, info, (ftnlen)1);
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__5 = i__;
i__3 = i__;
i__4 = i__;
q__1.r = rwork[i__3] * work[i__4].r, q__1.i = rwork[i__3]
* work[i__4].i;
work[i__5].r = q__1.r, work[i__5].i = q__1.i;
/* L110: */
}
} else if (kase == 2) {
/* Multiply by inv(A)*diag(W). */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__5 = i__;
i__3 = i__;
i__4 = i__;
q__1.r = rwork[i__3] * work[i__4].r, q__1.i = rwork[i__3]
* work[i__4].i;
work[i__5].r = q__1.r, work[i__5].i = q__1.i;
/* L120: */
}
cpbtrs_(uplo, n, kd, &c__1, &afb[afb_offset], ldafb, &work[1],
n, info, (ftnlen)1);
}
goto L100;
}
/* Normalize error. */
lstres = 0.f;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
/* Computing MAX */
i__5 = i__ + j * x_dim1;
r__3 = lstres, r__4 = (r__1 = x[i__5].r, dabs(r__1)) + (r__2 =
r_imag(&x[i__ + j * x_dim1]), dabs(r__2));
lstres = dmax(r__3,r__4);
/* L130: */
}
if (lstres != 0.f) {
ferr[j] /= lstres;
}
/* L140: */
}
return 0;
/* End of CPBRFS */
} /* cpbrfs_ */
/* Subroutine */ int ctrrfs_(char *uplo, char *trans, char *diag, integer *n,
integer *nrhs, complex *a, integer *lda, complex *b, integer *ldb,
complex *x, integer *ldx, real *ferr, real *berr, complex *work, real
*rwork, integer *info, ftnlen uplo_len, ftnlen trans_len, ftnlen
diag_len)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, x_dim1, x_offset, i__1, i__2,
i__3, i__4, i__5;
real r__1, r__2, r__3, r__4;
complex q__1;
/* Builtin functions */
double r_imag(complex *);
/* Local variables */
static integer i__, j, k;
static real s, xk;
static integer nz;
static real eps;
static integer kase;
static real safe1, safe2;
extern logical lsame_(char *, char *, ftnlen, ftnlen);
extern /* Subroutine */ int ccopy_(integer *, complex *, integer *,
complex *, integer *), caxpy_(integer *, complex *, complex *,
integer *, complex *, integer *);
static logical upper;
extern /* Subroutine */ int ctrmv_(char *, char *, char *, integer *,
complex *, integer *, complex *, integer *, ftnlen, ftnlen,
ftnlen), ctrsv_(char *, char *, char *, integer *, complex *,
integer *, complex *, integer *, ftnlen, ftnlen, ftnlen), clacon_(
integer *, complex *, complex *, real *, integer *);
extern doublereal slamch_(char *, ftnlen);
static real safmin;
extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
static logical notran;
static char transn[1], transt[1];
static logical nounit;
static real lstres;
/* -- LAPACK routine (version 3.0) -- */
/* Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., */
/* Courant Institute, Argonne National Lab, and Rice University */
/* September 30, 1994 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* CTRRFS provides error bounds and backward error estimates for the */
/* solution to a system of linear equations with a triangular */
/* coefficient matrix. */
/* The solution matrix X must be computed by CTRTRS or some other */
/* means before entering this routine. CTRRFS does not do iterative */
/* refinement because doing so cannot improve the backward error. */
/* Arguments */
/* ========= */
/* UPLO (input) CHARACTER*1 */
/* = 'U': A is upper triangular; */
/* = 'L': A is lower triangular. */
/* TRANS (input) CHARACTER*1 */
/* Specifies the form of the system of equations: */
/* = 'N': A * X = B (No transpose) */
/* = 'T': A**T * X = B (Transpose) */
/* = 'C': A**H * X = B (Conjugate transpose) */
/* DIAG (input) CHARACTER*1 */
/* = 'N': A is non-unit triangular; */
/* = 'U': A is unit triangular. */
/* N (input) INTEGER */
/* The order of the matrix A. N >= 0. */
/* NRHS (input) INTEGER */
/* The number of right hand sides, i.e., the number of columns */
/* of the matrices B and X. NRHS >= 0. */
/* A (input) COMPLEX array, dimension (LDA,N) */
/* The triangular matrix A. If UPLO = 'U', the leading N-by-N */
/* upper triangular part of the array A contains the upper */
/* triangular matrix, and the strictly lower triangular part of */
/* A is not referenced. If UPLO = 'L', the leading N-by-N lower */
/* triangular part of the array A contains the lower triangular */
/* matrix, and the strictly upper triangular part of A is not */
/* referenced. If DIAG = 'U', the diagonal elements of A are */
/* also not referenced and are assumed to be 1. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1,N). */
/* B (input) COMPLEX array, dimension (LDB,NRHS) */
/* The right hand side matrix B. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= max(1,N). */
/* X (input) COMPLEX array, dimension (LDX,NRHS) */
/* The solution matrix X. */
/* LDX (input) INTEGER */
/* The leading dimension of the array X. LDX >= max(1,N). */
/* FERR (output) REAL array, dimension (NRHS) */
/* The estimated forward error bound for each solution vector */
/* X(j) (the j-th column of the solution matrix X). */
/* If XTRUE is the true solution corresponding to X(j), FERR(j) */
/* is an estimated upper bound for the magnitude of the largest */
/* element in (X(j) - XTRUE) divided by the magnitude of the */
/* largest element in X(j). The estimate is as reliable as */
/* the estimate for RCOND, and is almost always a slight */
/* overestimate of the true error. */
/* BERR (output) REAL array, dimension (NRHS) */
/* The componentwise relative backward error of each solution */
/* vector X(j) (i.e., the smallest relative change in */
/* any element of A or B that makes X(j) an exact solution). */
/* WORK (workspace) COMPLEX array, dimension (2*N) */
/* RWORK (workspace) REAL array, dimension (N) */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Statement Functions .. */
/* .. */
/* .. Statement Function definitions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
x_dim1 = *ldx;
x_offset = 1 + x_dim1;
x -= x_offset;
--ferr;
--berr;
--work;
--rwork;
/* Function Body */
*info = 0;
upper = lsame_(uplo, "U", (ftnlen)1, (ftnlen)1);
notran = lsame_(trans, "N", (ftnlen)1, (ftnlen)1);
nounit = lsame_(diag, "N", (ftnlen)1, (ftnlen)1);
if (! upper && ! lsame_(uplo, "L", (ftnlen)1, (ftnlen)1)) {
*info = -1;
} else if (! notran && ! lsame_(trans, "T", (ftnlen)1, (ftnlen)1) && !
