void dngeres(int *n, int *nev, int *aptr, int *aind,
double *avals, int *bptr, int *bind, double *bvals,
double *dr, double *di, double *z, int *ldz,
double *res)
{
int i, first=1, ione=1, j;
double md, rnrm, *ax, *bx;
ax = (double*)malloc((*n)*sizeof(double));
bx = (double*)malloc((*n)*sizeof(double));
for (i = 0; i<(*nev); i++) {
if (di[i] == 0.0) {
dmvm_(n, avals, aind, aptr, &z(1,i+1), ax, &ione);
dmvm_(n, bvals, bind, bptr, &z(1,i+1), bx, &ione);
md = -dr[i];
daxpy_(n, &md, bx, &ione, ax, &ione);
res[i] = dnrm2_(n, ax, &ione);
res[i] = res[i]/abs(dr[i]);
}
else if (first) {
dmvm_(n, avals, aind, aptr, &z(1,i+1), ax, &ione);
dmvm_(n, bvals, bind, bptr, &z(1,i+1), bx, &ione);
md = -dr[i];
daxpy_(n, &md, bx, &ione, ax, &ione);
dmvm_(n, bvals, bind, bptr, &z(1,i+2), bx, &ione);
daxpy_(n, &di[i], bx, &ione, ax, &ione);
rnrm = dnrm2_(n, ax, &ione);
res[i] = rnrm*rnrm;
dmvm_(n, avals, aind, aptr, &z(1,i+2), ax, &ione);
dmvm_(n, bvals, bind, bptr, &z(1,i+2), bx, &ione);
md = -dr[i];
daxpy_(n, &md, bx, &ione, ax, &ione);
dmvm_(n, bvals, bind, bptr, &z(1,i+1), bx, &ione);
md = -di[i];
daxpy_(n, &md, bx, &ione, ax, &ione);
rnrm = dnrm2_(n, ax, &ione);
res[i] = dlapy2_(&res[i], &rnrm);
res[i] = res[i]/dlapy2_(&dr[i],&di[i]);
res[i+1] = res[i];
first = 0;
}
else {
first = 1;
}
}
free(ax);
free(bx);
}
/* Subroutine */ int dhseqr_(char *job, char *compz, integer *n, integer *ilo,
integer *ihi, doublereal *h, integer *ldh, doublereal *wr,
doublereal *wi, doublereal *z, integer *ldz, doublereal *work,
integer *lwork, integer *info)
{
/* -- LAPACK routine (version 2.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
September 30, 1994
Purpose
=======
DHSEQR computes the eigenvalues of a real upper Hessenberg matrix H
and, optionally, the matrices T and Z from the Schur decomposition
H = Z T Z**T, where T is an upper quasi-triangular matrix (the Schur
form), and Z is the orthogonal matrix of Schur vectors.
Optionally Z may be postmultiplied into an input orthogonal matrix Q,
so that this routine can give the Schur factorization of a matrix A
which has been reduced to the Hessenberg form H by the orthogonal
matrix Q: A = Q*H*Q**T = (QZ)*T*(QZ)**T.
Arguments
=========
JOB (input) CHARACTER*1
= 'E': compute eigenvalues only;
= 'S': compute eigenvalues and the Schur form T.
COMPZ (input) CHARACTER*1
= 'N': no Schur vectors are computed;
= 'I': Z is initialized to the unit matrix and the matrix Z
of Schur vectors of H is returned;
= 'V': Z must contain an orthogonal matrix Q on entry, and
the product Q*Z is returned.
N (input) INTEGER
The order of the matrix H. N >= 0.
ILO (input) INTEGER
IHI (input) INTEGER
It is assumed that H is already upper triangular in rows
and columns 1:ILO-1 and IHI+1:N. ILO and IHI are normally
set by a previous call to DGEBAL, and then passed to SGEHRD
when the matrix output by DGEBAL is reduced to Hessenberg
form. Otherwise ILO and IHI should be set to 1 and N
respectively.
