inline size_t amax(size_t N, const double* x)
{
lapack_int num = (lapack_int)(N);
lapack_int incx = 1;
return idamax_(&num, (double*)(x), &incx);
}
/*! return its largest absolute value */
inline double damax(const _dgematrix& mat)
{ VERBOSE_REPORT;
double val( mat.array[idamax_(mat.m*mat.n, mat.array, 1) -1] );
mat.destroy();
return val;
}
/*! search the index of element having the largest absolute value
in 0-based numbering system */
inline void idamax(long& i, long& j, const _dgematrix& mat)
{ VERBOSE_REPORT;
long index( idamax_(mat.m*mat.n, mat.array, 1) -1 );
i =index%mat.m;
j =index/mat.m;
mat.destroy();
}
/*! return its largest absolute value */
inline double damax(const dssmatrix& mat)
{
#ifdef CPPL_VERBOSE
std::cerr << "# [MARK] damax(const dssmatrix&)"
<< std::endl;
#endif//CPPL_VERBOSE
return mat.Array[idamax_(mat.VOL, mat.Array, 1) -1];
}
/* ************************************************************
PROCEDURE maxabs - computes inf-norm using BLAS
INPUT
x - vector of length n
n - length of x.
RETURNS y = norm(x,inf).
************************************************************ */
double maxabs(const double *x,const mwIndex n)
{
mwIndex one=1;
#ifdef PC
return fabs(x[idamax(&n,x,&one)]);
#endif
#ifdef UNIX
return fabs(x[idamax_(&n,x,&one)]);
#endif
}
/*! search the index of element having the largest absolute value
in 0-based numbering system */
inline void idamax(long& i, long& j, const dssmatrix& mat)
{
#ifdef CPPL_VERBOSE
std::cerr << "# [MARK] idamax(long&, long&, const dssmatrix&)"
<< std::endl;
#endif//CPPL_VERBOSE
long index( idamax_(mat.VOL, mat.Array, 1) -1 );
i =mat.Indx[index];
j =mat.Jndx[index];
}
/*! return its largest absolute value */
inline double damax(const _drovector& vec)
{
#ifdef CPPL_VERBOSE
std::cerr << "# [MARK] damax(const _drovector&)"
<< std::endl;
#endif//CPPL_VERBOSE
double val( vec.Array[idamax_(vec.L, vec.Array, 1)-1] );
vec.destroy();
return val;
}
/*! return the index of element having the largest absolute value
in 0-based numbering system */
inline long idamax(const _drovector& vec)
{
#ifdef CPPL_VERBOSE
std::cerr << "# [MARK] idamax(const _drovector&)"
<< std::endl;
#endif//CPPL_VERBOSE
long i( idamax_(vec.L, vec.Array, 1) -1 );
vec.destroy();
return i;
}
/*! search the index of element having the largest absolute value
in 0-based numbering system */
inline void idamax(long& i, long& j, const dsymatrix& mat)
{VERBOSE_REPORT;
i=j=0;
double val(0.);
for(long J=0; J<mat.n; J++){
long I( J +idamax_(mat.m-J, mat.darray[J]+J, 1) -1 );
if( val<fabs(mat.darray[J][I]) ){
val =fabs(mat.darray[J][I]);
i=I;
j=J;
}
}
}
/* Subroutine */ int dsytf2_(char *uplo, integer *n, doublereal *a, integer *
lda, integer *ipiv, integer *info)
{
/* -- LAPACK routine (version 3.1) --
Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd..
November 2006
Purpose
=======
DSYTF2 computes the factorization of a real symmetric matrix A using
the Bunch-Kaufman diagonal pivoting method:
A = U*D*U' or A = L*D*L'
where U (or L) is a product of permutation and unit upper (lower)
triangular matrices, U' is the transpose of U, and D is symmetric and
block diagonal with 1-by-1 and 2-by-2 diagonal blocks.
This is the unblocked version of the algorithm, calling Level 2 BLAS.
Arguments
=========
UPLO (input) CHARACTER*1
Specifies whether the upper or lower triangular part of the
symmetric matrix A is stored:
= 'U': Upper triangular
= 'L': Lower triangular
N (input) INTEGER
The order of the matrix A. N >= 0.
A (input/output) DOUBLE PRECISION array, dimension (LDA,N)
On entry, the symmetric 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.
On exit, the block diagonal matrix D and the multipliers used
to obtain the factor U or L (see below for further details).
LDA (input) INTEGER
The leading dimension of the array A. LDA >= max(1,N).
IPIV (output) INTEGER array, dimension (N)
Details of the interchanges and the block structure of D.
If IPIV(k) > 0, then rows and columns k and IPIV(k) were
interchanged and D(k,k) is a 1-by-1 diagonal block.
If UPLO = 'U' and IPIV(k) = IPIV(k-1) < 0, then rows and
columns k-1 and -IPIV(k) were interchanged and D(k-1:k,k-1:k)
is a 2-by-2 diagonal block. If UPLO = 'L' and IPIV(k) =
IPIV(k+1) < 0, then rows and columns k+1 and -IPIV(k) were
interchanged and D(k:k+1,k:k+1) is a 2-by-2 diagonal block.
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -k, the k-th argument had an illegal value
> 0: if INFO = k, D(k,k) is exactly zero. The factorization
has been completed, but the block diagonal matrix D is
exactly singular, and division by zero will occur if it
is used to solve a system of equations.
Further Details
===============
09-29-06 - patch from
Bobby Cheng, MathWorks
Replace l.204 and l.372
IF( MAX( ABSAKK, COLMAX ).EQ.ZERO ) THEN
by
IF( (MAX( ABSAKK, COLMAX ).EQ.ZERO) .OR. DISNAN(ABSAKK) ) THEN
01-01-96 - Based on modifications by
J. Lewis, Boeing Computer Services Company
A. Petitet, Computer Science Dept., Univ. of Tenn., Knoxville, USA
1-96 - Based on modifications by J. Lewis, Boeing Computer Services
Company
If UPLO = 'U', then A = U*D*U', where
U = P(n)*U(n)* ... *P(k)U(k)* ...,
i.e., U is a product of terms P(k)*U(k), where k decreases from n to
1 in steps of 1 or 2, and D is a block diagonal matrix with 1-by-1
and 2-by-2 diagonal blocks D(k). P(k) is a permutation matrix as
defined by IPIV(k), and U(k) is a unit upper triangular matrix, such
that if the diagonal block D(k) is of order s (s = 1 or 2), then
( I v 0 ) k-s
U(k) = ( 0 I 0 ) s
( 0 0 I ) n-k
k-s s n-k
If s = 1, D(k) overwrites A(k,k), and v overwrites A(1:k-1,k).
If s = 2, the upper triangle of D(k) overwrites A(k-1,k-1), A(k-1,k),
and A(k,k), and v overwrites A(1:k-2,k-1:k).
If UPLO = 'L', then A = L*D*L', where
L = P(1)*L(1)* ... *P(k)*L(k)* ...,
i.e., L is a product of terms P(k)*L(k), where k increases from 1 to
n in steps of 1 or 2, and D is a block diagonal matrix with 1-by-1
and 2-by-2 diagonal blocks D(k). P(k) is a permutation matrix as
defined by IPIV(k), and L(k) is a unit lower triangular matrix, such
that if the diagonal block D(k) is of order s (s = 1 or 2), then
( I 0 0 ) k-1
L(k) = ( 0 I 0 ) s
( 0 v I ) n-k-s+1
k-1 s n-k-s+1
If s = 1, D(k) overwrites A(k,k), and v overwrites A(k+1:n,k).
If s = 2, the lower triangle of D(k) overwrites A(k,k), A(k+1,k),
and A(k+1,k+1), and v overwrites A(k+2:n,k:k+1).
=====================================================================
Test the input parameters.
Parameter adjustments */
/* Table of constant values */
static const integer c__1 = 1;
/* System generated locals */
integer a_dim1, a_offset, i__1, i__2;
doublereal d__1, d__2, d__3;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
_THREAD_STATIC_ integer i__, j, k;
_THREAD_STATIC_ doublereal t, r1, d11, d12, d21, d22;
_THREAD_STATIC_ integer kk, kp;
_THREAD_STATIC_ doublereal wk, wkm1, wkp1;
_THREAD_STATIC_ integer imax, jmax;
extern /* Subroutine */ int dsyr_(char *, integer *, doublereal *,
doublereal *, integer *, doublereal *, integer *);
_THREAD_STATIC_ doublereal alpha;
extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
integer *);
extern logical lsame_(char *, char *);
extern /* Subroutine */ int dswap_(integer *, doublereal *, integer *,
doublereal *, integer *);
_THREAD_STATIC_ integer kstep;
_THREAD_STATIC_ logical upper;
_THREAD_STATIC_ doublereal absakk;
extern integer idamax_(integer *, doublereal *, integer *);
extern logical disnan_(doublereal *);
extern /* Subroutine */ int xerbla_(char *, integer *);
_THREAD_STATIC_ doublereal colmax, rowmax;
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
--ipiv;
/* 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;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("DSYTF2", &i__1);
return 0;
}
/* Initialize ALPHA for use in choosing pivot block size. */
alpha = (sqrt(17.) + 1.) / 8.;
if (upper) {
/* Factorize A as U*D*U' using the upper triangle of A
K is the main loop index, decreasing from N to 1 in steps of
1 or 2 */
k = *n;
L10:
/* If K < 1, exit from loop */
if (k < 1) {
goto L70;
}
kstep = 1;
/* Determine rows and columns to be interchanged and whether
a 1-by-1 or 2-by-2 pivot block will be used */
absakk = (d__1 = a[k + k * a_dim1], abs(d__1));
/* IMAX is the row-index of the largest off-diagonal element in
column K, and COLMAX is its absolute value */
if (k > 1) {
i__1 = k - 1;
imax = idamax_(&i__1, &a[k * a_dim1 + 1], &c__1);
colmax = (d__1 = a[imax + k * a_dim1], abs(d__1));
} else {
colmax = 0.;
}
if (max(absakk,colmax) == 0. || disnan_(&absakk)) {
/* Column K is zero or contains a NaN: set INFO and continue */
if (*info == 0) {
*info = k;
}
kp = k;
} else {
if (absakk >= alpha * colmax) {
/* no interchange, use 1-by-1 pivot block */
kp = k;
} else {
/* JMAX is the column-index of the largest off-diagonal
element in row IMAX, and ROWMAX is its absolute value */
i__1 = k - imax;
jmax = imax + idamax_(&i__1, &a[imax + (imax + 1) * a_dim1],
lda);
rowmax = (d__1 = a[imax + jmax * a_dim1], abs(d__1));
if (imax > 1) {
i__1 = imax - 1;
jmax = idamax_(&i__1, &a[imax * a_dim1 + 1], &c__1);
/* Computing MAX */
d__2 = rowmax, d__3 = (d__1 = a[jmax + imax * a_dim1],
abs(d__1));
rowmax = max(d__2,d__3);
}
if (absakk >= alpha * colmax * (colmax / rowmax)) {
/* no interchange, use 1-by-1 pivot block */
kp = k;
} else if ((d__1 = a[imax + imax * a_dim1], abs(d__1)) >=
alpha * rowmax) {
/* interchange rows and columns K and IMAX, use 1-by-1
pivot block */
kp = imax;
} else {
/* interchange rows and columns K-1 and IMAX, use 2-by-2
pivot block */
kp = imax;
kstep = 2;
}
}
kk = k - kstep + 1;
if (kp != kk) {
/* Interchange rows and columns KK and KP in the leading
submatrix A(1:k,1:k) */
i__1 = kp - 1;
dswap_(&i__1, &a[kk * a_dim1 + 1], &c__1, &a[kp * a_dim1 + 1],
&c__1);
i__1 = kk - kp - 1;
dswap_(&i__1, &a[kp + 1 + kk * a_dim1], &c__1, &a[kp + (kp +
1) * a_dim1], lda);
t = a[kk + kk * a_dim1];
a[kk + kk * a_dim1] = a[kp + kp * a_dim1];
a[kp + kp * a_dim1] = t;
if (kstep == 2) {
t = a[k - 1 + k * a_dim1];
a[k - 1 + k * a_dim1] = a[kp + k * a_dim1];
a[kp + k * a_dim1] = t;
}
}
/* Update the leading submatrix */
if (kstep == 1) {
/* 1-by-1 pivot block D(k): column k now holds
W(k) = U(k)*D(k)
where U(k) is the k-th column of U
Perform a rank-1 update of A(1:k-1,1:k-1) as
A := A - U(k)*D(k)*U(k)' = A - W(k)*1/D(k)*W(k)' */
r1 = 1. / a[k + k * a_dim1];
i__1 = k - 1;
d__1 = -r1;
dsyr_(uplo, &i__1, &d__1, &a[k * a_dim1 + 1], &c__1, &a[
a_offset], lda);
/* Store U(k) in column k */
i__1 = k - 1;
dscal_(&i__1, &r1, &a[k * a_dim1 + 1], &c__1);
} else {
/* 2-by-2 pivot block D(k): columns k and k-1 now hold
( W(k-1) W(k) ) = ( U(k-1) U(k) )*D(k)
where U(k) and U(k-1) are the k-th and (k-1)-th columns
of U
Perform a rank-2 update of A(1:k-2,1:k-2) as
A := A - ( U(k-1) U(k) )*D(k)*( U(k-1) U(k) )'
= A - ( W(k-1) W(k) )*inv(D(k))*( W(k-1) W(k) )' */
if (k > 2) {
d12 = a[k - 1 + k * a_dim1];
d22 = a[k - 1 + (k - 1) * a_dim1] / d12;
d11 = a[k + k * a_dim1] / d12;
t = 1. / (d11 * d22 - 1.);
d12 = t / d12;
for (j = k - 2; j >= 1; --j) {
wkm1 = d12 * (d11 * a[j + (k - 1) * a_dim1] - a[j + k
* a_dim1]);
wk = d12 * (d22 * a[j + k * a_dim1] - a[j + (k - 1) *
a_dim1]);
for (i__ = j; i__ >= 1; --i__) {
a[i__ + j * a_dim1] = a[i__ + j * a_dim1] - a[i__
+ k * a_dim1] * wk - a[i__ + (k - 1) *
a_dim1] * wkm1;
/* L20: */
}
a[j + k * a_dim1] = wk;
a[j + (k - 1) * a_dim1] = wkm1;
/* L30: */
}
}
}
}
/* Store details of the interchanges in IPIV */
if (kstep == 1) {
ipiv[k] = kp;
} else {
ipiv[k] = -kp;
ipiv[k - 1] = -kp;
}
/* Decrease K and return to the start of the main loop */
k -= kstep;
goto L10;
} else {
/* Factorize A as L*D*L' using the lower triangle of A
K is the main loop index, increasing from 1 to N in steps of
1 or 2 */
k = 1;
L40:
/* If K > N, exit from loop */
if (k > *n) {
goto L70;
}
kstep = 1;
/* Determine rows and columns to be interchanged and whether
a 1-by-1 or 2-by-2 pivot block will be used */
absakk = (d__1 = a[k + k * a_dim1], abs(d__1));
/* IMAX is the row-index of the largest off-diagonal element in
column K, and COLMAX is its absolute value */
if (k < *n) {
i__1 = *n - k;
imax = k + idamax_(&i__1, &a[k + 1 + k * a_dim1], &c__1);
colmax = (d__1 = a[imax + k * a_dim1], abs(d__1));
} else {
colmax = 0.;
}
if (max(absakk,colmax) == 0. || disnan_(&absakk)) {
/* Column K is zero or contains a NaN: set INFO and continue */
if (*info == 0) {
*info = k;
}
kp = k;
} else {
if (absakk >= alpha * colmax) {
/* no interchange, use 1-by-1 pivot block */
kp = k;
} else {
/* JMAX is the column-index of the largest off-diagonal
element in row IMAX, and ROWMAX is its absolute value */
i__1 = imax - k;
jmax = k - 1 + idamax_(&i__1, &a[imax + k * a_dim1], lda);
rowmax = (d__1 = a[imax + jmax * a_dim1], abs(d__1));
if (imax < *n) {
i__1 = *n - imax;
jmax = imax + idamax_(&i__1, &a[imax + 1 + imax * a_dim1],
&c__1);
/* Computing MAX */
d__2 = rowmax, d__3 = (d__1 = a[jmax + imax * a_dim1],
abs(d__1));
rowmax = max(d__2,d__3);
}
if (absakk >= alpha * colmax * (colmax / rowmax)) {
/* no interchange, use 1-by-1 pivot block */
kp = k;
} else if ((d__1 = a[imax + imax * a_dim1], abs(d__1)) >=
alpha * rowmax) {
/* interchange rows and columns K and IMAX, use 1-by-1
pivot block */
kp = imax;
} else {
/* interchange rows and columns K+1 and IMAX, use 2-by-2
pivot block */
kp = imax;
kstep = 2;
}
}
kk = k + kstep - 1;
if (kp != kk) {
/* Interchange rows and columns KK and KP in the trailing
submatrix A(k:n,k:n) */
if (kp < *n) {
i__1 = *n - kp;
dswap_(&i__1, &a[kp + 1 + kk * a_dim1], &c__1, &a[kp + 1
+ kp * a_dim1], &c__1);
}
i__1 = kp - kk - 1;
dswap_(&i__1, &a[kk + 1 + kk * a_dim1], &c__1, &a[kp + (kk +
1) * a_dim1], lda);
t = a[kk + kk * a_dim1];
a[kk + kk * a_dim1] = a[kp + kp * a_dim1];
a[kp + kp * a_dim1] = t;
if (kstep == 2) {
t = a[k + 1 + k * a_dim1];
a[k + 1 + k * a_dim1] = a[kp + k * a_dim1];
a[kp + k * a_dim1] = t;
}
}
/* Update the trailing submatrix */
if (kstep == 1) {
/* 1-by-1 pivot block D(k): column k now holds
W(k) = L(k)*D(k)
where L(k) is the k-th column of L */
if (k < *n) {
/* Perform a rank-1 update of A(k+1:n,k+1:n) as
A := A - L(k)*D(k)*L(k)' = A - W(k)*(1/D(k))*W(k)' */
d11 = 1. / a[k + k * a_dim1];
i__1 = *n - k;
d__1 = -d11;
dsyr_(uplo, &i__1, &d__1, &a[k + 1 + k * a_dim1], &c__1, &
a[k + 1 + (k + 1) * a_dim1], lda);
/* Store L(k) in column K */
i__1 = *n - k;
dscal_(&i__1, &d11, &a[k + 1 + k * a_dim1], &c__1);
}
} else {
/* 2-by-2 pivot block D(k) */
if (k < *n - 1) {
/* Perform a rank-2 update of A(k+2:n,k+2:n) as
A := A - ( (A(k) A(k+1))*D(k)**(-1) ) * (A(k) A(k+1))'
where L(k) and L(k+1) are the k-th and (k+1)-th
columns of L */
d21 = a[k + 1 + k * a_dim1];
d11 = a[k + 1 + (k + 1) * a_dim1] / d21;
d22 = a[k + k * a_dim1] / d21;
t = 1. / (d11 * d22 - 1.);
d21 = t / d21;
i__1 = *n;
for (j = k + 2; j <= i__1; ++j) {
wk = d21 * (d11 * a[j + k * a_dim1] - a[j + (k + 1) *
a_dim1]);
wkp1 = d21 * (d22 * a[j + (k + 1) * a_dim1] - a[j + k
* a_dim1]);
i__2 = *n;
for (i__ = j; i__ <= i__2; ++i__) {
a[i__ + j * a_dim1] = a[i__ + j * a_dim1] - a[i__
+ k * a_dim1] * wk - a[i__ + (k + 1) *
a_dim1] * wkp1;
/* L50: */
}
a[j + k * a_dim1] = wk;
a[j + (k + 1) * a_dim1] = wkp1;
/* L60: */
}
}
}
}
/* Store details of the interchanges in IPIV */
if (kstep == 1) {
ipiv[k] = kp;
} else {
ipiv[k] = -kp;
ipiv[k + 1] = -kp;
}
/* Increase K and return to the start of the main loop */
k += kstep;
goto L40;
}
L70:
return 0;
/* End of DSYTF2 */
} /* dsytf2_ */
/* Subroutine */ int zptcon_(integer *n, doublereal *d__, doublecomplex *e,
doublereal *anorm, doublereal *rcond, doublereal *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
=======
ZPTCON computes the reciprocal of the condition number (in the
1-norm) of a complex Hermitian positive definite tridiagonal matrix
using the factorization A = L*D*L**H or A = U**H*D*U computed by
ZPTTRF.
Norm(inv(A)) is computed by a direct method, and the reciprocal of
the condition number is computed as
RCOND = 1 / (ANORM * norm(inv(A))).
Arguments
=========
N (input) INTEGER
The order of the matrix A. N >= 0.
D (input) DOUBLE PRECISION array, dimension (N)
The n diagonal elements of the diagonal matrix D from the
factorization of A, as computed by ZPTTRF.
E (input) COMPLEX*16 array, dimension (N-1)
The (n-1) off-diagonal elements of the unit bidiagonal factor
U or L from the factorization of A, as computed by ZPTTRF.
ANORM (input) DOUBLE PRECISION
The 1-norm of the original matrix A.
RCOND (output) DOUBLE PRECISION
The reciprocal of the condition number of the matrix A,
computed as RCOND = 1/(ANORM * AINVNM), where AINVNM is the
1-norm of inv(A) computed in this routine.
RWORK (workspace) DOUBLE PRECISION array, dimension (N)
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
Further Details
===============
The method used is described in Nicholas J. Higham, "Efficient
Algorithms for Computing the Condition Number of a Tridiagonal
Matrix", SIAM J. Sci. Stat. Comput., Vol. 7, No. 1, January 1986.
=====================================================================
Test the input arguments.
Parameter adjustments */
/* Table of constant values */
static integer c__1 = 1;
/* System generated locals */
integer i__1;
doublereal d__1;
/* Builtin functions */
double z_abs(doublecomplex *);
/* Local variables */
static integer i__, ix;
extern integer idamax_(integer *, doublereal *, integer *);
extern /* Subroutine */ int xerbla_(char *, integer *);
static doublereal ainvnm;
--rwork;
--e;
--d__;
/* Function Body */
*info = 0;
if (*n < 0) {
*info = -1;
} else if (*anorm < 0.) {
*info = -4;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("ZPTCON", &i__1);
return 0;
}
/* Quick return if possible */
*rcond = 0.;
if (*n == 0) {
*rcond = 1.;
return 0;
} else if (*anorm == 0.) {
return 0;
}
/* Check that D(1:N) is positive. */
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
if (d__[i__] <= 0.) {
return 0;
}
/* L10: */
}
/* Solve M(A) * x = e, where M(A) = (m(i,j)) is given by
m(i,j) = abs(A(i,j)), i = j,
m(i,j) = -abs(A(i,j)), i .ne. j,
and e = [ 1, 1, ..., 1 ]'. Note M(A) = M(L)*D*M(L)'.
Solve M(L) * x = e. */
rwork[1] = 1.;
i__1 = *n;
for (i__ = 2; i__ <= i__1; ++i__) {
rwork[i__] = rwork[i__ - 1] * z_abs(&e[i__ - 1]) + 1.;
/* L20: */
}
/* Solve D * M(L)' * x = b. */
rwork[*n] /= d__[*n];
for (i__ = *n - 1; i__ >= 1; --i__) {
rwork[i__] = rwork[i__] / d__[i__] + rwork[i__ + 1] * z_abs(&e[i__]);
/* L30: */
}
/* Compute AINVNM = max(x(i)), 1<=i<=n. */
ix = idamax_(n, &rwork[1], &c__1);
ainvnm = (d__1 = rwork[ix], abs(d__1));
/* Compute the reciprocal condition number. */
if (ainvnm != 0.) {
*rcond = 1. / ainvnm / *anorm;
}
return 0;
/* End of ZPTCON */
} /* zptcon_ */
/*< SUBROUTINE DSPTRF( UPLO, N, AP, IPIV, INFO ) >*/
/* Subroutine */ int dsptrf_(char *uplo, integer *n, doublereal *ap, integer *ipiv,
integer *info, ftnlen uplo_len)
{
/* System generated locals */
integer i__1;
doublereal d__1, d__2, d__3;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
doublereal c__;
integer j, k;
doublereal s, t, r1, r2;
integer kc, kk, kp, kx, knc, kpc=0, npp, imax=0, jmax;
extern /* Subroutine */ int drot_(integer *, doublereal *, integer *,
doublereal *, integer *, doublereal *, doublereal *), dspr_(char *
, integer *, doublereal *, doublereal *, integer *, doublereal *,
ftnlen);
doublereal alpha;
extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
integer *);
extern logical lsame_(const char *, const char *, ftnlen, ftnlen);
extern /* Subroutine */ int dswap_(integer *, doublereal *, integer *,
doublereal *, integer *);
integer kstep;
logical upper;
extern /* Subroutine */ int dlaev2_(doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *, doublereal *,
doublereal *);
doublereal absakk;
extern integer idamax_(integer *, doublereal *, integer *);
extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
doublereal colmax, rowmax;
(void)uplo_len;
/* -- LAPACK routine (version 2.0) -- */
/* Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., */
/* Courant Institute, Argonne National Lab, and Rice University */
/* March 31, 1993 */
/* .. Scalar Arguments .. */
/*< CHARACTER UPLO >*/
/*< INTEGER INFO, N >*/
/* .. */
/* .. Array Arguments .. */
/*< INTEGER IPIV( * ) >*/
/*< DOUBLE PRECISION AP( * ) >*/
/* .. */
/* Purpose */
/* ======= */
/* DSPTRF computes the factorization of a real symmetric matrix A stored */
/* in packed format using the Bunch-Kaufman diagonal pivoting method: */
/* A = U*D*U**T or A = L*D*L**T */
/* where U (or L) is a product of permutation and unit upper (lower) */
/* triangular matrices, and D is symmetric and block diagonal with */
/* 1-by-1 and 2-by-2 diagonal blocks. */
/* 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. */
/* AP (input/output) DOUBLE PRECISION array, dimension (N*(N+1)/2) */
/* On entry, the upper or lower triangle of the symmetric 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. */
/* On exit, the block diagonal matrix D and the multipliers used */
/* to obtain the factor U or L, stored as a packed triangular */
/* matrix overwriting A (see below for further details). */
/* IPIV (output) INTEGER array, dimension (N) */
/* Details of the interchanges and the block structure of D. */
/* If IPIV(k) > 0, then rows and columns k and IPIV(k) were */
/* interchanged and D(k,k) is a 1-by-1 diagonal block. */
/* If UPLO = 'U' and IPIV(k) = IPIV(k-1) < 0, then rows and */
/* columns k-1 and -IPIV(k) were interchanged and D(k-1:k,k-1:k) */
/* is a 2-by-2 diagonal block. If UPLO = 'L' and IPIV(k) = */
/* IPIV(k+1) < 0, then rows and columns k+1 and -IPIV(k) were */
/* interchanged and D(k:k+1,k:k+1) is a 2-by-2 diagonal block. */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* > 0: if INFO = i, D(i,i) is exactly zero. The factorization */
/* has been completed, but the block diagonal matrix D is */
/* exactly singular, and division by zero will occur if it */
/* is used to solve a system of equations. */
/* Further Details */
/* =============== */
/* If UPLO = 'U', then A = U*D*U', where */
/* U = P(n)*U(n)* ... *P(k)U(k)* ..., */
/* i.e., U is a product of terms P(k)*U(k), where k decreases from n to */
/* 1 in steps of 1 or 2, and D is a block diagonal matrix with 1-by-1 */
/* and 2-by-2 diagonal blocks D(k). P(k) is a permutation matrix as */
/* defined by IPIV(k), and U(k) is a unit upper triangular matrix, such */
/* that if the diagonal block D(k) is of order s (s = 1 or 2), then */
/* ( I v 0 ) k-s */
/* U(k) = ( 0 I 0 ) s */
/* ( 0 0 I ) n-k */
/* k-s s n-k */
/* If s = 1, D(k) overwrites A(k,k), and v overwrites A(1:k-1,k). */
/* If s = 2, the upper triangle of D(k) overwrites A(k-1,k-1), A(k-1,k), */
/* and A(k,k), and v overwrites A(1:k-2,k-1:k). */
/* If UPLO = 'L', then A = L*D*L', where */
/* L = P(1)*L(1)* ... *P(k)*L(k)* ..., */
/* i.e., L is a product of terms P(k)*L(k), where k increases from 1 to */
/* n in steps of 1 or 2, and D is a block diagonal matrix with 1-by-1 */
/* and 2-by-2 diagonal blocks D(k). P(k) is a permutation matrix as */
/* defined by IPIV(k), and L(k) is a unit lower triangular matrix, such */
/* that if the diagonal block D(k) is of order s (s = 1 or 2), then */
/* ( I 0 0 ) k-1 */
/* L(k) = ( 0 I 0 ) s */
/* ( 0 v I ) n-k-s+1 */
/* k-1 s n-k-s+1 */
/* If s = 1, D(k) overwrites A(k,k), and v overwrites A(k+1:n,k). */
/* If s = 2, the lower triangle of D(k) overwrites A(k,k), A(k+1,k), */
/* and A(k+1,k+1), and v overwrites A(k+2:n,k:k+1). */
/* ===================================================================== */
/* .. Parameters .. */
/*< DOUBLE PRECISION ZERO, ONE >*/
/*< PARAMETER ( ZERO = 0.0D+0, ONE = 1.0D+0 ) >*/
/*< DOUBLE PRECISION EIGHT, SEVTEN >*/
/*< PARAMETER ( EIGHT = 8.0D+0, SEVTEN = 17.0D+0 ) >*/
/* .. */
/* .. Local Scalars .. */
/*< LOGICAL UPPER >*/
/*< >*/
/*< DOUBLE PRECISION ABSAKK, ALPHA, C, COLMAX, R1, R2, ROWMAX, S, T >*/
/* .. */
/* .. External Functions .. */
/*< LOGICAL LSAME >*/
/*< INTEGER IDAMAX >*/
/*< EXTERNAL LSAME, IDAMAX >*/
/* .. */
/* .. External Subroutines .. */
/*< EXTERNAL DLAEV2, DROT, DSCAL, DSPR, DSWAP, XERBLA >*/
/* .. */
/* .. Intrinsic Functions .. */
/*< INTRINSIC ABS, MAX, SQRT >*/
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/*< INFO = 0 >*/
/* Parameter adjustments */
--ipiv;
--ap;
/* Function Body */
*info = 0;
/*< UPPER = LSAME( UPLO, 'U' ) >*/
upper = lsame_(uplo, "U", (ftnlen)1, (ftnlen)1);
/*< IF( .NOT.UPPER .AND. .NOT.LSAME( UPLO, 'L' ) ) THEN >*/
if (! upper && ! lsame_(uplo, "L", (ftnlen)1, (ftnlen)1)) {
/*< INFO = -1 >*/
*info = -1;
/*< ELSE IF( N.LT.0 ) THEN >*/
} else if (*n < 0) {
/*< INFO = -2 >*/
*info = -2;
/*< END IF >*/
}
/*< IF( INFO.NE.0 ) THEN >*/
if (*info != 0) {
/*< CALL XERBLA( 'DSPTRF', -INFO ) >*/
i__1 = -(*info);
xerbla_("DSPTRF", &i__1, (ftnlen)6);
/*< RETURN >*/
return 0;
/*< END IF >*/
}
/* Initialize ALPHA for use in choosing pivot block size. */
/*< ALPHA = ( ONE+SQRT( SEVTEN ) ) / EIGHT >*/
alpha = (sqrt(17.) + 1.) / 8.;
/*< IF( UPPER ) THEN >*/
if (upper) {
/* Factorize A as U*D*U' using the upper triangle of A */
/* K is the main loop index, decreasing from N to 1 in steps of */
/* 1 or 2 */
/*< K = N >*/
k = *n;
/*< KC = ( N-1 )*N / 2 + 1 >*/
kc = (*n - 1) * *n / 2 + 1;
/*< 10 CONTINUE >*/
L10:
/*< KNC = KC >*/
knc = kc;
/* If K < 1, exit from loop */
/*< >*/
if (k < 1) {
goto L70;
}
/*< KSTEP = 1 >*/
kstep = 1;
/* Determine rows and columns to be interchanged and whether */
/* a 1-by-1 or 2-by-2 pivot block will be used */
/*< ABSAKK = ABS( AP( KC+K-1 ) ) >*/
absakk = (d__1 = ap[kc + k - 1], abs(d__1));
/* IMAX is the row-index of the largest off-diagonal element in */
/* column K, and COLMAX is its absolute value */
/*< IF( K.GT.1 ) THEN >*/
if (k > 1) {
/*< IMAX = IDAMAX( K-1, AP( KC ), 1 ) >*/
i__1 = k - 1;
imax = idamax_(&i__1, &ap[kc], &c__1);
/*< COLMAX = ABS( AP( KC+IMAX-1 ) ) >*/
colmax = (d__1 = ap[kc + imax - 1], abs(d__1));
/*< ELSE >*/
} else {
/*< COLMAX = ZERO >*/
colmax = 0.;
/*< END IF >*/
}
/*< IF( MAX( ABSAKK, COLMAX ).EQ.ZERO ) THEN >*/
if (max(absakk,colmax) == 0.) {
/* Column K is zero: set INFO and continue */
/*< >*/
if (*info == 0) {
*info = k;
}
/*< KP = K >*/
kp = k;
/*< ELSE >*/
} else {
/*< IF( ABSAKK.GE.ALPHA*COLMAX ) THEN >*/
if (absakk >= alpha * colmax) {
/* no interchange, use 1-by-1 pivot block */
/*< KP = K >*/
kp = k;
/*< ELSE >*/
} else {
/* JMAX is the column-index of the largest off-diagonal */
/* element in row IMAX, and ROWMAX is its absolute value */
/*< ROWMAX = ZERO >*/
rowmax = 0.;
/*< JMAX = IMAX >*/
jmax = imax;
/*< KX = IMAX*( IMAX+1 ) / 2 + IMAX >*/
kx = imax * (imax + 1) / 2 + imax;
/*< DO 20 J = IMAX + 1, K >*/