lsame_(trans, "C", (ftnlen)1, (ftnlen)1)) {
*info = -2;
} else if (! nounit && ! lsame_(diag, "U", (ftnlen)1, (ftnlen)1)) {
*info = -3;
} else if (*n < 0) {
*info = -4;
} else if (*nrhs < 0) {
*info = -5;
} else if (*lda < max(1,*n)) {
*info = -7;
} else if (*ldb < max(1,*n)) {
*info = -9;
} else if (*ldx < max(1,*n)) {
*info = -11;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CTRRFS", &i__1, (ftnlen)6);
return 0;
}
/* Quick return if possible */
if (*n == 0 || *nrhs == 0) {
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
ferr[j] = 0.f;
berr[j] = 0.f;
/* L10: */
}
return 0;
}
if (notran) {
*(unsigned char *)transn = 'N';
*(unsigned char *)transt = 'C';
} else {
*(unsigned char *)transn = 'C';
*(unsigned char *)transt = 'N';
}
/* NZ = maximum number of nonzero elements in each row of A, plus 1 */
nz = *n + 1;
eps = slamch_("Epsilon", (ftnlen)7);
safmin = slamch_("Safe minimum", (ftnlen)12);
safe1 = nz * safmin;
safe2 = safe1 / eps;
/* Do for each right hand side */
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
/* Compute residual R = B - op(A) * X, */
/* where op(A) = A, A**T, or A**H, depending on TRANS. */
ccopy_(n, &x[j * x_dim1 + 1], &c__1, &work[1], &c__1);
ctrmv_(uplo, trans, diag, n, &a[a_offset], lda, &work[1], &c__1, (
ftnlen)1, (ftnlen)1, (ftnlen)1);
q__1.r = -1.f, q__1.i = -0.f;
caxpy_(n, &q__1, &b[j * b_dim1 + 1], &c__1, &work[1], &c__1);
/* Compute componentwise relative backward error from formula */
/* max(i) ( abs(R(i)) / ( abs(op(A))*abs(X) + abs(B) )(i) ) */
/* where abs(Z) is the componentwise absolute value of the matrix */
/* or vector Z. If the i-th component of the denominator is less */
/* than SAFE2, then SAFE1 is added to the i-th components of the */
/* numerator and denominator before dividing. */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__ + j * b_dim1;
rwork[i__] = (r__1 = b[i__3].r, dabs(r__1)) + (r__2 = r_imag(&b[
i__ + j * b_dim1]), dabs(r__2));
/* L20: */
}
if (notran) {
/* Compute abs(A)*abs(X) + abs(B). */
if (upper) {
if (nounit) {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
i__3 = k + j * x_dim1;
xk = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
x[k + j * x_dim1]), dabs(r__2));
i__3 = k;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = i__ + k * a_dim1;
rwork[i__] += ((r__1 = a[i__4].r, dabs(r__1)) + (
r__2 = r_imag(&a[i__ + k * a_dim1]), dabs(
r__2))) * xk;
/* L30: */
}
/* L40: */
}
} else {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
i__3 = k + j * x_dim1;
xk = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
x[k + j * x_dim1]), dabs(r__2));
i__3 = k - 1;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = i__ + k * a_dim1;
rwork[i__] += ((r__1 = a[i__4].r, dabs(r__1)) + (
r__2 = r_imag(&a[i__ + k * a_dim1]), dabs(
r__2))) * xk;
/* L50: */
}
rwork[k] += xk;
/* L60: */
}
}
} else {
if (nounit) {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
i__3 = k + j * x_dim1;
xk = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
x[k + j * x_dim1]), dabs(r__2));
i__3 = *n;
for (i__ = k; i__ <= i__3; ++i__) {
i__4 = i__ + k * a_dim1;
rwork[i__] += ((r__1 = a[i__4].r, dabs(r__1)) + (
r__2 = r_imag(&a[i__ + k * a_dim1]), dabs(
r__2))) * xk;
/* L70: */
}
/* L80: */
}
} else {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
i__3 = k + j * x_dim1;
xk = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
x[k + j * x_dim1]), dabs(r__2));
i__3 = *n;
for (i__ = k + 1; i__ <= i__3; ++i__) {
i__4 = i__ + k * a_dim1;
rwork[i__] += ((r__1 = a[i__4].r, dabs(r__1)) + (
r__2 = r_imag(&a[i__ + k * a_dim1]), dabs(
r__2))) * xk;
/* L90: */
}
rwork[k] += xk;
/* L100: */
}
}
}
} else {
/* Compute abs(A**H)*abs(X) + abs(B). */
if (upper) {
if (nounit) {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
s = 0.f;
i__3 = k;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = i__ + k * a_dim1;
i__5 = i__ + j * x_dim1;
s += ((r__1 = a[i__4].r, dabs(r__1)) + (r__2 =
r_imag(&a[i__ + k * a_dim1]), dabs(r__2)))
* ((r__3 = x[i__5].r, dabs(r__3)) + (
r__4 = r_imag(&x[i__ + j * x_dim1]), dabs(
r__4)));
/* L110: */
}
rwork[k] += s;
/* L120: */
}
} else {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
i__3 = k + j * x_dim1;
s = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
x[k + j * x_dim1]), dabs(r__2));
i__3 = k - 1;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = i__ + k * a_dim1;
i__5 = i__ + j * x_dim1;
s += ((r__1 = a[i__4].r, dabs(r__1)) + (r__2 =
r_imag(&a[i__ + k * a_dim1]), dabs(r__2)))
* ((r__3 = x[i__5].r, dabs(r__3)) + (
r__4 = r_imag(&x[i__ + j * x_dim1]), dabs(
r__4)));
/* L130: */
}
rwork[k] += s;
/* L140: */
}
}
} else {
if (nounit) {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
s = 0.f;
i__3 = *n;
for (i__ = k; i__ <= i__3; ++i__) {
i__4 = i__ + k * a_dim1;
i__5 = i__ + j * x_dim1;
s += ((r__1 = a[i__4].r, dabs(r__1)) + (r__2 =
r_imag(&a[i__ + k * a_dim1]), dabs(r__2)))
* ((r__3 = x[i__5].r, dabs(r__3)) + (
r__4 = r_imag(&x[i__ + j * x_dim1]), dabs(
r__4)));
/* L150: */
}
rwork[k] += s;
/* L160: */
}
} else {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
i__3 = k + j * x_dim1;
s = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
x[k + j * x_dim1]), dabs(r__2));
i__3 = *n;
for (i__ = k + 1; i__ <= i__3; ++i__) {
i__4 = i__ + k * a_dim1;
i__5 = i__ + j * x_dim1;
s += ((r__1 = a[i__4].r, dabs(r__1)) + (r__2 =
r_imag(&a[i__ + k * a_dim1]), dabs(r__2)))
* ((r__3 = x[i__5].r, dabs(r__3)) + (
r__4 = r_imag(&x[i__ + j * x_dim1]), dabs(
r__4)));
/* L170: */
}
rwork[k] += s;
/* L180: */
}
}
}
}
s = 0.f;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
/* Computing MAX */
i__3 = i__;
r__3 = s, r__4 = ((r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2))) / rwork[i__];
s = dmax(r__3,r__4);
} else {
/* Computing MAX */
i__3 = i__;
r__3 = s, r__4 = ((r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + safe1) / (rwork[i__]
+ safe1);
s = dmax(r__3,r__4);
}
/* L190: */
}
berr[j] = s;
/* Bound error from formula */
/* norm(X - XTRUE) / norm(X) .le. FERR = */
/* norm( abs(inv(op(A)))* */
/* ( abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) / norm(X) */
/* where */
/* norm(Z) is the magnitude of the largest component of Z */
/* inv(op(A)) is the inverse of op(A) */
/* abs(Z) is the componentwise absolute value of the matrix or */
/* vector Z */
/* NZ is the maximum number of nonzeros in any row of A, plus 1 */
/* EPS is machine epsilon */
/* The i-th component of abs(R)+NZ*EPS*(abs(op(A))*abs(X)+abs(B)) */
/* is incremented by SAFE1 if the i-th component of */
/* abs(op(A))*abs(X) + abs(B) is less than SAFE2. */
/* Use CLACON to estimate the infinity-norm of the matrix */
/* inv(op(A)) * diag(W), */
/* where W = abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
i__3 = i__;
rwork[i__] = (r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + nz * eps * rwork[
i__];
} else {
i__3 = i__;
rwork[i__] = (r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + nz * eps * rwork[
i__] + safe1;
}
/* L200: */
}
kase = 0;
L210:
clacon_(n, &work[*n + 1], &work[1], &ferr[j], &kase);
if (kase != 0) {
if (kase == 1) {
/* Multiply by diag(W)*inv(op(A)**H). */
ctrsv_(uplo, transt, diag, n, &a[a_offset], lda, &work[1], &
c__1, (ftnlen)1, (ftnlen)1, (ftnlen)1);
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__;
i__4 = i__;
i__5 = i__;
q__1.r = rwork[i__4] * work[i__5].r, q__1.i = rwork[i__4]
* work[i__5].i;
work[i__3].r = q__1.r, work[i__3].i = q__1.i;
/* L220: */
}
} else {
/* Multiply by inv(op(A))*diag(W). */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__;
i__4 = i__;
i__5 = i__;
q__1.r = rwork[i__4] * work[i__5].r, q__1.i = rwork[i__4]
* work[i__5].i;
work[i__3].r = q__1.r, work[i__3].i = q__1.i;
/* L230: */
}
ctrsv_(uplo, transn, diag, n, &a[a_offset], lda, &work[1], &
c__1, (ftnlen)1, (ftnlen)1, (ftnlen)1);
}
goto L210;
}
/* Normalize error. */
lstres = 0.f;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
/* Computing MAX */
i__3 = i__ + j * x_dim1;
r__3 = lstres, r__4 = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&x[i__ + j * x_dim1]), dabs(r__2));
lstres = dmax(r__3,r__4);
/* L240: */
}
if (lstres != 0.f) {
ferr[j] /= lstres;
}
/* L250: */
}
return 0;
/* End of CTRRFS */
} /* ctrrfs_ */
/* Subroutine */ int cpocon_(char *uplo, integer *n, complex *a, integer *lda,