1 <= ILO <= IHI <= N, if N > 0; ILO=1 and IHI=0, if N=0.
H (input/output) DOUBLE PRECISION array, dimension (LDH,N)
On entry, the upper Hessenberg matrix H.
On exit, if JOB = 'S', H contains the upper quasi-triangular
matrix T from the Schur decomposition (the Schur form);
2-by-2 diagonal blocks (corresponding to complex conjugate
pairs of eigenvalues) are returned in standard form, with
H(i,i) = H(i+1,i+1) and H(i+1,i)*H(i,i+1) < 0. If JOB = 'E',
the contents of H are unspecified on exit.
LDH (input) INTEGER
The leading dimension of the array H. LDH >= max(1,N).
WR (output) DOUBLE PRECISION array, dimension (N)
WI (output) DOUBLE PRECISION array, dimension (N)
The real and imaginary parts, respectively, of the computed
eigenvalues. If two eigenvalues are computed as a complex
conjugate pair, they are stored in consecutive elements of
WR and WI, say the i-th and (i+1)th, with WI(i) > 0 and
WI(i+1) < 0. If JOB = 'S', the eigenvalues are stored in the
same order as on the diagonal of the Schur form returned in
H, with WR(i) = H(i,i) and, if H(i:i+1,i:i+1) is a 2-by-2
diagonal block, WI(i) = sqrt(H(i+1,i)*H(i,i+1)) and
WI(i+1) = -WI(i).
Z (input/output) DOUBLE PRECISION array, dimension (LDZ,N)
If COMPZ = 'N': Z is not referenced.
If COMPZ = 'I': on entry, Z need not be set, and on exit, Z
contains the orthogonal matrix Z of the Schur vectors of H.
If COMPZ = 'V': on entry Z must contain an N-by-N matrix Q,
which is assumed to be equal to the unit matrix except for
the submatrix Z(ILO:IHI,ILO:IHI); on exit Z contains Q*Z.
Normally Q is the orthogonal matrix generated by DORGHR after
the call to DGEHRD which formed the Hessenberg matrix H.
LDZ (input) INTEGER
The leading dimension of the array Z.
LDZ >= max(1,N) if COMPZ = 'I' or 'V'; LDZ >= 1 otherwise.
WORK (workspace) DOUBLE PRECISION array, dimension (N)
LWORK (input) INTEGER
This argument is currently redundant.
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
> 0: if INFO = i, DHSEQR failed to compute all of the
eigenvalues in a total of 30*(IHI-ILO+1) iterations;
elements 1:ilo-1 and i+1:n of WR and WI contain those
eigenvalues which have been successfully computed.
=====================================================================
Decode and test the input parameters
Parameter adjustments
Function Body */
/* Table of constant values */
static doublereal c_b9 = 0.;
static doublereal c_b10 = 1.;
static integer c__4 = 4;
static integer c_n1 = -1;
static integer c__2 = 2;
static integer c__8 = 8;
static integer c__15 = 15;
static logical c_false = FALSE_;
static integer c__1 = 1;
/* System generated locals */
address a__1[2];
integer h_dim1, h_offset, z_dim1, z_offset, i__1, i__2, i__3[2], i__4,
i__5;
doublereal d__1, d__2;
char ch__1[2];
/* Builtin functions
Subroutine */ int s_cat(char *, char **, integer *, integer *, ftnlen);
/* Local variables */
static integer maxb;
static doublereal absw;
static integer ierr;
static doublereal unfl, temp, ovfl;
static integer i, j, k, l;
static doublereal s[225] /* was [15][15] */, v[16];
extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
integer *);
extern logical lsame_(char *, char *);