i__1 = k;
for (j = imax + 1; j <= i__1; ++j) {
/*< IF( ABS( AP( KX ) ).GT.ROWMAX ) THEN >*/
if ((d__1 = ap[kx], abs(d__1)) > rowmax) {
/*< ROWMAX = ABS( AP( KX ) ) >*/
rowmax = (d__1 = ap[kx], abs(d__1));
/*< JMAX = J >*/
jmax = j;
/*< END IF >*/
}
/*< KX = KX + J >*/
kx += j;
/*< 20 CONTINUE >*/
/* L20: */
}
/*< KPC = ( IMAX-1 )*IMAX / 2 + 1 >*/
kpc = (imax - 1) * imax / 2 + 1;
/*< IF( IMAX.GT.1 ) THEN >*/
if (imax > 1) {
/*< JMAX = IDAMAX( IMAX-1, AP( KPC ), 1 ) >*/
i__1 = imax - 1;
jmax = idamax_(&i__1, &ap[kpc], &c__1);
/*< ROWMAX = MAX( ROWMAX, ABS( AP( KPC+JMAX-1 ) ) ) >*/
/* Computing MAX */
d__2 = rowmax, d__3 = (d__1 = ap[kpc + jmax - 1], abs(
d__1));
rowmax = max(d__2,d__3);
/*< END IF >*/
}
/*< IF( ABSAKK.GE.ALPHA*COLMAX*( COLMAX / ROWMAX ) ) THEN >*/
if (absakk >= alpha * colmax * (colmax / rowmax)) {
/* no interchange, use 1-by-1 pivot block */
/*< KP = K >*/
kp = k;
/*< ELSE IF( ABS( AP( KPC+IMAX-1 ) ).GE.ALPHA*ROWMAX ) THEN >*/
} else if ((d__1 = ap[kpc + imax - 1], abs(d__1)) >= alpha *
rowmax) {
/* interchange rows and columns K and IMAX, use 1-by-1 */
/* pivot block */
/*< KP = IMAX >*/
kp = imax;
/*< ELSE >*/
} else {
/* interchange rows and columns K-1 and IMAX, use 2-by-2 */
/* pivot block */
/*< KP = IMAX >*/
kp = imax;
/*< KSTEP = 2 >*/
kstep = 2;
/*< END IF >*/
}
/*< END IF >*/
}
/*< KK = K - KSTEP + 1 >*/
kk = k - kstep + 1;
/*< >*/
if (kstep == 2) {
knc = knc - k + 1;
}
/*< IF( KP.NE.KK ) THEN >*/
if (kp != kk) {
/* Interchange rows and columns KK and KP in the leading */
/* submatrix A(1:k,1:k) */
/*< CALL DSWAP( KP-1, AP( KNC ), 1, AP( KPC ), 1 ) >*/
i__1 = kp - 1;
dswap_(&i__1, &ap[knc], &c__1, &ap[kpc], &c__1);
/*< KX = KPC + KP - 1 >*/
kx = kpc + kp - 1;
/*< DO 30 J = KP + 1, KK - 1 >*/
i__1 = kk - 1;
for (j = kp + 1; j <= i__1; ++j) {
/*< KX = KX + J - 1 >*/
kx = kx + j - 1;
/*< T = AP( KNC+J-1 ) >*/
t = ap[knc + j - 1];
/*< AP( KNC+J-1 ) = AP( KX ) >*/
ap[knc + j - 1] = ap[kx];
/*< AP( KX ) = T >*/
ap[kx] = t;
/*< 30 CONTINUE >*/
/* L30: */
}
/*< T = AP( KNC+KK-1 ) >*/
t = ap[knc + kk - 1];
/*< AP( KNC+KK-1 ) = AP( KPC+KP-1 ) >*/
ap[knc + kk - 1] = ap[kpc + kp - 1];
/*< AP( KPC+KP-1 ) = T >*/
ap[kpc + kp - 1] = t;
/*< IF( KSTEP.EQ.2 ) THEN >*/
if (kstep == 2) {
/*< T = AP( KC+K-2 ) >*/
t = ap[kc + k - 2];
/*< AP( KC+K-2 ) = AP( KC+KP-1 ) >*/
ap[kc + k - 2] = ap[kc + kp - 1];
/*< AP( KC+KP-1 ) = T >*/
ap[kc + kp - 1] = t;
/*< END IF >*/
}
/*< END IF >*/
}
/* Update the leading submatrix */
/*< IF( KSTEP.EQ.1 ) THEN >*/
if (kstep == 1) {
/* 1-by-1 pivot block D(k): column k now holds */
/* W(k) = U(k)*D(k) */
/* where U(k) is the k-th column of U */
/* Perform a rank-1 update of A(1:k-1,1:k-1) as */
/* A := A - U(k)*D(k)*U(k)' = A - W(k)*1/D(k)*W(k)' */
/*< R1 = ONE / AP( KC+K-1 ) >*/
r1 = 1. / ap[kc + k - 1];
/*< CALL DSPR( UPLO, K-1, -R1, AP( KC ), 1, AP ) >*/
i__1 = k - 1;
d__1 = -r1;
dspr_(uplo, &i__1, &d__1, &ap[kc], &c__1, &ap[1], (ftnlen)1);
/* Store U(k) in column k */
/*< CALL DSCAL( K-1, R1, AP( KC ), 1 ) >*/
i__1 = k - 1;
dscal_(&i__1, &r1, &ap[kc], &c__1);
/*< ELSE >*/
} else {
/* 2-by-2 pivot block D(k): columns k and k-1 now hold */
/* ( W(k-1) W(k) ) = ( U(k-1) U(k) )*D(k) */
/* where U(k) and U(k-1) are the k-th and (k-1)-th columns */
/* of U */
/* Perform a rank-2 update of A(1:k-2,1:k-2) as */
/* A := A - ( U(k-1) U(k) )*D(k)*( U(k-1) U(k) )' */
/* = A - ( W(k-1) W(k) )*inv(D(k))*( W(k-1) W(k) )' */
/* Convert this to two rank-1 updates by using the eigen- */
/* decomposition of D(k) */
/*< >*/
dlaev2_(&ap[kc - 1], &ap[kc + k - 2], &ap[kc + k - 1], &r1, &
r2, &c__, &s);
/*< R1 = ONE / R1 >*/
r1 = 1. / r1;
/*< R2 = ONE / R2 >*/
r2 = 1. / r2;
/*< CALL DROT( K-2, AP( KNC ), 1, AP( KC ), 1, C, S ) >*/
i__1 = k - 2;
drot_(&i__1, &ap[knc], &c__1, &ap[kc], &c__1, &c__, &s);
/*< CALL DSPR( UPLO, K-2, -R1, AP( KNC ), 1, AP ) >*/
i__1 = k - 2;
d__1 = -r1;
dspr_(uplo, &i__1, &d__1, &ap[knc], &c__1, &ap[1], (ftnlen)1);
/*< CALL DSPR( UPLO, K-2, -R2, AP( KC ), 1, AP ) >*/
i__1 = k - 2;
d__1 = -r2;
dspr_(uplo, &i__1, &d__1, &ap[kc], &c__1, &ap[1], (ftnlen)1);
/* Store U(k) and U(k-1) in columns k and k-1 */
/*< CALL DSCAL( K-2, R1, AP( KNC ), 1 ) >*/
i__1 = k - 2;
dscal_(&i__1, &r1, &ap[knc], &c__1);
/*< CALL DSCAL( K-2, R2, AP( KC ), 1 ) >*/
i__1 = k - 2;
dscal_(&i__1, &r2, &ap[kc], &c__1);
/*< CALL DROT( K-2, AP( KNC ), 1, AP( KC ), 1, C, -S ) >*/
i__1 = k - 2;
d__1 = -s;
drot_(&i__1, &ap[knc], &c__1, &ap[kc], &c__1, &c__, &d__1);
/*< END IF >*/
}
/*< END IF >*/
}
/* Store details of the interchanges in IPIV */
/*< IF( KSTEP.EQ.1 ) THEN >*/
if (kstep == 1) {
/*< IPIV( K ) = KP >*/
ipiv[k] = kp;
/*< ELSE >*/
} else {
/*< IPIV( K ) = -KP >*/
ipiv[k] = -kp;
/*< IPIV( K-1 ) = -KP >*/
ipiv[k - 1] = -kp;
/*< END IF >*/
}
/* Decrease K and return to the start of the main loop */
/*< K = K - KSTEP >*/
k -= kstep;
/*< KC = KNC - K >*/
kc = knc - k;
/*< GO TO 10 >*/
goto L10;
/*< ELSE >*/
} else {
/* Factorize A as L*D*L' using the lower triangle of A */
/* K is the main loop index, increasing from 1 to N in steps of */
/* 1 or 2 */
/*< K = 1 >*/
k = 1;
/*< KC = 1 >*/
kc = 1;
/*< NPP = N*( N+1 ) / 2 >*/
npp = *n * (*n + 1) / 2;
/*< 40 CONTINUE >*/
L40:
/*< KNC = KC >*/
knc = kc;
/* If K > N, exit from loop */
/*< >*/
if (k > *n) {
goto L70;
}
/*< KSTEP = 1 >*/
kstep = 1;
/* Determine rows and columns to be interchanged and whether */
/* a 1-by-1 or 2-by-2 pivot block will be used */
/*< ABSAKK = ABS( AP( KC ) ) >*/
absakk = (d__1 = ap[kc], abs(d__1));
/* IMAX is the row-index of the largest off-diagonal element in */
/* column K, and COLMAX is its absolute value */
/*< IF( K.LT.N ) THEN >*/
if (k < *n) {
/*< IMAX = K + IDAMAX( N-K, AP( KC+1 ), 1 ) >*/
i__1 = *n - k;
imax = k + idamax_(&i__1, &ap[kc + 1], &c__1);
/*< COLMAX = ABS( AP( KC+IMAX-K ) ) >*/
colmax = (d__1 = ap[kc + imax - k], abs(d__1));
/*< ELSE >*/
} else {
/*< COLMAX = ZERO >*/
colmax = 0.;
/*< END IF >*/
}
/*< IF( MAX( ABSAKK, COLMAX ).EQ.ZERO ) THEN >*/
if (max(absakk,colmax) == 0.) {
/* Column K is zero: set INFO and continue */
/*< >*/
if (*info == 0) {
*info = k;
}
/*< KP = K >*/
kp = k;
/*< ELSE >*/
} else {
/*< IF( ABSAKK.GE.ALPHA*COLMAX ) THEN >*/
if (absakk >= alpha * colmax) {
/* no interchange, use 1-by-1 pivot block */
/*< KP = K >*/
kp = k;
/*< ELSE >*/
} else {
/* JMAX is the column-index of the largest off-diagonal */
/* element in row IMAX, and ROWMAX is its absolute value */
/*< ROWMAX = ZERO >*/
rowmax = 0.;
/*< KX = KC + IMAX - K >*/
kx = kc + imax - k;
/*< DO 50 J = K, IMAX - 1 >*/
i__1 = imax - 1;
for (j = k; j <= i__1; ++j) {
/*< IF( ABS( AP( KX ) ).GT.ROWMAX ) THEN >*/
if ((d__1 = ap[kx], abs(d__1)) > rowmax) {
/*< ROWMAX = ABS( AP( KX ) ) >*/
rowmax = (d__1 = ap[kx], abs(d__1));
/*< JMAX = J >*/
jmax = j;
/*< END IF >*/
}
/*< KX = KX + N - J >*/
kx = kx + *n - j;
/*< 50 CONTINUE >*/
/* L50: */
}
/*< KPC = NPP - ( N-IMAX+1 )*( N-IMAX+2 ) / 2 + 1 >*/
kpc = npp - (*n - imax + 1) * (*n - imax + 2) / 2 + 1;
/*< IF( IMAX.LT.N ) THEN >*/
if (imax < *n) {
/*< JMAX = IMAX + IDAMAX( N-IMAX, AP( KPC+1 ), 1 ) >*/
i__1 = *n - imax;
jmax = imax + idamax_(&i__1, &ap[kpc + 1], &c__1);
/*< ROWMAX = MAX( ROWMAX, ABS( AP( KPC+JMAX-IMAX ) ) ) >*/
/* Computing MAX */
d__2 = rowmax, d__3 = (d__1 = ap[kpc + jmax - imax], abs(
d__1));
rowmax = max(d__2,d__3);
/*< END IF >*/
}
/*< IF( ABSAKK.GE.ALPHA*COLMAX*( COLMAX / ROWMAX ) ) THEN >*/
if (absakk >= alpha * colmax * (colmax / rowmax)) {
/* no interchange, use 1-by-1 pivot block */
/*< KP = K >*/
kp = k;
/*< ELSE IF( ABS( AP( KPC ) ).GE.ALPHA*ROWMAX ) THEN >*/
} else if ((d__1 = ap[kpc], abs(d__1)) >= alpha * rowmax) {
/* interchange rows and columns K and IMAX, use 1-by-1 */
/* pivot block */
/*< KP = IMAX >*/
kp = imax;
/*< ELSE >*/
} else {
/* interchange rows and columns K+1 and IMAX, use 2-by-2 */
/* pivot block */
/*< KP = IMAX >*/
kp = imax;
/*< KSTEP = 2 >*/
kstep = 2;
/*< END IF >*/
}
/*< END IF >*/
}
/*< KK = K + KSTEP - 1 >*/
kk = k + kstep - 1;
/*< >*/
if (kstep == 2) {
knc = knc + *n - k + 1;
}
/*< IF( KP.NE.KK ) THEN >*/
if (kp != kk) {
/* Interchange rows and columns KK and KP in the trailing */
/* submatrix A(k:n,k:n) */
/*< >*/
if (kp < *n) {
i__1 = *n - kp;
dswap_(&i__1, &ap[knc + kp - kk + 1], &c__1, &ap[kpc + 1],
&c__1);
}
/*< KX = KNC + KP - KK >*/
kx = knc + kp - kk;
/*< DO 60 J = KK + 1, KP - 1 >*/
i__1 = kp - 1;
for (j = kk + 1; j <= i__1; ++j) {
/*< KX = KX + N - J + 1 >*/
kx = kx + *n - j + 1;
/*< T = AP( KNC+J-KK ) >*/
t = ap[knc + j - kk];
/*< AP( KNC+J-KK ) = AP( KX ) >*/
ap[knc + j - kk] = ap[kx];
/*< AP( KX ) = T >*/
ap[kx] = t;
/*< 60 CONTINUE >*/
/* L60: */
}
/*< T = AP( KNC ) >*/
t = ap[knc];
/*< AP( KNC ) = AP( KPC ) >*/
ap[knc] = ap[kpc];
/*< AP( KPC ) = T >*/
ap[kpc] = t;
/*< IF( KSTEP.EQ.2 ) THEN >*/
if (kstep == 2) {
/*< T = AP( KC+1 ) >*/
t = ap[kc + 1];
/*< AP( KC+1 ) = AP( KC+KP-K ) >*/
ap[kc + 1] = ap[kc + kp - k];
/*< AP( KC+KP-K ) = T >*/
ap[kc + kp - k] = t;
/*< END IF >*/
}
/*< END IF >*/
}
/* Update the trailing submatrix */
/*< IF( KSTEP.EQ.1 ) THEN >*/
if (kstep == 1) {
/* 1-by-1 pivot block D(k): column k now holds */
/* W(k) = L(k)*D(k) */
/* where L(k) is the k-th column of L */
/*< IF( K.LT.N ) THEN >*/
if (k < *n) {
/* Perform a rank-1 update of A(k+1:n,k+1:n) as */
/* A := A - L(k)*D(k)*L(k)' = A - W(k)*(1/D(k))*W(k)' */
/*< R1 = ONE / AP( KC ) >*/
r1 = 1. / ap[kc];
/*< >*/
i__1 = *n - k;
d__1 = -r1;
dspr_(uplo, &i__1, &d__1, &ap[kc + 1], &c__1, &ap[kc + *n
- k + 1], (ftnlen)1);
/* Store L(k) in column K */
/*< CALL DSCAL( N-K, R1, AP( KC+1 ), 1 ) >*/
i__1 = *n - k;
dscal_(&i__1, &r1, &ap[kc + 1], &c__1);
/*< END IF >*/
}
/*< ELSE >*/
} else {
/* 2-by-2 pivot block D(k): columns K and K+1 now hold */
/* ( W(k) W(k+1) ) = ( L(k) L(k+1) )*D(k) */
/* where L(k) and L(k+1) are the k-th and (k+1)-th columns */
/* of L */
/*< IF( K.LT.N-1 ) THEN >*/
if (k < *n - 1) {
/* Perform a rank-2 update of A(k+2:n,k+2:n) as */
/* A := A - ( L(k) L(k+1) )*D(k)*( L(k) L(k+1) )' */
/* = A - ( W(k) W(k+1) )*inv(D(k))*( W(k) W(k+1) )' */
/* Convert this to two rank-1 updates by using the eigen- */
/* decomposition of D(k) */
/*< >*/
dlaev2_(&ap[kc], &ap[kc + 1], &ap[knc], &r1, &r2, &c__, &
s);
/*< R1 = ONE / R1 >*/
r1 = 1. / r1;
/*< R2 = ONE / R2 >*/
r2 = 1. / r2;
/*< >*/
i__1 = *n - k - 1;
drot_(&i__1, &ap[kc + 2], &c__1, &ap[knc + 1], &c__1, &
c__, &s);
/*< >*/
i__1 = *n - k - 1;
d__1 = -r1;
dspr_(uplo, &i__1, &d__1, &ap[kc + 2], &c__1, &ap[knc + *
n - k], (ftnlen)1);
/*< >*/
i__1 = *n - k - 1;
d__1 = -r2;
dspr_(uplo, &i__1, &d__1, &ap[knc + 1], &c__1, &ap[knc + *
n - k], (ftnlen)1);
/* Store L(k) and L(k+1) in columns k and k+1 */
/*< CALL DSCAL( N-K-1, R1, AP( KC+2 ), 1 ) >*/
i__1 = *n - k - 1;
dscal_(&i__1, &r1, &ap[kc + 2], &c__1);
/*< CALL DSCAL( N-K-1, R2, AP( KNC+1 ), 1 ) >*/
i__1 = *n - k - 1;
dscal_(&i__1, &r2, &ap[knc + 1], &c__1);
/*< >*/
i__1 = *n - k - 1;
d__1 = -s;
drot_(&i__1, &ap[kc + 2], &c__1, &ap[knc + 1], &c__1, &
c__, &d__1);
/*< END IF >*/
}
/*< END IF >*/
}
/*< END IF >*/
}
/* Store details of the interchanges in IPIV */
/*< IF( KSTEP.EQ.1 ) THEN >*/
if (kstep == 1) {
/*< IPIV( K ) = KP >*/
ipiv[k] = kp;
/*< ELSE >*/
} else {
/*< IPIV( K ) = -KP >*/
ipiv[k] = -kp;
/*< IPIV( K+1 ) = -KP >*/
ipiv[k + 1] = -kp;
/*< END IF >*/
}
/* Increase K and return to the start of the main loop */
/*< K = K + KSTEP >*/
k += kstep;
/*< KC = KNC + N - K + 2 >*/
kc = knc + *n - k + 2;
/*< GO TO 40 >*/
goto L40;
/*< END IF >*/
}
/*< 70 CONTINUE >*/
L70:
/*< RETURN >*/
return 0;
/* End of DSPTRF */
/*< END >*/
} /* dsptrf_ */
/* Subroutine */ int dgbtrf_(integer *m, integer *n, integer *kl, integer *ku,
doublereal *ab, integer *ldab, integer *ipiv, integer *info)
{
/* System generated locals */
integer ab_dim1, ab_offset, i__1, i__2, i__3, i__4, i__5, i__6;
doublereal d__1;
/* Local variables */
integer i__, j, i2, i3, j2, j3, k2, jb, nb, ii, jj, jm, ip, jp, km, ju,
kv, nw;
extern /* Subroutine */ int dger_(integer *, integer *, doublereal *,
doublereal *, integer *, doublereal *, integer *, doublereal *,
integer *);
doublereal temp;
extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
integer *), dgemm_(char *, char *, integer *, integer *, integer *
, doublereal *, doublereal *, integer *, doublereal *, integer *,
doublereal *, doublereal *, integer *), dcopy_(
integer *, doublereal *, integer *, doublereal *, integer *),
dswap_(integer *, doublereal *, integer *, doublereal *, integer *
);
doublereal work13[4160] /* was [65][64] */, work31[4160] /*
was [65][64] */;
extern /* Subroutine */ int dtrsm_(char *, char *, char *, char *,
integer *, integer *, doublereal *, doublereal *, integer *,
doublereal *, integer *), dgbtf2_(
integer *, integer *, integer *, integer *, doublereal *, integer
*, integer *, integer *);
extern integer idamax_(integer *, doublereal *, integer *);
extern /* Subroutine */ int xerbla_(char *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *);
extern /* Subroutine */ int dlaswp_(integer *, doublereal *, integer *,
integer *, integer *, integer *, integer *);
/* -- LAPACK routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* DGBTRF computes an LU factorization of a real m-by-n band matrix A */
/* using partial pivoting with row interchanges. */
/* This is the blocked version of the algorithm, calling Level 3 BLAS. */
/* Arguments */
/* ========= */
/* M (input) INTEGER */
/* The number of rows of the matrix A. M >= 0. */
/* N (input) INTEGER */
/* The number of columns 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. */
/* AB (input/output) DOUBLE PRECISION array, dimension (LDAB,N) */
/* On entry, the matrix A in band storage, in rows KL+1 to */
/* 2*KL+KU+1; rows 1 to KL of the array need not be set. */
/* The j-th column of A is stored in the j-th column of the */
/* array AB as follows: */
/* AB(kl+ku+1+i-j,j) = A(i,j) for max(1,j-ku)<=i<=min(m,j+kl) */
/* On exit, details of the factorization: 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. */
/* See below for further details. */
/* LDAB (input) INTEGER */
/* The leading dimension of the array AB. LDAB >= 2*KL+KU+1. */
/* IPIV (output) INTEGER array, dimension (min(M,N)) */
/* The pivot indices; for 1 <= i <= min(M,N), row i of the */
/* matrix was interchanged with row IPIV(i). */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* > 0: if INFO = +i, U(i,i) is exactly zero. The factorization */
/* has been completed, but the factor U is exactly */
/* singular, and division by zero will occur if it is used */
/* to solve a system of equations. */
/* Further Details */
/* =============== */
/* The band storage scheme is illustrated by the following example, when */
/* M = N = 6, KL = 2, KU = 1: */
/* On entry: On exit: */
/* * * * + + + * * * u14 u25 u36 */
/* * * + + + + * * u13 u24 u35 u46 */
/* * a12 a23 a34 a45 a56 * u12 u23 u34 u45 u56 */
/* a11 a22 a33 a44 a55 a66 u11 u22 u33 u44 u55 u66 */
/* a21 a32 a43 a54 a65 * m21 m32 m43 m54 m65 * */
/* a31 a42 a53 a64 * * m31 m42 m53 m64 * * */
/* Array elements marked * are not used by the routine; elements marked */
/* + need not be set on entry, but are required by the routine to store */
/* elements of U because of fill-in resulting from the row interchanges. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. Local Arrays .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* KV is the number of superdiagonals in the factor U, allowing for */
/* fill-in */
/* Parameter adjustments */
ab_dim1 = *ldab;
ab_offset = 1 + ab_dim1;
ab -= ab_offset;
--ipiv;
/* Function Body */
kv = *ku + *kl;
/* Test the input parameters. */
*info = 0;
if (*m < 0) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*kl < 0) {
*info = -3;
} else if (*ku < 0) {
*info = -4;
} else if (*ldab < *kl + kv + 1) {
*info = -6;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("DGBTRF", &i__1);
return 0;
}
/* Quick return if possible */
if (*m == 0 || *n == 0) {
return 0;
}
/* Determine the block size for this environment */
nb = ilaenv_(&c__1, "DGBTRF", " ", m, n, kl, ku);
/* The block size must not exceed the limit set by the size of the */
/* local arrays WORK13 and WORK31. */
nb = min(nb,64);
if (nb <= 1 || nb > *kl) {
/* Use unblocked code */
dgbtf2_(m, n, kl, ku, &ab[ab_offset], ldab, &ipiv[1], info);
} else {
/* Use blocked code */
/* Zero the superdiagonal elements of the work array WORK13 */
i__1 = nb;
for (j = 1; j <= i__1; ++j) {
i__2 = j - 1;
for (i__ = 1; i__ <= i__2; ++i__) {
work13[i__ + j * 65 - 66] = 0.;
/* L10: */
}
/* L20: */
}
/* Zero the subdiagonal elements of the work array WORK31 */
i__1 = nb;
for (j = 1; j <= i__1; ++j) {
i__2 = nb;
for (i__ = j + 1; i__ <= i__2; ++i__) {
work31[i__ + j * 65 - 66] = 0.;
/* L30: */
}
/* L40: */
}
/* Gaussian elimination with partial pivoting */
/* Set fill-in elements in columns KU+2 to KV to zero */
i__1 = min(kv,*n);
for (j = *ku + 2; j <= i__1; ++j) {
i__2 = *kl;
for (i__ = kv - j + 2; i__ <= i__2; ++i__) {
ab[i__ + j * ab_dim1] = 0.;
/* L50: */
}
/* L60: */
}
/* JU is the index of the last column affected by the current */
/* stage of the factorization */
ju = 1;
i__1 = min(*m,*n);
i__2 = nb;
for (j = 1; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
/* Computing MIN */
i__3 = nb, i__4 = min(*m,*n) - j + 1;
jb = min(i__3,i__4);
/* The active part of the matrix is partitioned */
/* A11 A12 A13 */
/* A21 A22 A23 */
/* A31 A32 A33 */
/* Here A11, A21 and A31 denote the current block of JB columns */
/* which is about to be factorized. The number of rows in the */
/* partitioning are JB, I2, I3 respectively, and the numbers */
/* of columns are JB, J2, J3. The superdiagonal elements of A13 */
/* and the subdiagonal elements of A31 lie outside the band. */
/* Computing MIN */
i__3 = *kl - jb, i__4 = *m - j - jb + 1;
i2 = min(i__3,i__4);
/* Computing MIN */
i__3 = jb, i__4 = *m - j - *kl + 1;
i3 = min(i__3,i__4);
/* J2 and J3 are computed after JU has been updated. */
/* Factorize the current block of JB columns */
i__3 = j + jb - 1;
for (jj = j; jj <= i__3; ++jj) {
/* Set fill-in elements in column JJ+KV to zero */
if (jj + kv <= *n) {
i__4 = *kl;
for (i__ = 1; i__ <= i__4; ++i__) {
ab[i__ + (jj + kv) * ab_dim1] = 0.;
/* L70: */
}
}
/* Find pivot and test for singularity. KM is the number of */
/* subdiagonal elements in the current column. */
/* Computing MIN */
i__4 = *kl, i__5 = *m - jj;
km = min(i__4,i__5);
i__4 = km + 1;
jp = idamax_(&i__4, &ab[kv + 1 + jj * ab_dim1], &c__1);
ipiv[jj] = jp + jj - j;
if (ab[kv + jp + jj * ab_dim1] != 0.) {
/* Computing MAX */
/* Computing MIN */
i__6 = jj + *ku + jp - 1;
i__4 = ju, i__5 = min(i__6,*n);
ju = max(i__4,i__5);
if (jp != 1) {
/* Apply interchange to columns J to J+JB-1 */
if (jp + jj - 1 < j + *kl) {
i__4 = *ldab - 1;
i__5 = *ldab - 1;
dswap_(&jb, &ab[kv + 1 + jj - j + j * ab_dim1], &
i__4, &ab[kv + jp + jj - j + j * ab_dim1],
&i__5);
} else {
/* The interchange affects columns J to JJ-1 of A31 */
/* which are stored in the work array WORK31 */
i__4 = jj - j;
i__5 = *ldab - 1;
dswap_(&i__4, &ab[kv + 1 + jj - j + j * ab_dim1],
&i__5, &work31[jp + jj - j - *kl - 1], &
c__65);
i__4 = j + jb - jj;
i__5 = *ldab - 1;
i__6 = *ldab - 1;
dswap_(&i__4, &ab[kv + 1 + jj * ab_dim1], &i__5, &
ab[kv + jp + jj * ab_dim1], &i__6);
}
}
/* Compute multipliers */
d__1 = 1. / ab[kv + 1 + jj * ab_dim1];
dscal_(&km, &d__1, &ab[kv + 2 + jj * ab_dim1], &c__1);
/* Update trailing submatrix within the band and within */
/* the current block. JM is the index of the last column */
/* which needs to be updated. */
/* Computing MIN */
i__4 = ju, i__5 = j + jb - 1;
jm = min(i__4,i__5);
if (jm > jj) {
i__4 = jm - jj;
i__5 = *ldab - 1;
i__6 = *ldab - 1;
dger_(&km, &i__4, &c_b18, &ab[kv + 2 + jj * ab_dim1],
&c__1, &ab[kv + (jj + 1) * ab_dim1], &i__5, &
ab[kv + 1 + (jj + 1) * ab_dim1], &i__6);
}
} else {
/* If pivot is zero, set INFO to the index of the pivot */
/* unless a zero pivot has already been found. */
if (*info == 0) {
*info = jj;
}
}
/* Copy current column of A31 into the work array WORK31 */
/* Computing MIN */
i__4 = jj - j + 1;
nw = min(i__4,i3);
if (nw > 0) {
dcopy_(&nw, &ab[kv + *kl + 1 - jj + j + jj * ab_dim1], &
c__1, &work31[(jj - j + 1) * 65 - 65], &c__1);
}
/* L80: */
}
if (j + jb <= *n) {
/* Apply the row interchanges to the other blocks. */
/* Computing MIN */
i__3 = ju - j + 1;
j2 = min(i__3,kv) - jb;
/* Computing MAX */
i__3 = 0, i__4 = ju - j - kv + 1;
j3 = max(i__3,i__4);
/* Use DLASWP to apply the row interchanges to A12, A22, and */
/* A32. */
i__3 = *ldab - 1;
dlaswp_(&j2, &ab[kv + 1 - jb + (j + jb) * ab_dim1], &i__3, &
c__1, &jb, &ipiv[j], &c__1);
/* Adjust the pivot indices. */
i__3 = j + jb - 1;
for (i__ = j; i__ <= i__3; ++i__) {
ipiv[i__] = ipiv[i__] + j - 1;
/* L90: */
}
/* Apply the row interchanges to A13, A23, and A33 */
/* columnwise. */
k2 = j - 1 + jb + j2;
i__3 = j3;
for (i__ = 1; i__ <= i__3; ++i__) {
jj = k2 + i__;
i__4 = j + jb - 1;
for (ii = j + i__ - 1; ii <= i__4; ++ii) {
ip = ipiv[ii];
if (ip != ii) {
temp = ab[kv + 1 + ii - jj + jj * ab_dim1];
ab[kv + 1 + ii - jj + jj * ab_dim1] = ab[kv + 1 +
ip - jj + jj * ab_dim1];
ab[kv + 1 + ip - jj + jj * ab_dim1] = temp;
}
/* L100: */
}
/* L110: */
}
/* Update the relevant part of the trailing submatrix */
if (j2 > 0) {
/* Update A12 */
i__3 = *ldab - 1;
i__4 = *ldab - 1;
dtrsm_("Left", "Lower", "No transpose", "Unit", &jb, &j2,
&c_b31, &ab[kv + 1 + j * ab_dim1], &i__3, &ab[kv
+ 1 - jb + (j + jb) * ab_dim1], &i__4);
if (i2 > 0) {
/* Update A22 */
i__3 = *ldab - 1;
i__4 = *ldab - 1;
i__5 = *ldab - 1;
dgemm_("No transpose", "No transpose", &i2, &j2, &jb,
&c_b18, &ab[kv + 1 + jb + j * ab_dim1], &i__3,
&ab[kv + 1 - jb + (j + jb) * ab_dim1], &i__4,
&c_b31, &ab[kv + 1 + (j + jb) * ab_dim1], &
i__5);
}
if (i3 > 0) {
/* Update A32 */
i__3 = *ldab - 1;
i__4 = *ldab - 1;
dgemm_("No transpose", "No transpose", &i3, &j2, &jb,
&c_b18, work31, &c__65, &ab[kv + 1 - jb + (j
+ jb) * ab_dim1], &i__3, &c_b31, &ab[kv + *kl
+ 1 - jb + (j + jb) * ab_dim1], &i__4);
}
}
if (j3 > 0) {
/* Copy the lower triangle of A13 into the work array */
/* WORK13 */
i__3 = j3;
for (jj = 1; jj <= i__3; ++jj) {
i__4 = jb;
for (ii = jj; ii <= i__4; ++ii) {
work13[ii + jj * 65 - 66] = ab[ii - jj + 1 + (jj
+ j + kv - 1) * ab_dim1];
/* L120: */
}
/* L130: */
}
/* Update A13 in the work array */
i__3 = *ldab - 1;
dtrsm_("Left", "Lower", "No transpose", "Unit", &jb, &j3,
&c_b31, &ab[kv + 1 + j * ab_dim1], &i__3, work13,
&c__65);
if (i2 > 0) {
/* Update A23 */
i__3 = *ldab - 1;
i__4 = *ldab - 1;
dgemm_("No transpose", "No transpose", &i2, &j3, &jb,
&c_b18, &ab[kv + 1 + jb + j * ab_dim1], &i__3,
work13, &c__65, &c_b31, &ab[jb + 1 + (j + kv)
* ab_dim1], &i__4);
}
if (i3 > 0) {
/* Update A33 */
i__3 = *ldab - 1;
dgemm_("No transpose", "No transpose", &i3, &j3, &jb,
&c_b18, work31, &c__65, work13, &c__65, &
c_b31, &ab[*kl + 1 + (j + kv) * ab_dim1], &
i__3);
}
/* Copy the lower triangle of A13 back into place */
i__3 = j3;
for (jj = 1; jj <= i__3; ++jj) {
i__4 = jb;
for (ii = jj; ii <= i__4; ++ii) {
ab[ii - jj + 1 + (jj + j + kv - 1) * ab_dim1] =
work13[ii + jj * 65 - 66];
/* L140: */
}
/* L150: */
}
}
} else {
/* Adjust the pivot indices. */
i__3 = j + jb - 1;
for (i__ = j; i__ <= i__3; ++i__) {
ipiv[i__] = ipiv[i__] + j - 1;
/* L160: */
}
}
/* Partially undo the interchanges in the current block to */
/* restore the upper triangular form of A31 and copy the upper */
/* triangle of A31 back into place */
i__3 = j;
for (jj = j + jb - 1; jj >= i__3; --jj) {
jp = ipiv[jj] - jj + 1;
if (jp != 1) {
/* Apply interchange to columns J to JJ-1 */
if (jp + jj - 1 < j + *kl) {
/* The interchange does not affect A31 */
i__4 = jj - j;
i__5 = *ldab - 1;
i__6 = *ldab - 1;
dswap_(&i__4, &ab[kv + 1 + jj - j + j * ab_dim1], &
i__5, &ab[kv + jp + jj - j + j * ab_dim1], &
i__6);
} else {
/* The interchange does affect A31 */
i__4 = jj - j;
i__5 = *ldab - 1;
dswap_(&i__4, &ab[kv + 1 + jj - j + j * ab_dim1], &
i__5, &work31[jp + jj - j - *kl - 1], &c__65);
}
}
/* Copy the current column of A31 back into place */
/* Computing MIN */
i__4 = i3, i__5 = jj - j + 1;
nw = min(i__4,i__5);
if (nw > 0) {
dcopy_(&nw, &work31[(jj - j + 1) * 65 - 65], &c__1, &ab[
kv + *kl + 1 - jj + j + jj * ab_dim1], &c__1);
}
/* L170: */
}
/* L180: */
}
}
return 0;
/* End of DGBTRF */
} /* dgbtrf_ */
/* Subroutine */ int dstein_(integer * n, doublereal * d__, doublereal * e,
integer * m, doublereal * w, integer * iblock,
integer * isplit, doublereal * z__, integer * ldz,
doublereal * work, integer * iwork,
integer * ifail, integer * info)
{
/* System generated locals */
integer z_dim1, z_offset, i__1, i__2, i__3;
doublereal d__1, d__2, d__3, d__4, d__5;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
static integer jblk, nblk;
extern doublereal ddot_(integer *, doublereal *, integer *,
doublereal *, integer *);
static integer jmax;
extern doublereal dnrm2_(integer *, doublereal *, integer *);
static integer i__, j;
extern /* Subroutine */ int dscal_(integer *, doublereal *,
doublereal *,
integer *);
static integer iseed[4], gpind, iinfo;
extern doublereal dasum_(integer *, doublereal *, integer *);
extern /* Subroutine */ int dcopy_(integer *, doublereal *, integer *,
doublereal *, integer *);
static integer b1;
extern /* Subroutine */ int daxpy_(integer *, doublereal *,
doublereal *,
integer *, doublereal *, integer *);
static integer j1;
static doublereal ortol;
static integer indrv1, indrv2, indrv3, indrv4, indrv5, bn;
extern doublereal dlamch_(char *);
extern /* Subroutine */ int dlagtf_(integer *, doublereal *,
doublereal *,
doublereal *, doublereal *,
doublereal *, doublereal *,
integer *, integer *);
static doublereal xj;
extern integer idamax_(integer *, doublereal *, integer *);
extern /* Subroutine */ int xerbla_(char *, integer *), dlagts_(
integer
*,
integer
*,
doublereal
*,
doublereal
*,
doublereal
*,
doublereal
*,
integer
*,
doublereal
*,
doublereal
*,
integer
*);
static integer nrmchk;
extern /* Subroutine */ int dlarnv_(integer *, integer *, integer *,
doublereal *);
static integer blksiz;
static doublereal onenrm, dtpcrt, pertol, scl, eps, sep, nrm, tol;
static integer its;
static doublereal xjm, ztr, eps1;
#define z___ref(a_1,a_2) z__[(a_2)*z_dim1 + a_1]
/* -- LAPACK routine (instrumented to count operations, version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
September 30, 1994
Common block to return operation count and iteration count
ITCNT is initialized to 0, OPS is only incremented
Purpose
=======
DSTEIN computes the eigenvectors of a real symmetric tridiagonal
matrix T corresponding to specified eigenvalues, using inverse
iteration.