real *anorm, real *rcond, complex *work, real *rwork, integer *info,
ftnlen uplo_len)
{
/* System generated locals */
integer a_dim1, a_offset, i__1;
real r__1, r__2;
/* Builtin functions */
double r_imag(complex *);
/* Local variables */
static integer ix, kase;
static real scale;
extern logical lsame_(char *, char *, ftnlen, ftnlen);
static logical upper;
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *);
extern integer icamax_(integer *, complex *, integer *);
static real scalel;
extern doublereal slamch_(char *, ftnlen);
static real scaleu;
extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
static real ainvnm;
extern /* Subroutine */ int clatrs_(char *, char *, char *, char *,
integer *, complex *, integer *, complex *, real *, real *,
integer *, ftnlen, ftnlen, ftnlen, ftnlen), csrscl_(integer *,
real *, complex *, integer *);
static char normin[1];
static real smlnum;
/* -- LAPACK routine (version 3.0) -- */
/* Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., */
/* Courant Institute, Argonne National Lab, and Rice University */
/* March 31, 1993 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* CPOCON estimates the reciprocal of the condition number (in the */
/* 1-norm) of a complex Hermitian positive definite matrix using the */
/* Cholesky factorization A = U**H*U or A = L*L**H computed by CPOTRF. */
/* An estimate is obtained for norm(inv(A)), and the reciprocal of the */
/* condition number is computed as RCOND = 1 / (ANORM * norm(inv(A))). */
/* Arguments */
/* ========= */
/* UPLO (input) CHARACTER*1 */
/* = 'U': Upper triangle of A is stored; */
/* = 'L': Lower triangle of A is stored. */
/* N (input) INTEGER */
/* The order of the matrix A. N >= 0. */
/* A (input) COMPLEX array, dimension (LDA,N) */
/* The triangular factor U or L from the Cholesky factorization */
/* A = U**H*U or A = L*L**H, as computed by CPOTRF. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1,N). */
/* ANORM (input) REAL */
/* The 1-norm (or infinity-norm) of the Hermitian matrix A. */
/* RCOND (output) REAL */
/* The reciprocal of the condition number of the matrix A, */
/* computed as RCOND = 1/(ANORM * AINVNM), where AINVNM is an */
/* estimate of the 1-norm of inv(A) computed in this routine. */
/* WORK (workspace) COMPLEX array, dimension (2*N) */
/* RWORK (workspace) REAL array, dimension (N) */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Statement Functions .. */
/* .. */
/* .. Statement Function definitions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
--work;
--rwork;
/* Function Body */
*info = 0;
upper = lsame_(uplo, "U", (ftnlen)1, (ftnlen)1);
if (! upper && ! lsame_(uplo, "L", (ftnlen)1, (ftnlen)1)) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*lda < max(1,*n)) {
*info = -4;
} else if (*anorm < 0.f) {
*info = -5;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CPOCON", &i__1, (ftnlen)6);
return 0;
}
/* Quick return if possible */
*rcond = 0.f;
if (*n == 0) {
*rcond = 1.f;
return 0;
} else if (*anorm == 0.f) {
return 0;
}
smlnum = slamch_("Safe minimum", (ftnlen)12);
/* Estimate the 1-norm of inv(A). */
kase = 0;
*(unsigned char *)normin = 'N';
L10:
clacon_(n, &work[*n + 1], &work[1], &ainvnm, &kase);
if (kase != 0) {
if (upper) {
/* Multiply by inv(U'). */
clatrs_("Upper", "Conjugate transpose", "Non-unit", normin, n, &a[
a_offset], lda, &work[1], &scalel, &rwork[1], info, (
ftnlen)5, (ftnlen)19, (ftnlen)8, (ftnlen)1);
*(unsigned char *)normin = 'Y';
/* Multiply by inv(U). */
clatrs_("Upper", "No transpose", "Non-unit", normin, n, &a[
a_offset], lda, &work[1], &scaleu, &rwork[1], info, (
ftnlen)5, (ftnlen)12, (ftnlen)8, (ftnlen)1);
} else {
/* Multiply by inv(L). */
clatrs_("Lower", "No transpose", "Non-unit", normin, n, &a[
a_offset], lda, &work[1], &scalel, &rwork[1], info, (
ftnlen)5, (ftnlen)12, (ftnlen)8, (ftnlen)1);
*(unsigned char *)normin = 'Y';
/* Multiply by inv(L'). */
clatrs_("Lower", "Conjugate transpose", "Non-unit", normin, n, &a[
a_offset], lda, &work[1], &scaleu, &rwork[1], info, (
ftnlen)5, (ftnlen)19, (ftnlen)8, (ftnlen)1);
}
/* Multiply by 1/SCALE if doing so will not cause overflow. */
scale = scalel * scaleu;
if (scale != 1.f) {
ix = icamax_(n, &work[1], &c__1);
i__1 = ix;
if (scale < ((r__1 = work[i__1].r, dabs(r__1)) + (r__2 = r_imag(&
work[ix]), dabs(r__2))) * smlnum || scale == 0.f) {
goto L20;
}
csrscl_(n, &scale, &work[1], &c__1);
}
goto L10;
}
/* Compute the estimate of the reciprocal condition number. */
if (ainvnm != 0.f) {
*rcond = 1.f / ainvnm / *anorm;
}
L20:
return 0;
/* End of CPOCON */
} /* cpocon_ */
void
cgsrfs(trans_t trans, SuperMatrix *A, SuperMatrix *L, SuperMatrix *U,
int *perm_r, int *perm_c, equed_t equed, float *R, float *C,
SuperMatrix *B, SuperMatrix *X, float *ferr, float *berr,
Gstat_t *Gstat, int *info)
{
/*
* -- SuperLU MT routine (version 2.0) --
* Lawrence Berkeley National Lab, Univ. of California Berkeley,
* and Xerox Palo Alto Research Center.
* September 10, 2007
*
*
* Purpose
* =======
*
* cgsrfs improves the computed solution to a system of linear
* equations and provides error bounds and backward error estimates for
* the solution.
*
* See supermatrix.h for the definition of 'SuperMatrix' structure.
*
* Arguments
* =========
*
* trans (input) trans_t
* Specifies the form of the system of equations:
* = NOTRANS: A * X = B (No transpose)
* = TRANS: A**T * X = B (Transpose)
* = CONJ: A**H * X = B (Conjugate transpose = Transpose)
*
* A (input) SuperMatrix*
* The original matrix A in the system, or the scaled A if
* equilibration was done. The type of A can be:
* Stype = NC, Dtype = _D, Mtype = GE.
*
* L (input) SuperMatrix*
* The factor L from the factorization Pr*A*Pc=L*U. Use
* compressed row subscripts storage for supernodes,
* i.e., L has types: Stype = SCP, Dtype = _D, Mtype = TRLU.
*
* U (input) SuperMatrix*
* The factor U from the factorization Pr*A*Pc=L*U as computed by
* dgstrf(). Use column-wise storage scheme,
* i.e., U has types: Stype = NCP, Dtype = _D, Mtype = TRU.
*
* perm_r (input) int*, dimension (A->nrow)
* Row permutation vector, which defines the permutation matrix Pr;
* perm_r[i] = j means row i of A is in position j in Pr*A.
*
* perm_c (input) int*, dimension (A->ncol)
* Column permutation vector, which defines the
* permutation matrix Pc; perm_c[i] = j means column i of A is
* in position j in A*Pc.
*
* equed (input) equed_t
* Specifies the form of equilibration that was done.
* = NOEQUIL: No equilibration.
* = ROW: Row equilibration, i.e., A was premultiplied by diag(R).
* = COL: Column equilibration, i.e., A was postmultiplied by
* diag(C).
* = BOTH: Both row and column equilibration, i.e., A was replaced
* by diag(R)*A*diag(C).
*
* R (input) double*, dimension (A->nrow)
* The row scale factors for A.
* If equed = ROW or BOTH, A is premultiplied by diag(R).
* If equed = NOEQUIL or COL, R is not accessed.
*
* C (input) double*, dimension (A->ncol)
* The column scale factors for A.
* If equed = COL or BOTH, A is postmultiplied by diag(C).
* If equed = NOEQUIL or ROW, C is not accessed.
*
* B (input) SuperMatrix*
* B has types: Stype = DN, Dtype = _D, Mtype = GE.
* The right hand side matrix B.
*
* X (input/output) SuperMatrix*
* X has types: Stype = DN, Dtype = _D, Mtype = GE.
* On entry, the solution matrix X, as computed by dgstrs().
* On exit, the improved solution matrix X.
*
* FERR (output) double*, dimension (B->ncol)
* The estimated forward error bound for each solution vector
* X(j) (the j-th column of the solution matrix X).
* If XTRUE is the true solution corresponding to X(j), FERR(j)
* is an estimated upper bound for the magnitude of the largest
* element in (X(j) - XTRUE) divided by the magnitude of the
* largest element in X(j). The estimate is as reliable as
* the estimate for RCOND, and is almost always a slight
* overestimate of the true error.
*
* BERR (output) double*, dimension (B->ncol)
* The componentwise relative backward error of each solution
* vector X(j) (i.e., the smallest relative change in
* any element of A or B that makes X(j) an exact solution).
*
* info (output) int*
* = 0: successful exit
* < 0: if INFO = -i, the i-th argument had an illegal value
*
* Internal Parameters
* ===================
*
* ITMAX is the maximum number of steps of iterative refinement.