extern /* Subroutine */ int dgemv_(char *, integer *, integer *,
doublereal *, doublereal *, integer *, doublereal *, integer *,
doublereal *, doublereal *, integer *);
static integer itemp;
extern /* Subroutine */ int dcopy_(integer *, doublereal *, integer *,
doublereal *, integer *);
static integer i1, i2;
static logical initz, wantt, wantz;
extern doublereal dlapy2_(doublereal *, doublereal *);
extern /* Subroutine */ int dlabad_(doublereal *, doublereal *);
static integer ii, nh;
extern doublereal dlamch_(char *);
extern /* Subroutine */ int dlarfg_(integer *, doublereal *, doublereal *,
integer *, doublereal *);
static integer nr, ns;
extern integer idamax_(integer *, doublereal *, integer *);
static integer nv;
extern doublereal dlanhs_(char *, integer *, doublereal *, integer *,
doublereal *);
extern /* Subroutine */ int dlahqr_(logical *, logical *, integer *,
integer *, integer *, doublereal *, integer *, doublereal *,
doublereal *, integer *, integer *, doublereal *, integer *,
integer *);
static doublereal vv[16];
extern /* Subroutine */ int dlacpy_(char *, integer *, integer *,
doublereal *, integer *, doublereal *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *, ftnlen, ftnlen);
extern /* Subroutine */ int dlaset_(char *, integer *, integer *,
doublereal *, doublereal *, doublereal *, integer *),
dlarfx_(char *, integer *, integer *, doublereal *, doublereal *,
doublereal *, integer *, doublereal *), xerbla_(char *,
integer *);
static doublereal smlnum;
static integer itn;
static doublereal tau;
static integer its;
static doublereal ulp, tst1;
#define S(I) s[(I)]
#define WAS(I) was[(I)]
#define V(I) v[(I)]
#define VV(I) vv[(I)]
#define WR(I) wr[(I)-1]
#define WI(I) wi[(I)-1]
#define WORK(I) work[(I)-1]
#define H(I,J) h[(I)-1 + ((J)-1)* ( *ldh)]
#define Z(I,J) z[(I)-1 + ((J)-1)* ( *ldz)]
wantt = lsame_(job, "S");
initz = lsame_(compz, "I");
wantz = initz || lsame_(compz, "V");
*info = 0;
if (! lsame_(job, "E") && ! wantt) {
*info = -1;
} else if (! lsame_(compz, "N") && ! wantz) {
*info = -2;
} else if (*n < 0) {
*info = -3;
} else if (*ilo < 1 || *ilo > max(1,*n)) {
*info = -4;
} else if (*ihi < min(*ilo,*n) || *ihi > *n) {
*info = -5;
} else if (*ldh < max(1,*n)) {
*info = -7;
} else if (*ldz < 1 || wantz && *ldz < max(1,*n)) {
*info = -11;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("DHSEQR", &i__1);
return 0;
}
/* Initialize Z, if necessary */
if (initz) {
dlaset_("Full", n, n, &c_b9, &c_b10, &Z(1,1), ldz);
}
/* Store the eigenvalues isolated by DGEBAL. */
i__1 = *ilo - 1;
for (i = 1; i <= *ilo-1; ++i) {
WR(i) = H(i,i);
WI(i) = 0.;
/* L10: */
}
i__1 = *n;
for (i = *ihi + 1; i <= *n; ++i) {
WR(i) = H(i,i);
WI(i) = 0.;
/* L20: */
}
/* Quick return if possible. */
if (*n == 0) {
return 0;
}
if (*ilo == *ihi) {
WR(*ilo) = H(*ilo,*ilo);
WI(*ilo) = 0.;
return 0;
}
/* Set rows and columns ILO to IHI to zero below the first
subdiagonal. */
i__1 = *ihi - 2;
for (j = *ilo; j <= *ihi-2; ++j) {
i__2 = *n;
for (i = j + 2; i <= *n; ++i) {
H(i,j) = 0.;
/* L30: */
}
/* L40: */
}
nh = *ihi - *ilo + 1;
/* Determine the order of the multi-shift QR algorithm to be used.