The maximum number of iterations allowed for each eigenvector is
specified by an internal parameter MAXITS (currently set to 5).
Arguments
=========
N (input) INTEGER
The order of the matrix. N >= 0.
D (input) DOUBLE PRECISION array, dimension (N)
The n diagonal elements of the tridiagonal matrix T.
E (input) DOUBLE PRECISION array, dimension (N)
The (n-1) subdiagonal elements of the tridiagonal matrix
T, in elements 1 to N-1. E(N) need not be set.
M (input) INTEGER
The number of eigenvectors to be found. 0 <= M <= N.
W (input) DOUBLE PRECISION array, dimension (N)
The first M elements of W contain the eigenvalues for
which eigenvectors are to be computed. The eigenvalues
should be grouped by split-off block and ordered from
smallest to largest within the block. ( The output array
W from DSTEBZ with ORDER = 'B' is expected here. )
IBLOCK (input) INTEGER array, dimension (N)
The submatrix indices associated with the corresponding
eigenvalues in W; IBLOCK(i)=1 if eigenvalue W(i) belongs to
the first submatrix from the top, =2 if W(i) belongs to
the second submatrix, etc. ( The output array IBLOCK
from DSTEBZ is expected here. )
ISPLIT (input) INTEGER array, dimension (N)
The splitting points, at which T breaks up into submatrices.
The first submatrix consists of rows/columns 1 to
ISPLIT( 1 ), the second of rows/columns ISPLIT( 1 )+1
through ISPLIT( 2 ), etc.
( The output array ISPLIT from DSTEBZ is expected here. )
Z (output) DOUBLE PRECISION array, dimension (LDZ, M)
The computed eigenvectors. The eigenvector associated
with the eigenvalue W(i) is stored in the i-th column of
Z. Any vector which fails to converge is set to its current
iterate after MAXITS iterations.
LDZ (input) INTEGER
The leading dimension of the array Z. LDZ >= max(1,N).
WORK (workspace) DOUBLE PRECISION array, dimension (5*N)
IWORK (workspace) INTEGER array, dimension (N)
IFAIL (output) INTEGER array, dimension (M)
On normal exit, all elements of IFAIL are zero.
If one or more eigenvectors fail to converge after
MAXITS iterations, then their indices are stored in
array IFAIL.
INFO (output) INTEGER
= 0: successful exit.
< 0: if INFO = -i, the i-th argument had an illegal value
> 0: if INFO = i, then i eigenvectors failed to converge
in MAXITS iterations. Their indices are stored in
array IFAIL.
Internal Parameters
===================
MAXITS INTEGER, default = 5
The maximum number of iterations performed.
EXTRA INTEGER, default = 2
The number of iterations performed after norm growth
criterion is satisfied, should be at least 1.
=====================================================================
Test the input parameters.
Parameter adjustments */
--d__;
--e;
--w;
--iblock;
--isplit;
z_dim1 = *ldz;
z_offset = 1 + z_dim1 * 1;
z__ -= z_offset;
--work;
--iwork;
--ifail;
/* Function Body */
*info = 0;
i__1 = *m;
for (i__ = 1; i__ <= i__1; ++i__) {
ifail[i__] = 0;
/* L10: */
}
if (*n < 0) {
*info = -1;
} else if (*m < 0 || *m > *n) {
*info = -4;
} else if (*ldz < max(1, *n)) {
*info = -9;
} else {
i__1 = *m;
for (j = 2; j <= i__1; ++j) {
if (iblock[j] < iblock[j - 1]) {
*info = -6;
goto L30;
}
if (iblock[j] == iblock[j - 1] && w[j] < w[j - 1]) {
*info = -5;
goto L30;
}
/* L20: */
}
L30:
;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("DSTEIN", &i__1);
return 0;
}
/* Initialize iteration count. */
latime_1.itcnt = 0.;
/* Quick return if possible */
if (*n == 0 || *m == 0) {
return 0;
} else if (*n == 1) {
z___ref(1, 1) = 1.;
return 0;
}
/* Get machine constants. */
eps = dlamch_("Precision");
/* Initialize seed for random number generator DLARNV. */
for (i__ = 1; i__ <= 4; ++i__) {
iseed[i__ - 1] = 1;
/* L40: */
}
/* Initialize pointers. */
indrv1 = 0;
indrv2 = indrv1 + *n;
indrv3 = indrv2 + *n;
indrv4 = indrv3 + *n;
indrv5 = indrv4 + *n;
/* Compute eigenvectors of matrix blocks. */
j1 = 1;
i__1 = iblock[*m];
for (nblk = 1; nblk <= i__1; ++nblk) {
/* Find starting and ending indices of block nblk. */
if (nblk == 1) {
b1 = 1;
} else {
b1 = isplit[nblk - 1] + 1;
}
bn = isplit[nblk];
blksiz = bn - b1 + 1;
if (blksiz == 1) {
goto L60;
}
gpind = b1;
/* Compute reorthogonalization criterion and stopping criterion. */
onenrm = (d__1 = d__[b1], abs(d__1)) + (d__2 =
e[b1], abs(d__2));
/* Computing MAX */
d__3 = onenrm, d__4 = (d__1 = d__[bn], abs(d__1)) + (d__2 =
e[bn - 1],
abs(d__2));
onenrm = max(d__3, d__4);
i__2 = bn - 1;
for (i__ = b1 + 1; i__ <= i__2; ++i__) {
/* Computing MAX */
d__4 = onenrm, d__5 = (d__1 =
d__[i__], abs(d__1)) + (d__2 =
e[i__ -
1],
abs
(d__2)) +
(d__3 = e[i__], abs(d__3));
onenrm = max(d__4, d__5);
/* L50: */
}
ortol = onenrm * .001;
dtpcrt = sqrt(.1 / blksiz);
/* Increment opcount for computing criteria. */
latime_1.ops = latime_1.ops + ((bn - b1) << 1) + 3;
/* Loop through eigenvalues of block nblk. */
L60:
jblk = 0;
i__2 = *m;
for (j = j1; j <= i__2; ++j) {
if (iblock[j] != nblk) {
j1 = j;
goto L160;
}
++jblk;
xj = w[j];
/* Skip all the work if the block size is one. */
if (blksiz == 1) {
work[indrv1 + 1] = 1.;
goto L120;
}
/* If eigenvalues j and j-1 are too close, add a relatively
small perturbation. */
if (jblk > 1) {
eps1 = (d__1 = eps * xj, abs(d__1));
pertol = eps1 * 10.;
sep = xj - xjm;
if (sep < pertol) {
xj = xjm + pertol;
}
}
its = 0;
nrmchk = 0;
/* Get random starting vector. */
dlarnv_(&c__2, iseed, &blksiz, &work[indrv1 + 1]);
/* Increment opcount for getting random starting vector.
( DLARND(2,.) requires 9 flops. ) */
latime_1.ops += blksiz * 9;
/* Copy the matrix T so it won't be destroyed in factorization. */
dcopy_(&blksiz, &d__[b1], &c__1, &work[indrv4 + 1],
&c__1);
i__3 = blksiz - 1;
dcopy_(&i__3, &e[b1], &c__1, &work[indrv2 + 2], &c__1);
i__3 = blksiz - 1;
dcopy_(&i__3, &e[b1], &c__1, &work[indrv3 + 1], &c__1);
/* Compute LU factors with partial pivoting ( PT = LU ) */
tol = 0.;
dlagtf_(&blksiz, &work[indrv4 + 1], &xj,
&work[indrv2 + 2], &work[indrv3 + 1], &tol,
&work[indrv5 + 1], &iwork[1], &iinfo);
/* Increment opcount for computing LU factors.
( DLAGTF(BLKSIZ,...) requires about 8*BLKSIZ flops. ) */
latime_1.ops += blksiz << 3;
/* Update iteration count. */
L70:
++its;
if (its > 5) {
goto L100;
}
/* Normalize and scale the righthand side vector Pb.
Computing MAX */
d__2 = eps, d__3 = (d__1 =
work[indrv4 + blksiz], abs(d__1));
scl =
blksiz * onenrm * max(d__2, d__3) / dasum_(&blksiz,
&work
[indrv1 +
1],
&c__1);
dscal_(&blksiz, &scl, &work[indrv1 + 1], &c__1);
/* Solve the system LU = Pb. */
dlagts_(&c_n1, &blksiz, &work[indrv4 + 1],
&work[indrv2 + 2], &work[indrv3 + 1],
&work[indrv5 + 1], &iwork[1], &work[indrv1 + 1],
&tol, &iinfo);
/* Increment opcount for scaling and solving linear system.
( DLAGTS(-1,BLKSIZ,...) requires about 8*BLKSIZ flops. ) */
latime_1.ops = latime_1.ops + 3 + blksiz * 10;
/* Reorthogonalize by modified Gram-Schmidt if eigenvalues are
close enough. */
if (jblk == 1) {
goto L90;
}
if ((d__1 = xj - xjm, abs(d__1)) > ortol) {
gpind = j;
}
if (gpind != j) {
i__3 = j - 1;
for (i__ = gpind; i__ <= i__3; ++i__) {
ztr =
-ddot_(&blksiz, &work[indrv1 + 1],
&c__1, &z___ref(b1, i__),
&c__1);
daxpy_(&blksiz, &ztr, &z___ref(b1, i__),
&c__1, &work[indrv1 + 1], &c__1);
/* L80: */
}
/* Increment opcount for reorthogonalizing. */
latime_1.ops += (j - gpind) * blksiz << 2;
}
/* Check the infinity norm of the iterate. */
L90:
jmax = idamax_(&blksiz, &work[indrv1 + 1], &c__1);
nrm = (d__1 = work[indrv1 + jmax], abs(d__1));
/* Continue for additional iterations after norm reaches
stopping criterion. */
if (nrm < dtpcrt) {
goto L70;
}
++nrmchk;
if (nrmchk < 3) {
goto L70;
}
goto L110;
/* If stopping criterion was not satisfied, update info and
store eigenvector number in array ifail. */
L100:
++(*info);
ifail[*info] = j;
/* Accept iterate as jth eigenvector. */
L110:
scl = 1. / dnrm2_(&blksiz, &work[indrv1 + 1], &c__1);
jmax = idamax_(&blksiz, &work[indrv1 + 1], &c__1);
if (work[indrv1 + jmax] < 0.) {
scl = -scl;
}
dscal_(&blksiz, &scl, &work[indrv1 + 1], &c__1);
/* Increment opcount for scaling. */
latime_1.ops += blksiz * 3;
L120:
i__3 = *n;
for (i__ = 1; i__ <= i__3; ++i__) {
z___ref(i__, j) = 0.;
/* L130: */
}
i__3 = blksiz;
for (i__ = 1; i__ <= i__3; ++i__) {
z___ref(b1 + i__ - 1, j) = work[indrv1 + i__];
/* L140: */
}
/* Save the shift to check eigenvalue spacing at next
iteration. */
xjm = xj;
/* L150: */
}
L160:
;
}
return 0;
/* End of DSTEIN */
} /* dstein_ */
/* Subroutine */ int dlalsd_(char *uplo, integer *smlsiz, integer *n, integer
*nrhs, doublereal *d__, doublereal *e, doublereal *b, integer *ldb,
doublereal *rcond, integer *rank, doublereal *work, integer *iwork,
integer *info)
{
/* System generated locals */
integer b_dim1, b_offset, i__1, i__2;
doublereal d__1;
/* Builtin functions */
double log(doublereal), d_sign(doublereal *, doublereal *);
/* Local variables */
integer c__, i__, j, k;
doublereal r__;
integer s, u, z__;
doublereal cs;
integer bx;
doublereal sn;
integer st, vt, nm1, st1;
doublereal eps;
integer iwk;
doublereal tol;
integer difl, difr;
doublereal rcnd;
integer perm, nsub;
extern /* Subroutine */ int drot_(integer *, doublereal *, integer *,
doublereal *, integer *, doublereal *, doublereal *);
integer nlvl, sqre, bxst;
extern /* Subroutine */ int dgemm_(char *, char *, integer *, integer *,
integer *, doublereal *, doublereal *, integer *, doublereal *,
integer *, doublereal *, doublereal *, integer *),
dcopy_(integer *, doublereal *, integer *, doublereal *, integer
*);
integer poles, sizei, nsize, nwork, icmpq1, icmpq2;
extern doublereal dlamch_(char *);
extern /* Subroutine */ int dlasda_(integer *, integer *, integer *,
integer *, doublereal *, doublereal *, doublereal *, integer *,
doublereal *, integer *, doublereal *, doublereal *, doublereal *,
doublereal *, integer *, integer *, integer *, integer *,
doublereal *, doublereal *, doublereal *, doublereal *, integer *,
integer *), dlalsa_(integer *, integer *, integer *, integer *,
doublereal *, integer *, doublereal *, integer *, doublereal *,
integer *, doublereal *, integer *, doublereal *, doublereal *,
doublereal *, doublereal *, integer *, integer *, integer *,
integer *, doublereal *, doublereal *, doublereal *, doublereal *,
integer *, integer *), dlascl_(char *, integer *, integer *,
doublereal *, doublereal *, integer *, integer *, doublereal *,
integer *, integer *);
extern integer idamax_(integer *, doublereal *, integer *);
extern /* Subroutine */ int dlasdq_(char *, integer *, integer *, integer
*, integer *, integer *, doublereal *, doublereal *, doublereal *,
integer *, doublereal *, integer *, doublereal *, integer *,
doublereal *, integer *), dlacpy_(char *, integer *,
integer *, doublereal *, integer *, doublereal *, integer *), dlartg_(doublereal *, doublereal *, doublereal *,
doublereal *, doublereal *), dlaset_(char *, integer *, integer *,
doublereal *, doublereal *, doublereal *, integer *),
xerbla_(char *, integer *);
integer givcol;
extern doublereal dlanst_(char *, integer *, doublereal *, doublereal *);
extern /* Subroutine */ int dlasrt_(char *, integer *, doublereal *,
integer *);
doublereal orgnrm;
integer givnum, givptr, smlszp;
/* -- LAPACK routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* DLALSD uses the singular value decomposition of A to solve the least */
/* squares problem of finding X to minimize the Euclidean norm of each */
/* column of A*X-B, where A is N-by-N upper bidiagonal, and X and B */
/* are N-by-NRHS. The solution X overwrites B. */
/* The singular values of A smaller than RCOND times the largest */
/* singular value are treated as zero in solving the least squares */
/* problem; in this case a minimum norm solution is returned. */
/* The actual singular values are returned in D in ascending order. */
/* This code makes very mild assumptions about floating point */
/* arithmetic. It will work on machines with a guard digit in */
/* add/subtract, or on those binary machines without guard digits */
/* which subtract like the Cray XMP, Cray YMP, Cray C 90, or Cray 2. */
/* It could conceivably fail on hexadecimal or decimal machines */
/* without guard digits, but we know of none. */
/* Arguments */
/* ========= */
/* UPLO (input) CHARACTER*1 */
/* = 'U': D and E define an upper bidiagonal matrix. */
/* = 'L': D and E define a lower bidiagonal matrix. */
/* SMLSIZ (input) INTEGER */
/* The maximum size of the subproblems at the bottom of the */
/* computation tree. */
/* N (input) INTEGER */
/* The dimension of the bidiagonal matrix. N >= 0. */
/* NRHS (input) INTEGER */
/* The number of columns of B. NRHS must be at least 1. */
/* D (input/output) DOUBLE PRECISION array, dimension (N) */
/* On entry D contains the main diagonal of the bidiagonal */
/* matrix. On exit, if INFO = 0, D contains its singular values. */
/* E (input/output) DOUBLE PRECISION array, dimension (N-1) */
/* Contains the super-diagonal entries of the bidiagonal matrix. */
/* On exit, E has been destroyed. */
/* B (input/output) DOUBLE PRECISION array, dimension (LDB,NRHS) */
/* On input, B contains the right hand sides of the least */
/* squares problem. On output, B contains the solution X. */
/* LDB (input) INTEGER */
/* The leading dimension of B in the calling subprogram. */
/* LDB must be at least max(1,N). */
/* RCOND (input) DOUBLE PRECISION */
/* The singular values of A less than or equal to RCOND times */
/* the largest singular value are treated as zero in solving */
/* the least squares problem. If RCOND is negative, */
/* machine precision is used instead. */
/* For example, if diag(S)*X=B were the least squares problem, */
/* where diag(S) is a diagonal matrix of singular values, the */
/* solution would be X(i) = B(i) / S(i) if S(i) is greater than */
/* RCOND*max(S), and X(i) = 0 if S(i) is less than or equal to */
/* RCOND*max(S). */
/* RANK (output) INTEGER */
/* The number of singular values of A greater than RCOND times */
/* the largest singular value. */
/* WORK (workspace) DOUBLE PRECISION array, dimension at least */
/* (9*N + 2*N*SMLSIZ + 8*N*NLVL + N*NRHS + (SMLSIZ+1)**2), */
/* where NLVL = max(0, INT(log_2 (N/(SMLSIZ+1))) + 1). */
/* IWORK (workspace) INTEGER array, dimension at least */
/* (3*N*NLVL + 11*N) */
/* INFO (output) INTEGER */
/* = 0: successful exit. */
/* < 0: if INFO = -i, the i-th argument had an illegal value. */
/* > 0: The algorithm failed to compute an singular value while */
/* working on the submatrix lying in rows and columns */
/* INFO/(N+1) through MOD(INFO,N+1). */
/* Further Details */
/* =============== */
/* Based on contributions by */
/* Ming Gu and Ren-Cang Li, Computer Science Division, University of */
/* California at Berkeley, USA */
/* Osni Marques, LBNL/NERSC, USA */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
--d__;
--e;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
--work;
--iwork;
/* Function Body */
*info = 0;
if (*n < 0) {
*info = -3;
} else if (*nrhs < 1) {
*info = -4;
} else if (*ldb < 1 || *ldb < *n) {
*info = -8;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("DLALSD", &i__1);
return 0;
}
eps = dlamch_("Epsilon");
/* Set up the tolerance. */
if (*rcond <= 0. || *rcond >= 1.) {
rcnd = eps;
} else {
rcnd = *rcond;
}
*rank = 0;
/* Quick return if possible. */
if (*n == 0) {
return 0;
} else if (*n == 1) {
if (d__[1] == 0.) {
dlaset_("A", &c__1, nrhs, &c_b6, &c_b6, &b[b_offset], ldb);
} else {
*rank = 1;
dlascl_("G", &c__0, &c__0, &d__[1], &c_b11, &c__1, nrhs, &b[
b_offset], ldb, info);
d__[1] = abs(d__[1]);
}
return 0;
}
/* Rotate the matrix if it is lower bidiagonal. */
if (*(unsigned char *)uplo == 'L') {
i__1 = *n - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
dlartg_(&d__[i__], &e[i__], &cs, &sn, &r__);
d__[i__] = r__;
e[i__] = sn * d__[i__ + 1];
d__[i__ + 1] = cs * d__[i__ + 1];
if (*nrhs == 1) {
drot_(&c__1, &b[i__ + b_dim1], &c__1, &b[i__ + 1 + b_dim1], &
c__1, &cs, &sn);
} else {
work[(i__ << 1) - 1] = cs;
work[i__ * 2] = sn;
}
/* L10: */
}
if (*nrhs > 1) {
i__1 = *nrhs;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = *n - 1;
for (j = 1; j <= i__2; ++j) {
cs = work[(j << 1) - 1];
sn = work[j * 2];
drot_(&c__1, &b[j + i__ * b_dim1], &c__1, &b[j + 1 + i__ *
b_dim1], &c__1, &cs, &sn);
/* L20: */
}
/* L30: */
}
}
}
/* Scale. */
nm1 = *n - 1;
orgnrm = dlanst_("M", n, &d__[1], &e[1]);
if (orgnrm == 0.) {
dlaset_("A", n, nrhs, &c_b6, &c_b6, &b[b_offset], ldb);
return 0;
}
dlascl_("G", &c__0, &c__0, &orgnrm, &c_b11, n, &c__1, &d__[1], n, info);
dlascl_("G", &c__0, &c__0, &orgnrm, &c_b11, &nm1, &c__1, &e[1], &nm1,
info);
/* If N is smaller than the minimum divide size SMLSIZ, then solve */
/* the problem with another solver. */
if (*n <= *smlsiz) {
nwork = *n * *n + 1;
dlaset_("A", n, n, &c_b6, &c_b11, &work[1], n);
dlasdq_("U", &c__0, n, n, &c__0, nrhs, &d__[1], &e[1], &work[1], n, &
work[1], n, &b[b_offset], ldb, &work[nwork], info);
if (*info != 0) {
return 0;
}
tol = rcnd * (d__1 = d__[idamax_(n, &d__[1], &c__1)], abs(d__1));
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
if (d__[i__] <= tol) {
dlaset_("A", &c__1, nrhs, &c_b6, &c_b6, &b[i__ + b_dim1], ldb);
} else {
dlascl_("G", &c__0, &c__0, &d__[i__], &c_b11, &c__1, nrhs, &b[
i__ + b_dim1], ldb, info);
++(*rank);
}
/* L40: */
}
dgemm_("T", "N", n, nrhs, n, &c_b11, &work[1], n, &b[b_offset], ldb, &
c_b6, &work[nwork], n);
dlacpy_("A", n, nrhs, &work[nwork], n, &b[b_offset], ldb);
/* Unscale. */
dlascl_("G", &c__0, &c__0, &c_b11, &orgnrm, n, &c__1, &d__[1], n,
info);
dlasrt_("D", n, &d__[1], info);
dlascl_("G", &c__0, &c__0, &orgnrm, &c_b11, n, nrhs, &b[b_offset],
ldb, info);
return 0;
}
/* Book-keeping and setting up some constants. */
nlvl = (integer) (log((doublereal) (*n) / (doublereal) (*smlsiz + 1)) /
log(2.)) + 1;
smlszp = *smlsiz + 1;
u = 1;
vt = *smlsiz * *n + 1;
difl = vt + smlszp * *n;
difr = difl + nlvl * *n;
z__ = difr + (nlvl * *n << 1);
c__ = z__ + nlvl * *n;
s = c__ + *n;
poles = s + *n;
givnum = poles + (nlvl << 1) * *n;
bx = givnum + (nlvl << 1) * *n;
nwork = bx + *n * *nrhs;
sizei = *n + 1;
k = sizei + *n;
givptr = k + *n;
perm = givptr + *n;
givcol = perm + nlvl * *n;
iwk = givcol + (nlvl * *n << 1);
st = 1;
sqre = 0;
icmpq1 = 1;
icmpq2 = 0;
nsub = 0;
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
if ((d__1 = d__[i__], abs(d__1)) < eps) {
d__[i__] = d_sign(&eps, &d__[i__]);
}
/* L50: */
}
i__1 = nm1;
for (i__ = 1; i__ <= i__1; ++i__) {
if ((d__1 = e[i__], abs(d__1)) < eps || i__ == nm1) {
++nsub;
iwork[nsub] = st;
/* Subproblem found. First determine its size and then */
/* apply divide and conquer on it. */
if (i__ < nm1) {
/* A subproblem with E(I) small for I < NM1. */
nsize = i__ - st + 1;
iwork[sizei + nsub - 1] = nsize;
} else if ((d__1 = e[i__], abs(d__1)) >= eps) {
/* A subproblem with E(NM1) not too small but I = NM1. */
nsize = *n - st + 1;
iwork[sizei + nsub - 1] = nsize;
} else {
/* A subproblem with E(NM1) small. This implies an */
/* 1-by-1 subproblem at D(N), which is not solved */
/* explicitly. */
nsize = i__ - st + 1;
iwork[sizei + nsub - 1] = nsize;
++nsub;
iwork[nsub] = *n;
iwork[sizei + nsub - 1] = 1;
dcopy_(nrhs, &b[*n + b_dim1], ldb, &work[bx + nm1], n);
}
st1 = st - 1;
if (nsize == 1) {
/* This is a 1-by-1 subproblem and is not solved */
/* explicitly. */
dcopy_(nrhs, &b[st + b_dim1], ldb, &work[bx + st1], n);
} else if (nsize <= *smlsiz) {
/* This is a small subproblem and is solved by DLASDQ. */
dlaset_("A", &nsize, &nsize, &c_b6, &c_b11, &work[vt + st1],
n);
dlasdq_("U", &c__0, &nsize, &nsize, &c__0, nrhs, &d__[st], &e[
st], &work[vt + st1], n, &work[nwork], n, &b[st +
b_dim1], ldb, &work[nwork], info);
if (*info != 0) {
return 0;
}
dlacpy_("A", &nsize, nrhs, &b[st + b_dim1], ldb, &work[bx +
st1], n);
} else {
/* A large problem. Solve it using divide and conquer. */
dlasda_(&icmpq1, smlsiz, &nsize, &sqre, &d__[st], &e[st], &
work[u + st1], n, &work[vt + st1], &iwork[k + st1], &
work[difl + st1], &work[difr + st1], &work[z__ + st1],
&work[poles + st1], &iwork[givptr + st1], &iwork[
givcol + st1], n, &iwork[perm + st1], &work[givnum +
st1], &work[c__ + st1], &work[s + st1], &work[nwork],
&iwork[iwk], info);
if (*info != 0) {
return 0;
}
bxst = bx + st1;
dlalsa_(&icmpq2, smlsiz, &nsize, nrhs, &b[st + b_dim1], ldb, &
work[bxst], n, &work[u + st1], n, &work[vt + st1], &
iwork[k + st1], &work[difl + st1], &work[difr + st1],
&work[z__ + st1], &work[poles + st1], &iwork[givptr +
st1], &iwork[givcol + st1], n, &iwork[perm + st1], &
work[givnum + st1], &work[c__ + st1], &work[s + st1],
&work[nwork], &iwork[iwk], info);
if (*info != 0) {
return 0;
}
}
st = i__ + 1;
}
/* L60: */
}
/* Apply the singular values and treat the tiny ones as zero. */
tol = rcnd * (d__1 = d__[idamax_(n, &d__[1], &c__1)], abs(d__1));
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
/* Some of the elements in D can be negative because 1-by-1 */
/* subproblems were not solved explicitly. */
if ((d__1 = d__[i__], abs(d__1)) <= tol) {
dlaset_("A", &c__1, nrhs, &c_b6, &c_b6, &work[bx + i__ - 1], n);
} else {
++(*rank);
dlascl_("G", &c__0, &c__0, &d__[i__], &c_b11, &c__1, nrhs, &work[
bx + i__ - 1], n, info);
}
d__[i__] = (d__1 = d__[i__], abs(d__1));
/* L70: */
}
/* Now apply back the right singular vectors. */
icmpq2 = 1;
i__1 = nsub;
for (i__ = 1; i__ <= i__1; ++i__) {
st = iwork[i__];
st1 = st - 1;
nsize = iwork[sizei + i__ - 1];
bxst = bx + st1;
if (nsize == 1) {
dcopy_(nrhs, &work[bxst], n, &b[st + b_dim1], ldb);
} else if (nsize <= *smlsiz) {
dgemm_("T", "N", &nsize, nrhs, &nsize, &c_b11, &work[vt + st1], n,
&work[bxst], n, &c_b6, &b[st + b_dim1], ldb);
} else {
dlalsa_(&icmpq2, smlsiz, &nsize, nrhs, &work[bxst], n, &b[st +
b_dim1], ldb, &work[u + st1], n, &work[vt + st1], &iwork[
k + st1], &work[difl + st1], &work[difr + st1], &work[z__
+ st1], &work[poles + st1], &iwork[givptr + st1], &iwork[
givcol + st1], n, &iwork[perm + st1], &work[givnum + st1],
&work[c__ + st1], &work[s + st1], &work[nwork], &iwork[
iwk], info);
if (*info != 0) {
return 0;
}
}
/* L80: */
}
/* Unscale and sort the singular values. */
dlascl_("G", &c__0, &c__0, &c_b11, &orgnrm, n, &c__1, &d__[1], n, info);
dlasrt_("D", n, &d__[1], info);
dlascl_("G", &c__0, &c__0, &orgnrm, &c_b11, n, nrhs, &b[b_offset], ldb,
info);
return 0;
/* End of DLALSD */
} /* dlalsd_ */
/*! return its largest absolute value */
inline double damax(const _dcovector& vec)
{ VERBOSE_REPORT;
double val( vec.array[idamax_(vec.l, vec.array, 1)-1] );
vec.destroy();
return val;
}
int
dlacon_(int *n, double *v, double *x, int *isgn, double *est, int *kase)
{
/*
Purpose
=======
DLACON estimates the 1-norm of a square matrix A.
Reverse communication is used for evaluating matrix-vector products.
Arguments
=========
N (input) INT
The order of the matrix. N >= 1.
V (workspace) DOUBLE PRECISION array, dimension (N)
On the final return, V = A*W, where EST = norm(V)/norm(W)
(W is not returned).
X (input/output) DOUBLE PRECISION array, dimension (N)
On an intermediate return, X should be overwritten by
A * X, if KASE=1,
A' * X, if KASE=2,
and DLACON must be re-called with all the other parameters
unchanged.
ISGN (workspace) INT array, dimension (N)
EST (output) DOUBLE PRECISION
An estimate (a lower bound) for norm(A).
KASE (input/output) INT
On the initial call to DLACON, KASE should be 0.
On an intermediate return, KASE will be 1 or 2, indicating
whether X should be overwritten by A * X or A' * X.
On the final return from DLACON, KASE will again be 0.
Further Details
======= =======
Contributed by Nick Higham, University of Manchester.
Originally named CONEST, dated March 16, 1988.