*
*/
#define ITMAX 5
/* Table of constant values */
int ione = 1;
complex ndone = {-1., 0.};
complex done = {1., 0.};
/* Local variables */
NCformat *Astore;
complex *Aval;
SuperMatrix Bjcol;
DNformat *Bstore, *Xstore, *Bjcol_store;
complex *Bmat, *Xmat, *Bptr, *Xptr;
int kase;
float safe1, safe2;
int i, j, k, irow, nz, count, notran, rowequ, colequ;
int ldb, ldx, nrhs;
float s, xk, lstres, eps, safmin;
char transc[1];
trans_t transt;
complex *work;
float *rwork;
int *iwork;
extern double slamch_(char *);
extern int clacon_(int *, complex *, complex *, float *, int *);
#ifdef _CRAY
extern int CCOPY(int *, complex *, int *, complex *, int *);
extern int CSAXPY(int *, complex *, complex *, int *, complex *, int *);
#else
extern int ccopy_(int *, complex *, int *, complex *, int *);
extern int caxpy_(int *, complex *, complex *, int *, complex *, int *);
#endif
Astore = A->Store;
Aval = Astore->nzval;
Bstore = B->Store;
Xstore = X->Store;
Bmat = Bstore->nzval;
Xmat = Xstore->nzval;
ldb = Bstore->lda;
ldx = Xstore->lda;
nrhs = B->ncol;
/* Test the input parameters */
*info = 0;
notran = (trans == NOTRANS);
if ( !notran && trans != TRANS && trans != CONJ ) *info = -1;
else if ( A->nrow != A->ncol || A->nrow < 0 ||
A->Stype != SLU_NC || A->Dtype != SLU_C || A->Mtype != SLU_GE )
*info = -2;
else if ( L->nrow != L->ncol || L->nrow < 0 ||
L->Stype != SLU_SCP || L->Dtype != SLU_C || L->Mtype != SLU_TRLU )
*info = -3;
else if ( U->nrow != U->ncol || U->nrow < 0 ||
U->Stype != SLU_NCP || U->Dtype != SLU_C || U->Mtype != SLU_TRU )
*info = -4;
else if ( ldb < SUPERLU_MAX(0, A->nrow) ||
B->Stype != SLU_DN || B->Dtype != SLU_C || B->Mtype != SLU_GE )
*info = -10;
else if ( ldx < SUPERLU_MAX(0, A->nrow) ||
X->Stype != SLU_DN || X->Dtype != SLU_C || X->Mtype != SLU_GE )
*info = -11;
if (*info != 0) {
i = -(*info);
xerbla_("cgsrfs", &i);
return;
}
/* Quick return if possible */
if ( A->nrow == 0 || nrhs == 0) {
for (j = 0; j < nrhs; ++j) {
ferr[j] = 0.;
berr[j] = 0.;
}
return;
}
rowequ = (equed == ROW) || (equed == BOTH);
colequ = (equed == COL) || (equed == BOTH);
/* Allocate working space */
work = complexMalloc(2*A->nrow);
rwork = (float *) SUPERLU_MALLOC( (size_t) A->nrow * sizeof(float) );
iwork = intMalloc(A->nrow);
if ( !work || !rwork || !iwork )
SUPERLU_ABORT("Malloc fails for work/rwork/iwork.");
if ( notran ) {
*(unsigned char *)transc = 'N';
transt = TRANS;
} else {
*(unsigned char *)transc = 'T';
transt = NOTRANS;
}
/* NZ = maximum number of nonzero elements in each row of A, plus 1 */
nz = A->ncol + 1;
eps = slamch_("Epsilon");
safmin = slamch_("Safe minimum");
/* Set SAFE1 essentially to be the underflow threshold times the
number of additions in each row. */
safe1 = nz * safmin;
safe2 = safe1 / eps;
/* Compute the number of nonzeros in each row (or column) of A */
for (i = 0; i < A->nrow; ++i) iwork[i] = 0;
if ( notran ) {
for (k = 0; k < A->ncol; ++k)
for (i = Astore->colptr[k]; i < Astore->colptr[k+1]; ++i)
++iwork[Astore->rowind[i]];
} else {
for (k = 0; k < A->ncol; ++k)
iwork[k] = Astore->colptr[k+1] - Astore->colptr[k];
}
/* Copy one column of RHS B into Bjcol. */
Bjcol.Stype = B->Stype;
Bjcol.Dtype = B->Dtype;
Bjcol.Mtype = B->Mtype;
Bjcol.nrow = B->nrow;
Bjcol.ncol = 1;
Bjcol.Store = (void *) SUPERLU_MALLOC( sizeof(DNformat) );
if ( !Bjcol.Store ) SUPERLU_ABORT("SUPERLU_MALLOC fails for Bjcol.Store");
Bjcol_store = Bjcol.Store;
Bjcol_store->lda = ldb;
Bjcol_store->nzval = work; /* address aliasing */
/* Do for each right hand side ... */
for (j = 0; j < nrhs; ++j) {
count = 0;
lstres = 3.;
Bptr = &Bmat[j*ldb];
Xptr = &Xmat[j*ldx];
while (1) { /* Loop until stopping criterion is satisfied. */
/* Compute residual R = B - op(A) * X,
where op(A) = A, A**T, or A**H, depending on TRANS. */
#ifdef _CRAY
CCOPY(&A->nrow, Bptr, &ione, work, &ione);
#else
ccopy_(&A->nrow, Bptr, &ione, work, &ione);
#endif
sp_cgemv(transc, ndone, A, Xptr, ione, done, work, ione);
/* Compute componentwise relative backward error from formula
max(i) ( abs(R(i)) / ( abs(op(A))*abs(X) + abs(B) )(i) )
where abs(Z) is the componentwise absolute value of the matrix
or vector Z. If the i-th component of the denominator is less
than SAFE2, then SAFE1 is added to the i-th component of the
numerator before dividing. */
for (i = 0; i < A->nrow; ++i) rwork[i] = c_abs1( &Bptr[i] );
/* Compute abs(op(A))*abs(X) + abs(B). */
if (notran) {
for (k = 0; k < A->ncol; ++k) {
xk = c_abs1( &Xptr[k] );
for (i = Astore->colptr[k]; i < Astore->colptr[k+1]; ++i)
rwork[Astore->rowind[i]] += c_abs1(&Aval[i]) * xk;
}
} else {
for (k = 0; k < A->ncol; ++k) {
s = 0.;
for (i = Astore->colptr[k]; i < Astore->colptr[k+1]; ++i) {
irow = Astore->rowind[i];
s += c_abs1(&Aval[i]) * c_abs1(&Xptr[irow]);
}
rwork[k] += s;
}
}
s = 0.;
for (i = 0; i < A->nrow; ++i) {
if (rwork[i] > safe2) {
s = SUPERLU_MAX( s, c_abs1(&work[i]) / rwork[i] );
} else if ( rwork[i] != 0.0 ) {
s = SUPERLU_MAX( s, (c_abs1(&work[i]) + safe1) / rwork[i] );
}
/* If rwork[i] is exactly 0.0, then we know the true
residual also must be exactly 0.0. */
}
berr[j] = s;
/* Test stopping criterion. Continue iterating if
1) The residual BERR(J) is larger than machine epsilon, and
2) BERR(J) decreased by at least a factor of 2 during the
last iteration, and
3) At most ITMAX iterations tried. */
if (berr[j] > eps && berr[j] * 2. <= lstres && count < ITMAX) {
/* Update solution and try again. */
cgstrs (trans, L, U, perm_r, perm_c, &Bjcol, Gstat, info);
#ifdef _CRAY
CAXPY(&A->nrow, &done, work, &ione,
&Xmat[j*ldx], &ione);
#else
caxpy_(&A->nrow, &done, work, &ione,
&Xmat[j*ldx], &ione);
#endif
lstres = berr[j];
++count;
} else {
break;
}
} /* end while */
/* Bound error from formula:
norm(X - XTRUE) / norm(X) .le. FERR = norm( abs(inv(op(A)))*
( abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) / norm(X)
where
norm(Z) is the magnitude of the largest component of Z
inv(op(A)) is the inverse of op(A)
abs(Z) is the componentwise absolute value of the matrix or
vector Z
NZ is the maximum number of nonzeros in any row of A, plus 1
EPS is machine epsilon
The i-th component of abs(R)+NZ*EPS*(abs(op(A))*abs(X)+abs(B))
is incremented by SAFE1 if the i-th component of
abs(op(A))*abs(X) + abs(B) is less than SAFE2.
Use CLACON to estimate the infinity-norm of the matrix
inv(op(A)) * diag(W),
where W = abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) */
for (i = 0; i < A->nrow; ++i) rwork[i] = c_abs1( &Bptr[i] );
/* Compute abs(op(A))*abs(X) + abs(B). */
if ( notran ) {
for (k = 0; k < A->ncol; ++k) {
xk = c_abs1( &Xptr[k] );
for (i = Astore->colptr[k]; i < Astore->colptr[k+1]; ++i)
rwork[Astore->rowind[i]] += c_abs1(&Aval[i]) * xk;
}
} else {
for (k = 0; k < A->ncol; ++k) {
s = 0.;
for (i = Astore->colptr[k]; i < Astore->colptr[k+1]; ++i) {
irow = Astore->rowind[i];
xk = c_abs1( &Xptr[irow] );
s += c_abs1(&Aval[i]) * xk;
}
rwork[k] += s;
}
}
for (i = 0; i < A->nrow; ++i)
if (rwork[i] > safe2)
rwork[i] = c_abs(&work[i]) + (iwork[i]+1)*eps*rwork[i];
else
rwork[i] = c_abs(&work[i])+(iwork[i]+1)*eps*rwork[i]+safe1;
kase = 0;
do {
clacon_(&A->nrow, &work[A->nrow], work,
&ferr[j], &kase);
if (kase == 0) break;
if (kase == 1) {
/* Multiply by diag(W)*inv(op(A)**T)*(diag(C) or diag(R)). */
if ( notran && colequ )
for (i = 0; i < A->ncol; ++i) {
cs_mult(&work[i], &work[i], C[i]);
}
else if ( !notran && rowequ )
for (i = 0; i < A->nrow; ++i) {
cs_mult(&work[i], &work[i], R[i]);
}
cgstrs (transt, L, U, perm_r, perm_c, &Bjcol, Gstat, info);
for (i = 0; i < A->nrow; ++i) {
cs_mult(&work[i], &work[i], rwork[i]);
}
} else {
/* Multiply by (diag(C) or diag(R))*inv(op(A))*diag(W). */
for (i = 0; i < A->nrow; ++i) {
cs_mult(&work[i], &work[i], rwork[i]);
}
cgstrs (trans, L, U, perm_r, perm_c, &Bjcol, Gstat, info);
if ( notran && colequ )
for (i = 0; i < A->ncol; ++i) {
cs_mult(&work[i], &work[i], C[i]);
}
else if ( !notran && rowequ )
for (i = 0; i < A->ncol; ++i) {
cs_mult(&work[i], &work[i], R[i]);
}
}
} while ( kase != 0 );
/* Normalize error. */
lstres = 0.;
if ( notran && colequ ) {
for (i = 0; i < A->nrow; ++i)
lstres = SUPERLU_MAX( lstres, C[i] * c_abs1( &Xptr[i]) );
} else if ( !notran && rowequ ) {
for (i = 0; i < A->nrow; ++i)
lstres = SUPERLU_MAX( lstres, R[i] * c_abs1( &Xptr[i]) );
} else {
for (i = 0; i < A->nrow; ++i)
lstres = SUPERLU_MAX( lstres, c_abs1( &Xptr[i]) );
}
if ( lstres != 0. )
ferr[j] /= lstres;
} /* for each RHS j ... */
SUPERLU_FREE(work);
SUPERLU_FREE(rwork);
SUPERLU_FREE(iwork);
SUPERLU_FREE(Bjcol.Store);
return;
} /* cgsrfs */
/* Subroutine */ int ctrcon_(char *norm, char *uplo, char *diag, integer *n,
complex *a, integer *lda, real *rcond, complex *work, real *rwork,
integer *info)
{
/* -- LAPACK routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
March 31, 1993
Purpose
=======
CTRCON estimates the reciprocal of the condition number of a
triangular matrix A, in either the 1-norm or the infinity-norm.