Writing concatenation */
i__3[0] = 1, a__1[0] = job;
i__3[1] = 1, a__1[1] = compz;
s_cat(ch__1, a__1, i__3, &c__2, 2L);
ns = ilaenv_(&c__4, "DHSEQR", ch__1, n, ilo, ihi, &c_n1, 6L, 2L);
/* Writing concatenation */
i__3[0] = 1, a__1[0] = job;
i__3[1] = 1, a__1[1] = compz;
s_cat(ch__1, a__1, i__3, &c__2, 2L);
maxb = ilaenv_(&c__8, "DHSEQR", ch__1, n, ilo, ihi, &c_n1, 6L, 2L);
if (ns <= 2 || ns > nh || maxb >= nh) {
/* Use the standard double-shift algorithm */
dlahqr_(&wantt, &wantz, n, ilo, ihi, &H(1,1), ldh, &WR(1), &WI(1)
, ilo, ihi, &Z(1,1), ldz, info);
return 0;
}
maxb = max(3,maxb);
/* Computing MIN */
i__1 = min(ns,maxb);
ns = min(i__1,15);
/* Now 2 < NS <= MAXB < NH.
Set machine-dependent constants for the stopping criterion.
If norm(H) <= sqrt(OVFL), overflow should not occur. */
unfl = dlamch_("Safe minimum");
ovfl = 1. / unfl;
dlabad_(&unfl, &ovfl);
ulp = dlamch_("Precision");
smlnum = unfl * (nh / ulp);
/* I1 and I2 are the indices of the first row and last column of H
to which transformations must be applied. If eigenvalues only are
being computed, I1 and I2 are set inside the main loop. */
if (wantt) {
i1 = 1;
i2 = *n;
}
/* ITN is the total number of multiple-shift QR iterations allowed. */
itn = nh * 30;
/* The main loop begins here. I is the loop index and decreases from
IHI to ILO in steps of at most MAXB. Each iteration of the loop
works with the active submatrix in rows and columns L to I.
Eigenvalues I+1 to IHI have already converged. Either L = ILO or
H(L,L-1) is negligible so that the matrix splits. */
i = *ihi;
L50:
l = *ilo;
if (i < *ilo) {
goto L170;
}
/* Perform multiple-shift QR iterations on rows and columns ILO to I
until a submatrix of order at most MAXB splits off at the bottom
because a subdiagonal element has become negligible. */
i__1 = itn;
for (its = 0; its <= itn; ++its) {
/* Look for a single small subdiagonal element. */
i__2 = l + 1;
for (k = i; k >= l+1; --k) {
tst1 = (d__1 = H(k-1,k-1), abs(d__1)) + (d__2 =
H(k,k), abs(d__2));
if (tst1 == 0.) {
i__4 = i - l + 1;
tst1 = dlanhs_("1", &i__4, &H(l,l), ldh, &WORK(1));
}
/* Computing MAX */
d__2 = ulp * tst1;
if ((d__1 = H(k,k-1), abs(d__1)) <= max(d__2,
smlnum)) {
goto L70;
}
/* L60: */
}
L70:
l = k;
if (l > *ilo) {
/* H(L,L-1) is negligible. */
H(l,l-1) = 0.;
}
/* Exit from loop if a submatrix of order <= MAXB has split off
. */
if (l >= i - maxb + 1) {
goto L160;
}
/* Now the active submatrix is in rows and columns L to I. If
eigenvalues only are being computed, only the active submatr
ix
need be transformed. */
if (! wantt) {
i1 = l;
i2 = i;
}
if (its == 20 || its == 30) {
/* Exceptional shifts. */
i__2 = i;