Reference: N.J. Higham, "FORTRAN codes for estimating the one-norm of
a real or complex matrix, with applications to condition estimation",
ACM Trans. Math. Soft., vol. 14, no. 4, pp. 381-396, December 1988.
=====================================================================
*/
/* Table of constant values */
int c__1 = 1;
double zero = 0.0;
double one = 1.0;
/* Local variables */
static int iter;
static int jump, jlast;
static double altsgn, estold;
static int i, j;
double temp;
extern int idamax_(int *, double *, int *);
extern double dasum_(int *, double *, int *);
extern int dcopy_(int *, double *, int *, double *, int *);
#define d_sign(a, b) (b >= 0 ? fabs(a) : -fabs(a)) /* Copy sign */
#define i_dnnt(a) \
( a>=0 ? floor(a+.5) : -floor(.5-a) ) /* Round to nearest integer */
if ( *kase == 0 ) {
for (i = 0; i < *n; ++i) {
x[i] = 1. / (double) (*n);
}
*kase = 1;
jump = 1;
return 0;
}
switch (jump) {
case 1: goto L20;
case 2: goto L40;
case 3: goto L70;
case 4: goto L110;
case 5: goto L140;
}
/* ................ ENTRY (JUMP = 1)
FIRST ITERATION. X HAS BEEN OVERWRITTEN BY A*X. */
L20:
if (*n == 1) {
v[0] = x[0];
*est = fabs(v[0]);
/* ... QUIT */
goto L150;
}
*est = dasum_(n, x, &c__1);
for (i = 0; i < *n; ++i) {
x[i] = d_sign(one, x[i]);
isgn[i] = i_dnnt(x[i]);
}
*kase = 2;
jump = 2;
return 0;
/* ................ ENTRY (JUMP = 2)
FIRST ITERATION. X HAS BEEN OVERWRITTEN BY TRANSPOSE(A)*X. */
L40:
j = idamax_(n, &x[0], &c__1);
--j;
iter = 2;
/* MAIN LOOP - ITERATIONS 2,3,...,ITMAX. */
L50:
for (i = 0; i < *n; ++i) x[i] = zero;
x[j] = one;
*kase = 1;
jump = 3;
return 0;
/* ................ ENTRY (JUMP = 3)
X HAS BEEN OVERWRITTEN BY A*X. */
L70:
dcopy_(n, &x[0], &c__1, &v[0], &c__1);
estold = *est;
*est = dasum_(n, v, &c__1);
for (i = 0; i < *n; ++i)
if (i_dnnt(d_sign(one, x[i])) != isgn[i])
goto L90;
/* REPEATED SIGN VECTOR DETECTED, HENCE ALGORITHM HAS CONVERGED. */
goto L120;
L90:
/* TEST FOR CYCLING. */
if (*est <= estold) goto L120;
for (i = 0; i < *n; ++i) {
x[i] = d_sign(one, x[i]);
isgn[i] = i_dnnt(x[i]);
}
*kase = 2;
jump = 4;
return 0;
/* ................ ENTRY (JUMP = 4)
X HAS BEEN OVERWRITTEN BY TRANDPOSE(A)*X. */
L110:
jlast = j;
j = idamax_(n, &x[0], &c__1);
--j;
if (x[jlast] != fabs(x[j]) && iter < 5) {
++iter;
goto L50;
}
/* ITERATION COMPLETE. FINAL STAGE. */
L120:
altsgn = 1.;
for (i = 1; i <= *n; ++i) {
x[i-1] = altsgn * ((double) (i - 1) / (double) (*n - 1) + 1.);
altsgn = -altsgn;
}
*kase = 1;
jump = 5;
return 0;
/* ................ ENTRY (JUMP = 5)
X HAS BEEN OVERWRITTEN BY A*X. */
L140:
temp = dasum_(n, x, &c__1) / (double) (*n * 3) * 2.;
if (temp > *est) {
dcopy_(n, &x[0], &c__1, &v[0], &c__1);
*est = temp;
}
L150:
*kase = 0;
return 0;
} /* dlacon_ */
/* Subroutine */ int zlaqps_(integer *m, integer *n, integer *offset, integer
*nb, integer *kb, doublecomplex *a, integer *lda, integer *jpvt,
doublecomplex *tau, doublereal *vn1, doublereal *vn2, doublecomplex *
auxv, doublecomplex *f, integer *ldf)
{
/* System generated locals */
integer a_dim1, a_offset, f_dim1, f_offset, i__1, i__2, i__3;
doublereal d__1, d__2;
doublecomplex z__1;
/* Builtin functions */
double sqrt(doublereal);
void d_cnjg(doublecomplex *, doublecomplex *);
double z_abs(doublecomplex *);
integer i_dnnt(doublereal *);
/* Local variables */
integer j, k, rk;
doublecomplex akk;
integer pvt;
doublereal temp, temp2, tol3z;
integer itemp;
extern /* Subroutine */ int zgemm_(char *, char *, integer *, integer *,
integer *, doublecomplex *, doublecomplex *, integer *,
doublecomplex *, integer *, doublecomplex *, doublecomplex *,
integer *), zgemv_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublecomplex *, doublecomplex *, integer *),
zswap_(integer *, doublecomplex *, integer *, doublecomplex *,
integer *);
extern doublereal dznrm2_(integer *, doublecomplex *, integer *), dlamch_(
char *);
extern integer idamax_(integer *, doublereal *, integer *);
integer lsticc;
extern /* Subroutine */ int zlarfp_(integer *, doublecomplex *,
doublecomplex *, integer *, doublecomplex *);
integer lastrk;
/* -- LAPACK auxiliary routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* ZLAQPS computes a step of QR factorization with column pivoting */
/* of a complex M-by-N matrix A by using Blas-3. It tries to factorize */
/* NB columns from A starting from the row OFFSET+1, and updates all */
/* of the matrix with Blas-3 xGEMM. */
/* In some cases, due to catastrophic cancellations, it cannot */
/* factorize NB columns. Hence, the actual number of factorized */
/* columns is returned in KB. */
/* Block A(1:OFFSET,1:N) is accordingly pivoted, but not factorized. */
/* Arguments */
/* ========= */
/* M (input) INTEGER */
/* The number of rows of the matrix A. M >= 0. */
/* N (input) INTEGER */
/* The number of columns of the matrix A. N >= 0 */
/* OFFSET (input) INTEGER */
/* The number of rows of A that have been factorized in */
/* previous steps. */
/* NB (input) INTEGER */
/* The number of columns to factorize. */
/* KB (output) INTEGER */
/* The number of columns actually factorized. */
/* A (input/output) COMPLEX*16 array, dimension (LDA,N) */
/* On entry, the M-by-N matrix A. */
/* On exit, block A(OFFSET+1:M,1:KB) is the triangular */
/* factor obtained and block A(1:OFFSET,1:N) has been */
/* accordingly pivoted, but no factorized. */
/* The rest of the matrix, block A(OFFSET+1:M,KB+1:N) has */
/* been updated. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1,M). */
/* JPVT (input/output) INTEGER array, dimension (N) */
/* JPVT(I) = K <==> Column K of the full matrix A has been */
/* permuted into position I in AP. */
/* TAU (output) COMPLEX*16 array, dimension (KB) */
/* The scalar factors of the elementary reflectors. */
/* VN1 (input/output) DOUBLE PRECISION array, dimension (N) */
/* The vector with the partial column norms. */
/* VN2 (input/output) DOUBLE PRECISION array, dimension (N) */
/* The vector with the exact column norms. */
/* AUXV (input/output) COMPLEX*16 array, dimension (NB) */
/* Auxiliar vector. */
/* F (input/output) COMPLEX*16 array, dimension (LDF,NB) */
/* Matrix F' = L*Y'*A. */
/* LDF (input) INTEGER */
/* The leading dimension of the array F. LDF >= max(1,N). */
/* Further Details */
/* =============== */
/* Based on contributions by */
/* G. Quintana-Orti, Depto. de Informatica, Universidad Jaime I, Spain */
/* X. Sun, Computer Science Dept., Duke University, USA */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
--jpvt;
--tau;
--vn1;
--vn2;
--auxv;
f_dim1 = *ldf;
f_offset = 1 + f_dim1;
f -= f_offset;
/* Function Body */
/* Computing MIN */
i__1 = *m, i__2 = *n + *offset;
lastrk = min(i__1,i__2);
lsticc = 0;
k = 0;
tol3z = sqrt(dlamch_("Epsilon"));
/* Beginning of while loop. */
L10:
if (k < *nb && lsticc == 0) {
++k;
rk = *offset + k;
/* Determine ith pivot column and swap if necessary */
i__1 = *n - k + 1;
pvt = k - 1 + idamax_(&i__1, &vn1[k], &c__1);
if (pvt != k) {
zswap_(m, &a[pvt * a_dim1 + 1], &c__1, &a[k * a_dim1 + 1], &c__1);
i__1 = k - 1;
zswap_(&i__1, &f[pvt + f_dim1], ldf, &f[k + f_dim1], ldf);
itemp = jpvt[pvt];
jpvt[pvt] = jpvt[k];
jpvt[k] = itemp;
vn1[pvt] = vn1[k];
vn2[pvt] = vn2[k];
}
/* Apply previous Householder reflectors to column K: */
/* A(RK:M,K) := A(RK:M,K) - A(RK:M,1:K-1)*F(K,1:K-1)'. */
if (k > 1) {
i__1 = k - 1;
for (j = 1; j <= i__1; ++j) {
i__2 = k + j * f_dim1;
d_cnjg(&z__1, &f[k + j * f_dim1]);
f[i__2].r = z__1.r, f[i__2].i = z__1.i;
/* L20: */
}
i__1 = *m - rk + 1;
i__2 = k - 1;
z__1.r = -1., z__1.i = -0.;
zgemv_("No transpose", &i__1, &i__2, &z__1, &a[rk + a_dim1], lda,
&f[k + f_dim1], ldf, &c_b2, &a[rk + k * a_dim1], &c__1);
i__1 = k - 1;
for (j = 1; j <= i__1; ++j) {
i__2 = k + j * f_dim1;
d_cnjg(&z__1, &f[k + j * f_dim1]);
f[i__2].r = z__1.r, f[i__2].i = z__1.i;
/* L30: */
}
}
/* Generate elementary reflector H(k). */
if (rk < *m) {
i__1 = *m - rk + 1;
zlarfp_(&i__1, &a[rk + k * a_dim1], &a[rk + 1 + k * a_dim1], &
c__1, &tau[k]);
} else {
zlarfp_(&c__1, &a[rk + k * a_dim1], &a[rk + k * a_dim1], &c__1, &
tau[k]);
}
i__1 = rk + k * a_dim1;
akk.r = a[i__1].r, akk.i = a[i__1].i;
i__1 = rk + k * a_dim1;
a[i__1].r = 1., a[i__1].i = 0.;
/* Compute Kth column of F: */
/* Compute F(K+1:N,K) := tau(K)*A(RK:M,K+1:N)'*A(RK:M,K). */
if (k < *n) {
i__1 = *m - rk + 1;
i__2 = *n - k;
zgemv_("Conjugate transpose", &i__1, &i__2, &tau[k], &a[rk + (k +
1) * a_dim1], lda, &a[rk + k * a_dim1], &c__1, &c_b1, &f[
k + 1 + k * f_dim1], &c__1);
}
/* Padding F(1:K,K) with zeros. */
i__1 = k;
for (j = 1; j <= i__1; ++j) {
i__2 = j + k * f_dim1;
f[i__2].r = 0., f[i__2].i = 0.;
/* L40: */
}
/* Incremental updating of F: */
/* F(1:N,K) := F(1:N,K) - tau(K)*F(1:N,1:K-1)*A(RK:M,1:K-1)' */
/* *A(RK:M,K). */
if (k > 1) {
i__1 = *m - rk + 1;
i__2 = k - 1;
i__3 = k;
z__1.r = -tau[i__3].r, z__1.i = -tau[i__3].i;
zgemv_("Conjugate transpose", &i__1, &i__2, &z__1, &a[rk + a_dim1]
, lda, &a[rk + k * a_dim1], &c__1, &c_b1, &auxv[1], &c__1);
i__1 = k - 1;
zgemv_("No transpose", n, &i__1, &c_b2, &f[f_dim1 + 1], ldf, &
auxv[1], &c__1, &c_b2, &f[k * f_dim1 + 1], &c__1);
}
/* Update the current row of A: */
/* A(RK,K+1:N) := A(RK,K+1:N) - A(RK,1:K)*F(K+1:N,1:K)'. */
if (k < *n) {
i__1 = *n - k;
z__1.r = -1., z__1.i = -0.;
zgemm_("No transpose", "Conjugate transpose", &c__1, &i__1, &k, &
z__1, &a[rk + a_dim1], lda, &f[k + 1 + f_dim1], ldf, &
c_b2, &a[rk + (k + 1) * a_dim1], lda);
}
/* Update partial column norms. */
if (rk < lastrk) {
i__1 = *n;
for (j = k + 1; j <= i__1; ++j) {
if (vn1[j] != 0.) {
/* NOTE: The following 4 lines follow from the analysis in */
/* Lapack Working Note 176. */
temp = z_abs(&a[rk + j * a_dim1]) / vn1[j];
/* Computing MAX */
d__1 = 0., d__2 = (temp + 1.) * (1. - temp);
temp = max(d__1,d__2);
/* Computing 2nd power */
d__1 = vn1[j] / vn2[j];
temp2 = temp * (d__1 * d__1);
if (temp2 <= tol3z) {
vn2[j] = (doublereal) lsticc;
lsticc = j;
} else {
vn1[j] *= sqrt(temp);
}
}
/* L50: */
}
}
i__1 = rk + k * a_dim1;
a[i__1].r = akk.r, a[i__1].i = akk.i;
/* End of while loop. */
goto L10;
}
*kb = k;
rk = *offset + *kb;
/* Apply the block reflector to the rest of the matrix: */
/* A(OFFSET+KB+1:M,KB+1:N) := A(OFFSET+KB+1:M,KB+1:N) - */
/* A(OFFSET+KB+1:M,1:KB)*F(KB+1:N,1:KB)'. */
/* Computing MIN */
i__1 = *n, i__2 = *m - *offset;
if (*kb < min(i__1,i__2)) {
i__1 = *m - rk;
i__2 = *n - *kb;
z__1.r = -1., z__1.i = -0.;
zgemm_("No transpose", "Conjugate transpose", &i__1, &i__2, kb, &z__1,
&a[rk + 1 + a_dim1], lda, &f[*kb + 1 + f_dim1], ldf, &c_b2, &
a[rk + 1 + (*kb + 1) * a_dim1], lda);
}
/* Recomputation of difficult columns. */
L60:
if (lsticc > 0) {
itemp = i_dnnt(&vn2[lsticc]);
i__1 = *m - rk;
vn1[lsticc] = dznrm2_(&i__1, &a[rk + 1 + lsticc * a_dim1], &c__1);
/* NOTE: The computation of VN1( LSTICC ) relies on the fact that */
/* SNRM2 does not fail on vectors with norm below the value of */
/* SQRT(DLAMCH('S')) */
vn2[lsticc] = vn1[lsticc];
lsticc = itemp;
goto L60;
}
return 0;
/* End of ZLAQPS */
} /* zlaqps_ */
/* Subroutine */extern "C" int dlaqp2_(integer *m, integer *n, integer *offset,
doublereal *a, integer *lda, integer *jpvt, doublereal *tau,
doublereal *vn1, doublereal *vn2, doublereal *work)
{
/* -- LAPACK auxiliary 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
=======
DLAQP2 computes a QR factorization with column pivoting of
the block A(OFFSET+1:M,1:N).
The block A(1:OFFSET,1:N) is accordingly pivoted, but not factorized.
Arguments
=========
M (input) INTEGER
The number of rows of the matrix A. M >= 0.
N (input) INTEGER
The number of columns of the matrix A. N >= 0.
OFFSET (input) INTEGER
The number of rows of the matrix A that must be pivoted
but no factorized. OFFSET >= 0.
A (input/output) DOUBLE PRECISION array, dimension (LDA,N)
On entry, the M-by-N matrix A.
On exit, the upper triangle of block A(OFFSET+1:M,1:N) is
the triangular factor obtained; the elements in block
A(OFFSET+1:M,1:N) below the diagonal, together with the
array TAU, represent the orthogonal matrix Q as a product of
elementary reflectors. Block A(1:OFFSET,1:N) has been
accordingly pivoted, but no factorized.
LDA (input) INTEGER
The leading dimension of the array A. LDA >= max(1,M).
JPVT (input/output) INTEGER array, dimension (N)
On entry, if JPVT(i) .ne. 0, the i-th column of A is permuted
to the front of A*P (a leading column); if JPVT(i) = 0,
the i-th column of A is a free column.
On exit, if JPVT(i) = k, then the i-th column of A*P
was the k-th column of A.
TAU (output) DOUBLE PRECISION array, dimension (min(M,N))
The scalar factors of the elementary reflectors.
VN1 (input/output) DOUBLE PRECISION array, dimension (N)
The vector with the partial column norms.
VN2 (input/output) DOUBLE PRECISION array, dimension (N)
The vector with the exact column norms.
WORK (workspace) DOUBLE PRECISION array, dimension (N)
Further Details
===============
Based on contributions by
G. Quintana-Orti, Depto. de Informatica, Universidad Jaime I, Spain
X. Sun, Computer Science Dept., Duke University, USA
=====================================================================
Parameter adjustments */
/* Table of constant values */
static integer c__1 = 1;
/* System generated locals */
integer a_dim1, a_offset, i__1, i__2, i__3;
doublereal d__1, d__2;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
static doublereal temp;
extern doublereal dnrm2_(integer *, doublereal *, integer *);
static doublereal temp2;
static integer i__, j;
extern /* Subroutine */ int dlarf_(char *, integer *, integer *,
doublereal *, integer *, doublereal *, doublereal *, integer *,
doublereal *);
static integer offpi, itemp;
extern /* Subroutine */ int dswap_(integer *, doublereal *, integer *,
doublereal *, integer *);
static integer mn;
extern /* Subroutine */ int dlarfg_(integer *, doublereal *, doublereal *,
integer *, doublereal *);
extern integer idamax_(integer *, doublereal *, integer *);
static doublereal aii;
static integer pvt;
#define a_ref(a_1,a_2) a[(a_2)*a_dim1 + a_1]
a_dim1 = *lda;
a_offset = 1 + a_dim1 * 1;
a -= a_offset;
--jpvt;
--tau;
--vn1;
--vn2;
--work;
/* Function Body
Computing MIN */
i__1 = *m - *offset;
mn = min(i__1,*n);
/* Compute factorization. */
i__1 = mn;
for (i__ = 1; i__ <= i__1; ++i__) {
offpi = *offset + i__;
/* Determine ith pivot column and swap if necessary. */
i__2 = *n - i__ + 1;
int ttt =idamax_(&i__2, &vn1[i__], &c__1);//zfm 可删 2003-1-8
pvt = i__ - 1 + idamax_(&i__2, &vn1[i__], &c__1);
//pvt的值决定了交换顺序 //zfm 2003-2-8
if (pvt != i__) {
dswap_(m, &a_ref(1, pvt), &c__1, &a_ref(1, i__), &c__1);
itemp = jpvt[pvt];
jpvt[pvt] = jpvt[i__];
jpvt[i__] = itemp;
vn1[pvt] = vn1[i__];
vn2[pvt] = vn2[i__];
}
/* Generate elementary reflector H(i). */
if (offpi < *m) {
i__2 = *m - offpi + 1;
dlarfg_(&i__2, &a_ref(offpi, i__), &a_ref(offpi + 1, i__), &c__1,
&tau[i__]);
} else {
dlarfg_(&c__1, &a_ref(*m, i__), &a_ref(*m, i__), &c__1, &tau[i__])
;
}
if (i__ < *n) {
/* Apply H(i)' to A(offset+i:m,i+1:n) from the left. */
aii = a_ref(offpi, i__);
a_ref(offpi, i__) = 1.;
i__2 = *m - offpi + 1;
i__3 = *n - i__;
dlarf_("Left", &i__2, &i__3, &a_ref(offpi, i__), &c__1, &tau[i__],
&a_ref(offpi, i__ + 1), lda, &work[1]);
a_ref(offpi, i__) = aii;
}
/* Update partial column norms. */
i__2 = *n;
for (j = i__ + 1; j <= i__2; ++j) {
if (vn1[j] != 0.) {
/* Computing 2nd power */
d__2 = (d__1 = a_ref(offpi, j), abs(d__1)) / vn1[j];
temp = 1. - d__2 * d__2;
temp = max(temp,0.);
/* Computing 2nd power */
d__1 = vn1[j] / vn2[j];
temp2 = temp * .05 * (d__1 * d__1) + 1.;
if (temp2 == 1.) {
if (offpi < *m) {
i__3 = *m - offpi;
vn1[j] = dnrm2_(&i__3, &a_ref(offpi + 1, j), &c__1);
vn2[j] = vn1[j];
} else {
vn1[j] = 0.;
vn2[j] = 0.;
}
} else {
vn1[j] *= sqrt(temp);
}
}
/* L10: */
}
/* L20: */
}
return 0;
/* End of DLAQP2 */
} /* dlaqp2_ */
size_t idamax(const int N,
const double *X,
const int incX) {
return idamax_(&N, X, &incX);
}
/* Subroutine */ int dlaqtr_(logical *ltran, logical *lreal, integer *n,
doublereal *t, integer *ldt, doublereal *b, doublereal *w, doublereal
*scale, doublereal *x, doublereal *work, integer *info)
{
/* System generated locals */
integer t_dim1, t_offset, i__1, i__2;
doublereal d__1, d__2, d__3, d__4, d__5, d__6;
/* Local variables */
doublereal d__[4] /* was [2][2] */;
integer i__, j, k;
doublereal v[4] /* was [2][2] */, z__;
integer j1, j2, n1, n2;
doublereal si, xj, sr, rec, eps, tjj, tmp;
extern doublereal ddot_(integer *, doublereal *, integer *, doublereal *,
integer *);
integer ierr;
doublereal smin, xmax;
extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
integer *);
extern doublereal dasum_(integer *, doublereal *, integer *);
extern /* Subroutine */ int daxpy_(integer *, doublereal *, doublereal *,
integer *, doublereal *, integer *);
integer jnext;
doublereal sminw, xnorm;
extern /* Subroutine */ int dlaln2_(logical *, integer *, integer *,
doublereal *, doublereal *, doublereal *, integer *, doublereal *,
doublereal *, doublereal *, integer *, doublereal *, doublereal *
, doublereal *, integer *, doublereal *, doublereal *, integer *);
extern doublereal dlamch_(char *), dlange_(char *, integer *,
integer *, doublereal *, integer *, doublereal *);
extern integer idamax_(integer *, doublereal *, integer *);
doublereal scaloc;
extern /* Subroutine */ int dladiv_(doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *, doublereal *);
doublereal bignum;
logical notran;
doublereal smlnum;
/* -- LAPACK auxiliary routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* DLAQTR solves the real quasi-triangular system */
/* op(T)*p = scale*c, if LREAL = .TRUE. */
/* or the complex quasi-triangular systems */
/* op(T + iB)*(p+iq) = scale*(c+id), if LREAL = .FALSE. */
/* in real arithmetic, where T is upper quasi-triangular. */
/* If LREAL = .FALSE., then the first diagonal block of T must be */
/* 1 by 1, B is the specially structured matrix */
/* B = [ b(1) b(2) ... b(n) ] */
/* [ w ] */
/* [ w ] */
/* [ . ] */
/* [ w ] */
/* op(A) = A or A', A' denotes the conjugate transpose of */
/* matrix A. */
/* On input, X = [ c ]. On output, X = [ p ]. */
/* [ d ] [ q ] */
/* This subroutine is designed for the condition number estimation */
/* in routine DTRSNA. */
/* Arguments */
/* ========= */
/* LTRAN (input) LOGICAL */
/* On entry, LTRAN specifies the option of conjugate transpose: */
/* = .FALSE., op(T+i*B) = T+i*B, */
/* = .TRUE., op(T+i*B) = (T+i*B)'. */
/* LREAL (input) LOGICAL */
/* On entry, LREAL specifies the input matrix structure: */
/* = .FALSE., the input is complex */
/* = .TRUE., the input is real */
/* N (input) INTEGER */
/* On entry, N specifies the order of T+i*B. N >= 0. */
/* T (input) DOUBLE PRECISION array, dimension (LDT,N) */
/* On entry, T contains a matrix in Schur canonical form. */
/* If LREAL = .FALSE., then the first diagonal block of T mu */
/* be 1 by 1. */
/* LDT (input) INTEGER */
/* The leading dimension of the matrix T. LDT >= max(1,N). */
/* B (input) DOUBLE PRECISION array, dimension (N) */
/* On entry, B contains the elements to form the matrix */
/* B as described above. */
/* If LREAL = .TRUE., B is not referenced. */
/* W (input) DOUBLE PRECISION */
/* On entry, W is the diagonal element of the matrix B. */
/* If LREAL = .TRUE., W is not referenced. */
/* SCALE (output) DOUBLE PRECISION */
/* On exit, SCALE is the scale factor. */
/* X (input/output) DOUBLE PRECISION array, dimension (2*N) */
/* On entry, X contains the right hand side of the system. */
/* On exit, X is overwritten by the solution. */
/* WORK (workspace) DOUBLE PRECISION array, dimension (N) */
/* INFO (output) INTEGER */
/* On exit, INFO is set to */
/* 0: successful exit. */
/* 1: the some diagonal 1 by 1 block has been perturbed by */
/* a small number SMIN to keep nonsingularity. */
/* 2: the some diagonal 2 by 2 block has been perturbed by */
/* a small number in DLALN2 to keep nonsingularity. */
/* NOTE: In the interests of speed, this routine does not */
/* check the inputs for errors. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. Local Arrays .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Do not test the input parameters for errors */
/* Parameter adjustments */
t_dim1 = *ldt;
t_offset = 1 + t_dim1;
t -= t_offset;
--b;
--x;
--work;
/* Function Body */
notran = ! (*ltran);
*info = 0;
/* Quick return if possible */
if (*n == 0) {
return 0;
}
/* Set constants to control overflow */
eps = dlamch_("P");
smlnum = dlamch_("S") / eps;
bignum = 1. / smlnum;
xnorm = dlange_("M", n, n, &t[t_offset], ldt, d__);
if (! (*lreal)) {
/* Computing MAX */
d__1 = xnorm, d__2 = abs(*w), d__1 = max(d__1,d__2), d__2 = dlange_(
"M", n, &c__1, &b[1], n, d__);
xnorm = max(d__1,d__2);
}
/* Computing MAX */
d__1 = smlnum, d__2 = eps * xnorm;
smin = max(d__1,d__2);
/* Compute 1-norm of each column of strictly upper triangular */
/* part of T to control overflow in triangular solver. */
work[1] = 0.;
i__1 = *n;
for (j = 2; j <= i__1; ++j) {
i__2 = j - 1;
work[j] = dasum_(&i__2, &t[j * t_dim1 + 1], &c__1);
/* L10: */
}
if (! (*lreal)) {
i__1 = *n;
for (i__ = 2; i__ <= i__1; ++i__) {
work[i__] += (d__1 = b[i__], abs(d__1));
/* L20: */
}
}
n2 = *n << 1;
n1 = *n;
if (! (*lreal)) {
n1 = n2;
}
k = idamax_(&n1, &x[1], &c__1);
xmax = (d__1 = x[k], abs(d__1));
*scale = 1.;
if (xmax > bignum) {
*scale = bignum / xmax;
dscal_(&n1, scale, &x[1], &c__1);
xmax = bignum;
}
if (*lreal) {
if (notran) {
/* Solve T*p = scale*c */
jnext = *n;
for (j = *n; j >= 1; --j) {
if (j > jnext) {
goto L30;
}
j1 = j;
j2 = j;
jnext = j - 1;
if (j > 1) {
if (t[j + (j - 1) * t_dim1] != 0.) {
j1 = j - 1;
jnext = j - 2;
}
}
if (j1 == j2) {
/* Meet 1 by 1 diagonal block */
/* Scale to avoid overflow when computing */
/* x(j) = b(j)/T(j,j) */
xj = (d__1 = x[j1], abs(d__1));
tjj = (d__1 = t[j1 + j1 * t_dim1], abs(d__1));
tmp = t[j1 + j1 * t_dim1];
if (tjj < smin) {
tmp = smin;
tjj = smin;
*info = 1;
}
if (xj == 0.) {
goto L30;
}
if (tjj < 1.) {
if (xj > bignum * tjj) {
rec = 1. / xj;
dscal_(n, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
}
x[j1] /= tmp;
xj = (d__1 = x[j1], abs(d__1));
/* Scale x if necessary to avoid overflow when adding a */
/* multiple of column j1 of T. */
if (xj > 1.) {
rec = 1. / xj;
if (work[j1] > (bignum - xmax) * rec) {
dscal_(n, &rec, &x[1], &c__1);
*scale *= rec;
}
}
if (j1 > 1) {
i__1 = j1 - 1;
d__1 = -x[j1];
daxpy_(&i__1, &d__1, &t[j1 * t_dim1 + 1], &c__1, &x[1]
, &c__1);
i__1 = j1 - 1;
k = idamax_(&i__1, &x[1], &c__1);
xmax = (d__1 = x[k], abs(d__1));
}
} else {
/* Meet 2 by 2 diagonal block */
/* Call 2 by 2 linear system solve, to take */
/* care of possible overflow by scaling factor. */
d__[0] = x[j1];
d__[1] = x[j2];
dlaln2_(&c_false, &c__2, &c__1, &smin, &c_b21, &t[j1 + j1
* t_dim1], ldt, &c_b21, &c_b21, d__, &c__2, &
c_b25, &c_b25, v, &c__2, &scaloc, &xnorm, &ierr);
if (ierr != 0) {
*info = 2;
}
if (scaloc != 1.) {
dscal_(n, &scaloc, &x[1], &c__1);
*scale *= scaloc;
}
x[j1] = v[0];
x[j2] = v[1];
/* Scale V(1,1) (= X(J1)) and/or V(2,1) (=X(J2)) */
/* to avoid overflow in updating right-hand side. */
/* Computing MAX */
d__1 = abs(v[0]), d__2 = abs(v[1]);
xj = max(d__1,d__2);
if (xj > 1.) {
rec = 1. / xj;
/* Computing MAX */
d__1 = work[j1], d__2 = work[j2];
if (max(d__1,d__2) > (bignum - xmax) * rec) {
dscal_(n, &rec, &x[1], &c__1);
*scale *= rec;
}
}
/* Update right-hand side */
if (j1 > 1) {
i__1 = j1 - 1;
d__1 = -x[j1];
daxpy_(&i__1, &d__1, &t[j1 * t_dim1 + 1], &c__1, &x[1]
, &c__1);
i__1 = j1 - 1;
d__1 = -x[j2];
daxpy_(&i__1, &d__1, &t[j2 * t_dim1 + 1], &c__1, &x[1]
, &c__1);
i__1 = j1 - 1;
k = idamax_(&i__1, &x[1], &c__1);
xmax = (d__1 = x[k], abs(d__1));
}
}
L30:
;
}
} else {
/* Solve T'*p = scale*c */
jnext = 1;
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
if (j < jnext) {
goto L40;
}
j1 = j;
j2 = j;
jnext = j + 1;
if (j < *n) {
if (t[j + 1 + j * t_dim1] != 0.) {
j2 = j + 1;
jnext = j + 2;
}
}
if (j1 == j2) {
/* 1 by 1 diagonal block */
/* Scale if necessary to avoid overflow in forming the */
/* right-hand side element by inner product. */
xj = (d__1 = x[j1], abs(d__1));
if (xmax > 1.) {
rec = 1. / xmax;
if (work[j1] > (bignum - xj) * rec) {
dscal_(n, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
}
i__2 = j1 - 1;
x[j1] -= ddot_(&i__2, &t[j1 * t_dim1 + 1], &c__1, &x[1], &
c__1);
xj = (d__1 = x[j1], abs(d__1));
tjj = (d__1 = t[j1 + j1 * t_dim1], abs(d__1));
tmp = t[j1 + j1 * t_dim1];
if (tjj < smin) {
tmp = smin;
tjj = smin;
*info = 1;
}
if (tjj < 1.) {
if (xj > bignum * tjj) {
rec = 1. / xj;
dscal_(n, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
}
x[j1] /= tmp;
/* Computing MAX */
d__2 = xmax, d__3 = (d__1 = x[j1], abs(d__1));
xmax = max(d__2,d__3);
} else {
/* 2 by 2 diagonal block */
/* Scale if necessary to avoid overflow in forming the */
/* right-hand side elements by inner product. */
/* Computing MAX */
d__3 = (d__1 = x[j1], abs(d__1)), d__4 = (d__2 = x[j2],
abs(d__2));
xj = max(d__3,d__4);
if (xmax > 1.) {
rec = 1. / xmax;
/* Computing MAX */
d__1 = work[j2], d__2 = work[j1];