The norm of A is computed and an estimate is obtained for
norm(inv(A)), then the reciprocal of the condition number is
computed as
RCOND = 1 / ( norm(A) * norm(inv(A)) ).
Arguments
=========
NORM (input) CHARACTER*1
Specifies whether the 1-norm condition number or the
infinity-norm condition number is required:
= '1' or 'O': 1-norm;
= 'I': Infinity-norm.
UPLO (input) CHARACTER*1
= 'U': A is upper triangular;
= 'L': A is lower triangular.
DIAG (input) CHARACTER*1
= 'N': A is non-unit triangular;
= 'U': A is unit triangular.
N (input) INTEGER
The order of the matrix A. N >= 0.
A (input) COMPLEX array, dimension (LDA,N)
The triangular matrix A. If UPLO = 'U', the leading N-by-N
upper triangular part of the array A contains the upper
triangular matrix, and the strictly lower triangular part of
A is not referenced. If UPLO = 'L', the leading N-by-N lower
triangular part of the array A contains the lower triangular
matrix, and the strictly upper triangular part of A is not
referenced. If DIAG = 'U', the diagonal elements of A are
also not referenced and are assumed to be 1.
LDA (input) INTEGER
The leading dimension of the array A. LDA >= max(1,N).
RCOND (output) REAL
The reciprocal of the condition number of the matrix A,
computed as RCOND = 1/(norm(A) * norm(inv(A))).
WORK (workspace) COMPLEX array, dimension (2*N)
RWORK (workspace) REAL array, dimension (N)
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
=====================================================================
Test the input parameters.
Parameter adjustments */
/* Table of constant values */
static integer c__1 = 1;
/* System generated locals */
integer a_dim1, a_offset, i__1;
real r__1, r__2;
/* Builtin functions */
double r_imag(complex *);
/* Local variables */
static integer kase, kase1;
static real scale;
extern logical lsame_(char *, char *);
static real anorm;
static logical upper;
static real xnorm;
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *);
static integer ix;
extern integer icamax_(integer *, complex *, integer *);
extern doublereal slamch_(char *);
extern /* Subroutine */ int xerbla_(char *, integer *);
extern doublereal clantr_(char *, char *, char *, integer *, integer *,
complex *, integer *, real *);
static real ainvnm;
extern /* Subroutine */ int clatrs_(char *, char *, char *, char *,
integer *, complex *, integer *, complex *, real *, real *,
integer *), csrscl_(integer *,
real *, complex *, integer *);
static logical onenrm;
static char normin[1];
static real smlnum;
static logical nounit;
a_dim1 = *lda;
a_offset = 1 + a_dim1 * 1;
a -= a_offset;
--work;
--rwork;
/* Function Body */
*info = 0;
upper = lsame_(uplo, "U");
onenrm = *(unsigned char *)norm == '1' || lsame_(norm, "O");
nounit = lsame_(diag, "N");
if (! onenrm && ! lsame_(norm, "I")) {
*info = -1;
} else if (! upper && ! lsame_(uplo, "L")) {
*info = -2;
} else if (! nounit && ! lsame_(diag, "U")) {
*info = -3;
} else if (*n < 0) {
*info = -4;
} else if (*lda < max(1,*n)) {
*info = -6;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CTRCON", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0) {
*rcond = 1.f;
return 0;
}
*rcond = 0.f;
smlnum = slamch_("Safe minimum") * (real) max(1,*n);
/* Compute the norm of the triangular matrix A. */
anorm = clantr_(norm, uplo, diag, n, n, &a[a_offset], lda, &rwork[1]);
/* Continue only if ANORM > 0. */
if (anorm > 0.f) {
/* Estimate the norm of the inverse of A. */
ainvnm = 0.f;
*(unsigned char *)normin = 'N';
if (onenrm) {
kase1 = 1;
} else {
kase1 = 2;
}
kase = 0;
L10:
clacon_(n, &work[*n + 1], &work[1], &ainvnm, &kase);
if (kase != 0) {
if (kase == kase1) {
/* Multiply by inv(A). */
clatrs_(uplo, "No transpose", diag, normin, n, &a[a_offset],
lda, &work[1], &scale, &rwork[1], info);
} else {
/* Multiply by inv(A'). */
clatrs_(uplo, "Conjugate transpose", diag, normin, n, &a[
a_offset], lda, &work[1], &scale, &rwork[1], info);
}
*(unsigned char *)normin = 'Y';
/* Multiply by 1/SCALE if doing so will not cause overflow. */
if (scale != 1.f) {
ix = icamax_(n, &work[1], &c__1);
i__1 = ix;
xnorm = (r__1 = work[i__1].r, dabs(r__1)) + (r__2 = r_imag(&
work[ix]), dabs(r__2));
if (scale < xnorm * smlnum || scale == 0.f) {
goto L20;
}
csrscl_(n, &scale, &work[1], &c__1);
}
goto L10;
}
/* Compute the estimate of the reciprocal condition number. */
if (ainvnm != 0.f) {
*rcond = 1.f / anorm / ainvnm;
}
}
L20:
return 0;
/* End of CTRCON */
} /* ctrcon_ */
/* Subroutine */ int cherfs_(char *uplo, integer *n, integer *nrhs, complex *
a, integer *lda, complex *af, integer *ldaf, integer *ipiv, complex *
b, integer *ldb, complex *x, integer *ldx, real *ferr, real *berr,
complex *work, real *rwork, integer *info)
{
/* -- LAPACK routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
September 30, 1994
Purpose
=======
CHERFS improves the computed solution to a system of linear
equations when the coefficient matrix is Hermitian indefinite, and
provides error bounds and backward error estimates for the solution.
Arguments
=========
UPLO (input) CHARACTER*1
= 'U': Upper triangle of A is stored;
= 'L': Lower triangle of A is stored.
N (input) INTEGER
The order of the matrix A. N >= 0.
NRHS (input) INTEGER
The number of right hand sides, i.e., the number of columns
of the matrices B and X. NRHS >= 0.
A (input) COMPLEX array, dimension (LDA,N)
The Hermitian matrix A. If UPLO = 'U', the leading N-by-N
upper triangular part of A contains the upper triangular part
of the matrix A, and the strictly lower triangular part of A
is not referenced. If UPLO = 'L', the leading N-by-N lower
triangular part of A contains the lower triangular part of
the matrix A, and the strictly upper triangular part of A is
not referenced.
LDA (input) INTEGER
The leading dimension of the array A. LDA >= max(1,N).
AF (input) COMPLEX array, dimension (LDAF,N)
The factored form of the matrix A. AF contains the block
diagonal matrix D and the multipliers used to obtain the
factor U or L from the factorization A = U*D*U**H or
A = L*D*L**H as computed by CHETRF.
LDAF (input) INTEGER
The leading dimension of the array AF. LDAF >= max(1,N).
IPIV (input) INTEGER array, dimension (N)
Details of the interchanges and the block structure of D
as determined by CHETRF.
B (input) COMPLEX array, dimension (LDB,NRHS)
The right hand side matrix B.
LDB (input) INTEGER
The leading dimension of the array B. LDB >= max(1,N).
X (input/output) COMPLEX array, dimension (LDX,NRHS)
On entry, the solution matrix X, as computed by CHETRS.
On exit, the improved solution matrix X.
LDX (input) INTEGER
The leading dimension of the array X. LDX >= max(1,N).
FERR (output) REAL array, dimension (NRHS)
The estimated forward error bound for each solution vector
X(j) (the j-th column of the solution matrix X).
If XTRUE is the true solution corresponding to X(j), FERR(j)
is an estimated upper bound for the magnitude of the largest
element in (X(j) - XTRUE) divided by the magnitude of the
largest element in X(j). The estimate is as reliable as
the estimate for RCOND, and is almost always a slight
overestimate of the true error.
BERR (output) REAL array, dimension (NRHS)
The componentwise relative backward error of each solution
vector X(j) (i.e., the smallest relative change in
any element of A or B that makes X(j) an exact solution).
WORK (workspace) COMPLEX array, dimension (2*N)
RWORK (workspace) REAL array, dimension (N)
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
Internal Parameters
===================
ITMAX is the maximum number of steps of iterative refinement.
=====================================================================
Test the input parameters.