for (ii = i - ns + 1; ii <= i; ++ii) {
WR(ii) = ((d__1 = H(ii,ii-1), abs(d__1)) + (
d__2 = H(ii,ii), abs(d__2))) * 1.5;
WI(ii) = 0.;
/* L80: */
}
} else {
/* Use eigenvalues of trailing submatrix of order NS as
shifts. */
dlacpy_("Full", &ns, &ns, &H(i-ns+1,i-ns+1),
ldh, s, &c__15);
dlahqr_(&c_false, &c_false, &ns, &c__1, &ns, s, &c__15, &WR(i -
ns + 1), &WI(i - ns + 1), &c__1, &ns, &Z(1,1), ldz, &
ierr);
if (ierr > 0) {
/* If DLAHQR failed to compute all NS eigenvalues
, use the
unconverged diagonal elements as the remaining
shifts. */
i__2 = ierr;
for (ii = 1; ii <= ierr; ++ii) {
WR(i - ns + ii) = S(ii + ii * 15 - 16);
WI(i - ns + ii) = 0.;
/* L90: */
}
}
}
/* Form the first column of (G-w(1)) (G-w(2)) . . . (G-w(ns))
where G is the Hessenberg submatrix H(L:I,L:I) and w is
the vector of shifts (stored in WR and WI). The result is
stored in the local array V. */
V(0) = 1.;
i__2 = ns + 1;
for (ii = 2; ii <= ns+1; ++ii) {
V(ii - 1) = 0.;
/* L100: */
}
nv = 1;
i__2 = i;
for (j = i - ns + 1; j <= i; ++j) {
if (WI(j) >= 0.) {
if (WI(j) == 0.) {
/* real shift */
i__4 = nv + 1;
dcopy_(&i__4, v, &c__1, vv, &c__1);
i__4 = nv + 1;
d__1 = -WR(j);
dgemv_("No transpose", &i__4, &nv, &c_b10, &H(l,l), ldh, vv, &c__1, &d__1, v, &c__1);
++nv;
} else if (WI(j) > 0.) {
/* complex conjugate pair of shifts */
i__4 = nv + 1;
dcopy_(&i__4, v, &c__1, vv, &c__1);
i__4 = nv + 1;
d__1 = WR(j) * -2.;
dgemv_("No transpose", &i__4, &nv, &c_b10, &H(l,l), ldh, v, &c__1, &d__1, vv, &c__1);
i__4 = nv + 1;
itemp = idamax_(&i__4, vv, &c__1);
/* Computing MAX */
d__2 = (d__1 = VV(itemp - 1), abs(d__1));
temp = 1. / max(d__2,smlnum);
i__4 = nv + 1;
dscal_(&i__4, &temp, vv, &c__1);
absw = dlapy2_(&WR(j), &WI(j));
temp = temp * absw * absw;
i__4 = nv + 2;
i__5 = nv + 1;
dgemv_("No transpose", &i__4, &i__5, &c_b10, &H(l,l), ldh, vv, &c__1, &temp, v, &c__1);
nv += 2;
}
/* Scale V(1:NV) so that max(abs(V(i))) = 1. If V
is zero,
reset it to the unit vector. */
itemp = idamax_(&nv, v, &c__1);
temp = (d__1 = V(itemp - 1), abs(d__1));
if (temp == 0.) {
V(0) = 1.;
i__4 = nv;
for (ii = 2; ii <= nv; ++ii) {
V(ii - 1) = 0.;
/* L110: */
}
} else {
temp = max(temp,smlnum);
d__1 = 1. / temp;
dscal_(&nv, &d__1, v, &c__1);
}
}
/* L120: */
}
/* Multiple-shift QR step */
i__2 = i - 1;
for (k = l; k <= i-1; ++k) {
/* The first iteration of this loop determines a reflect
ion G
from the vector V and applies it from left and right
to H,
thus creating a nonzero bulge below the subdiagonal.
Each subsequent iteration determines a reflection G t
o
restore the Hessenberg form in the (K-1)th column, an
d thus
chases the bulge one step toward the bottom of the ac
tive
submatrix. NR is the order of G.