if (max(d__1,d__2) > (bignum - xj) * rec) {
dscal_(n, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
}
i__2 = j1 - 1;
d__[0] = x[j1] - ddot_(&i__2, &t[j1 * t_dim1 + 1], &c__1,
&x[1], &c__1);
i__2 = j1 - 1;
d__[1] = x[j2] - ddot_(&i__2, &t[j2 * t_dim1 + 1], &c__1,
&x[1], &c__1);
dlaln2_(&c_true, &c__2, &c__1, &smin, &c_b21, &t[j1 + j1 *
t_dim1], ldt, &c_b21, &c_b21, d__, &c__2, &c_b25,
&c_b25, v, &c__2, &scaloc, &xnorm, &ierr);
if (ierr != 0) {
*info = 2;
}
if (scaloc != 1.) {
dscal_(n, &scaloc, &x[1], &c__1);
*scale *= scaloc;
}
x[j1] = v[0];
x[j2] = v[1];
/* Computing MAX */
d__3 = (d__1 = x[j1], abs(d__1)), d__4 = (d__2 = x[j2],
abs(d__2)), d__3 = max(d__3,d__4);
xmax = max(d__3,xmax);
}
L40:
;
}
}
} else {
/* Computing MAX */
d__1 = eps * abs(*w);
sminw = max(d__1,smin);
if (notran) {
/* Solve (T + iB)*(p+iq) = c+id */
jnext = *n;
for (j = *n; j >= 1; --j) {
if (j > jnext) {
goto L70;
}
j1 = j;
j2 = j;
jnext = j - 1;
if (j > 1) {
if (t[j + (j - 1) * t_dim1] != 0.) {
j1 = j - 1;
jnext = j - 2;
}
}
if (j1 == j2) {
/* 1 by 1 diagonal block */
/* Scale if necessary to avoid overflow in division */
z__ = *w;
if (j1 == 1) {
z__ = b[1];
}
xj = (d__1 = x[j1], abs(d__1)) + (d__2 = x[*n + j1], abs(
d__2));
tjj = (d__1 = t[j1 + j1 * t_dim1], abs(d__1)) + abs(z__);
tmp = t[j1 + j1 * t_dim1];
if (tjj < sminw) {
tmp = sminw;
tjj = sminw;
*info = 1;
}
if (xj == 0.) {
goto L70;
}
if (tjj < 1.) {
if (xj > bignum * tjj) {
rec = 1. / xj;
dscal_(&n2, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
}
dladiv_(&x[j1], &x[*n + j1], &tmp, &z__, &sr, &si);
x[j1] = sr;
x[*n + j1] = si;
xj = (d__1 = x[j1], abs(d__1)) + (d__2 = x[*n + j1], abs(
d__2));
/* Scale x if necessary to avoid overflow when adding a */
/* multiple of column j1 of T. */
if (xj > 1.) {
rec = 1. / xj;
if (work[j1] > (bignum - xmax) * rec) {
dscal_(&n2, &rec, &x[1], &c__1);
*scale *= rec;
}
}
if (j1 > 1) {
i__1 = j1 - 1;
d__1 = -x[j1];
daxpy_(&i__1, &d__1, &t[j1 * t_dim1 + 1], &c__1, &x[1]
, &c__1);
i__1 = j1 - 1;
d__1 = -x[*n + j1];
daxpy_(&i__1, &d__1, &t[j1 * t_dim1 + 1], &c__1, &x[*
n + 1], &c__1);
x[1] += b[j1] * x[*n + j1];
x[*n + 1] -= b[j1] * x[j1];
xmax = 0.;
i__1 = j1 - 1;
for (k = 1; k <= i__1; ++k) {
/* Computing MAX */
d__3 = xmax, d__4 = (d__1 = x[k], abs(d__1)) + (
d__2 = x[k + *n], abs(d__2));
xmax = max(d__3,d__4);
/* L50: */
}
}
} else {
/* Meet 2 by 2 diagonal block */
d__[0] = x[j1];
d__[1] = x[j2];
d__[2] = x[*n + j1];
d__[3] = x[*n + j2];
d__1 = -(*w);
dlaln2_(&c_false, &c__2, &c__2, &sminw, &c_b21, &t[j1 +
j1 * t_dim1], ldt, &c_b21, &c_b21, d__, &c__2, &
c_b25, &d__1, v, &c__2, &scaloc, &xnorm, &ierr);
if (ierr != 0) {
*info = 2;
}
if (scaloc != 1.) {
i__1 = *n << 1;
dscal_(&i__1, &scaloc, &x[1], &c__1);
*scale = scaloc * *scale;
}
x[j1] = v[0];
x[j2] = v[1];
x[*n + j1] = v[2];
x[*n + j2] = v[3];
/* Scale X(J1), .... to avoid overflow in */
/* updating right hand side. */
/* Computing MAX */
d__1 = abs(v[0]) + abs(v[2]), d__2 = abs(v[1]) + abs(v[3])
;
xj = max(d__1,d__2);
if (xj > 1.) {
rec = 1. / xj;
/* Computing MAX */
d__1 = work[j1], d__2 = work[j2];
if (max(d__1,d__2) > (bignum - xmax) * rec) {
dscal_(&n2, &rec, &x[1], &c__1);
*scale *= rec;
}
}
/* Update the right-hand side. */
if (j1 > 1) {
i__1 = j1 - 1;
d__1 = -x[j1];
daxpy_(&i__1, &d__1, &t[j1 * t_dim1 + 1], &c__1, &x[1]
, &c__1);
i__1 = j1 - 1;
d__1 = -x[j2];
daxpy_(&i__1, &d__1, &t[j2 * t_dim1 + 1], &c__1, &x[1]
, &c__1);
i__1 = j1 - 1;
d__1 = -x[*n + j1];
daxpy_(&i__1, &d__1, &t[j1 * t_dim1 + 1], &c__1, &x[*
n + 1], &c__1);
i__1 = j1 - 1;
d__1 = -x[*n + j2];
daxpy_(&i__1, &d__1, &t[j2 * t_dim1 + 1], &c__1, &x[*
n + 1], &c__1);
x[1] = x[1] + b[j1] * x[*n + j1] + b[j2] * x[*n + j2];
x[*n + 1] = x[*n + 1] - b[j1] * x[j1] - b[j2] * x[j2];
xmax = 0.;
i__1 = j1 - 1;
for (k = 1; k <= i__1; ++k) {
/* Computing MAX */
d__3 = (d__1 = x[k], abs(d__1)) + (d__2 = x[k + *
n], abs(d__2));
xmax = max(d__3,xmax);
/* L60: */
}
}
}
L70:
;
}
} else {
/* Solve (T + iB)'*(p+iq) = c+id */
jnext = 1;
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
if (j < jnext) {
goto L80;
}
j1 = j;
j2 = j;
jnext = j + 1;
if (j < *n) {
if (t[j + 1 + j * t_dim1] != 0.) {
j2 = j + 1;
jnext = j + 2;
}
}
if (j1 == j2) {
/* 1 by 1 diagonal block */
/* Scale if necessary to avoid overflow in forming the */
/* right-hand side element by inner product. */
xj = (d__1 = x[j1], abs(d__1)) + (d__2 = x[j1 + *n], abs(
d__2));
if (xmax > 1.) {
rec = 1. / xmax;
if (work[j1] > (bignum - xj) * rec) {
dscal_(&n2, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
}
i__2 = j1 - 1;
x[j1] -= ddot_(&i__2, &t[j1 * t_dim1 + 1], &c__1, &x[1], &
c__1);
i__2 = j1 - 1;
x[*n + j1] -= ddot_(&i__2, &t[j1 * t_dim1 + 1], &c__1, &x[
*n + 1], &c__1);
if (j1 > 1) {
x[j1] -= b[j1] * x[*n + 1];
x[*n + j1] += b[j1] * x[1];
}
xj = (d__1 = x[j1], abs(d__1)) + (d__2 = x[j1 + *n], abs(
d__2));
z__ = *w;
if (j1 == 1) {
z__ = b[1];
}
/* Scale if necessary to avoid overflow in */
/* complex division */
tjj = (d__1 = t[j1 + j1 * t_dim1], abs(d__1)) + abs(z__);
tmp = t[j1 + j1 * t_dim1];
if (tjj < sminw) {
tmp = sminw;
tjj = sminw;
*info = 1;
}
if (tjj < 1.) {
if (xj > bignum * tjj) {
rec = 1. / xj;
dscal_(&n2, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
}
d__1 = -z__;
dladiv_(&x[j1], &x[*n + j1], &tmp, &d__1, &sr, &si);
x[j1] = sr;
x[j1 + *n] = si;
/* Computing MAX */
d__3 = (d__1 = x[j1], abs(d__1)) + (d__2 = x[j1 + *n],
abs(d__2));
xmax = max(d__3,xmax);
} else {
/* 2 by 2 diagonal block */
/* Scale if necessary to avoid overflow in forming the */
/* right-hand side element by inner product. */
/* Computing MAX */
d__5 = (d__1 = x[j1], abs(d__1)) + (d__2 = x[*n + j1],
abs(d__2)), d__6 = (d__3 = x[j2], abs(d__3)) + (
d__4 = x[*n + j2], abs(d__4));
xj = max(d__5,d__6);
if (xmax > 1.) {
rec = 1. / xmax;
/* Computing MAX */
d__1 = work[j1], d__2 = work[j2];
if (max(d__1,d__2) > (bignum - xj) / xmax) {
dscal_(&n2, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
}
i__2 = j1 - 1;
d__[0] = x[j1] - ddot_(&i__2, &t[j1 * t_dim1 + 1], &c__1,
&x[1], &c__1);
i__2 = j1 - 1;
d__[1] = x[j2] - ddot_(&i__2, &t[j2 * t_dim1 + 1], &c__1,
&x[1], &c__1);
i__2 = j1 - 1;
d__[2] = x[*n + j1] - ddot_(&i__2, &t[j1 * t_dim1 + 1], &
c__1, &x[*n + 1], &c__1);
i__2 = j1 - 1;
d__[3] = x[*n + j2] - ddot_(&i__2, &t[j2 * t_dim1 + 1], &
c__1, &x[*n + 1], &c__1);
d__[0] -= b[j1] * x[*n + 1];
d__[1] -= b[j2] * x[*n + 1];
d__[2] += b[j1] * x[1];
d__[3] += b[j2] * x[1];
dlaln2_(&c_true, &c__2, &c__2, &sminw, &c_b21, &t[j1 + j1
* t_dim1], ldt, &c_b21, &c_b21, d__, &c__2, &
c_b25, w, v, &c__2, &scaloc, &xnorm, &ierr);
if (ierr != 0) {
*info = 2;
}
if (scaloc != 1.) {
dscal_(&n2, &scaloc, &x[1], &c__1);
*scale = scaloc * *scale;
}
x[j1] = v[0];
x[j2] = v[1];
x[*n + j1] = v[2];
x[*n + j2] = v[3];
/* Computing MAX */
d__5 = (d__1 = x[j1], abs(d__1)) + (d__2 = x[*n + j1],
abs(d__2)), d__6 = (d__3 = x[j2], abs(d__3)) + (
d__4 = x[*n + j2], abs(d__4)), d__5 = max(d__5,
d__6);
xmax = max(d__5,xmax);
}
L80:
;
}
}
}
return 0;
/* End of DLAQTR */
} /* dlaqtr_ */
/* Subroutine */ int dsgesv_(integer *n, integer *nrhs, doublereal *a,
integer *lda, integer *ipiv, doublereal *b, integer *ldb, doublereal *
x, integer *ldx, doublereal *work, real *swork, integer *iter,
integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, work_dim1, work_offset,
x_dim1, x_offset, i__1;
doublereal d__1;
/* Local variables */
integer i__;
doublereal cte, eps, anrm;
integer ptsa;
doublereal rnrm, xnrm;
integer ptsx;
integer iiter;
/* -- LAPACK PROTOTYPE driver routine (version 3.2) -- */
/* February 2007 */
/* Purpose */
/* ======= */
/* DSGESV computes the solution to a real system of linear equations */
/* A * X = B, */
/* where A is an N-by-N matrix and X and B are N-by-NRHS matrices. */
/* DSGESV first attempts to factorize the matrix in SINGLE PRECISION */
/* and use this factorization within an iterative refinement procedure */
/* to produce a solution with DOUBLE PRECISION normwise backward error */
/* quality (see below). If the approach fails the method switches to a */
/* DOUBLE PRECISION factorization and solve. */
/* The iterative refinement is not going to be a winning strategy if */
/* the ratio SINGLE PRECISION performance over DOUBLE PRECISION */
/* performance is too small. A reasonable strategy should take the */
/* number of right-hand sides and the size of the matrix into account. */
/* This might be done with a call to ILAENV in the future. Up to now, we */
/* always try iterative refinement. */
/* The iterative refinement process is stopped if */
/* ITER > ITERMAX */
/* or for all the RHS we have: */
/* RNRM < SQRT(N)*XNRM*ANRM*EPS*BWDMAX */
/* where */
/* o ITER is the number of the current iteration in the iterative */
/* refinement process */
/* o RNRM is the infinity-norm of the residual */
/* o XNRM is the infinity-norm of the solution */
/* o ANRM is the infinity-operator-norm of the matrix A */
/* o EPS is the machine epsilon returned by DLAMCH('Epsilon') */
/* The value ITERMAX and BWDMAX are fixed to 30 and 1.0D+00 */
/* respectively. */
/* Arguments */
/* ========= */
/* N (input) INTEGER */
/* The number of linear equations, i.e., 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. */
/* A (input or input/ouptut) DOUBLE PRECISION array, */
/* dimension (LDA,N) */
/* On entry, the N-by-N coefficient matrix A. */
/* On exit, if iterative refinement has been successfully used */
/* (INFO.EQ.0 and ITER.GE.0, see description below), then A is */
/* unchanged, if double precision factorization has been used */
/* (INFO.EQ.0 and ITER.LT.0, see description below), then the */
/* array A contains the factors L and U from the factorization */
/* A = P*L*U; the unit diagonal elements of L are not stored. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1,N). */
/* IPIV (output) INTEGER array, dimension (N) */
/* The pivot indices that define the permutation matrix P; */
/* row i of the matrix was interchanged with row IPIV(i). */
/* Corresponds either to the single precision factorization */
/* (if INFO.EQ.0 and ITER.GE.0) or the double precision */
/* factorization (if INFO.EQ.0 and ITER.LT.0). */
/* B (input) DOUBLE PRECISION array, dimension (LDB,NRHS) */
/* The N-by-NRHS right hand side matrix B. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= max(1,N). */
/* X (output) DOUBLE PRECISION array, dimension (LDX,NRHS) */
/* If INFO = 0, the N-by-NRHS solution matrix X. */
/* LDX (input) INTEGER */
/* The leading dimension of the array X. LDX >= max(1,N). */
/* WORK (workspace) DOUBLE PRECISION array, dimension (N*NRHS) */
/* This array is used to hold the residual vectors. */
/* SWORK (workspace) REAL array, dimension (N*(N+NRHS)) */
/* This array is used to use the single precision matrix and the */
/* right-hand sides or solutions in single precision. */
/* ITER (output) INTEGER */
/* < 0: iterative refinement has failed, double precision */
/* factorization has been performed */
/* -1 : the routine fell back to full precision for */
/* implementation- or machine-specific reasons */
/* -2 : narrowing the precision induced an overflow, */
/* the routine fell back to full precision */
/* -3 : failure of SGETRF */
/* -31: stop the iterative refinement after the 30th */
/* iterations */
/* > 0: iterative refinement has been sucessfully used. */
/* Returns the number of iterations */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* > 0: if INFO = i, U(i,i) computed in DOUBLE PRECISION is */
/* exactly zero. The factorization has been completed, */
/* but the factor U is exactly singular, so the solution */
/* could not be computed. */
/* ========= */
/* Parameter adjustments */
work_dim1 = *n;
work_offset = 1 + work_dim1;
work -= work_offset;
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
--ipiv;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
x_dim1 = *ldx;
x_offset = 1 + x_dim1;
x -= x_offset;
--swork;
/* Function Body */
*info = 0;
*iter = 0;
/* Test the input parameters. */
if (*n < 0) {
*info = -1;
} else if (*nrhs < 0) {
*info = -2;
} else if (*lda < max(1,*n)) {
*info = -4;
} else if (*ldb < max(1,*n)) {
*info = -7;
} else if (*ldx < max(1,*n)) {
*info = -9;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("DSGESV", &i__1);
return 0;
}
/* Quick return if (N.EQ.0). */
if (*n == 0) {
return 0;
}
/* Skip single precision iterative refinement if a priori slower */
/* than double precision factorization. */
if (FALSE_) {
*iter = -1;
goto L40;
}
/* Compute some constants. */
anrm = dlange_("I", n, n, &a[a_offset], lda, &work[work_offset]);
eps = dlamch_("Epsilon");
cte = anrm * eps * sqrt((doublereal) (*n)) * 1.;
/* Set the indices PTSA, PTSX for referencing SA and SX in SWORK. */
ptsa = 1;
ptsx = ptsa + *n * *n;
/* Convert B from double precision to single precision and store the */
/* result in SX. */
dlag2s_(n, nrhs, &b[b_offset], ldb, &swork[ptsx], n, info);
if (*info != 0) {
*iter = -2;
goto L40;
}
/* Convert A from double precision to single precision and store the */
/* result in SA. */
dlag2s_(n, n, &a[a_offset], lda, &swork[ptsa], n, info);
if (*info != 0) {
*iter = -2;
goto L40;
}
/* Compute the LU factorization of SA. */
sgetrf_(n, n, &swork[ptsa], n, &ipiv[1], info);
if (*info != 0) {
*iter = -3;
goto L40;
}
/* Solve the system SA*SX = SB. */
sgetrs_("No transpose", n, nrhs, &swork[ptsa], n, &ipiv[1], &swork[ptsx],
n, info);
/* Convert SX back to double precision */
slag2d_(n, nrhs, &swork[ptsx], n, &x[x_offset], ldx, info);
/* Compute R = B - AX (R is WORK). */
dlacpy_("All", n, nrhs, &b[b_offset], ldb, &work[work_offset], n);
dgemm_("No Transpose", "No Transpose", n, nrhs, n, &c_b10, &a[a_offset],
lda, &x[x_offset], ldx, &c_b11, &work[work_offset], n);
/* Check whether the NRHS normwise backward errors satisfy the */
/* stopping criterion. If yes, set ITER=0 and return. */
i__1 = *nrhs;
for (i__ = 1; i__ <= i__1; ++i__) {
xnrm = (d__1 = x[idamax_(n, &x[i__ * x_dim1 + 1], &c__1) + i__ *
x_dim1], abs(d__1));
rnrm = (d__1 = work[idamax_(n, &work[i__ * work_dim1 + 1], &c__1) +
i__ * work_dim1], abs(d__1));
if (rnrm > xnrm * cte) {
goto L10;
}
}
/* If we are here, the NRHS normwise backward errors satisfy the */
/* stopping criterion. We are good to exit. */
*iter = 0;
return 0;
L10:
for (iiter = 1; iiter <= 30; ++iiter) {
/* Convert R (in WORK) from double precision to single precision */
/* and store the result in SX. */
dlag2s_(n, nrhs, &work[work_offset], n, &swork[ptsx], n, info);
if (*info != 0) {
*iter = -2;
goto L40;
}
/* Solve the system SA*SX = SR. */
sgetrs_("No transpose", n, nrhs, &swork[ptsa], n, &ipiv[1], &swork[
ptsx], n, info);
/* Convert SX back to double precision and update the current */
/* iterate. */
slag2d_(n, nrhs, &swork[ptsx], n, &work[work_offset], n, info);
i__1 = *nrhs;
for (i__ = 1; i__ <= i__1; ++i__) {
daxpy_(n, &c_b11, &work[i__ * work_dim1 + 1], &c__1, &x[i__ *
x_dim1 + 1], &c__1);
}
/* Compute R = B - AX (R is WORK). */
dlacpy_("All", n, nrhs, &b[b_offset], ldb, &work[work_offset], n);
dgemm_("No Transpose", "No Transpose", n, nrhs, n, &c_b10, &a[
a_offset], lda, &x[x_offset], ldx, &c_b11, &work[work_offset],
n);
/* Check whether the NRHS normwise backward errors satisfy the */
/* stopping criterion. If yes, set ITER=IITER>0 and return. */
i__1 = *nrhs;
for (i__ = 1; i__ <= i__1; ++i__) {
xnrm = (d__1 = x[idamax_(n, &x[i__ * x_dim1 + 1], &c__1) + i__ *
x_dim1], abs(d__1));
rnrm = (d__1 = work[idamax_(n, &work[i__ * work_dim1 + 1], &c__1)
+ i__ * work_dim1], abs(d__1));
if (rnrm > xnrm * cte) {
goto L20;
}
}
/* If we are here, the NRHS normwise backward errors satisfy the */
/* stopping criterion, we are good to exit. */
*iter = iiter;
return 0;
L20:
;
}
/* If we are at this place of the code, this is because we have */
/* performed ITER=ITERMAX iterations and never satisified the */
/* stopping criterion, set up the ITER flag accordingly and follow up */
/* on double precision routine. */
*iter = -31;
L40:
/* Single-precision iterative refinement failed to converge to a */
/* satisfactory solution, so we resort to double precision. */
dgetrf_(n, n, &a[a_offset], lda, &ipiv[1], info);
if (*info != 0) {
return 0;
}
dlacpy_("All", n, nrhs, &b[b_offset], ldb, &x[x_offset], ldx);
dgetrs_("No transpose", n, nrhs, &a[a_offset], lda, &ipiv[1], &x[x_offset]
, ldx, info);
return 0;
/* End of DSGESV. */
} /* dsgesv_ */
int dsptrf_(char *uplo, int *n, double *ap, int *
ipiv, int *info)
{
/* System generated locals */
int i__1, i__2;
double d__1, d__2, d__3;
/* Builtin functions */
double sqrt(double);
/* Local variables */
int i__, j, k;
double t, r1, d11, d12, d21, d22;
int kc, kk, kp;
double wk;
int kx, knc, kpc, npp;
double wkm1, wkp1;
int imax, jmax;
extern int dspr_(char *, int *, double *,
double *, int *, double *);
double alpha;
extern int dscal_(int *, double *, double *,
int *);
extern int lsame_(char *, char *);
extern int dswap_(int *, double *, int *,
double *, int *);
int kstep;
int upper;
double absakk;
extern int idamax_(int *, double *, int *);
extern int xerbla_(char *, int *);
double colmax, rowmax;
/* -- LAPACK routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* DSPTRF computes the factorization of a float symmetric matrix A stored */
/* in packed format using the Bunch-Kaufman diagonal pivoting method: */
/* A = U*D*U**T or A = L*D*L**T */
/* where U (or L) is a product of permutation and unit upper (lower) */
/* triangular matrices, and D is symmetric and block diagonal with */
/* 1-by-1 and 2-by-2 diagonal blocks. */
/* 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. */
/* AP (input/output) DOUBLE PRECISION array, dimension (N*(N+1)/2) */
/* On entry, the upper or lower triangle of the symmetric 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. */
/* On exit, the block diagonal matrix D and the multipliers used */
/* to obtain the factor U or L, stored as a packed triangular */
/* matrix overwriting A (see below for further details). */
/* IPIV (output) INTEGER array, dimension (N) */
/* Details of the interchanges and the block structure of D. */
/* If IPIV(k) > 0, then rows and columns k and IPIV(k) were */
/* interchanged and D(k,k) is a 1-by-1 diagonal block. */
/* If UPLO = 'U' and IPIV(k) = IPIV(k-1) < 0, then rows and */
/* columns k-1 and -IPIV(k) were interchanged and D(k-1:k,k-1:k) */
/* is a 2-by-2 diagonal block. If UPLO = 'L' and IPIV(k) = */
/* IPIV(k+1) < 0, then rows and columns k+1 and -IPIV(k) were */
/* interchanged and D(k:k+1,k:k+1) is a 2-by-2 diagonal block. */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* > 0: if INFO = i, D(i,i) is exactly zero. The factorization */
/* has been completed, but the block diagonal matrix D is */
/* exactly singular, and division by zero will occur if it */
/* is used to solve a system of equations. */
/* Further Details */
/* =============== */
/* 5-96 - Based on modifications by J. Lewis, Boeing Computer Services */
/* Company */
/* If UPLO = 'U', then A = U*D*U', where */
/* U = P(n)*U(n)* ... *P(k)U(k)* ..., */
/* i.e., U is a product of terms P(k)*U(k), where k decreases from n to */
/* 1 in steps of 1 or 2, and D is a block diagonal matrix with 1-by-1 */
/* and 2-by-2 diagonal blocks D(k). P(k) is a permutation matrix as */
/* defined by IPIV(k), and U(k) is a unit upper triangular matrix, such */
/* that if the diagonal block D(k) is of order s (s = 1 or 2), then */
/* ( I v 0 ) k-s */
/* U(k) = ( 0 I 0 ) s */
/* ( 0 0 I ) n-k */
/* k-s s n-k */
/* If s = 1, D(k) overwrites A(k,k), and v overwrites A(1:k-1,k). */
/* If s = 2, the upper triangle of D(k) overwrites A(k-1,k-1), A(k-1,k), */
/* and A(k,k), and v overwrites A(1:k-2,k-1:k). */
/* If UPLO = 'L', then A = L*D*L', where */
/* L = P(1)*L(1)* ... *P(k)*L(k)* ..., */
/* i.e., L is a product of terms P(k)*L(k), where k increases from 1 to */
/* n in steps of 1 or 2, and D is a block diagonal matrix with 1-by-1 */
/* and 2-by-2 diagonal blocks D(k). P(k) is a permutation matrix as */
/* defined by IPIV(k), and L(k) is a unit lower triangular matrix, such */
/* that if the diagonal block D(k) is of order s (s = 1 or 2), then */
/* ( I 0 0 ) k-1 */
/* L(k) = ( 0 I 0 ) s */
/* ( 0 v I ) n-k-s+1 */
/* k-1 s n-k-s+1 */
/* If s = 1, D(k) overwrites A(k,k), and v overwrites A(k+1:n,k). */
/* If s = 2, the lower triangle of D(k) overwrites A(k,k), A(k+1,k), */
/* and A(k+1,k+1), and v overwrites A(k+2:n,k:k+1). */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
--ipiv;
--ap;
/* Function Body */
*info = 0;
upper = lsame_(uplo, "U");
if (! upper && ! lsame_(uplo, "L")) {
*info = -1;
} else if (*n < 0) {
*info = -2;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("DSPTRF", &i__1);
return 0;
}
/* Initialize ALPHA for use in choosing pivot block size. */
alpha = (sqrt(17.) + 1.) / 8.;
if (upper) {
/* Factorize A as U*D*U' using the upper triangle of A */
/* K is the main loop index, decreasing from N to 1 in steps of */
/* 1 or 2 */
k = *n;
kc = (*n - 1) * *n / 2 + 1;
L10:
knc = kc;
/* If K < 1, exit from loop */
if (k < 1) {
goto L110;
}
kstep = 1;
/* Determine rows and columns to be interchanged and whether */
/* a 1-by-1 or 2-by-2 pivot block will be used */
absakk = (d__1 = ap[kc + k - 1], ABS(d__1));
/* IMAX is the row-index of the largest off-diagonal element in */
/* column K, and COLMAX is its absolute value */
if (k > 1) {
i__1 = k - 1;
imax = idamax_(&i__1, &ap[kc], &c__1);
colmax = (d__1 = ap[kc + imax - 1], ABS(d__1));
} else {
colmax = 0.;
}
if (MAX(absakk,colmax) == 0.) {
/* Column K is zero: set INFO and continue */
if (*info == 0) {
*info = k;
}
kp = k;
} else {
if (absakk >= alpha * colmax) {
/* no interchange, use 1-by-1 pivot block */
kp = k;
} else {
/* JMAX is the column-index of the largest off-diagonal */
/* element in row IMAX, and ROWMAX is its absolute value */
rowmax = 0.;
jmax = imax;
kx = imax * (imax + 1) / 2 + imax;
i__1 = k;
for (j = imax + 1; j <= i__1; ++j) {
if ((d__1 = ap[kx], ABS(d__1)) > rowmax) {
rowmax = (d__1 = ap[kx], ABS(d__1));
jmax = j;
}
kx += j;
/* L20: */
}
kpc = (imax - 1) * imax / 2 + 1;
if (imax > 1) {
i__1 = imax - 1;
jmax = idamax_(&i__1, &ap[kpc], &c__1);
/* Computing MAX */
d__2 = rowmax, d__3 = (d__1 = ap[kpc + jmax - 1], ABS(
d__1));
rowmax = MAX(d__2,d__3);
}
if (absakk >= alpha * colmax * (colmax / rowmax)) {
/* no interchange, use 1-by-1 pivot block */
kp = k;
} else if ((d__1 = ap[kpc + imax - 1], ABS(d__1)) >= alpha *
rowmax) {
/* interchange rows and columns K and IMAX, use 1-by-1 */
/* pivot block */
kp = imax;
} else {
/* interchange rows and columns K-1 and IMAX, use 2-by-2 */
/* pivot block */
kp = imax;
kstep = 2;
}
}
kk = k - kstep + 1;
if (kstep == 2) {
knc = knc - k + 1;
}
if (kp != kk) {
/* Interchange rows and columns KK and KP in the leading */
/* submatrix A(1:k,1:k) */
i__1 = kp - 1;
dswap_(&i__1, &ap[knc], &c__1, &ap[kpc], &c__1);
kx = kpc + kp - 1;
i__1 = kk - 1;
for (j = kp + 1; j <= i__1; ++j) {
kx = kx + j - 1;
t = ap[knc + j - 1];
ap[knc + j - 1] = ap[kx];
ap[kx] = t;
/* L30: */
}
t = ap[knc + kk - 1];
ap[knc + kk - 1] = ap[kpc + kp - 1];
ap[kpc + kp - 1] = t;
if (kstep == 2) {
t = ap[kc + k - 2];
ap[kc + k - 2] = ap[kc + kp - 1];
ap[kc + kp - 1] = t;
}
}
/* Update the leading submatrix */
if (kstep == 1) {
/* 1-by-1 pivot block D(k): column k now holds */
/* W(k) = U(k)*D(k) */
/* where U(k) is the k-th column of U */
/* Perform a rank-1 update of A(1:k-1,1:k-1) as */
/* A := A - U(k)*D(k)*U(k)' = A - W(k)*1/D(k)*W(k)' */
r1 = 1. / ap[kc + k - 1];
i__1 = k - 1;
d__1 = -r1;
dspr_(uplo, &i__1, &d__1, &ap[kc], &c__1, &ap[1]);
/* Store U(k) in column k */
i__1 = k - 1;
dscal_(&i__1, &r1, &ap[kc], &c__1);
} else {
/* 2-by-2 pivot block D(k): columns k and k-1 now hold */
/* ( W(k-1) W(k) ) = ( U(k-1) U(k) )*D(k) */
/* where U(k) and U(k-1) are the k-th and (k-1)-th columns */
/* of U */
/* Perform a rank-2 update of A(1:k-2,1:k-2) as */
/* A := A - ( U(k-1) U(k) )*D(k)*( U(k-1) U(k) )' */
/* = A - ( W(k-1) W(k) )*inv(D(k))*( W(k-1) W(k) )' */
if (k > 2) {
d12 = ap[k - 1 + (k - 1) * k / 2];
d22 = ap[k - 1 + (k - 2) * (k - 1) / 2] / d12;
d11 = ap[k + (k - 1) * k / 2] / d12;
t = 1. / (d11 * d22 - 1.);
d12 = t / d12;
for (j = k - 2; j >= 1; --j) {
wkm1 = d12 * (d11 * ap[j + (k - 2) * (k - 1) / 2] -
ap[j + (k - 1) * k / 2]);
wk = d12 * (d22 * ap[j + (k - 1) * k / 2] - ap[j + (k
- 2) * (k - 1) / 2]);
for (i__ = j; i__ >= 1; --i__) {
ap[i__ + (j - 1) * j / 2] = ap[i__ + (j - 1) * j /
2] - ap[i__ + (k - 1) * k / 2] * wk - ap[
i__ + (k - 2) * (k - 1) / 2] * wkm1;
/* L40: */
}
ap[j + (k - 1) * k / 2] = wk;
ap[j + (k - 2) * (k - 1) / 2] = wkm1;
/* L50: */
}
}
}
}
/* Store details of the interchanges in IPIV */
if (kstep == 1) {
ipiv[k] = kp;
} else {
ipiv[k] = -kp;
ipiv[k - 1] = -kp;
}
/* Decrease K and return to the start of the main loop */
k -= kstep;
kc = knc - k;
goto L10;
} else {
/* Factorize A as L*D*L' using the lower triangle of A */
/* K is the main loop index, increasing from 1 to N in steps of */