Parameter adjustments */
/* Table of constant values */
static complex c_b1 = {1.f,0.f};
static integer c__1 = 1;
/* System generated locals */
integer a_dim1, a_offset, af_dim1, af_offset, b_dim1, b_offset, x_dim1,
x_offset, i__1, i__2, i__3, i__4, i__5;
real r__1, r__2, r__3, r__4;
complex q__1;
/* Builtin functions */
double r_imag(complex *);
/* Local variables */
static integer kase;
static real safe1, safe2;
static integer i__, j, k;
static real s;
extern logical lsame_(char *, char *);
extern /* Subroutine */ int chemv_(char *, integer *, complex *, complex *
, integer *, complex *, integer *, complex *, complex *, integer *
), ccopy_(integer *, complex *, integer *, complex *,
integer *), caxpy_(integer *, complex *, complex *, integer *,
complex *, integer *);
static integer count;
static logical upper;
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *);
static real xk;
extern doublereal slamch_(char *);
static integer nz;
static real safmin;
extern /* Subroutine */ int xerbla_(char *, integer *), chetrs_(
char *, integer *, integer *, complex *, integer *, integer *,
complex *, integer *, integer *);
static real lstres, eps;
#define a_subscr(a_1,a_2) (a_2)*a_dim1 + a_1
#define a_ref(a_1,a_2) a[a_subscr(a_1,a_2)]
#define b_subscr(a_1,a_2) (a_2)*b_dim1 + a_1
#define b_ref(a_1,a_2) b[b_subscr(a_1,a_2)]
#define x_subscr(a_1,a_2) (a_2)*x_dim1 + a_1
#define x_ref(a_1,a_2) x[x_subscr(a_1,a_2)]
a_dim1 = *lda;
a_offset = 1 + a_dim1 * 1;
a -= a_offset;
af_dim1 = *ldaf;
af_offset = 1 + af_dim1 * 1;
af -= af_offset;
--ipiv;
b_dim1 = *ldb;
b_offset = 1 + b_dim1 * 1;
b -= b_offset;
x_dim1 = *ldx;
x_offset = 1 + x_dim1 * 1;
x -= x_offset;
--ferr;
--berr;
--work;
--rwork;
/* Function Body */
*info = 0;
upper = lsame_(uplo, "U");
if (! upper && ! lsame_(uplo, "L")) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*nrhs < 0) {
*info = -3;
} else if (*lda < max(1,*n)) {
*info = -5;
} else if (*ldaf < max(1,*n)) {
*info = -7;
} else if (*ldb < max(1,*n)) {
*info = -10;
} else if (*ldx < max(1,*n)) {
*info = -12;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CHERFS", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0 || *nrhs == 0) {
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
ferr[j] = 0.f;
berr[j] = 0.f;
/* L10: */
}
return 0;
}
/* NZ = maximum number of nonzero elements in each row of A, plus 1 */
nz = *n + 1;
eps = slamch_("Epsilon");
safmin = slamch_("Safe minimum");
safe1 = nz * safmin;
safe2 = safe1 / eps;
/* Do for each right hand side */
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
count = 1;
lstres = 3.f;
L20:
/* Loop until stopping criterion is satisfied.
Compute residual R = B - A * X */
ccopy_(n, &b_ref(1, j), &c__1, &work[1], &c__1);
q__1.r = -1.f, q__1.i = 0.f;
chemv_(uplo, n, &q__1, &a[a_offset], lda, &x_ref(1, j), &c__1, &c_b1,
&work[1], &c__1);
/* Compute componentwise relative backward error from formula
max(i) ( abs(R(i)) / ( abs(A)*abs(X) + abs(B) )(i) )
where abs(Z) is the componentwise absolute value of the matrix
or vector Z. If the i-th component of the denominator is less
than SAFE2, then SAFE1 is added to the i-th components of the
numerator and denominator before dividing. */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = b_subscr(i__, j);
rwork[i__] = (r__1 = b[i__3].r, dabs(r__1)) + (r__2 = r_imag(&
b_ref(i__, j)), dabs(r__2));
/* L30: */
}
/* Compute abs(A)*abs(X) + abs(B). */
if (upper) {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
s = 0.f;
i__3 = x_subscr(k, j);
xk = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&x_ref(k,
j)), dabs(r__2));
i__3 = k - 1;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = a_subscr(i__, k);
rwork[i__] += ((r__1 = a[i__4].r, dabs(r__1)) + (r__2 =
r_imag(&a_ref(i__, k)), dabs(r__2))) * xk;
i__4 = a_subscr(i__, k);
i__5 = x_subscr(i__, j);
s += ((r__1 = a[i__4].r, dabs(r__1)) + (r__2 = r_imag(&
a_ref(i__, k)), dabs(r__2))) * ((r__3 = x[i__5].r,
dabs(r__3)) + (r__4 = r_imag(&x_ref(i__, j)),
dabs(r__4)));
/* L40: */
}
i__3 = a_subscr(k, k);
rwork[k] = rwork[k] + (r__1 = a[i__3].r, dabs(r__1)) * xk + s;
/* L50: */
}
} else {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
s = 0.f;
i__3 = x_subscr(k, j);
xk = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 = r_imag(&x_ref(k,
j)), dabs(r__2));
i__3 = a_subscr(k, k);
rwork[k] += (r__1 = a[i__3].r, dabs(r__1)) * xk;
i__3 = *n;
for (i__ = k + 1; i__ <= i__3; ++i__) {
i__4 = a_subscr(i__, k);
rwork[i__] += ((r__1 = a[i__4].r, dabs(r__1)) + (r__2 =
r_imag(&a_ref(i__, k)), dabs(r__2))) * xk;
i__4 = a_subscr(i__, k);
i__5 = x_subscr(i__, j);
s += ((r__1 = a[i__4].r, dabs(r__1)) + (r__2 = r_imag(&
a_ref(i__, k)), dabs(r__2))) * ((r__3 = x[i__5].r,
dabs(r__3)) + (r__4 = r_imag(&x_ref(i__, j)),
dabs(r__4)));
/* L60: */
}
rwork[k] += s;
/* L70: */
}
}
s = 0.f;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
/* Computing MAX */
i__3 = i__;
r__3 = s, r__4 = ((r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2))) / rwork[i__];
s = dmax(r__3,r__4);
} else {
/* Computing MAX */
i__3 = i__;
r__3 = s, r__4 = ((r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + safe1) / (rwork[i__]
+ safe1);
s = dmax(r__3,r__4);
}
/* L80: */
}
berr[j] = s;
/* Test stopping criterion. Continue iterating if
1) The residual BERR(J) is larger than machine epsilon, and
2) BERR(J) decreased by at least a factor of 2 during the
last iteration, and
3) At most ITMAX iterations tried. */
if (berr[j] > eps && berr[j] * 2.f <= lstres && count <= 5) {
/* Update solution and try again. */
chetrs_(uplo, n, &c__1, &af[af_offset], ldaf, &ipiv[1], &work[1],
n, info);
caxpy_(n, &c_b1, &work[1], &c__1, &x_ref(1, j), &c__1);
lstres = berr[j];
++count;
goto L20;
}
/* Bound error from formula
norm(X - XTRUE) / norm(X) .le. FERR =
norm( abs(inv(A))*
( abs(R) + NZ*EPS*( abs(A)*abs(X)+abs(B) ))) / norm(X)
where
norm(Z) is the magnitude of the largest component of Z
inv(A) is the inverse of A
abs(Z) is the componentwise absolute value of the matrix or
vector Z
NZ is the maximum number of nonzeros in any row of A, plus 1
EPS is machine epsilon
The i-th component of abs(R)+NZ*EPS*(abs(A)*abs(X)+abs(B))
is incremented by SAFE1 if the i-th component of
abs(A)*abs(X) + abs(B) is less than SAFE2.
Use CLACON to estimate the infinity-norm of the matrix
inv(A) * diag(W),
where W = abs(R) + NZ*EPS*( abs(A)*abs(X)+abs(B) ))) */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
i__3 = i__;
rwork[i__] = (r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + nz * eps * rwork[
i__];
} else {
i__3 = i__;
rwork[i__] = (r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + nz * eps * rwork[
i__] + safe1;
}
/* L90: */
}
kase = 0;
L100:
clacon_(n, &work[*n + 1], &work[1], &ferr[j], &kase);
if (kase != 0) {
if (kase == 1) {
/* Multiply by diag(W)*inv(A'). */
chetrs_(uplo, n, &c__1, &af[af_offset], ldaf, &ipiv[1], &work[
1], n, info);
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__;
i__4 = i__;
i__5 = i__;
q__1.r = rwork[i__4] * work[i__5].r, q__1.i = rwork[i__4]
* work[i__5].i;
work[i__3].r = q__1.r, work[i__3].i = q__1.i;
/* L110: */
}
} else if (kase == 2) {
/* Multiply by inv(A)*diag(W). */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__;
i__4 = i__;
i__5 = i__;
q__1.r = rwork[i__4] * work[i__5].r, q__1.i = rwork[i__4]
* work[i__5].i;
work[i__3].r = q__1.r, work[i__3].i = q__1.i;
/* L120: */
}
chetrs_(uplo, n, &c__1, &af[af_offset], ldaf, &ipiv[1], &work[
1], n, info);
}
goto L100;
}
/* Normalize error. */
lstres = 0.f;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
/* Computing MAX */
i__3 = x_subscr(i__, j);
r__3 = lstres, r__4 = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&x_ref(i__, j)), dabs(r__2));
lstres = dmax(r__3,r__4);
/* L130: */
}
if (lstres != 0.f) {
ferr[j] /= lstres;
}
/* L140: */
}
return 0;
/* End of CHERFS */
} /* cherfs_ */
/* Subroutine */ int cgtrfs_(char *trans, integer *n, integer *nrhs, complex *
dl, complex *d__, complex *du, complex *dlf, complex *df, complex *
duf, complex *du2, integer *ipiv, complex *b, integer *ldb, complex *
x, integer *ldx, real *ferr, real *berr, complex *work, real *rwork,
integer *info)
{
/* -- LAPACK routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
September 30, 1994
Purpose
=======
CGTRFS improves the computed solution to a system of linear
equations when the coefficient matrix is tridiagonal, and provides
error bounds and backward error estimates for the solution.