Computing MIN */
i__4 = ns + 1, i__5 = i - k + 1;
nr = min(i__4,i__5);
if (k > l) {
dcopy_(&nr, &H(k,k-1), &c__1, v, &c__1);
}
dlarfg_(&nr, v, &V(1), &c__1, &tau);
if (k > l) {
H(k,k-1) = V(0);
i__4 = i;
for (ii = k + 1; ii <= i; ++ii) {
H(ii,k-1) = 0.;
/* L130: */
}
}
V(0) = 1.;
/* Apply G from the left to transform the rows of the ma
trix in
columns K to I2. */
i__4 = i2 - k + 1;
dlarfx_("Left", &nr, &i__4, v, &tau, &H(k,k), ldh, &
WORK(1));
/* Apply G from the right to transform the columns of th
e
matrix in rows I1 to min(K+NR,I).
Computing MIN */
i__5 = k + nr;
i__4 = min(i__5,i) - i1 + 1;
dlarfx_("Right", &i__4, &nr, v, &tau, &H(i1,k), ldh, &
WORK(1));
if (wantz) {
/* Accumulate transformations in the matrix Z */
dlarfx_("Right", &nh, &nr, v, &tau, &Z(*ilo,k),
ldz, &WORK(1));
}
/* L140: */
}
/* L150: */
}
/* Failure to converge in remaining number of iterations */
*info = i;
return 0;
L160:
/* A submatrix of order <= MAXB in rows and columns L to I has split
off. Use the double-shift QR algorithm to handle it. */
dlahqr_(&wantt, &wantz, n, &l, &i, &H(1,1), ldh, &WR(1), &WI(1), ilo,
ihi, &Z(1,1), ldz, info);
if (*info > 0) {
return 0;
}
/* Decrement number of remaining iterations, and return to start of
the main loop with a new value of I. */
itn -= its;
i = l - 1;
goto L50;
L170:
return 0;
/* End of DHSEQR */
} /* dhseqr_ */
/* Subroutine */
int zlarfgp_(integer *n, doublecomplex *alpha, doublecomplex *x, integer *incx, doublecomplex *tau)
{
/* System generated locals */
integer i__1, i__2;
doublereal d__1, d__2;
doublecomplex z__1, z__2;
/* Builtin functions */
double d_imag(doublecomplex *), d_sign(doublereal *, doublereal *), z_abs( doublecomplex *);
/* Local variables */
integer j;
doublecomplex savealpha;
integer knt;
doublereal beta, alphi, alphr;
extern /* Subroutine */
int zscal_(integer *, doublecomplex *, doublecomplex *, integer *);
doublereal xnorm;
extern doublereal dlapy2_(doublereal *, doublereal *), dlapy3_(doublereal *, doublereal *, doublereal *), dznrm2_(integer *, doublecomplex * , integer *), dlamch_(char *);
extern /* Subroutine */
int zdscal_(integer *, doublereal *, doublecomplex *, integer *);
doublereal bignum;
extern /* Double Complex */
VOID zladiv_(doublecomplex *, doublecomplex *, doublecomplex *);
doublereal smlnum;
/* -- 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 .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Executable Statements .. */
/* Parameter adjustments */
--x;
/* Function Body */
if (*n <= 0)
{
tau->r = 0., tau->i = 0.;
return 0;
}
i__1 = *n - 1;
xnorm = dznrm2_(&i__1, &x[1], incx);
alphr = alpha->r;
alphi = d_imag(alpha);
if (xnorm == 0.)
{
/* H = [1-alpha/abs(alpha) 0;
0 I], sign chosen so ALPHA >= 0. */
if (alphi == 0.)
{
if (alphr >= 0.)