/* 1 or 2 */
k = 1;
kc = 1;
npp = *n * (*n + 1) / 2;
L60:
knc = kc;
/* If K > N, exit from loop */
if (k > *n) {
goto L110;
}
kstep = 1;
/* Determine rows and columns to be interchanged and whether */
/* a 1-by-1 or 2-by-2 pivot block will be used */
absakk = (d__1 = ap[kc], ABS(d__1));
/* IMAX is the row-index of the largest off-diagonal element in */
/* column K, and COLMAX is its absolute value */
if (k < *n) {
i__1 = *n - k;
imax = k + idamax_(&i__1, &ap[kc + 1], &c__1);
colmax = (d__1 = ap[kc + imax - k], ABS(d__1));
} else {
colmax = 0.;
}
if (MAX(absakk,colmax) == 0.) {
/* Column K is zero: set INFO and continue */
if (*info == 0) {
*info = k;
}
kp = k;
} else {
if (absakk >= alpha * colmax) {
/* no interchange, use 1-by-1 pivot block */
kp = k;
} else {
/* JMAX is the column-index of the largest off-diagonal */
/* element in row IMAX, and ROWMAX is its absolute value */
rowmax = 0.;
kx = kc + imax - k;
i__1 = imax - 1;
for (j = k; j <= i__1; ++j) {
if ((d__1 = ap[kx], ABS(d__1)) > rowmax) {
rowmax = (d__1 = ap[kx], ABS(d__1));
jmax = j;
}
kx = kx + *n - j;
/* L70: */
}
kpc = npp - (*n - imax + 1) * (*n - imax + 2) / 2 + 1;
if (imax < *n) {
i__1 = *n - imax;
jmax = imax + idamax_(&i__1, &ap[kpc + 1], &c__1);
/* Computing MAX */
d__2 = rowmax, d__3 = (d__1 = ap[kpc + jmax - imax], ABS(
d__1));
rowmax = MAX(d__2,d__3);
}
if (absakk >= alpha * colmax * (colmax / rowmax)) {
/* no interchange, use 1-by-1 pivot block */
kp = k;
} else if ((d__1 = ap[kpc], ABS(d__1)) >= alpha * rowmax) {
/* interchange rows and columns K and IMAX, use 1-by-1 */
/* pivot block */
kp = imax;
} else {
/* interchange rows and columns K+1 and IMAX, use 2-by-2 */
/* pivot block */
kp = imax;
kstep = 2;
}
}
kk = k + kstep - 1;
if (kstep == 2) {
knc = knc + *n - k + 1;
}
if (kp != kk) {
/* Interchange rows and columns KK and KP in the trailing */
/* submatrix A(k:n,k:n) */
if (kp < *n) {
i__1 = *n - kp;
dswap_(&i__1, &ap[knc + kp - kk + 1], &c__1, &ap[kpc + 1],
&c__1);
}
kx = knc + kp - kk;
i__1 = kp - 1;
for (j = kk + 1; j <= i__1; ++j) {
kx = kx + *n - j + 1;
t = ap[knc + j - kk];
ap[knc + j - kk] = ap[kx];
ap[kx] = t;
/* L80: */
}
t = ap[knc];
ap[knc] = ap[kpc];
ap[kpc] = t;
if (kstep == 2) {
t = ap[kc + 1];
ap[kc + 1] = ap[kc + kp - k];
ap[kc + kp - k] = t;
}
}
/* Update the trailing submatrix */
if (kstep == 1) {
/* 1-by-1 pivot block D(k): column k now holds */
/* W(k) = L(k)*D(k) */
/* where L(k) is the k-th column of L */
if (k < *n) {
/* Perform a rank-1 update of A(k+1:n,k+1:n) as */
/* A := A - L(k)*D(k)*L(k)' = A - W(k)*(1/D(k))*W(k)' */
r1 = 1. / ap[kc];
i__1 = *n - k;
d__1 = -r1;
dspr_(uplo, &i__1, &d__1, &ap[kc + 1], &c__1, &ap[kc + *n
- k + 1]);
/* Store L(k) in column K */
i__1 = *n - k;
dscal_(&i__1, &r1, &ap[kc + 1], &c__1);
}
} else {
/* 2-by-2 pivot block D(k): columns K and K+1 now hold */
/* ( W(k) W(k+1) ) = ( L(k) L(k+1) )*D(k) */
/* where L(k) and L(k+1) are the k-th and (k+1)-th columns */
/* of L */
if (k < *n - 1) {
/* Perform a rank-2 update of A(k+2:n,k+2:n) as */
/* A := A - ( L(k) L(k+1) )*D(k)*( L(k) L(k+1) )' */
/* = A - ( W(k) W(k+1) )*inv(D(k))*( W(k) W(k+1) )' */
d21 = ap[k + 1 + (k - 1) * ((*n << 1) - k) / 2];
d11 = ap[k + 1 + k * ((*n << 1) - k - 1) / 2] / d21;
d22 = ap[k + (k - 1) * ((*n << 1) - k) / 2] / d21;
t = 1. / (d11 * d22 - 1.);
d21 = t / d21;
i__1 = *n;
for (j = k + 2; j <= i__1; ++j) {
wk = d21 * (d11 * ap[j + (k - 1) * ((*n << 1) - k) /
2] - ap[j + k * ((*n << 1) - k - 1) / 2]);
wkp1 = d21 * (d22 * ap[j + k * ((*n << 1) - k - 1) /
2] - ap[j + (k - 1) * ((*n << 1) - k) / 2]);
i__2 = *n;
for (i__ = j; i__ <= i__2; ++i__) {
ap[i__ + (j - 1) * ((*n << 1) - j) / 2] = ap[i__
+ (j - 1) * ((*n << 1) - j) / 2] - ap[i__
+ (k - 1) * ((*n << 1) - k) / 2] * wk -
ap[i__ + k * ((*n << 1) - k - 1) / 2] *
wkp1;
/* L90: */
}
ap[j + (k - 1) * ((*n << 1) - k) / 2] = wk;
ap[j + k * ((*n << 1) - k - 1) / 2] = wkp1;
/* L100: */
}
}
}
}
/* Store details of the interchanges in IPIV */
if (kstep == 1) {
ipiv[k] = kp;
} else {
ipiv[k] = -kp;
ipiv[k + 1] = -kp;
}
/* Increase K and return to the start of the main loop */
k += kstep;
kc = knc + *n - k + 2;
goto L60;
}
L110:
return 0;
/* End of DSPTRF */
} /* dsptrf_ */
/* Subroutine */ int zlatrs_(char *uplo, char *trans, char *diag, char *
normin, integer *n, doublecomplex *a, integer *lda, doublecomplex *x,
doublereal *scale, doublereal *cnorm, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, i__1, i__2, i__3, i__4, i__5;
doublereal d__1, d__2, d__3, d__4;
doublecomplex z__1, z__2, z__3, z__4;
/* Local variables */
integer i__, j;
doublereal xj, rec, tjj;
integer jinc;
doublereal xbnd;
integer imax;
doublereal tmax;
doublecomplex tjjs;
doublereal xmax, grow;
doublereal tscal;
doublecomplex uscal;
integer jlast;
doublecomplex csumj;
logical upper;
doublereal bignum;
logical notran;
integer jfirst;
doublereal smlnum;
logical nounit;
/* -- LAPACK auxiliary routine (version 3.2) -- */
/* November 2006 */
/* Purpose */
/* ======= */
/* ZLATRS solves one of the triangular systems */
/* A * x = s*b, A**T * x = s*b, or A**H * x = s*b, */
/* with scaling to prevent overflow. Here A is an upper or lower */
/* triangular matrix, A**T denotes the transpose of A, A**H denotes the */
/* conjugate transpose of A, x and b are n-element vectors, and s is a */
/* scaling factor, usually less than or equal to 1, chosen so that the */
/* components of x will be less than the overflow threshold. If the */
/* unscaled problem will not cause overflow, the Level 2 BLAS routine */
/* ZTRSV is called. If the matrix A is singular (A(j,j) = 0 for some j), */
/* then s is set to 0 and a non-trivial solution to A*x = 0 is returned. */
/* Arguments */
/* ========= */
/* UPLO (input) CHARACTER*1 */
/* Specifies whether the matrix A is upper or lower triangular. */
/* = 'U': Upper triangular */
/* = 'L': Lower triangular */
/* TRANS (input) CHARACTER*1 */
/* Specifies the operation applied to A. */
/* = 'N': Solve A * x = s*b (No transpose) */
/* = 'T': Solve A**T * x = s*b (Transpose) */
/* = 'C': Solve A**H * x = s*b (Conjugate transpose) */
/* DIAG (input) CHARACTER*1 */
/* Specifies whether or not the matrix A is unit triangular. */
/* = 'N': Non-unit triangular */
/* = 'U': Unit triangular */
/* NORMIN (input) CHARACTER*1 */
/* Specifies whether CNORM has been set or not. */
/* = 'Y': CNORM contains the column norms on entry */
/* = 'N': CNORM is not set on entry. On exit, the norms will */
/* be computed and stored in CNORM. */
/* N (input) INTEGER */
/* The order of the matrix A. N >= 0. */
/* A (input) COMPLEX*16 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). */
/* X (input/output) COMPLEX*16 array, dimension (N) */
/* On entry, the right hand side b of the triangular system. */
/* On exit, X is overwritten by the solution vector x. */
/* SCALE (output) DOUBLE PRECISION */
/* The scaling factor s for the triangular system */
/* A * x = s*b, A**T * x = s*b, or A**H * x = s*b. */
/* If SCALE = 0, the matrix A is singular or badly scaled, and */
/* the vector x is an exact or approximate solution to A*x = 0. */
/* CNORM (input or output) DOUBLE PRECISION array, dimension (N) */
/* If NORMIN = 'Y', CNORM is an input argument and CNORM(j) */
/* contains the norm of the off-diagonal part of the j-th column */
/* of A. If TRANS = 'N', CNORM(j) must be greater than or equal */
/* to the infinity-norm, and if TRANS = 'T' or 'C', CNORM(j) */
/* must be greater than or equal to the 1-norm. */
/* If NORMIN = 'N', CNORM is an output argument and CNORM(j) */
/* returns the 1-norm of the offdiagonal part of the j-th column */
/* of A. */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -k, the k-th argument had an illegal value */
/* Further Details */
/* ======= ======= */
/* A rough bound on x is computed; if that is less than overflow, ZTRSV */
/* is called, otherwise, specific code is used which checks for possible */
/* overflow or divide-by-zero at every operation. */
/* A columnwise scheme is used for solving A*x = b. The basic algorithm */
/* if A is lower triangular is */
/* x[1:n] := b[1:n] */
/* x(j) := x(j) / A(j,j) */
/* x[j+1:n] := x[j+1:n] - x(j) * A[j+1:n,j] */
/* end */
/* Define bounds on the components of x after j iterations of the loop: */
/* M(j) = bound on x[1:j] */
/* G(j) = bound on x[j+1:n] */
/* Then for iteration j+1 we have */
/* M(j+1) <= G(j) / | A(j+1,j+1) | */
/* G(j+1) <= G(j) + M(j+1) * | A[j+2:n,j+1] | */
/* <= G(j) ( 1 + CNORM(j+1) / | A(j+1,j+1) | ) */
/* where CNORM(j+1) is greater than or equal to the infinity-norm of */
/* column j+1 of A, not counting the diagonal. Hence */
/* G(j) <= G(0) product ( 1 + CNORM(i) / | A(i,i) | ) */
/* 1<=i<=j */
/* and */
/* |x(j)| <= ( G(0) / |A(j,j)| ) product ( 1 + CNORM(i) / |A(i,i)| ) */
/* 1<=i< j */
/* Since |x(j)| <= M(j), we use the Level 2 BLAS routine ZTRSV if the */
/* max(underflow, 1/overflow). */
/* The bound on x(j) is also used to determine when a step in the */
/* columnwise method can be performed without fear of overflow. If */
/* the computed bound is greater than a large constant, x is scaled to */
/* prevent overflow, but if the bound overflows, x is set to 0, x(j) to */
/* 1, and scale to 0, and a non-trivial solution to A*x = 0 is found. */
/* Similarly, a row-wise scheme is used to solve A**T *x = b or */
/* A**H *x = b. The basic algorithm for A upper triangular is */
/* x(j) := ( b(j) - A[1:j-1,j]' * x[1:j-1] ) / A(j,j) */
/* end */
/* We simultaneously compute two bounds */
/* G(j) = bound on ( b(i) - A[1:i-1,i]' * x[1:i-1] ), 1<=i<=j */
/* M(j) = bound on x(i), 1<=i<=j */
/* add the constraint G(j) >= G(j-1) and M(j) >= M(j-1) for j >= 1. */
/* Then the bound on x(j) is */
/* M(j) <= M(j-1) * ( 1 + CNORM(j) ) / | A(j,j) | */
/* <= M(0) * product ( ( 1 + CNORM(i) ) / |A(i,i)| ) */
/* 1<=i<=j */
/* and we can safely call ZTRSV if 1/M(n) and 1/G(n) are both greater */
/* than max(underflow, 1/overflow). */
/* ===================================================================== */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
--x;
--cnorm;
/* Function Body */
*info = 0;
upper = lsame_(uplo, "U");
notran = lsame_(trans, "N");
nounit = lsame_(diag, "N");
/* Test the input parameters. */
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 (! lsame_(normin, "Y") && ! lsame_(normin,
"N")) {
*info = -4;
} else if (*n < 0) {
*info = -5;
} else if (*lda < max(1,*n)) {
*info = -7;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("ZLATRS", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
/* Determine machine dependent parameters to control overflow. */
smlnum = dlamch_("Safe minimum");
bignum = 1. / smlnum;
dlabad_(&smlnum, &bignum);
smlnum /= dlamch_("Precision");
bignum = 1. / smlnum;
*scale = 1.;
if (lsame_(normin, "N")) {
/* Compute the 1-norm of each column, not including the diagonal. */
if (upper) {
/* A is upper triangular. */
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
i__2 = j - 1;
cnorm[j] = dzasum_(&i__2, &a[j * a_dim1 + 1], &c__1);
}
} else {
/* A is lower triangular. */
i__1 = *n - 1;
for (j = 1; j <= i__1; ++j) {
i__2 = *n - j;
cnorm[j] = dzasum_(&i__2, &a[j + 1 + j * a_dim1], &c__1);
}
cnorm[*n] = 0.;
}
}
/* Scale the column norms by TSCAL if the maximum element in CNORM is */
/* greater than BIGNUM/2. */
imax = idamax_(n, &cnorm[1], &c__1);
tmax = cnorm[imax];
if (tmax <= bignum * .5) {
tscal = 1.;
} else {
tscal = .5 / (smlnum * tmax);
dscal_(n, &tscal, &cnorm[1], &c__1);
}
/* Compute a bound on the computed solution vector to see if the */
/* Level 2 BLAS routine ZTRSV can be used. */
xmax = 0.;
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
/* Computing MAX */
i__2 = j;
d__3 = xmax, d__4 = (d__1 = x[i__2].r / 2., abs(d__1)) + (d__2 =
d_imag(&x[j]) / 2., abs(d__2));
xmax = max(d__3,d__4);
}
xbnd = xmax;
if (notran) {
/* Compute the growth in A * x = b. */
if (upper) {
jfirst = *n;
jlast = 1;
jinc = -1;
} else {
jfirst = 1;
jlast = *n;
jinc = 1;
}
if (tscal != 1.) {
grow = 0.;
goto L60;
}
if (nounit) {
/* A is non-unit triangular. */
/* Compute GROW = 1/G(j) and XBND = 1/M(j). */
grow = .5 / max(xbnd,smlnum);
xbnd = grow;
i__1 = jlast;
i__2 = jinc;
for (j = jfirst; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
/* Exit the loop if the growth factor is too small. */
if (grow <= smlnum) {
goto L60;
}
i__3 = j + j * a_dim1;
tjjs.r = a[i__3].r, tjjs.i = a[i__3].i;
tjj = (d__1 = tjjs.r, abs(d__1)) + (d__2 = d_imag(&tjjs), abs(
d__2));
if (tjj >= smlnum) {
/* M(j) = G(j-1) / abs(A(j,j)) */
/* Computing MIN */
d__1 = xbnd, d__2 = min(1.,tjj) * grow;
xbnd = min(d__1,d__2);
} else {
/* M(j) could overflow, set XBND to 0. */
xbnd = 0.;
}
if (tjj + cnorm[j] >= smlnum) {
/* G(j) = G(j-1)*( 1 + CNORM(j) / abs(A(j,j)) ) */
grow *= tjj / (tjj + cnorm[j]);
} else {
/* G(j) could overflow, set GROW to 0. */
grow = 0.;
}
}
grow = xbnd;
} else {
/* A is unit triangular. */
/* Computing MIN */
d__1 = 1., d__2 = .5 / max(xbnd,smlnum);
grow = min(d__1,d__2);
i__2 = jlast;
i__1 = jinc;
for (j = jfirst; i__1 < 0 ? j >= i__2 : j <= i__2; j += i__1) {
/* Exit the loop if the growth factor is too small. */
if (grow <= smlnum) {
goto L60;
}
/* G(j) = G(j-1)*( 1 + CNORM(j) ) */
grow *= 1. / (cnorm[j] + 1.);
}
}
L60:
;
} else {
/* Compute the growth in A**T * x = b or A**H * x = b. */
if (upper) {
jfirst = 1;
jlast = *n;
jinc = 1;
} else {
jfirst = *n;
jlast = 1;
jinc = -1;
}
if (tscal != 1.) {
grow = 0.;
goto L90;
}
if (nounit) {
/* A is non-unit triangular. */
/* Compute GROW = 1/G(j) and XBND = 1/M(j). */
grow = .5 / max(xbnd,smlnum);
xbnd = grow;
i__1 = jlast;
i__2 = jinc;
for (j = jfirst; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
/* Exit the loop if the growth factor is too small. */
if (grow <= smlnum) {
goto L90;
}
/* G(j) = max( G(j-1), M(j-1)*( 1 + CNORM(j) ) ) */
xj = cnorm[j] + 1.;
/* Computing MIN */
d__1 = grow, d__2 = xbnd / xj;
grow = min(d__1,d__2);
i__3 = j + j * a_dim1;
tjjs.r = a[i__3].r, tjjs.i = a[i__3].i;
tjj = (d__1 = tjjs.r, abs(d__1)) + (d__2 = d_imag(&tjjs), abs(
d__2));
if (tjj >= smlnum) {
/* M(j) = M(j-1)*( 1 + CNORM(j) ) / abs(A(j,j)) */
if (xj > tjj) {
xbnd *= tjj / xj;
}
} else {
/* M(j) could overflow, set XBND to 0. */
xbnd = 0.;
}
}
grow = min(grow,xbnd);
} else {
/* A is unit triangular. */
/* Computing MIN */
d__1 = 1., d__2 = .5 / max(xbnd,smlnum);
grow = min(d__1,d__2);
i__2 = jlast;
i__1 = jinc;
for (j = jfirst; i__1 < 0 ? j >= i__2 : j <= i__2; j += i__1) {
/* Exit the loop if the growth factor is too small. */
if (grow <= smlnum) {
goto L90;
}
/* G(j) = ( 1 + CNORM(j) )*G(j-1) */
xj = cnorm[j] + 1.;
grow /= xj;
}
}
L90:
;
}
if (grow * tscal > smlnum) {
/* Use the Level 2 BLAS solve if the reciprocal of the bound on */
/* elements of X is not too small. */
ztrsv_(uplo, trans, diag, n, &a[a_offset], lda, &x[1], &c__1);
} else {
/* Use a Level 1 BLAS solve, scaling intermediate results. */
if (xmax > bignum * .5) {
/* Scale X so that its components are less than or equal to */
/* BIGNUM in absolute value. */
*scale = bignum * .5 / xmax;
zdscal_(n, scale, &x[1], &c__1);
xmax = bignum;
} else {
xmax *= 2.;
}
if (notran) {
/* Solve A * x = b */
i__1 = jlast;
i__2 = jinc;
for (j = jfirst; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
/* Compute x(j) = b(j) / A(j,j), scaling x if necessary. */
i__3 = j;
xj = (d__1 = x[i__3].r, abs(d__1)) + (d__2 = d_imag(&x[j]),
abs(d__2));
if (nounit) {
i__3 = j + j * a_dim1;
z__1.r = tscal * a[i__3].r, z__1.i = tscal * a[i__3].i;
tjjs.r = z__1.r, tjjs.i = z__1.i;
} else {
tjjs.r = tscal, tjjs.i = 0.;
if (tscal == 1.) {
goto L110;
}
}
tjj = (d__1 = tjjs.r, abs(d__1)) + (d__2 = d_imag(&tjjs), abs(
d__2));
if (tjj > smlnum) {
/* abs(A(j,j)) > SMLNUM: */
if (tjj < 1.) {
if (xj > tjj * bignum) {
/* Scale x by 1/b(j). */
rec = 1. / xj;
zdscal_(n, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
}
i__3 = j;
zladiv_(&z__1, &x[j], &tjjs);
x[i__3].r = z__1.r, x[i__3].i = z__1.i;
i__3 = j;
xj = (d__1 = x[i__3].r, abs(d__1)) + (d__2 = d_imag(&x[j])
, abs(d__2));
} else if (tjj > 0.) {
/* 0 < abs(A(j,j)) <= SMLNUM: */
if (xj > tjj * bignum) {
/* Scale x by (1/abs(x(j)))*abs(A(j,j))*BIGNUM */
/* to avoid overflow when dividing by A(j,j). */
rec = tjj * bignum / xj;
if (cnorm[j] > 1.) {
/* Scale by 1/CNORM(j) to avoid overflow when */
/* multiplying x(j) times column j. */
rec /= cnorm[j];
}
zdscal_(n, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
i__3 = j;
zladiv_(&z__1, &x[j], &tjjs);
x[i__3].r = z__1.r, x[i__3].i = z__1.i;
i__3 = j;
xj = (d__1 = x[i__3].r, abs(d__1)) + (d__2 = d_imag(&x[j])
, abs(d__2));
} else {
/* A(j,j) = 0: Set x(1:n) = 0, x(j) = 1, and */
/* scale = 0, and compute a solution to A*x = 0. */
i__3 = *n;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = i__;
x[i__4].r = 0., x[i__4].i = 0.;
}
i__3 = j;
x[i__3].r = 1., x[i__3].i = 0.;
xj = 1.;
*scale = 0.;
xmax = 0.;
}
L110:
/* Scale x if necessary to avoid overflow when adding a */
/* multiple of column j of A. */
if (xj > 1.) {
rec = 1. / xj;
if (cnorm[j] > (bignum - xmax) * rec) {
/* Scale x by 1/(2*abs(x(j))). */
rec *= .5;
zdscal_(n, &rec, &x[1], &c__1);
*scale *= rec;
}
} else if (xj * cnorm[j] > bignum - xmax) {
/* Scale x by 1/2. */
zdscal_(n, &c_b36, &x[1], &c__1);
*scale *= .5;
}
if (upper) {
if (j > 1) {
/* Compute the update */
/* x(1:j-1) := x(1:j-1) - x(j) * A(1:j-1,j) */
i__3 = j - 1;
i__4 = j;
z__2.r = -x[i__4].r, z__2.i = -x[i__4].i;
z__1.r = tscal * z__2.r, z__1.i = tscal * z__2.i;
zaxpy_(&i__3, &z__1, &a[j * a_dim1 + 1], &c__1, &x[1],
&c__1);
i__3 = j - 1;
i__ = izamax_(&i__3, &x[1], &c__1);
i__3 = i__;
xmax = (d__1 = x[i__3].r, abs(d__1)) + (d__2 = d_imag(
&x[i__]), abs(d__2));
}
} else {
if (j < *n) {
/* Compute the update */
/* x(j+1:n) := x(j+1:n) - x(j) * A(j+1:n,j) */
i__3 = *n - j;
i__4 = j;
z__2.r = -x[i__4].r, z__2.i = -x[i__4].i;
z__1.r = tscal * z__2.r, z__1.i = tscal * z__2.i;
zaxpy_(&i__3, &z__1, &a[j + 1 + j * a_dim1], &c__1, &
x[j + 1], &c__1);
i__3 = *n - j;
i__ = j + izamax_(&i__3, &x[j + 1], &c__1);
i__3 = i__;
xmax = (d__1 = x[i__3].r, abs(d__1)) + (d__2 = d_imag(
&x[i__]), abs(d__2));
}
}
}
} else if (lsame_(trans, "T")) {
/* Solve A**T * x = b */
i__2 = jlast;
i__1 = jinc;
for (j = jfirst; i__1 < 0 ? j >= i__2 : j <= i__2; j += i__1) {
/* Compute x(j) = b(j) - sum A(k,j)*x(k). */
/* k<>j */
i__3 = j;
xj = (d__1 = x[i__3].r, abs(d__1)) + (d__2 = d_imag(&x[j]),
abs(d__2));
uscal.r = tscal, uscal.i = 0.;
rec = 1. / max(xmax,1.);
if (cnorm[j] > (bignum - xj) * rec) {
/* If x(j) could overflow, scale x by 1/(2*XMAX). */
rec *= .5;
if (nounit) {
i__3 = j + j * a_dim1;
z__1.r = tscal * a[i__3].r, z__1.i = tscal * a[i__3]
.i;
tjjs.r = z__1.r, tjjs.i = z__1.i;
} else {
tjjs.r = tscal, tjjs.i = 0.;
}
tjj = (d__1 = tjjs.r, abs(d__1)) + (d__2 = d_imag(&tjjs),
abs(d__2));
if (tjj > 1.) {
/* Divide by A(j,j) when scaling x if A(j,j) > 1. */
/* Computing MIN */
d__1 = 1., d__2 = rec * tjj;
rec = min(d__1,d__2);
zladiv_(&z__1, &uscal, &tjjs);
uscal.r = z__1.r, uscal.i = z__1.i;
}
if (rec < 1.) {
zdscal_(n, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
}
csumj.r = 0., csumj.i = 0.;
if (uscal.r == 1. && uscal.i == 0.) {
/* If the scaling needed for A in the dot product is 1, */
/* call ZDOTU to perform the dot product. */
if (upper) {
i__3 = j - 1;
zdotu_(&z__1, &i__3, &a[j * a_dim1 + 1], &c__1, &x[1],
&c__1);
csumj.r = z__1.r, csumj.i = z__1.i;
} else if (j < *n) {
i__3 = *n - j;
zdotu_(&z__1, &i__3, &a[j + 1 + j * a_dim1], &c__1, &
x[j + 1], &c__1);
csumj.r = z__1.r, csumj.i = z__1.i;
}
} else {
/* Otherwise, use in-line code for the dot product. */
if (upper) {
i__3 = j - 1;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = i__ + j * a_dim1;
z__3.r = a[i__4].r * uscal.r - a[i__4].i *
uscal.i, z__3.i = a[i__4].r * uscal.i + a[
i__4].i * uscal.r;
i__5 = i__;
z__2.r = z__3.r * x[i__5].r - z__3.i * x[i__5].i,
z__2.i = z__3.r * x[i__5].i + z__3.i * x[
i__5].r;
z__1.r = csumj.r + z__2.r, z__1.i = csumj.i +
z__2.i;
csumj.r = z__1.r, csumj.i = z__1.i;
}
} else if (j < *n) {
i__3 = *n;
for (i__ = j + 1; i__ <= i__3; ++i__) {
i__4 = i__ + j * a_dim1;
z__3.r = a[i__4].r * uscal.r - a[i__4].i *
uscal.i, z__3.i = a[i__4].r * uscal.i + a[
i__4].i * uscal.r;
i__5 = i__;
z__2.r = z__3.r * x[i__5].r - z__3.i * x[i__5].i,
z__2.i = z__3.r * x[i__5].i + z__3.i * x[
i__5].r;
z__1.r = csumj.r + z__2.r, z__1.i = csumj.i +
z__2.i;
csumj.r = z__1.r, csumj.i = z__1.i;
}
}
}
z__1.r = tscal, z__1.i = 0.;
if (uscal.r == z__1.r && uscal.i == z__1.i) {
/* Compute x(j) := ( x(j) - CSUMJ ) / A(j,j) if 1/A(j,j) */
/* was not used to scale the dotproduct. */
i__3 = j;
i__4 = j;
z__1.r = x[i__4].r - csumj.r, z__1.i = x[i__4].i -
csumj.i;
x[i__3].r = z__1.r, x[i__3].i = z__1.i;
i__3 = j;
xj = (d__1 = x[i__3].r, abs(d__1)) + (d__2 = d_imag(&x[j])
, abs(d__2));
if (nounit) {
i__3 = j + j * a_dim1;
z__1.r = tscal * a[i__3].r, z__1.i = tscal * a[i__3]
.i;
tjjs.r = z__1.r, tjjs.i = z__1.i;
} else {
tjjs.r = tscal, tjjs.i = 0.;
if (tscal == 1.) {
goto L160;
}
}
/* Compute x(j) = x(j) / A(j,j), scaling if necessary. */
tjj = (d__1 = tjjs.r, abs(d__1)) + (d__2 = d_imag(&tjjs),
abs(d__2));
if (tjj > smlnum) {
/* abs(A(j,j)) > SMLNUM: */
if (tjj < 1.) {
if (xj > tjj * bignum) {
/* Scale X by 1/abs(x(j)). */
rec = 1. / xj;
zdscal_(n, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
}
i__3 = j;
zladiv_(&z__1, &x[j], &tjjs);
x[i__3].r = z__1.r, x[i__3].i = z__1.i;
} else if (tjj > 0.) {
/* 0 < abs(A(j,j)) <= SMLNUM: */
if (xj > tjj * bignum) {
/* Scale x by (1/abs(x(j)))*abs(A(j,j))*BIGNUM. */
rec = tjj * bignum / xj;
zdscal_(n, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
i__3 = j;
zladiv_(&z__1, &x[j], &tjjs);
x[i__3].r = z__1.r, x[i__3].i = z__1.i;
} else {
/* A(j,j) = 0: Set x(1:n) = 0, x(j) = 1, and */
/* scale = 0 and compute a solution to A**T *x = 0. */
i__3 = *n;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = i__;
x[i__4].r = 0., x[i__4].i = 0.;
}
i__3 = j;
x[i__3].r = 1., x[i__3].i = 0.;
*scale = 0.;
xmax = 0.;
}
L160:
;
} else {
/* Compute x(j) := x(j) / A(j,j) - CSUMJ if the dot */
/* product has already been divided by 1/A(j,j). */
i__3 = j;
zladiv_(&z__2, &x[j], &tjjs);
z__1.r = z__2.r - csumj.r, z__1.i = z__2.i - csumj.i;
x[i__3].r = z__1.r, x[i__3].i = z__1.i;
}
/* Computing MAX */
i__3 = j;
d__3 = xmax, d__4 = (d__1 = x[i__3].r, abs(d__1)) + (d__2 =
d_imag(&x[j]), abs(d__2));
xmax = max(d__3,d__4);
}
} else {
/* Solve A**H * x = b */
i__1 = jlast;
i__2 = jinc;
for (j = jfirst; i__2 < 0 ? j >= i__1 : j <= i__1; j += i__2) {
/* Compute x(j) = b(j) - sum A(k,j)*x(k). */
/* k<>j */
i__3 = j;
xj = (d__1 = x[i__3].r, abs(d__1)) + (d__2 = d_imag(&x[j]),
abs(d__2));
uscal.r = tscal, uscal.i = 0.;
rec = 1. / max(xmax,1.);
if (cnorm[j] > (bignum - xj) * rec) {
/* If x(j) could overflow, scale x by 1/(2*XMAX). */
rec *= .5;
if (nounit) {
d_cnjg(&z__2, &a[j + j * a_dim1]);
z__1.r = tscal * z__2.r, z__1.i = tscal * z__2.i;
tjjs.r = z__1.r, tjjs.i = z__1.i;
} else {
tjjs.r = tscal, tjjs.i = 0.;
}
tjj = (d__1 = tjjs.r, abs(d__1)) + (d__2 = d_imag(&tjjs),
abs(d__2));
if (tjj > 1.) {
/* Divide by A(j,j) when scaling x if A(j,j) > 1. */
/* Computing MIN */
d__1 = 1., d__2 = rec * tjj;
rec = min(d__1,d__2);
zladiv_(&z__1, &uscal, &tjjs);
uscal.r = z__1.r, uscal.i = z__1.i;
}
if (rec < 1.) {
zdscal_(n, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
}
csumj.r = 0., csumj.i = 0.;
if (uscal.r == 1. && uscal.i == 0.) {
/* If the scaling needed for A in the dot product is 1, */
/* call ZDOTC to perform the dot product. */
if (upper) {
i__3 = j - 1;
zdotc_(&z__1, &i__3, &a[j * a_dim1 + 1], &c__1, &x[1],
&c__1);
csumj.r = z__1.r, csumj.i = z__1.i;
} else if (j < *n) {
i__3 = *n - j;
zdotc_(&z__1, &i__3, &a[j + 1 + j * a_dim1], &c__1, &
x[j + 1], &c__1);
csumj.r = z__1.r, csumj.i = z__1.i;
}
} else {
/* Otherwise, use in-line code for the dot product. */
if (upper) {
i__3 = j - 1;
for (i__ = 1; i__ <= i__3; ++i__) {
d_cnjg(&z__4, &a[i__ + j * a_dim1]);
z__3.r = z__4.r * uscal.r - z__4.i * uscal.i,
z__3.i = z__4.r * uscal.i + z__4.i *
uscal.r;
i__4 = i__;
z__2.r = z__3.r * x[i__4].r - z__3.i * x[i__4].i,
z__2.i = z__3.r * x[i__4].i + z__3.i * x[
i__4].r;
z__1.r = csumj.r + z__2.r, z__1.i = csumj.i +
z__2.i;
csumj.r = z__1.r, csumj.i = z__1.i;
}
} else if (j < *n) {
i__3 = *n;
for (i__ = j + 1; i__ <= i__3; ++i__) {
d_cnjg(&z__4, &a[i__ + j * a_dim1]);
z__3.r = z__4.r * uscal.r - z__4.i * uscal.i,
z__3.i = z__4.r * uscal.i + z__4.i *
uscal.r;
i__4 = i__;
z__2.r = z__3.r * x[i__4].r - z__3.i * x[i__4].i,
z__2.i = z__3.r * x[i__4].i + z__3.i * x[
i__4].r;
z__1.r = csumj.r + z__2.r, z__1.i = csumj.i +
z__2.i;
csumj.r = z__1.r, csumj.i = z__1.i;
}
}
}
z__1.r = tscal, z__1.i = 0.;
if (uscal.r == z__1.r && uscal.i == z__1.i) {
/* Compute x(j) := ( x(j) - CSUMJ ) / A(j,j) if 1/A(j,j) */
/* was not used to scale the dotproduct. */
i__3 = j;
i__4 = j;
z__1.r = x[i__4].r - csumj.r, z__1.i = x[i__4].i -
csumj.i;
x[i__3].r = z__1.r, x[i__3].i = z__1.i;