Arguments
=========
TRANS (input) CHARACTER*1
Specifies the form of the system of equations:
= 'N': A * X = B (No transpose)
= 'T': A**T * X = B (Transpose)
= 'C': A**H * X = B (Conjugate transpose)
N (input) INTEGER
The order of the matrix A. N >= 0.
NRHS (input) INTEGER
The number of right hand sides, i.e., the number of columns
of the matrix B. NRHS >= 0.
DL (input) COMPLEX array, dimension (N-1)
The (n-1) subdiagonal elements of A.
D (input) COMPLEX array, dimension (N)
The diagonal elements of A.
DU (input) COMPLEX array, dimension (N-1)
The (n-1) superdiagonal elements of A.
DLF (input) COMPLEX array, dimension (N-1)
The (n-1) multipliers that define the matrix L from the
LU factorization of A as computed by CGTTRF.
DF (input) COMPLEX array, dimension (N)
The n diagonal elements of the upper triangular matrix U from
the LU factorization of A.
DUF (input) COMPLEX array, dimension (N-1)
The (n-1) elements of the first superdiagonal of U.
DU2 (input) COMPLEX array, dimension (N-2)
The (n-2) elements of the second superdiagonal of U.
IPIV (input) INTEGER array, dimension (N)
The pivot indices; for 1 <= i <= n, row i of the matrix was
interchanged with row IPIV(i). IPIV(i) will always be either
i or i+1; IPIV(i) = i indicates a row interchange was not
required.
B (input) COMPLEX array, dimension (LDB,NRHS)
The right hand side matrix B.
LDB (input) INTEGER
The leading dimension of the array B. LDB >= max(1,N).
X (input/output) COMPLEX array, dimension (LDX,NRHS)
On entry, the solution matrix X, as computed by CGTTRS.
On exit, the improved solution matrix X.
LDX (input) INTEGER
The leading dimension of the array X. LDX >= max(1,N).
FERR (output) REAL array, dimension (NRHS)
The estimated forward error bound for each solution vector
X(j) (the j-th column of the solution matrix X).
If XTRUE is the true solution corresponding to X(j), FERR(j)
is an estimated upper bound for the magnitude of the largest
element in (X(j) - XTRUE) divided by the magnitude of the
largest element in X(j). The estimate is as reliable as
the estimate for RCOND, and is almost always a slight
overestimate of the true error.
BERR (output) REAL array, dimension (NRHS)
The componentwise relative backward error of each solution
vector X(j) (i.e., the smallest relative change in
any element of A or B that makes X(j) an exact solution).
WORK (workspace) COMPLEX array, dimension (2*N)
RWORK (workspace) REAL array, dimension (N)
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
Internal Parameters
===================
ITMAX is the maximum number of steps of iterative refinement.
=====================================================================
Test the input parameters.
Parameter adjustments */
/* Table of constant values */
static integer c__1 = 1;
static real c_b18 = -1.f;
static real c_b19 = 1.f;
static complex c_b26 = {1.f,0.f};
/* System generated locals */
integer b_dim1, b_offset, x_dim1, x_offset, i__1, i__2, i__3, i__4, i__5,
i__6, i__7, i__8, i__9;
real r__1, r__2, r__3, r__4, r__5, r__6, r__7, r__8, r__9, r__10, r__11,
r__12, r__13, r__14;
complex q__1;
/* Builtin functions */
double r_imag(complex *);
/* Local variables */
static integer kase;
static real safe1, safe2;
static integer i__, j;
static real s;
extern logical lsame_(char *, char *);
extern /* Subroutine */ int ccopy_(integer *, complex *, integer *,
complex *, integer *), caxpy_(integer *, complex *, complex *,
integer *, complex *, integer *);
static integer count;
extern /* Subroutine */ int clacon_(integer *, complex *, complex *, real
*, integer *), clagtm_(char *, integer *, integer *, real *,
complex *, complex *, complex *, complex *, integer *, real *,
complex *, integer *);
static integer nz;
extern doublereal slamch_(char *);
static real safmin;
extern /* Subroutine */ int xerbla_(char *, integer *);
static logical notran;
static char transn[1];
extern /* Subroutine */ int cgttrs_(char *, integer *, integer *, complex
*, complex *, complex *, complex *, integer *, complex *, integer
*, integer *);
static char transt[1];
static real lstres, eps;
#define b_subscr(a_1,a_2) (a_2)*b_dim1 + a_1
#define b_ref(a_1,a_2) b[b_subscr(a_1,a_2)]
#define x_subscr(a_1,a_2) (a_2)*x_dim1 + a_1
#define x_ref(a_1,a_2) x[x_subscr(a_1,a_2)]
--dl;
--d__;
--du;
--dlf;
--df;
--duf;
--du2;
--ipiv;
b_dim1 = *ldb;
b_offset = 1 + b_dim1 * 1;
b -= b_offset;
x_dim1 = *ldx;
x_offset = 1 + x_dim1 * 1;
x -= x_offset;
--ferr;
--berr;
--work;
--rwork;
/* Function Body */
*info = 0;
notran = lsame_(trans, "N");
if (! notran && ! lsame_(trans, "T") && ! lsame_(
trans, "C")) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*nrhs < 0) {
*info = -3;
} else if (*ldb < max(1,*n)) {
*info = -13;
} else if (*ldx < max(1,*n)) {
*info = -15;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CGTRFS", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0 || *nrhs == 0) {
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
ferr[j] = 0.f;
berr[j] = 0.f;
/* L10: */
}
return 0;
}
if (notran) {
*(unsigned char *)transn = 'N';
*(unsigned char *)transt = 'C';
} else {
*(unsigned char *)transn = 'C';
*(unsigned char *)transt = 'N';
}
/* NZ = maximum number of nonzero elements in each row of A, plus 1 */
nz = 4;
eps = slamch_("Epsilon");
safmin = slamch_("Safe minimum");
safe1 = nz * safmin;
safe2 = safe1 / eps;
/* Do for each right hand side */
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
count = 1;
lstres = 3.f;
L20:
/* Loop until stopping criterion is satisfied.