{
/* When TAU.eq.ZERO, the vector is special-cased to be */
/* all zeros in the application routines. We do not need */
/* to clear it. */
tau->r = 0., tau->i = 0.;
}
else
{
/* However, the application routines rely on explicit */
/* zero checks when TAU.ne.ZERO, and we must clear X. */
tau->r = 2., tau->i = 0.;
i__1 = *n - 1;
for (j = 1;
j <= i__1;
++j)
{
i__2 = (j - 1) * *incx + 1;
x[i__2].r = 0.;
x[i__2].i = 0.; // , expr subst
}
z__1.r = -alpha->r;
z__1.i = -alpha->i; // , expr subst
alpha->r = z__1.r, alpha->i = z__1.i;
}
}
else
{
/* Only "reflecting" the diagonal entry to be real and non-negative. */
xnorm = dlapy2_(&alphr, &alphi);
d__1 = 1. - alphr / xnorm;
d__2 = -alphi / xnorm;
z__1.r = d__1;
z__1.i = d__2; // , expr subst
tau->r = z__1.r, tau->i = z__1.i;
i__1 = *n - 1;
for (j = 1;
j <= i__1;
++j)
{
i__2 = (j - 1) * *incx + 1;
x[i__2].r = 0.;
x[i__2].i = 0.; // , expr subst
}
alpha->r = xnorm, alpha->i = 0.;
}
}
else
{
/* general case */
d__1 = dlapy3_(&alphr, &alphi, &xnorm);
beta = d_sign(&d__1, &alphr);
smlnum = dlamch_("S") / dlamch_("E");
bignum = 1. / smlnum;
knt = 0;
if (abs(beta) < smlnum)
{
/* XNORM, BETA may be inaccurate;
scale X and recompute them */
L10:
++knt;
i__1 = *n - 1;
zdscal_(&i__1, &bignum, &x[1], incx);
beta *= bignum;
alphi *= bignum;
alphr *= bignum;
if (abs(beta) < smlnum)
{
goto L10;
}
/* New BETA is at most 1, at least SMLNUM */
i__1 = *n - 1;
xnorm = dznrm2_(&i__1, &x[1], incx);
z__1.r = alphr;
z__1.i = alphi; // , expr subst
alpha->r = z__1.r, alpha->i = z__1.i;
d__1 = dlapy3_(&alphr, &alphi, &xnorm);
beta = d_sign(&d__1, &alphr);
}
savealpha.r = alpha->r;
savealpha.i = alpha->i; // , expr subst
z__1.r = alpha->r + beta;
z__1.i = alpha->i; // , expr subst
alpha->r = z__1.r, alpha->i = z__1.i;
if (beta < 0.)
{
beta = -beta;
z__2.r = -alpha->r;
z__2.i = -alpha->i; // , expr subst
z__1.r = z__2.r / beta;
z__1.i = z__2.i / beta; // , expr subst
tau->r = z__1.r, tau->i = z__1.i;
}
else
{
alphr = alphi * (alphi / alpha->r);
alphr += xnorm * (xnorm / alpha->r);
d__1 = alphr / beta;
d__2 = -alphi / beta;
z__1.r = d__1;
z__1.i = d__2; // , expr subst
tau->r = z__1.r, tau->i = z__1.i;
d__1 = -alphr;
z__1.r = d__1;
z__1.i = alphi; // , expr subst
alpha->r = z__1.r, alpha->i = z__1.i;
}
zladiv_(&z__1, &c_b5, alpha);
alpha->r = z__1.r, alpha->i = z__1.i;
if (z_abs(tau) <= smlnum)
{
/* In the case where the computed TAU ends up being a denormalized number, */
/* it loses relative accuracy. This is a BIG problem. Solution: flush TAU */
/* to ZERO (or TWO or whatever makes a nonnegative real number for BETA). */
/* (Bug report provided by Pat Quillen from MathWorks on Jul 29, 2009.) */
/* (Thanks Pat. Thanks MathWorks.) */
alphr = savealpha.r;
alphi = d_imag(&savealpha);
if (alphi == 0.)
{
if (alphr >= 0.)