i__3 = j;
xj = (d__1 = x[i__3].r, abs(d__1)) + (d__2 = d_imag(&x[j])
, abs(d__2));
if (nounit) {
d_cnjg(&z__2, &a[j + j * a_dim1]);
z__1.r = tscal * z__2.r, z__1.i = tscal * z__2.i;
tjjs.r = z__1.r, tjjs.i = z__1.i;
} else {
tjjs.r = tscal, tjjs.i = 0.;
if (tscal == 1.) {
goto L210;
}
}
/* Compute x(j) = x(j) / A(j,j), scaling if necessary. */
tjj = (d__1 = tjjs.r, abs(d__1)) + (d__2 = d_imag(&tjjs),
abs(d__2));
if (tjj > smlnum) {
/* abs(A(j,j)) > SMLNUM: */
if (tjj < 1.) {
if (xj > tjj * bignum) {
/* Scale X by 1/abs(x(j)). */
rec = 1. / xj;
zdscal_(n, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
}
i__3 = j;
zladiv_(&z__1, &x[j], &tjjs);
x[i__3].r = z__1.r, x[i__3].i = z__1.i;
} else if (tjj > 0.) {
/* 0 < abs(A(j,j)) <= SMLNUM: */
if (xj > tjj * bignum) {
/* Scale x by (1/abs(x(j)))*abs(A(j,j))*BIGNUM. */
rec = tjj * bignum / xj;
zdscal_(n, &rec, &x[1], &c__1);
*scale *= rec;
xmax *= rec;
}
i__3 = j;
zladiv_(&z__1, &x[j], &tjjs);
x[i__3].r = z__1.r, x[i__3].i = z__1.i;
} else {
/* A(j,j) = 0: Set x(1:n) = 0, x(j) = 1, and */
/* scale = 0 and compute a solution to A**H *x = 0. */
i__3 = *n;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = i__;
x[i__4].r = 0., x[i__4].i = 0.;
}
i__3 = j;
x[i__3].r = 1., x[i__3].i = 0.;
*scale = 0.;
xmax = 0.;
}
L210:
;
} else {
/* Compute x(j) := x(j) / A(j,j) - CSUMJ if the dot */
/* product has already been divided by 1/A(j,j). */
i__3 = j;
zladiv_(&z__2, &x[j], &tjjs);
z__1.r = z__2.r - csumj.r, z__1.i = z__2.i - csumj.i;
x[i__3].r = z__1.r, x[i__3].i = z__1.i;
}
/* Computing MAX */
i__3 = j;
d__3 = xmax, d__4 = (d__1 = x[i__3].r, abs(d__1)) + (d__2 =
d_imag(&x[j]), abs(d__2));
xmax = max(d__3,d__4);
}
}
*scale /= tscal;
}
/* Scale the column norms by 1/TSCAL for return. */
if (tscal != 1.) {
d__1 = 1. / tscal;
dscal_(n, &d__1, &cnorm[1], &c__1);
}
return 0;
/* End of ZLATRS */
} /* zlatrs_ */
/* Subroutine */ int dlaed2_(integer *k, integer *n, integer *n1, doublereal *
d__, doublereal *q, integer *ldq, integer *indxq, doublereal *rho,
doublereal *z__, doublereal *dlamda, doublereal *w, doublereal *q2,
integer *indx, integer *indxc, integer *indxp, integer *coltyp,
integer *info)
{
/* System generated locals */
integer q_dim1, q_offset, i__1, i__2;
doublereal d__1, d__2, d__3, d__4;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
doublereal c__;
integer i__, j;
doublereal s, t;
integer k2, n2, ct, nj, pj, js, iq1, iq2, n1p1;
doublereal eps, tau, tol;
integer psm[4], imax, jmax;
extern /* Subroutine */ int drot_(integer *, doublereal *, integer *,
doublereal *, integer *, doublereal *, doublereal *);
integer ctot[4];
extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
integer *), dcopy_(integer *, doublereal *, integer *, doublereal
*, integer *);
extern doublereal dlapy2_(doublereal *, doublereal *), dlamch_(char *);
extern integer idamax_(integer *, doublereal *, integer *);
extern /* Subroutine */ int dlamrg_(integer *, integer *, doublereal *,
integer *, integer *, integer *), dlacpy_(char *, integer *,
integer *, doublereal *, integer *, doublereal *, integer *), xerbla_(char *, integer *);
/* -- LAPACK routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* DLAED2 merges the two sets of eigenvalues together into a single */
/* sorted set. Then it tries to deflate the size of the problem. */
/* There are two ways in which deflation can occur: when two or more */
/* eigenvalues are close together or if there is a tiny entry in the */
/* Z vector. For each such occurrence the order of the related secular */
/* equation problem is reduced by one. */
/* Arguments */
/* ========= */
/* K (output) INTEGER */
/* The number of non-deflated eigenvalues, and the order of the */
/* related secular equation. 0 <= K <=N. */
/* N (input) INTEGER */
/* The dimension of the symmetric tridiagonal matrix. N >= 0. */
/* N1 (input) INTEGER */
/* The location of the last eigenvalue in the leading sub-matrix. */
/* min(1,N) <= N1 <= N/2. */
/* D (input/output) DOUBLE PRECISION array, dimension (N) */
/* On entry, D contains the eigenvalues of the two submatrices to */
/* be combined. */
/* On exit, D contains the trailing (N-K) updated eigenvalues */
/* (those which were deflated) sorted into increasing order. */
/* Q (input/output) DOUBLE PRECISION array, dimension (LDQ, N) */
/* On entry, Q contains the eigenvectors of two submatrices in */
/* the two square blocks with corners at (1,1), (N1,N1) */
/* and (N1+1, N1+1), (N,N). */
/* On exit, Q contains the trailing (N-K) updated eigenvectors */
/* (those which were deflated) in its last N-K columns. */
/* LDQ (input) INTEGER */
/* The leading dimension of the array Q. LDQ >= max(1,N). */
/* INDXQ (input/output) INTEGER array, dimension (N) */
/* The permutation which separately sorts the two sub-problems */
/* in D into ascending order. Note that elements in the second */
/* half of this permutation must first have N1 added to their */
/* values. Destroyed on exit. */
/* RHO (input/output) DOUBLE PRECISION */
/* On entry, the off-diagonal element associated with the rank-1 */
/* cut which originally split the two submatrices which are now */
/* being recombined. */
/* On exit, RHO has been modified to the value required by */
/* DLAED3. */
/* Z (input) DOUBLE PRECISION array, dimension (N) */
/* On entry, Z contains the updating vector (the last */
/* row of the first sub-eigenvector matrix and the first row of */
/* the second sub-eigenvector matrix). */
/* On exit, the contents of Z have been destroyed by the updating */
/* process. */
/* DLAMDA (output) DOUBLE PRECISION array, dimension (N) */
/* A copy of the first K eigenvalues which will be used by */
/* DLAED3 to form the secular equation. */
/* W (output) DOUBLE PRECISION array, dimension (N) */
/* The first k values of the final deflation-altered z-vector */
/* which will be passed to DLAED3. */
/* Q2 (output) DOUBLE PRECISION array, dimension (N1**2+(N-N1)**2) */
/* A copy of the first K eigenvectors which will be used by */
/* DLAED3 in a matrix multiply (DGEMM) to solve for the new */
/* eigenvectors. */
/* INDX (workspace) INTEGER array, dimension (N) */
/* The permutation used to sort the contents of DLAMDA into */
/* ascending order. */
/* INDXC (output) INTEGER array, dimension (N) */
/* The permutation used to arrange the columns of the deflated */
/* Q matrix into three groups: the first group contains non-zero */
/* elements only at and above N1, the second contains */
/* non-zero elements only below N1, and the third is dense. */
/* INDXP (workspace) INTEGER array, dimension (N) */
/* The permutation used to place deflated values of D at the end */
/* of the array. INDXP(1:K) points to the nondeflated D-values */
/* and INDXP(K+1:N) points to the deflated eigenvalues. */
/* COLTYP (workspace/output) INTEGER array, dimension (N) */
/* During execution, a label which will indicate which of the */
/* following types a column in the Q2 matrix is: */
/* 1 : non-zero in the upper half only; */
/* 2 : dense; */
/* 3 : non-zero in the lower half only; */
/* 4 : deflated. */
/* On exit, COLTYP(i) is the number of columns of type i, */
/* for i=1 to 4 only. */
/* INFO (output) INTEGER */
/* = 0: successful exit. */
/* < 0: if INFO = -i, the i-th argument had an illegal value. */
/* Further Details */
/* =============== */
/* Based on contributions by */
/* Jeff Rutter, Computer Science Division, University of California */
/* at Berkeley, USA */
/* Modified by Francoise Tisseur, University of Tennessee. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Arrays .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
--d__;
q_dim1 = *ldq;
q_offset = 1 + q_dim1;
q -= q_offset;
--indxq;
--z__;
--dlamda;
--w;
--q2;
--indx;
--indxc;
--indxp;
--coltyp;
/* Function Body */
*info = 0;
if (*n < 0) {
*info = -2;
} else if (*ldq < max(1,*n)) {
*info = -6;
} else /* if(complicated condition) */ {
/* Computing MIN */
i__1 = 1, i__2 = *n / 2;
if (min(i__1,i__2) > *n1 || *n / 2 < *n1) {
*info = -3;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("DLAED2", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
n2 = *n - *n1;
n1p1 = *n1 + 1;
if (*rho < 0.) {
dscal_(&n2, &c_b3, &z__[n1p1], &c__1);
}
/* Normalize z so that norm(z) = 1. Since z is the concatenation of */
/* two normalized vectors, norm2(z) = sqrt(2). */
t = 1. / sqrt(2.);
dscal_(n, &t, &z__[1], &c__1);
/* RHO = ABS( norm(z)**2 * RHO ) */
*rho = (d__1 = *rho * 2., abs(d__1));
/* Sort the eigenvalues into increasing order */
i__1 = *n;
for (i__ = n1p1; i__ <= i__1; ++i__) {
indxq[i__] += *n1;
/* L10: */
}
/* re-integrate the deflated parts from the last pass */
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
dlamda[i__] = d__[indxq[i__]];
/* L20: */
}
dlamrg_(n1, &n2, &dlamda[1], &c__1, &c__1, &indxc[1]);
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
indx[i__] = indxq[indxc[i__]];
/* L30: */
}
/* Calculate the allowable deflation tolerance */
imax = idamax_(n, &z__[1], &c__1);
jmax = idamax_(n, &d__[1], &c__1);
eps = dlamch_("Epsilon");
/* Computing MAX */
d__3 = (d__1 = d__[jmax], abs(d__1)), d__4 = (d__2 = z__[imax], abs(d__2))
;
tol = eps * 8. * max(d__3,d__4);
/* If the rank-1 modifier is small enough, no more needs to be done */
/* except to reorganize Q so that its columns correspond with the */
/* elements in D. */
if (*rho * (d__1 = z__[imax], abs(d__1)) <= tol) {
*k = 0;
iq2 = 1;
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
i__ = indx[j];
dcopy_(n, &q[i__ * q_dim1 + 1], &c__1, &q2[iq2], &c__1);
dlamda[j] = d__[i__];
iq2 += *n;
/* L40: */
}
dlacpy_("A", n, n, &q2[1], n, &q[q_offset], ldq);
dcopy_(n, &dlamda[1], &c__1, &d__[1], &c__1);
goto L190;
}
/* If there are multiple eigenvalues then the problem deflates. Here */
/* the number of equal eigenvalues are found. As each equal */
/* eigenvalue is found, an elementary reflector is computed to rotate */
/* the corresponding eigensubspace so that the corresponding */
/* components of Z are zero in this new basis. */
i__1 = *n1;
for (i__ = 1; i__ <= i__1; ++i__) {
coltyp[i__] = 1;
/* L50: */
}
i__1 = *n;
for (i__ = n1p1; i__ <= i__1; ++i__) {
coltyp[i__] = 3;
/* L60: */
}
*k = 0;
k2 = *n + 1;
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
nj = indx[j];
if (*rho * (d__1 = z__[nj], abs(d__1)) <= tol) {
/* Deflate due to small z component. */
--k2;
coltyp[nj] = 4;
indxp[k2] = nj;
if (j == *n) {
goto L100;
}
} else {
pj = nj;
goto L80;
}
/* L70: */
}
L80:
++j;
nj = indx[j];
if (j > *n) {
goto L100;
}
if (*rho * (d__1 = z__[nj], abs(d__1)) <= tol) {
/* Deflate due to small z component. */
--k2;
coltyp[nj] = 4;
indxp[k2] = nj;
} else {
/* Check if eigenvalues are close enough to allow deflation. */
s = z__[pj];
c__ = z__[nj];
/* Find sqrt(a**2+b**2) without overflow or */
/* destructive underflow. */
tau = dlapy2_(&c__, &s);
t = d__[nj] - d__[pj];
c__ /= tau;
s = -s / tau;
if ((d__1 = t * c__ * s, abs(d__1)) <= tol) {
/* Deflation is possible. */
z__[nj] = tau;
z__[pj] = 0.;
if (coltyp[nj] != coltyp[pj]) {
coltyp[nj] = 2;
}
coltyp[pj] = 4;
drot_(n, &q[pj * q_dim1 + 1], &c__1, &q[nj * q_dim1 + 1], &c__1, &
c__, &s);
/* Computing 2nd power */
d__1 = c__;
/* Computing 2nd power */
d__2 = s;
t = d__[pj] * (d__1 * d__1) + d__[nj] * (d__2 * d__2);
/* Computing 2nd power */
d__1 = s;
/* Computing 2nd power */
d__2 = c__;
d__[nj] = d__[pj] * (d__1 * d__1) + d__[nj] * (d__2 * d__2);
d__[pj] = t;
--k2;
i__ = 1;
L90:
if (k2 + i__ <= *n) {
if (d__[pj] < d__[indxp[k2 + i__]]) {
indxp[k2 + i__ - 1] = indxp[k2 + i__];
indxp[k2 + i__] = pj;
++i__;
goto L90;
} else {
indxp[k2 + i__ - 1] = pj;
}
} else {
indxp[k2 + i__ - 1] = pj;
}
pj = nj;
} else {
++(*k);
dlamda[*k] = d__[pj];
w[*k] = z__[pj];
indxp[*k] = pj;
pj = nj;
}
}
goto L80;
L100:
/* Record the last eigenvalue. */
++(*k);
dlamda[*k] = d__[pj];
w[*k] = z__[pj];
indxp[*k] = pj;
/* Count up the total number of the various types of columns, then */
/* form a permutation which positions the four column types into */
/* four uniform groups (although one or more of these groups may be */
/* empty). */
for (j = 1; j <= 4; ++j) {
ctot[j - 1] = 0;
/* L110: */
}
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
ct = coltyp[j];
++ctot[ct - 1];
/* L120: */
}
/* PSM(*) = Position in SubMatrix (of types 1 through 4) */
psm[0] = 1;
psm[1] = ctot[0] + 1;
psm[2] = psm[1] + ctot[1];
psm[3] = psm[2] + ctot[2];
*k = *n - ctot[3];
/* Fill out the INDXC array so that the permutation which it induces */
/* will place all type-1 columns first, all type-2 columns next, */
/* then all type-3's, and finally all type-4's. */
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
js = indxp[j];
ct = coltyp[js];
indx[psm[ct - 1]] = js;
indxc[psm[ct - 1]] = j;
++psm[ct - 1];
/* L130: */
}
/* Sort the eigenvalues and corresponding eigenvectors into DLAMDA */
/* and Q2 respectively. The eigenvalues/vectors which were not */
/* deflated go into the first K slots of DLAMDA and Q2 respectively, */
/* while those which were deflated go into the last N - K slots. */
i__ = 1;
iq1 = 1;
iq2 = (ctot[0] + ctot[1]) * *n1 + 1;
i__1 = ctot[0];
for (j = 1; j <= i__1; ++j) {
js = indx[i__];
dcopy_(n1, &q[js * q_dim1 + 1], &c__1, &q2[iq1], &c__1);
z__[i__] = d__[js];
++i__;
iq1 += *n1;
/* L140: */
}
i__1 = ctot[1];
for (j = 1; j <= i__1; ++j) {
js = indx[i__];
dcopy_(n1, &q[js * q_dim1 + 1], &c__1, &q2[iq1], &c__1);
dcopy_(&n2, &q[*n1 + 1 + js * q_dim1], &c__1, &q2[iq2], &c__1);
z__[i__] = d__[js];
++i__;
iq1 += *n1;
iq2 += n2;
/* L150: */
}
i__1 = ctot[2];
for (j = 1; j <= i__1; ++j) {
js = indx[i__];
dcopy_(&n2, &q[*n1 + 1 + js * q_dim1], &c__1, &q2[iq2], &c__1);
z__[i__] = d__[js];
++i__;
iq2 += n2;
/* L160: */
}
iq1 = iq2;
i__1 = ctot[3];
for (j = 1; j <= i__1; ++j) {
js = indx[i__];
dcopy_(n, &q[js * q_dim1 + 1], &c__1, &q2[iq2], &c__1);
iq2 += *n;
z__[i__] = d__[js];
++i__;
/* L170: */
}
/* The deflated eigenvalues and their corresponding vectors go back */
/* into the last N - K slots of D and Q respectively. */
dlacpy_("A", n, &ctot[3], &q2[iq1], n, &q[(*k + 1) * q_dim1 + 1], ldq);
i__1 = *n - *k;
dcopy_(&i__1, &z__[*k + 1], &c__1, &d__[*k + 1], &c__1);
/* Copy CTOT into COLTYP for referencing in DLAED3. */
for (j = 1; j <= 4; ++j) {
coltyp[j] = ctot[j - 1];
/* L180: */
}
L190:
return 0;
/* End of DLAED2 */
} /* dlaed2_ */
/* Subroutine */
int dlaein_(logical *rightv, logical *noinit, integer *n, doublereal *h__, integer *ldh, doublereal *wr, doublereal *wi, doublereal *vr, doublereal *vi, doublereal *b, integer *ldb, doublereal *work, doublereal *eps3, doublereal *smlnum, doublereal * bignum, integer *info)
{
/* System generated locals */
integer b_dim1, b_offset, h_dim1, h_offset, i__1, i__2, i__3, i__4;
doublereal d__1, d__2, d__3, d__4;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
integer i__, j;
doublereal w, x, y;
integer i1, i2, i3;
doublereal w1, ei, ej, xi, xr, rec;
integer its, ierr;
doublereal temp, norm, vmax;
extern doublereal dnrm2_(integer *, doublereal *, integer *);
extern /* Subroutine */
int dscal_(integer *, doublereal *, doublereal *, integer *);
doublereal scale;
extern doublereal dasum_(integer *, doublereal *, integer *);
char trans[1];
doublereal vcrit, rootn, vnorm;
extern doublereal dlapy2_(doublereal *, doublereal *);
doublereal absbii, absbjj;
extern integer idamax_(integer *, doublereal *, integer *);
extern /* Subroutine */
int dladiv_(doublereal *, doublereal *, doublereal *, doublereal *, doublereal *, doublereal *), dlatrs_( char *, char *, char *, char *, integer *, doublereal *, integer * , doublereal *, doublereal *, doublereal *, integer *);
char normin[1];
doublereal nrmsml, growto;
/* -- LAPACK auxiliary routine (version 3.4.2) -- */
/* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
/* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
/* September 2012 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Parameter adjustments */
h_dim1 = *ldh;
h_offset = 1 + h_dim1;
h__ -= h_offset;
--vr;
--vi;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
--work;
/* Function Body */
*info = 0;
/* GROWTO is the threshold used in the acceptance test for an */
/* eigenvector. */
rootn = sqrt((doublereal) (*n));
growto = .1 / rootn;
/* Computing MAX */
d__1 = 1.;
d__2 = *eps3 * rootn; // , expr subst
nrmsml = max(d__1,d__2) * *smlnum;
/* Form B = H - (WR,WI)*I (except that the subdiagonal elements and */
/* the imaginary parts of the diagonal elements are not stored). */
i__1 = *n;
for (j = 1;
j <= i__1;
++j)
{
i__2 = j - 1;
for (i__ = 1;
i__ <= i__2;
++i__)
{
b[i__ + j * b_dim1] = h__[i__ + j * h_dim1];
/* L10: */
}
b[j + j * b_dim1] = h__[j + j * h_dim1] - *wr;
/* L20: */
}
if (*wi == 0.)
{
/* Real eigenvalue. */
if (*noinit)
{
/* Set initial vector. */
i__1 = *n;
for (i__ = 1;
i__ <= i__1;
++i__)
{
vr[i__] = *eps3;
/* L30: */
}
}
else
{
/* Scale supplied initial vector. */
vnorm = dnrm2_(n, &vr[1], &c__1);
d__1 = *eps3 * rootn / max(vnorm,nrmsml);
dscal_(n, &d__1, &vr[1], &c__1);
}
if (*rightv)
{
/* LU decomposition with partial pivoting of B, replacing zero */
/* pivots by EPS3. */
i__1 = *n - 1;
for (i__ = 1;
i__ <= i__1;
++i__)
{
ei = h__[i__ + 1 + i__ * h_dim1];
if ((d__1 = b[i__ + i__ * b_dim1], abs(d__1)) < abs(ei))
{
/* Interchange rows and eliminate. */
x = b[i__ + i__ * b_dim1] / ei;
b[i__ + i__ * b_dim1] = ei;
i__2 = *n;
for (j = i__ + 1;
j <= i__2;
++j)
{
temp = b[i__ + 1 + j * b_dim1];
b[i__ + 1 + j * b_dim1] = b[i__ + j * b_dim1] - x * temp;
b[i__ + j * b_dim1] = temp;
/* L40: */
}
}
else
{
/* Eliminate without interchange. */
if (b[i__ + i__ * b_dim1] == 0.)
{
b[i__ + i__ * b_dim1] = *eps3;
}
x = ei / b[i__ + i__ * b_dim1];
if (x != 0.)
{
i__2 = *n;
for (j = i__ + 1;
j <= i__2;
++j)
{
b[i__ + 1 + j * b_dim1] -= x * b[i__ + j * b_dim1] ;
/* L50: */
}
}
}
/* L60: */
}
if (b[*n + *n * b_dim1] == 0.)
{
b[*n + *n * b_dim1] = *eps3;
}
*(unsigned char *)trans = 'N';
}
else
{
/* UL decomposition with partial pivoting of B, replacing zero */
/* pivots by EPS3. */
for (j = *n;
j >= 2;
--j)
{
ej = h__[j + (j - 1) * h_dim1];
if ((d__1 = b[j + j * b_dim1], abs(d__1)) < abs(ej))
{
/* Interchange columns and eliminate. */
x = b[j + j * b_dim1] / ej;
b[j + j * b_dim1] = ej;
i__1 = j - 1;
for (i__ = 1;
i__ <= i__1;
++i__)
{
temp = b[i__ + (j - 1) * b_dim1];
b[i__ + (j - 1) * b_dim1] = b[i__ + j * b_dim1] - x * temp;
b[i__ + j * b_dim1] = temp;
/* L70: */
}
}
else
{
/* Eliminate without interchange. */
if (b[j + j * b_dim1] == 0.)
{
b[j + j * b_dim1] = *eps3;
}
x = ej / b[j + j * b_dim1];
if (x != 0.)
{
i__1 = j - 1;
for (i__ = 1;
i__ <= i__1;
++i__)
{
b[i__ + (j - 1) * b_dim1] -= x * b[i__ + j * b_dim1];
/* L80: */
}
}
}
/* L90: */
}
if (b[b_dim1 + 1] == 0.)
{
b[b_dim1 + 1] = *eps3;
}
*(unsigned char *)trans = 'T';
}
*(unsigned char *)normin = 'N';
i__1 = *n;
for (its = 1;
its <= i__1;
++its)
{
/* Solve U*x = scale*v for a right eigenvector */
/* or U**T*x = scale*v for a left eigenvector, */
/* overwriting x on v. */
dlatrs_("Upper", trans, "Nonunit", normin, n, &b[b_offset], ldb, & vr[1], &scale, &work[1], &ierr);
*(unsigned char *)normin = 'Y';
/* Test for sufficient growth in the norm of v. */
vnorm = dasum_(n, &vr[1], &c__1);
if (vnorm >= growto * scale)
{
goto L120;
}
/* Choose new orthogonal starting vector and try again. */
temp = *eps3 / (rootn + 1.);
vr[1] = *eps3;
i__2 = *n;
for (i__ = 2;
i__ <= i__2;
++i__)
{
vr[i__] = temp;
/* L100: */
}
vr[*n - its + 1] -= *eps3 * rootn;
/* L110: */
}
/* Failure to find eigenvector in N iterations. */
*info = 1;
L120: /* Normalize eigenvector. */
i__ = idamax_(n, &vr[1], &c__1);
d__2 = 1. / (d__1 = vr[i__], abs(d__1));
dscal_(n, &d__2, &vr[1], &c__1);
}
else
{
/* Complex eigenvalue. */
if (*noinit)
{
/* Set initial vector. */
i__1 = *n;
for (i__ = 1;
i__ <= i__1;
++i__)
{
vr[i__] = *eps3;
vi[i__] = 0.;
/* L130: */
}
}
else
{
/* Scale supplied initial vector. */
d__1 = dnrm2_(n, &vr[1], &c__1);
d__2 = dnrm2_(n, &vi[1], &c__1);
norm = dlapy2_(&d__1, &d__2);
rec = *eps3 * rootn / max(norm,nrmsml);
dscal_(n, &rec, &vr[1], &c__1);
dscal_(n, &rec, &vi[1], &c__1);
}
if (*rightv)
{
/* LU decomposition with partial pivoting of B, replacing zero */
/* pivots by EPS3. */
/* The imaginary part of the (i,j)-th element of U is stored in */
/* B(j+1,i). */
b[b_dim1 + 2] = -(*wi);
i__1 = *n;
for (i__ = 2;
i__ <= i__1;
++i__)
{
b[i__ + 1 + b_dim1] = 0.;
/* L140: */
}
i__1 = *n - 1;
for (i__ = 1;
i__ <= i__1;
++i__)
{
absbii = dlapy2_(&b[i__ + i__ * b_dim1], &b[i__ + 1 + i__ * b_dim1]);
ei = h__[i__ + 1 + i__ * h_dim1];
if (absbii < abs(ei))
{
/* Interchange rows and eliminate. */
xr = b[i__ + i__ * b_dim1] / ei;
xi = b[i__ + 1 + i__ * b_dim1] / ei;
b[i__ + i__ * b_dim1] = ei;
b[i__ + 1 + i__ * b_dim1] = 0.;
i__2 = *n;
for (j = i__ + 1;
j <= i__2;
++j)
{
temp = b[i__ + 1 + j * b_dim1];
b[i__ + 1 + j * b_dim1] = b[i__ + j * b_dim1] - xr * temp;
b[j + 1 + (i__ + 1) * b_dim1] = b[j + 1 + i__ * b_dim1] - xi * temp;
b[i__ + j * b_dim1] = temp;
b[j + 1 + i__ * b_dim1] = 0.;
/* L150: */
}
b[i__ + 2 + i__ * b_dim1] = -(*wi);
b[i__ + 1 + (i__ + 1) * b_dim1] -= xi * *wi;
b[i__ + 2 + (i__ + 1) * b_dim1] += xr * *wi;
}
else
{
/* Eliminate without interchanging rows. */
if (absbii == 0.)
{
b[i__ + i__ * b_dim1] = *eps3;
b[i__ + 1 + i__ * b_dim1] = 0.;
absbii = *eps3;
}
ei = ei / absbii / absbii;
xr = b[i__ + i__ * b_dim1] * ei;
xi = -b[i__ + 1 + i__ * b_dim1] * ei;
i__2 = *n;
for (j = i__ + 1;
j <= i__2;
++j)
{
b[i__ + 1 + j * b_dim1] = b[i__ + 1 + j * b_dim1] - xr * b[i__ + j * b_dim1] + xi * b[j + 1 + i__ * b_dim1];
b[j + 1 + (i__ + 1) * b_dim1] = -xr * b[j + 1 + i__ * b_dim1] - xi * b[i__ + j * b_dim1];
/* L160: */
}
b[i__ + 2 + (i__ + 1) * b_dim1] -= *wi;
}
/* Compute 1-norm of offdiagonal elements of i-th row. */
i__2 = *n - i__;
i__3 = *n - i__;
work[i__] = dasum_(&i__2, &b[i__ + (i__ + 1) * b_dim1], ldb) + dasum_(&i__3, &b[i__ + 2 + i__ * b_dim1], &c__1);
/* L170: */
}
if (b[*n + *n * b_dim1] == 0. && b[*n + 1 + *n * b_dim1] == 0.)
{
b[*n + *n * b_dim1] = *eps3;
}
work[*n] = 0.;
i1 = *n;
i2 = 1;
i3 = -1;
}
else
{
/* UL decomposition with partial pivoting of conjg(B), */
/* replacing zero pivots by EPS3. */
/* The imaginary part of the (i,j)-th element of U is stored in */
/* B(j+1,i). */
b[*n + 1 + *n * b_dim1] = *wi;
i__1 = *n - 1;
for (j = 1;
j <= i__1;
++j)
{
b[*n + 1 + j * b_dim1] = 0.;
/* L180: */
}
for (j = *n;
j >= 2;
--j)
{
ej = h__[j + (j - 1) * h_dim1];
absbjj = dlapy2_(&b[j + j * b_dim1], &b[j + 1 + j * b_dim1]);
if (absbjj < abs(ej))
{
/* Interchange columns and eliminate */
xr = b[j + j * b_dim1] / ej;
xi = b[j + 1 + j * b_dim1] / ej;
b[j + j * b_dim1] = ej;
b[j + 1 + j * b_dim1] = 0.;
i__1 = j - 1;
for (i__ = 1;
i__ <= i__1;
++i__)
{
temp = b[i__ + (j - 1) * b_dim1];
b[i__ + (j - 1) * b_dim1] = b[i__ + j * b_dim1] - xr * temp;
b[j + i__ * b_dim1] = b[j + 1 + i__ * b_dim1] - xi * temp;
b[i__ + j * b_dim1] = temp;
b[j + 1 + i__ * b_dim1] = 0.;
/* L190: */
}
b[j + 1 + (j - 1) * b_dim1] = *wi;
b[j - 1 + (j - 1) * b_dim1] += xi * *wi;
b[j + (j - 1) * b_dim1] -= xr * *wi;
}
else
{
/* Eliminate without interchange. */
if (absbjj == 0.)
{
b[j + j * b_dim1] = *eps3;
b[j + 1 + j * b_dim1] = 0.;
absbjj = *eps3;
}
ej = ej / absbjj / absbjj;
xr = b[j + j * b_dim1] * ej;
xi = -b[j + 1 + j * b_dim1] * ej;
i__1 = j - 1;
for (i__ = 1;
i__ <= i__1;
++i__)
{
b[i__ + (j - 1) * b_dim1] = b[i__ + (j - 1) * b_dim1] - xr * b[i__ + j * b_dim1] + xi * b[j + 1 + i__ * b_dim1];
b[j + i__ * b_dim1] = -xr * b[j + 1 + i__ * b_dim1] - xi * b[i__ + j * b_dim1];
/* L200: */
}
b[j + (j - 1) * b_dim1] += *wi;
}
/* Compute 1-norm of offdiagonal elements of j-th column. */
i__1 = j - 1;
i__2 = j - 1;
work[j] = dasum_(&i__1, &b[j * b_dim1 + 1], &c__1) + dasum_(& i__2, &b[j + 1 + b_dim1], ldb);
/* L210: */
}
if (b[b_dim1 + 1] == 0. && b[b_dim1 + 2] == 0.)
{
b[b_dim1 + 1] = *eps3;
}
work[1] = 0.;
i1 = 1;
i2 = *n;
i3 = 1;
}
i__1 = *n;
for (its = 1;
its <= i__1;
++its)
{
scale = 1.;
vmax = 1.;
vcrit = *bignum;
/* Solve U*(xr,xi) = scale*(vr,vi) for a right eigenvector, */
/* or U**T*(xr,xi) = scale*(vr,vi) for a left eigenvector, */
/* overwriting (xr,xi) on (vr,vi). */
i__2 = i2;
i__3 = i3;
for (i__ = i1;
i__3 < 0 ? i__ >= i__2 : i__ <= i__2;
i__ += i__3)
{
if (work[i__] > vcrit)
{
rec = 1. / vmax;
dscal_(n, &rec, &vr[1], &c__1);
dscal_(n, &rec, &vi[1], &c__1);
scale *= rec;
vmax = 1.;
vcrit = *bignum;
}
xr = vr[i__];
xi = vi[i__];
if (*rightv)
{
i__4 = *n;
for (j = i__ + 1;
j <= i__4;
++j)
{
xr = xr - b[i__ + j * b_dim1] * vr[j] + b[j + 1 + i__ * b_dim1] * vi[j];
xi = xi - b[i__ + j * b_dim1] * vi[j] - b[j + 1 + i__ * b_dim1] * vr[j];
/* L220: */
}
}
else
{
i__4 = i__ - 1;
for (j = 1;
j <= i__4;
++j)
{
xr = xr - b[j + i__ * b_dim1] * vr[j] + b[i__ + 1 + j * b_dim1] * vi[j];
xi = xi - b[j + i__ * b_dim1] * vi[j] - b[i__ + 1 + j * b_dim1] * vr[j];
/* L230: */
}
}
w = (d__1 = b[i__ + i__ * b_dim1], abs(d__1)) + (d__2 = b[i__ + 1 + i__ * b_dim1], abs(d__2));
if (w > *smlnum)
{
if (w < 1.)