Compute residual R = B - op(A) * X,
where op(A) = A, A**T, or A**H, depending on TRANS. */
ccopy_(n, &b_ref(1, j), &c__1, &work[1], &c__1);
clagtm_(trans, n, &c__1, &c_b18, &dl[1], &d__[1], &du[1], &x_ref(1, j)
, ldx, &c_b19, &work[1], n);
/* Compute abs(op(A))*abs(x) + abs(b) for use in the backward
error bound. */
if (notran) {
if (*n == 1) {
i__2 = b_subscr(1, j);
i__3 = x_subscr(1, j);
rwork[1] = (r__1 = b[i__2].r, dabs(r__1)) + (r__2 = r_imag(&
b_ref(1, j)), dabs(r__2)) + ((r__3 = d__[1].r, dabs(
r__3)) + (r__4 = r_imag(&d__[1]), dabs(r__4))) * ((
r__5 = x[i__3].r, dabs(r__5)) + (r__6 = r_imag(&x_ref(
1, j)), dabs(r__6)));
} else {
i__2 = b_subscr(1, j);
i__3 = x_subscr(1, j);
i__4 = x_subscr(2, j);
rwork[1] = (r__1 = b[i__2].r, dabs(r__1)) + (r__2 = r_imag(&
b_ref(1, j)), dabs(r__2)) + ((r__3 = d__[1].r, dabs(
r__3)) + (r__4 = r_imag(&d__[1]), dabs(r__4))) * ((
r__5 = x[i__3].r, dabs(r__5)) + (r__6 = r_imag(&x_ref(
1, j)), dabs(r__6))) + ((r__7 = du[1].r, dabs(r__7))
+ (r__8 = r_imag(&du[1]), dabs(r__8))) * ((r__9 = x[
i__4].r, dabs(r__9)) + (r__10 = r_imag(&x_ref(2, j)),
dabs(r__10)));
i__2 = *n - 1;
for (i__ = 2; i__ <= i__2; ++i__) {
i__3 = b_subscr(i__, j);
i__4 = i__ - 1;
i__5 = x_subscr(i__ - 1, j);
i__6 = i__;
i__7 = x_subscr(i__, j);
i__8 = i__;
i__9 = x_subscr(i__ + 1, j);
rwork[i__] = (r__1 = b[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&b_ref(i__, j)), dabs(r__2)) + ((r__3 = dl[
i__4].r, dabs(r__3)) + (r__4 = r_imag(&dl[i__ - 1]
), dabs(r__4))) * ((r__5 = x[i__5].r, dabs(r__5))
+ (r__6 = r_imag(&x_ref(i__ - 1, j)), dabs(r__6)))
+ ((r__7 = d__[i__6].r, dabs(r__7)) + (r__8 =
r_imag(&d__[i__]), dabs(r__8))) * ((r__9 = x[i__7]
.r, dabs(r__9)) + (r__10 = r_imag(&x_ref(i__, j)),
dabs(r__10))) + ((r__11 = du[i__8].r, dabs(r__11)
) + (r__12 = r_imag(&du[i__]), dabs(r__12))) * ((
r__13 = x[i__9].r, dabs(r__13)) + (r__14 = r_imag(
&x_ref(i__ + 1, j)), dabs(r__14)));
/* L30: */
}
i__2 = b_subscr(*n, j);
i__3 = *n - 1;
i__4 = x_subscr(*n - 1, j);
i__5 = *n;
i__6 = x_subscr(*n, j);
rwork[*n] = (r__1 = b[i__2].r, dabs(r__1)) + (r__2 = r_imag(&
b_ref(*n, j)), dabs(r__2)) + ((r__3 = dl[i__3].r,
dabs(r__3)) + (r__4 = r_imag(&dl[*n - 1]), dabs(r__4))
) * ((r__5 = x[i__4].r, dabs(r__5)) + (r__6 = r_imag(&
x_ref(*n - 1, j)), dabs(r__6))) + ((r__7 = d__[i__5]
.r, dabs(r__7)) + (r__8 = r_imag(&d__[*n]), dabs(r__8)
)) * ((r__9 = x[i__6].r, dabs(r__9)) + (r__10 =
r_imag(&x_ref(*n, j)), dabs(r__10)));
}
} else {
if (*n == 1) {
i__2 = b_subscr(1, j);
i__3 = x_subscr(1, j);
rwork[1] = (r__1 = b[i__2].r, dabs(r__1)) + (r__2 = r_imag(&
b_ref(1, j)), dabs(r__2)) + ((r__3 = d__[1].r, dabs(
r__3)) + (r__4 = r_imag(&d__[1]), dabs(r__4))) * ((
r__5 = x[i__3].r, dabs(r__5)) + (r__6 = r_imag(&x_ref(
1, j)), dabs(r__6)));
} else {
i__2 = b_subscr(1, j);
i__3 = x_subscr(1, j);
i__4 = x_subscr(2, j);
rwork[1] = (r__1 = b[i__2].r, dabs(r__1)) + (r__2 = r_imag(&
b_ref(1, j)), dabs(r__2)) + ((r__3 = d__[1].r, dabs(
r__3)) + (r__4 = r_imag(&d__[1]), dabs(r__4))) * ((
r__5 = x[i__3].r, dabs(r__5)) + (r__6 = r_imag(&x_ref(
1, j)), dabs(r__6))) + ((r__7 = dl[1].r, dabs(r__7))
+ (r__8 = r_imag(&dl[1]), dabs(r__8))) * ((r__9 = x[
i__4].r, dabs(r__9)) + (r__10 = r_imag(&x_ref(2, j)),
dabs(r__10)));
i__2 = *n - 1;
for (i__ = 2; i__ <= i__2; ++i__) {
i__3 = b_subscr(i__, j);
i__4 = i__ - 1;
i__5 = x_subscr(i__ - 1, j);
i__6 = i__;
i__7 = x_subscr(i__, j);
i__8 = i__;
i__9 = x_subscr(i__ + 1, j);
rwork[i__] = (r__1 = b[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&b_ref(i__, j)), dabs(r__2)) + ((r__3 = du[
i__4].r, dabs(r__3)) + (r__4 = r_imag(&du[i__ - 1]
), dabs(r__4))) * ((r__5 = x[i__5].r, dabs(r__5))
+ (r__6 = r_imag(&x_ref(i__ - 1, j)), dabs(r__6)))
+ ((r__7 = d__[i__6].r, dabs(r__7)) + (r__8 =
r_imag(&d__[i__]), dabs(r__8))) * ((r__9 = x[i__7]
.r, dabs(r__9)) + (r__10 = r_imag(&x_ref(i__, j)),
dabs(r__10))) + ((r__11 = dl[i__8].r, dabs(r__11)
) + (r__12 = r_imag(&dl[i__]), dabs(r__12))) * ((
r__13 = x[i__9].r, dabs(r__13)) + (r__14 = r_imag(
&x_ref(i__ + 1, j)), dabs(r__14)));
/* L40: */
}
i__2 = b_subscr(*n, j);
i__3 = *n - 1;
i__4 = x_subscr(*n - 1, j);
i__5 = *n;
i__6 = x_subscr(*n, j);
rwork[*n] = (r__1 = b[i__2].r, dabs(r__1)) + (r__2 = r_imag(&
b_ref(*n, j)), dabs(r__2)) + ((r__3 = du[i__3].r,
dabs(r__3)) + (r__4 = r_imag(&du[*n - 1]), dabs(r__4))
) * ((r__5 = x[i__4].r, dabs(r__5)) + (r__6 = r_imag(&
x_ref(*n - 1, j)), dabs(r__6))) + ((r__7 = d__[i__5]
.r, dabs(r__7)) + (r__8 = r_imag(&d__[*n]), dabs(r__8)
)) * ((r__9 = x[i__6].r, dabs(r__9)) + (r__10 =
r_imag(&x_ref(*n, j)), dabs(r__10)));
}
}
/* Compute componentwise relative backward error from formula
max(i) ( abs(R(i)) / ( abs(op(A))*abs(X) + abs(B) )(i) )
where abs(Z) is the componentwise absolute value of the matrix
or vector Z. If the i-th component of the denominator is less
than SAFE2, then SAFE1 is added to the i-th components of the
numerator and denominator before dividing. */
s = 0.f;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
/* Computing MAX */
i__3 = i__;
r__3 = s, r__4 = ((r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2))) / rwork[i__];
s = dmax(r__3,r__4);
} else {
/* Computing MAX */
i__3 = i__;
r__3 = s, r__4 = ((r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + safe1) / (rwork[i__]
+ safe1);
s = dmax(r__3,r__4);
}
/* L50: */
}
berr[j] = s;
/* Test stopping criterion. Continue iterating if
1) The residual BERR(J) is larger than machine epsilon, and
2) BERR(J) decreased by at least a factor of 2 during the
last iteration, and
3) At most ITMAX iterations tried. */
if (berr[j] > eps && berr[j] * 2.f <= lstres && count <= 5) {
/* Update solution and try again. */
cgttrs_(trans, n, &c__1, &dlf[1], &df[1], &duf[1], &du2[1], &ipiv[
1], &work[1], n, info);
caxpy_(n, &c_b26, &work[1], &c__1, &x_ref(1, j), &c__1);
lstres = berr[j];
++count;
goto L20;
}
/* Bound error from formula
norm(X - XTRUE) / norm(X) .le. FERR =
norm( abs(inv(op(A)))*
( abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) / norm(X)
where
norm(Z) is the magnitude of the largest component of Z
inv(op(A)) is the inverse of op(A)
abs(Z) is the componentwise absolute value of the matrix or
vector Z
NZ is the maximum number of nonzeros in any row of A, plus 1
EPS is machine epsilon
The i-th component of abs(R)+NZ*EPS*(abs(op(A))*abs(X)+abs(B))
is incremented by SAFE1 if the i-th component of
abs(op(A))*abs(X) + abs(B) is less than SAFE2.
Use CLACON to estimate the infinity-norm of the matrix
inv(op(A)) * diag(W),
where W = abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
i__3 = i__;
rwork[i__] = (r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + nz * eps * rwork[
i__];
} else {
i__3 = i__;
rwork[i__] = (r__1 = work[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&work[i__]), dabs(r__2)) + nz * eps * rwork[
i__] + safe1;
}
/* L60: */
}
kase = 0;
L70:
clacon_(n, &work[*n + 1], &work[1], &ferr[j], &kase);
if (kase != 0) {
if (kase == 1) {
/* Multiply by diag(W)*inv(op(A)**H). */
cgttrs_(transt, n, &c__1, &dlf[1], &df[1], &duf[1], &du2[1], &
ipiv[1], &work[1], n, info);
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__;
i__4 = i__;
i__5 = i__;
q__1.r = rwork[i__4] * work[i__5].r, q__1.i = rwork[i__4]
* work[i__5].i;
work[i__3].r = q__1.r, work[i__3].i = q__1.i;
/* L80: */
}
} else {
/* Multiply by inv(op(A))*diag(W). */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__;
i__4 = i__;
i__5 = i__;
q__1.r = rwork[i__4] * work[i__5].r, q__1.i = rwork[i__4]
* work[i__5].i;
work[i__3].r = q__1.r, work[i__3].i = q__1.i;
/* L90: */
}
cgttrs_(transn, n, &c__1, &dlf[1], &df[1], &duf[1], &du2[1], &
ipiv[1], &work[1], n, info);
}
goto L70;
}
/* Normalize error. */
lstres = 0.f;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
/* Computing MAX */
i__3 = x_subscr(i__, j);
r__3 = lstres, r__4 = (r__1 = x[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&x_ref(i__, j)), dabs(r__2));
lstres = dmax(r__3,r__4);
/* L100: */
}
if (lstres != 0.f) {
ferr[j] /= lstres;
}
/* L110: */
}
return 0;
/* End of CGTRFS */
} /* cgtrfs_ */