{
w1 = abs(xr) + abs(xi);
if (w1 > w * *bignum)
{
rec = 1. / w1;
dscal_(n, &rec, &vr[1], &c__1);
dscal_(n, &rec, &vi[1], &c__1);
xr = vr[i__];
xi = vi[i__];
scale *= rec;
vmax *= rec;
}
}
/* Divide by diagonal element of B. */
dladiv_(&xr, &xi, &b[i__ + i__ * b_dim1], &b[i__ + 1 + i__ * b_dim1], &vr[i__], &vi[i__]);
/* Computing MAX */
d__3 = (d__1 = vr[i__], abs(d__1)) + (d__2 = vi[i__], abs( d__2));
vmax = max(d__3,vmax);
vcrit = *bignum / vmax;
}
else
{
i__4 = *n;
for (j = 1;
j <= i__4;
++j)
{
vr[j] = 0.;
vi[j] = 0.;
/* L240: */
}
vr[i__] = 1.;
vi[i__] = 1.;
scale = 0.;
vmax = 1.;
vcrit = *bignum;
}
/* L250: */
}
/* Test for sufficient growth in the norm of (VR,VI). */
vnorm = dasum_(n, &vr[1], &c__1) + dasum_(n, &vi[1], &c__1);
if (vnorm >= growto * scale)
{
goto L280;
}
/* Choose a new orthogonal starting vector and try again. */
y = *eps3 / (rootn + 1.);
vr[1] = *eps3;
vi[1] = 0.;
i__3 = *n;
for (i__ = 2;
i__ <= i__3;
++i__)
{
vr[i__] = y;
vi[i__] = 0.;
/* L260: */
}
vr[*n - its + 1] -= *eps3 * rootn;
/* L270: */
}
/* Failure to find eigenvector in N iterations */
*info = 1;
L280: /* Normalize eigenvector. */
vnorm = 0.;
i__1 = *n;
for (i__ = 1;
i__ <= i__1;
++i__)
{
/* Computing MAX */
d__3 = vnorm;
d__4 = (d__1 = vr[i__], abs(d__1)) + (d__2 = vi[i__] , abs(d__2)); // , expr subst
vnorm = max(d__3,d__4);
/* L290: */
}
d__1 = 1. / vnorm;
dscal_(n, &d__1, &vr[1], &c__1);
d__1 = 1. / vnorm;
dscal_(n, &d__1, &vi[1], &c__1);
}
return 0;
/* End of DLAEIN */
}
/* Subroutine */ int dtbt05_(char *uplo, char *trans, char *diag, integer *n,
integer *kd, integer *nrhs, doublereal *ab, integer *ldab, doublereal
*b, integer *ldb, doublereal *x, integer *ldx, doublereal *xact,
integer *ldxact, doublereal *ferr, doublereal *berr, doublereal *
reslts)
{
/* System generated locals */
integer ab_dim1, ab_offset, b_dim1, b_offset, x_dim1, x_offset, xact_dim1,
xact_offset, i__1, i__2, i__3, i__4;
doublereal d__1, d__2, d__3;
/* Local variables */
integer i__, j, k, nz, ifu;
doublereal eps, tmp, diff, axbi;
integer imax;
doublereal unfl, ovfl;
logical unit;
extern logical lsame_(char *, char *);
logical upper;
doublereal xnorm;
extern doublereal dlamch_(char *);
extern integer idamax_(integer *, doublereal *, integer *);
doublereal errbnd;
logical notran;
/* -- LAPACK test routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* DTBT05 tests the error bounds from iterative refinement for the */
/* computed solution to a system of equations A*X = B, where A is a */
/* triangular band matrix. */
/* RESLTS(1) = test of the error bound */
/* = norm(X - XACT) / ( norm(X) * FERR ) */
/* A large value is returned if this ratio is not less than one. */
/* RESLTS(2) = residual from the iterative refinement routine */
/* = the maximum of BERR / ( NZ*EPS + (*) ), where */
/* (*) = NZ*UNFL / (min_i (abs(A)*abs(X) +abs(b))_i ) */
/* and NZ = max. number of nonzeros in any row of A, plus 1 */
/* Arguments */
/* ========= */
/* UPLO (input) CHARACTER*1 */
/* Specifies whether the matrix A is upper or lower triangular. */
/* = 'U': Upper triangular */
/* = 'L': Lower triangular */
/* TRANS (input) CHARACTER*1 */
/* Specifies the form of the system of equations. */
/* = 'N': A * X = B (No transpose) */
/* = 'T': A'* X = B (Transpose) */
/* = 'C': A'* X = B (Conjugate transpose = Transpose) */
/* DIAG (input) CHARACTER*1 */
/* Specifies whether or not the matrix A is unit triangular. */
/* = 'N': Non-unit triangular */
/* = 'U': Unit triangular */
/* N (input) INTEGER */
/* The number of rows of the matrices X, B, and XACT, and the */
/* order of the matrix A. N >= 0. */
/* KD (input) INTEGER */
/* The number of super-diagonals of the matrix A if UPLO = 'U', */
/* or the number of sub-diagonals if UPLO = 'L'. KD >= 0. */
/* NRHS (input) INTEGER */
/* The number of columns of the matrices X, B, and XACT. */
/* NRHS >= 0. */
/* AB (input) DOUBLE PRECISION array, dimension (LDAB,N) */
/* The upper or lower triangular 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). */
/* If DIAG = 'U', the diagonal elements of A are not referenced */
/* and are assumed to be 1. */
/* LDAB (input) INTEGER */
/* The leading dimension of the array AB. LDAB >= KD+1. */
/* B (input) DOUBLE PRECISION array, dimension (LDB,NRHS) */
/* The right hand side vectors for the system of linear */
/* equations. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= max(1,N). */
/* X (input) DOUBLE PRECISION array, dimension (LDX,NRHS) */
/* The computed solution vectors. Each vector is stored as a */
/* column of the matrix X. */
/* LDX (input) INTEGER */
/* The leading dimension of the array X. LDX >= max(1,N). */
/* XACT (input) DOUBLE PRECISION array, dimension (LDX,NRHS) */
/* The exact solution vectors. Each vector is stored as a */
/* column of the matrix XACT. */
/* LDXACT (input) INTEGER */
/* The leading dimension of the array XACT. LDXACT >= max(1,N). */
/* FERR (input) DOUBLE PRECISION array, dimension (NRHS) */
/* The estimated forward error bounds for each solution vector */
/* X. If XTRUE is the true solution, FERR bounds the magnitude */
/* of the largest entry in (X - XTRUE) divided by the magnitude */
/* of the largest entry in X. */
/* BERR (input) DOUBLE PRECISION array, dimension (NRHS) */
/* The componentwise relative backward error of each solution */
/* vector (i.e., the smallest relative change in any entry of A */
/* or B that makes X an exact solution). */
/* RESLTS (output) DOUBLE PRECISION array, dimension (2) */
/* The maximum over the NRHS solution vectors of the ratios: */
/* RESLTS(1) = norm(X - XACT) / ( norm(X) * FERR ) */
/* RESLTS(2) = BERR / ( NZ*EPS + (*) ) */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Quick exit if N = 0 or NRHS = 0. */
/* Parameter adjustments */
ab_dim1 = *ldab;
ab_offset = 1 + ab_dim1;
ab -= ab_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
x_dim1 = *ldx;
x_offset = 1 + x_dim1;
x -= x_offset;
xact_dim1 = *ldxact;
xact_offset = 1 + xact_dim1;
xact -= xact_offset;
--ferr;
--berr;
--reslts;
/* Function Body */
if (*n <= 0 || *nrhs <= 0) {
reslts[1] = 0.;
reslts[2] = 0.;
return 0;
}
eps = dlamch_("Epsilon");
unfl = dlamch_("Safe minimum");
ovfl = 1. / unfl;
upper = lsame_(uplo, "U");
notran = lsame_(trans, "N");
unit = lsame_(diag, "U");
/* Computing MIN */
i__1 = *kd, i__2 = *n - 1;
nz = min(i__1,i__2) + 1;
/* Test 1: Compute the maximum of */
/* norm(X - XACT) / ( norm(X) * FERR ) */
/* over all the vectors X and XACT using the infinity-norm. */
errbnd = 0.;
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
imax = idamax_(n, &x[j * x_dim1 + 1], &c__1);
/* Computing MAX */
d__2 = (d__1 = x[imax + j * x_dim1], abs(d__1));
xnorm = max(d__2,unfl);
diff = 0.;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
/* Computing MAX */
d__2 = diff, d__3 = (d__1 = x[i__ + j * x_dim1] - xact[i__ + j *
xact_dim1], abs(d__1));
diff = max(d__2,d__3);
/* L10: */
}
if (xnorm > 1.) {
goto L20;
} else if (diff <= ovfl * xnorm) {
goto L20;
} else {
errbnd = 1. / eps;
goto L30;
}
L20:
if (diff / xnorm <= ferr[j]) {
/* Computing MAX */
d__1 = errbnd, d__2 = diff / xnorm / ferr[j];
errbnd = max(d__1,d__2);
} else {
errbnd = 1. / eps;
}
L30:
;
}
reslts[1] = errbnd;
/* Test 2: Compute the maximum of BERR / ( NZ*EPS + (*) ), where */
/* (*) = NZ*UNFL / (min_i (abs(A)*abs(X) +abs(b))_i ) */
ifu = 0;
if (unit) {
ifu = 1;
}
i__1 = *nrhs;
for (k = 1; k <= i__1; ++k) {
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
tmp = (d__1 = b[i__ + k * b_dim1], abs(d__1));
if (upper) {
if (! notran) {
/* Computing MAX */
i__3 = i__ - *kd;
i__4 = i__ - ifu;
for (j = max(i__3,1); j <= i__4; ++j) {
tmp += (d__1 = ab[*kd + 1 - i__ + j + i__ * ab_dim1],
abs(d__1)) * (d__2 = x[j + k * x_dim1], abs(
d__2));
/* L40: */
}
if (unit) {
tmp += (d__1 = x[i__ + k * x_dim1], abs(d__1));
}
} else {
if (unit) {
tmp += (d__1 = x[i__ + k * x_dim1], abs(d__1));
}
/* Computing MIN */
i__3 = i__ + *kd;
i__4 = min(i__3,*n);
for (j = i__ + ifu; j <= i__4; ++j) {
tmp += (d__1 = ab[*kd + 1 + i__ - j + j * ab_dim1],
abs(d__1)) * (d__2 = x[j + k * x_dim1], abs(
d__2));
/* L50: */
}
}
} else {
if (notran) {
/* Computing MAX */
i__4 = i__ - *kd;
i__3 = i__ - ifu;
for (j = max(i__4,1); j <= i__3; ++j) {
tmp += (d__1 = ab[i__ + 1 - j + j * ab_dim1], abs(
d__1)) * (d__2 = x[j + k * x_dim1], abs(d__2))
;
/* L60: */
}
if (unit) {
tmp += (d__1 = x[i__ + k * x_dim1], abs(d__1));
}
} else {
if (unit) {
tmp += (d__1 = x[i__ + k * x_dim1], abs(d__1));
}
/* Computing MIN */
i__4 = i__ + *kd;
i__3 = min(i__4,*n);
for (j = i__ + ifu; j <= i__3; ++j) {
tmp += (d__1 = ab[j + 1 - i__ + i__ * ab_dim1], abs(
d__1)) * (d__2 = x[j + k * x_dim1], abs(d__2))
;
/* L70: */
}
}
}
if (i__ == 1) {
axbi = tmp;
} else {
axbi = min(axbi,tmp);
}
/* L80: */
}
/* Computing MAX */
d__1 = axbi, d__2 = nz * unfl;
tmp = berr[k] / (nz * eps + nz * unfl / max(d__1,d__2));
if (k == 1) {
reslts[2] = tmp;
} else {
reslts[2] = max(reslts[2],tmp);
}
/* L90: */
}
return 0;
/* End of DTBT05 */
} /* dtbt05_ */
/* Subroutine */ int zdrvpt_(logical *dotype, integer *nn, integer *nval,
integer *nrhs, doublereal *thresh, logical *tsterr, doublecomplex *a,
doublereal *d__, doublecomplex *e, doublecomplex *b, doublecomplex *x,
doublecomplex *xact, doublecomplex *work, doublereal *rwork, integer
*nout)
{
/* Initialized data */
static integer iseedy[4] = { 0,0,0,1 };
/* Format strings */
static char fmt_9999[] = "(1x,a6,\002, N =\002,i5,\002, type \002,i2,"
"\002, test \002,i2,\002, ratio = \002,g12.5)";
static char fmt_9998[] = "(1x,a6,\002, FACT='\002,a1,\002', N =\002,i5"
",\002, type \002,i2,\002, test \002,i2,\002, ratio = \002,g12.5)";
/* System generated locals */
integer i__1, i__2, i__3, i__4, i__5;
doublereal d__1, d__2;
/* Builtin functions */
/* Subroutine */ int s_copy(char *, char *, ftnlen, ftnlen);
double z_abs(doublecomplex *);
integer s_wsfe(cilist *), do_fio(integer *, char *, ftnlen), e_wsfe(void);
/* Local variables */
integer i__, j, k, n;
doublereal z__[3];
integer k1, ia, in, kl, ku, ix, nt, lda;
char fact[1];
doublereal cond;
integer mode;
doublereal dmax__;
integer imat, info;
char path[3], dist[1], type__[1];
integer nrun, ifact;
extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
integer *);
integer nfail, iseed[4];
extern doublereal dget06_(doublereal *, doublereal *);
doublereal rcond;
integer nimat;
doublereal anorm;
extern /* Subroutine */ int zget04_(integer *, integer *, doublecomplex *,
integer *, doublecomplex *, integer *, doublereal *, doublereal *
), dcopy_(integer *, doublereal *, integer *, doublereal *,
integer *);
integer izero, nerrs;
extern /* Subroutine */ int zptt01_(integer *, doublereal *,
doublecomplex *, doublereal *, doublecomplex *, doublecomplex *,
doublereal *);
logical zerot;
extern /* Subroutine */ int zcopy_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *), zptt02_(char *, integer *, integer *,
doublereal *, doublecomplex *, doublecomplex *, integer *,
doublecomplex *, integer *, doublereal *), zptt05_(
integer *, integer *, doublereal *, doublecomplex *,
doublecomplex *, integer *, doublecomplex *, integer *,
doublecomplex *, integer *, doublereal *, doublereal *,
doublereal *), zptsv_(integer *, integer *, doublereal *,
doublecomplex *, doublecomplex *, integer *, integer *), zlatb4_(
char *, integer *, integer *, integer *, char *, integer *,
integer *, doublereal *, integer *, doublereal *, char *), aladhd_(integer *, char *), alaerh_(char
*, char *, integer *, integer *, char *, integer *, integer *,
integer *, integer *, integer *, integer *, integer *, integer *,
integer *);
extern integer idamax_(integer *, doublereal *, integer *);
doublereal rcondc;
extern /* Subroutine */ int zdscal_(integer *, doublereal *,
doublecomplex *, integer *), alasvm_(char *, integer *, integer *,
integer *, integer *), dlarnv_(integer *, integer *,
integer *, doublereal *);
doublereal ainvnm;
extern doublereal zlanht_(char *, integer *, doublereal *, doublecomplex *
);
extern /* Subroutine */ int zlacpy_(char *, integer *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *);
extern doublereal dzasum_(integer *, doublecomplex *, integer *);
extern /* Subroutine */ int zlaset_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, doublecomplex *, integer *), zlaptm_(char *, integer *, integer *, doublereal *,
doublereal *, doublecomplex *, doublecomplex *, integer *,
doublereal *, doublecomplex *, integer *), zlatms_(
integer *, integer *, char *, integer *, char *, doublereal *,
integer *, doublereal *, doublereal *, integer *, integer *, char
*, doublecomplex *, integer *, doublecomplex *, integer *), zlarnv_(integer *, integer *, integer *,
doublecomplex *);
doublereal result[6];
extern /* Subroutine */ int zpttrf_(integer *, doublereal *,
doublecomplex *, integer *), zerrvx_(char *, integer *),
zpttrs_(char *, integer *, integer *, doublereal *, doublecomplex
*, doublecomplex *, integer *, integer *), zptsvx_(char *,
integer *, integer *, doublereal *, doublecomplex *, doublereal *
, doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublereal *, doublereal *, doublereal *,
doublecomplex *, doublereal *, integer *);
/* Fortran I/O blocks */
static cilist io___35 = { 0, 0, 0, fmt_9999, 0 };
static cilist io___38 = { 0, 0, 0, fmt_9998, 0 };
/* -- LAPACK test routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* ZDRVPT tests ZPTSV and -SVX. */
/* Arguments */
/* ========= */
/* DOTYPE (input) LOGICAL array, dimension (NTYPES) */
/* The matrix types to be used for testing. Matrices of type j */
/* (for 1 <= j <= NTYPES) are used for testing if DOTYPE(j) = */
/* .TRUE.; if DOTYPE(j) = .FALSE., then type j is not used. */
/* NN (input) INTEGER */
/* The number of values of N contained in the vector NVAL. */
/* NVAL (input) INTEGER array, dimension (NN) */
/* The values of the matrix dimension N. */
/* NRHS (input) INTEGER */
/* The number of right hand side vectors to be generated for */
/* each linear system. */
/* THRESH (input) DOUBLE PRECISION */
/* The threshold value for the test ratios. A result is */
/* included in the output file if RESULT >= THRESH. To have */
/* every test ratio printed, use THRESH = 0. */
/* TSTERR (input) LOGICAL */
/* Flag that indicates whether error exits are to be tested. */
/* A (workspace) COMPLEX*16 array, dimension (NMAX*2) */
/* D (workspace) DOUBLE PRECISION array, dimension (NMAX*2) */
/* E (workspace) COMPLEX*16 array, dimension (NMAX*2) */
/* B (workspace) COMPLEX*16 array, dimension (NMAX*NRHS) */
/* X (workspace) COMPLEX*16 array, dimension (NMAX*NRHS) */
/* XACT (workspace) COMPLEX*16 array, dimension (NMAX*NRHS) */
/* WORK (workspace) COMPLEX*16 array, dimension */
/* (NMAX*max(3,NRHS)) */
/* RWORK (workspace) DOUBLE PRECISION array, dimension (NMAX+2*NRHS) */
/* NOUT (input) INTEGER */
/* The unit number for output. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. Local Arrays .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Scalars in Common .. */
/* .. */
/* .. Common blocks .. */
/* .. */
/* .. Data statements .. */
/* Parameter adjustments */
--rwork;
--work;
--xact;
--x;
--b;
--e;
--d__;
--a;
--nval;
--dotype;
/* Function Body */
/* .. */
/* .. Executable Statements .. */
s_copy(path, "Zomplex precision", (ftnlen)1, (ftnlen)17);
s_copy(path + 1, "PT", (ftnlen)2, (ftnlen)2);
nrun = 0;
nfail = 0;
nerrs = 0;
for (i__ = 1; i__ <= 4; ++i__) {
iseed[i__ - 1] = iseedy[i__ - 1];
/* L10: */
}
/* Test the error exits */
if (*tsterr) {
zerrvx_(path, nout);
}
infoc_1.infot = 0;
i__1 = *nn;
for (in = 1; in <= i__1; ++in) {
/* Do for each value of N in NVAL. */
n = nval[in];
lda = max(1,n);
nimat = 12;
if (n <= 0) {
nimat = 1;
}
i__2 = nimat;
for (imat = 1; imat <= i__2; ++imat) {
/* Do the tests only if DOTYPE( IMAT ) is true. */
if (n > 0 && ! dotype[imat]) {
goto L110;
}
/* Set up parameters with ZLATB4. */
zlatb4_(path, &imat, &n, &n, type__, &kl, &ku, &anorm, &mode, &
cond, dist);
zerot = imat >= 8 && imat <= 10;
if (imat <= 6) {
/* Type 1-6: generate a symmetric tridiagonal matrix of */
/* known condition number in lower triangular band storage. */
s_copy(srnamc_1.srnamt, "ZLATMS", (ftnlen)6, (ftnlen)6);
zlatms_(&n, &n, dist, iseed, type__, &rwork[1], &mode, &cond,
&anorm, &kl, &ku, "B", &a[1], &c__2, &work[1], &info);
/* Check the error code from ZLATMS. */
if (info != 0) {
alaerh_(path, "ZLATMS", &info, &c__0, " ", &n, &n, &kl, &
ku, &c_n1, &imat, &nfail, &nerrs, nout);
goto L110;
}
izero = 0;
/* Copy the matrix to D and E. */
ia = 1;
i__3 = n - 1;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = i__;
i__5 = ia;
d__[i__4] = a[i__5].r;
i__4 = i__;
i__5 = ia + 1;
e[i__4].r = a[i__5].r, e[i__4].i = a[i__5].i;
ia += 2;
/* L20: */
}
if (n > 0) {
i__3 = n;
i__4 = ia;
d__[i__3] = a[i__4].r;
}
} else {
/* Type 7-12: generate a diagonally dominant matrix with */
/* unknown condition number in the vectors D and E. */
if (! zerot || ! dotype[7]) {
/* Let D and E have values from [-1,1]. */
dlarnv_(&c__2, iseed, &n, &d__[1]);
i__3 = n - 1;
zlarnv_(&c__2, iseed, &i__3, &e[1]);
/* Make the tridiagonal matrix diagonally dominant. */
if (n == 1) {
d__[1] = abs(d__[1]);
} else {
d__[1] = abs(d__[1]) + z_abs(&e[1]);
d__[n] = (d__1 = d__[n], abs(d__1)) + z_abs(&e[n - 1])
;
i__3 = n - 1;
for (i__ = 2; i__ <= i__3; ++i__) {
d__[i__] = (d__1 = d__[i__], abs(d__1)) + z_abs(&
e[i__]) + z_abs(&e[i__ - 1]);
/* L30: */
}
}
/* Scale D and E so the maximum element is ANORM. */
ix = idamax_(&n, &d__[1], &c__1);
dmax__ = d__[ix];
d__1 = anorm / dmax__;
dscal_(&n, &d__1, &d__[1], &c__1);
if (n > 1) {
i__3 = n - 1;
d__1 = anorm / dmax__;
zdscal_(&i__3, &d__1, &e[1], &c__1);
}
} else if (izero > 0) {
/* Reuse the last matrix by copying back the zeroed out */
/* elements. */
if (izero == 1) {
d__[1] = z__[1];
if (n > 1) {
e[1].r = z__[2], e[1].i = 0.;
}
} else if (izero == n) {
i__3 = n - 1;
e[i__3].r = z__[0], e[i__3].i = 0.;
d__[n] = z__[1];
} else {
i__3 = izero - 1;
e[i__3].r = z__[0], e[i__3].i = 0.;
d__[izero] = z__[1];
i__3 = izero;
e[i__3].r = z__[2], e[i__3].i = 0.;
}
}
/* For types 8-10, set one row and column of the matrix to */
/* zero. */
izero = 0;
if (imat == 8) {
izero = 1;
z__[1] = d__[1];
d__[1] = 0.;
if (n > 1) {
z__[2] = e[1].r;
e[1].r = 0., e[1].i = 0.;
}
} else if (imat == 9) {
izero = n;
if (n > 1) {
i__3 = n - 1;
z__[0] = e[i__3].r;
i__3 = n - 1;
e[i__3].r = 0., e[i__3].i = 0.;
}
z__[1] = d__[n];
d__[n] = 0.;
} else if (imat == 10) {
izero = (n + 1) / 2;
if (izero > 1) {
i__3 = izero - 1;
z__[0] = e[i__3].r;
i__3 = izero - 1;
e[i__3].r = 0., e[i__3].i = 0.;
i__3 = izero;
z__[2] = e[i__3].r;
i__3 = izero;
e[i__3].r = 0., e[i__3].i = 0.;
}
z__[1] = d__[izero];
d__[izero] = 0.;
}
}
/* Generate NRHS random solution vectors. */
ix = 1;
i__3 = *nrhs;
for (j = 1; j <= i__3; ++j) {
zlarnv_(&c__2, iseed, &n, &xact[ix]);
ix += lda;
/* L40: */
}
/* Set the right hand side. */
zlaptm_("Lower", &n, nrhs, &c_b24, &d__[1], &e[1], &xact[1], &lda,
&c_b25, &b[1], &lda);
for (ifact = 1; ifact <= 2; ++ifact) {
if (ifact == 1) {
*(unsigned char *)fact = 'F';
} else {
*(unsigned char *)fact = 'N';
}
/* Compute the condition number for comparison with */
/* the value returned by ZPTSVX. */
if (zerot) {
if (ifact == 1) {
goto L100;
}
rcondc = 0.;
} else if (ifact == 1) {
/* Compute the 1-norm of A. */
anorm = zlanht_("1", &n, &d__[1], &e[1]);
dcopy_(&n, &d__[1], &c__1, &d__[n + 1], &c__1);
if (n > 1) {
i__3 = n - 1;
zcopy_(&i__3, &e[1], &c__1, &e[n + 1], &c__1);
}
/* Factor the matrix A. */
zpttrf_(&n, &d__[n + 1], &e[n + 1], &info);
/* Use ZPTTRS to solve for one column at a time of */
/* inv(A), computing the maximum column sum as we go. */
ainvnm = 0.;
i__3 = n;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = n;
for (j = 1; j <= i__4; ++j) {
i__5 = j;
x[i__5].r = 0., x[i__5].i = 0.;
/* L50: */
}
i__4 = i__;
x[i__4].r = 1., x[i__4].i = 0.;
zpttrs_("Lower", &n, &c__1, &d__[n + 1], &e[n + 1], &
x[1], &lda, &info);
/* Computing MAX */
d__1 = ainvnm, d__2 = dzasum_(&n, &x[1], &c__1);
ainvnm = max(d__1,d__2);
/* L60: */
}
/* Compute the 1-norm condition number of A. */
if (anorm <= 0. || ainvnm <= 0.) {
rcondc = 1.;
} else {
rcondc = 1. / anorm / ainvnm;
}
}
if (ifact == 2) {
/* --- Test ZPTSV -- */
dcopy_(&n, &d__[1], &c__1, &d__[n + 1], &c__1);
if (n > 1) {
i__3 = n - 1;
zcopy_(&i__3, &e[1], &c__1, &e[n + 1], &c__1);
}
zlacpy_("Full", &n, nrhs, &b[1], &lda, &x[1], &lda);
/* Factor A as L*D*L' and solve the system A*X = B. */
s_copy(srnamc_1.srnamt, "ZPTSV ", (ftnlen)6, (ftnlen)6);
zptsv_(&n, nrhs, &d__[n + 1], &e[n + 1], &x[1], &lda, &
info);
/* Check error code from ZPTSV . */
if (info != izero) {
alaerh_(path, "ZPTSV ", &info, &izero, " ", &n, &n, &
c__1, &c__1, nrhs, &imat, &nfail, &nerrs,
nout);
}
nt = 0;
if (izero == 0) {
/* Check the factorization by computing the ratio */
/* norm(L*D*L' - A) / (n * norm(A) * EPS ) */
zptt01_(&n, &d__[1], &e[1], &d__[n + 1], &e[n + 1], &
work[1], result);
/* Compute the residual in the solution. */
zlacpy_("Full", &n, nrhs, &b[1], &lda, &work[1], &lda);
zptt02_("Lower", &n, nrhs, &d__[1], &e[1], &x[1], &
lda, &work[1], &lda, &result[1]);
/* Check solution from generated exact solution. */
zget04_(&n, nrhs, &x[1], &lda, &xact[1], &lda, &
rcondc, &result[2]);
nt = 3;
}
/* Print information about the tests that did not pass */
/* the threshold. */
i__3 = nt;
for (k = 1; k <= i__3; ++k) {
if (result[k - 1] >= *thresh) {
if (nfail == 0 && nerrs == 0) {
aladhd_(nout, path);
}
io___35.ciunit = *nout;
s_wsfe(&io___35);
do_fio(&c__1, "ZPTSV ", (ftnlen)6);
do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer))
;
do_fio(&c__1, (char *)&imat, (ftnlen)sizeof(
integer));
do_fio(&c__1, (char *)&k, (ftnlen)sizeof(integer))
;
do_fio(&c__1, (char *)&result[k - 1], (ftnlen)
sizeof(doublereal));
e_wsfe();
++nfail;
}
/* L70: */
}
nrun += nt;
}
/* --- Test ZPTSVX --- */
if (ifact > 1) {
/* Initialize D( N+1:2*N ) and E( N+1:2*N ) to zero. */
i__3 = n - 1;
for (i__ = 1; i__ <= i__3; ++i__) {
d__[n + i__] = 0.;
i__4 = n + i__;
e[i__4].r = 0., e[i__4].i = 0.;
/* L80: */
}
if (n > 0) {
d__[n + n] = 0.;
}
}
zlaset_("Full", &n, nrhs, &c_b62, &c_b62, &x[1], &lda);
/* Solve the system and compute the condition number and */
/* error bounds using ZPTSVX. */
s_copy(srnamc_1.srnamt, "ZPTSVX", (ftnlen)6, (ftnlen)6);
zptsvx_(fact, &n, nrhs, &d__[1], &e[1], &d__[n + 1], &e[n + 1]
, &b[1], &lda, &x[1], &lda, &rcond, &rwork[1], &rwork[
*nrhs + 1], &work[1], &rwork[(*nrhs << 1) + 1], &info);
/* Check the error code from ZPTSVX. */
if (info != izero) {
alaerh_(path, "ZPTSVX", &info, &izero, fact, &n, &n, &
c__1, &c__1, nrhs, &imat, &nfail, &nerrs, nout);
}
if (izero == 0) {
if (ifact == 2) {
/* Check the factorization by computing the ratio */
/* norm(L*D*L' - A) / (n * norm(A) * EPS ) */
k1 = 1;
zptt01_(&n, &d__[1], &e[1], &d__[n + 1], &e[n + 1], &
work[1], result);
} else {
k1 = 2;
}
/* Compute the residual in the solution. */
zlacpy_("Full", &n, nrhs, &b[1], &lda, &work[1], &lda);
zptt02_("Lower", &n, nrhs, &d__[1], &e[1], &x[1], &lda, &
work[1], &lda, &result[1]);
/* Check solution from generated exact solution. */
zget04_(&n, nrhs, &x[1], &lda, &xact[1], &lda, &rcondc, &
result[2]);
/* Check error bounds from iterative refinement. */
zptt05_(&n, nrhs, &d__[1], &e[1], &b[1], &lda, &x[1], &
lda, &xact[1], &lda, &rwork[1], &rwork[*nrhs + 1],
&result[3]);
} else {
k1 = 6;
}
/* Check the reciprocal of the condition number. */
result[5] = dget06_(&rcond, &rcondc);
/* Print information about the tests that did not pass */
/* the threshold. */
for (k = k1; k <= 6; ++k) {
if (result[k - 1] >= *thresh) {
if (nfail == 0 && nerrs == 0) {
aladhd_(nout, path);
}
io___38.ciunit = *nout;
s_wsfe(&io___38);
do_fio(&c__1, "ZPTSVX", (ftnlen)6);
do_fio(&c__1, fact, (ftnlen)1);
do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&imat, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&k, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&result[k - 1], (ftnlen)sizeof(
doublereal));
e_wsfe();
++nfail;
}
/* L90: */
}
nrun = nrun + 7 - k1;
L100:
;
}
L110:
;
}
/* L120: */
}
/* Print a summary of the results. */
alasvm_(path, nout, &nfail, &nrun, &nerrs);
return 0;
/* End of ZDRVPT */
} /* zdrvpt_ */
/* Subroutine */ int dgesc2_(integer *n, doublereal *a, integer *lda,
doublereal *rhs, integer *ipiv, integer *jpiv, doublereal *scale)
{
/* System generated locals */
integer a_dim1, a_offset, i__1, i__2;
doublereal d__1, d__2;
/* Local variables */
integer i__, j;
doublereal eps, temp;
extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
integer *), dlabad_(doublereal *, doublereal *);
extern doublereal dlamch_(char *);
extern integer idamax_(integer *, doublereal *, integer *);
doublereal bignum;
extern /* Subroutine */ int dlaswp_(integer *, doublereal *, integer *,
integer *, integer *, integer *, integer *);
doublereal smlnum;
/* -- LAPACK auxiliary routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* DGESC2 solves a system of linear equations */
/* A * X = scale* RHS */
/* with a general N-by-N matrix A using the LU factorization with */
/* complete pivoting computed by DGETC2. */
/* Arguments */
/* ========= */
/* N (input) INTEGER */
/* The order of the matrix A. */
/* A (input) DOUBLE PRECISION array, dimension (LDA,N) */
/* On entry, the LU part of the factorization of the n-by-n */
/* matrix A computed by DGETC2: A = P * L * U * Q */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1, N). */
/* RHS (input/output) DOUBLE PRECISION array, dimension (N). */
/* On entry, the right hand side vector b. */
/* On exit, the solution vector X. */
/* IPIV (input) INTEGER array, dimension (N). */
/* The pivot indices; for 1 <= i <= N, row i of the */
/* matrix has been interchanged with row IPIV(i). */
/* JPIV (input) INTEGER array, dimension (N). */
/* The pivot indices; for 1 <= j <= N, column j of the */
/* matrix has been interchanged with column JPIV(j). */
/* SCALE (output) DOUBLE PRECISION */
/* On exit, SCALE contains the scale factor. SCALE is chosen */
/* 0 <= SCALE <= 1 to prevent owerflow in the solution. */
/* Further Details */
/* =============== */
/* Based on contributions by */
/* Bo Kagstrom and Peter Poromaa, Department of Computing Science, */
/* Umea University, S-901 87 Umea, Sweden. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Set constant to control owerflow */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
--rhs;
--ipiv;
--jpiv;
/* Function Body */
eps = dlamch_("P");
smlnum = dlamch_("S") / eps;
bignum = 1. / smlnum;
dlabad_(&smlnum, &bignum);
/* Apply permutations IPIV to RHS */
i__1 = *n - 1;
dlaswp_(&c__1, &rhs[1], lda, &c__1, &i__1, &ipiv[1], &c__1);
/* Solve for L part */
i__1 = *n - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = *n;
for (j = i__ + 1; j <= i__2; ++j) {
rhs[j] -= a[j + i__ * a_dim1] * rhs[i__];
/* L10: */
}
/* L20: */
}
/* Solve for U part */
*scale = 1.;
/* Check for scaling */
i__ = idamax_(n, &rhs[1], &c__1);
if (smlnum * 2. * (d__1 = rhs[i__], abs(d__1)) > (d__2 = a[*n + *n *
a_dim1], abs(d__2))) {
temp = .5 / (d__1 = rhs[i__], abs(d__1));
dscal_(n, &temp, &rhs[1], &c__1);
*scale *= temp;
}
for (i__ = *n; i__ >= 1; --i__) {
temp = 1. / a[i__ + i__ * a_dim1];
rhs[i__] *= temp;
i__1 = *n;
for (j = i__ + 1; j <= i__1; ++j) {
rhs[i__] -= rhs[j] * (a[i__ + j * a_dim1] * temp);
/* L30: */
}
/* L40: */
}
/* Apply permutations JPIV to the solution (RHS) */
i__1 = *n - 1;
dlaswp_(&c__1, &rhs[1], lda, &c__1, &i__1, &jpiv[1], &c_n1);
return 0;
/* End of DGESC2 */
} /* dgesc2_ */
/*! return the index of element having the largest absolute value
in 0-based numbering system */
inline long idamax(const _dcovector& vec)
{ VERBOSE_REPORT;
long i( idamax_(vec.l, vec.array, 1) -1 );
vec.destroy();
return i;
}