{
tau->r = 0., tau->i = 0.;
}
else
{
tau->r = 2., tau->i = 0.;
i__1 = *n - 1;
for (j = 1;
j <= i__1;
++j)
{
i__2 = (j - 1) * *incx + 1;
x[i__2].r = 0.;
x[i__2].i = 0.; // , expr subst
}
z__1.r = -savealpha.r;
z__1.i = -savealpha.i; // , expr subst
beta = z__1.r;
}
}
else
{
xnorm = dlapy2_(&alphr, &alphi);
d__1 = 1. - alphr / xnorm;
d__2 = -alphi / xnorm;
z__1.r = d__1;
z__1.i = d__2; // , expr subst
tau->r = z__1.r, tau->i = z__1.i;
i__1 = *n - 1;
for (j = 1;
j <= i__1;
++j)
{
i__2 = (j - 1) * *incx + 1;
x[i__2].r = 0.;
x[i__2].i = 0.; // , expr subst
}
beta = xnorm;
}
}
else
{
/* This is the general case. */
i__1 = *n - 1;
zscal_(&i__1, alpha, &x[1], incx);
}
/* If BETA is subnormal, it may lose relative accuracy */
i__1 = knt;
for (j = 1;
j <= i__1;
++j)
{
beta *= smlnum;
/* L20: */
}
alpha->r = beta, alpha->i = 0.;
}
return 0;
/* End of ZLARFGP */
}
/* 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 zgetv0_(integer *ido, char *bmat, integer *itry, logical
*initv, integer *n, integer *j, doublecomplex *v, integer *ldv,
doublecomplex *resid, doublereal *rnorm, integer *ipntr,
doublecomplex *workd, integer *ierr, ftnlen bmat_len)
{
/* Initialized data */
static logical inits = TRUE_;
/* System generated locals */
integer v_dim1, v_offset, i__1, i__2;
doublereal d__1, d__2;
doublecomplex z__1;
/* Local variables */
static real t0, t1, t2, t3;
static integer jj, iter;
static logical orth;
static integer iseed[4], idist;
static doublecomplex cnorm;
extern /* Double Complex */ void zdotc_(doublecomplex *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *);
static logical first;
extern /* Subroutine */ int zgemv_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublecomplex *, doublecomplex *, integer *, ftnlen),
dvout_(integer *, integer *, doublereal *, integer *, char *,
ftnlen), zcopy_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *), zvout_(integer *, integer *,
doublecomplex *, integer *, char *, ftnlen);
extern doublereal dlapy2_(doublereal *, doublereal *), dznrm2_(integer *,
doublecomplex *, integer *);
static doublereal rnorm0;
extern /* Subroutine */ int arscnd_(real *);
static integer msglvl;
extern /* Subroutine */ int zlarnv_(integer *, integer *, integer *,
doublecomplex *);
/* %----------------------------------------------------% */
/* | Include files for debugging and timing information | */
/* %----------------------------------------------------% */
/* \SCCS Information: @(#) */
/* FILE: debug.h SID: 2.3 DATE OF SID: 11/16/95 RELEASE: 2 */
/* %---------------------------------% */
/* | See debug.doc for documentation | */
/* %---------------------------------% */
/* %------------------% */
/* | Scalar Arguments | */
/* %------------------% */
/* %--------------------------------% */
/* | See stat.doc for documentation | */
/* %--------------------------------% */
/* \SCCS Information: @(#) */
/* FILE: stat.h SID: 2.2 DATE OF SID: 11/16/95 RELEASE: 2 */
/* %-----------------% */
/* | Array Arguments | */
/* %-----------------% */
/* %------------% */
/* | Parameters | */
/* %------------% */
/* %------------------------% */
/* | Local Scalars & Arrays | */
/* %------------------------% */
/* %----------------------% */
/* | External Subroutines | */
/* %----------------------% */
/* %--------------------% */
/* | External Functions | */
/* %--------------------% */
/* %-----------------% */
/* | Data Statements | */
/* %-----------------% */
/* Parameter adjustments */
--workd;
--resid;
v_dim1 = *ldv;
v_offset = 1 + v_dim1;
v -= v_offset;
--ipntr;
/* Function Body */
/* %-----------------------% */
/* | Executable Statements | */
/* %-----------------------% */
/* %-----------------------------------% */
/* | Initialize the seed of the LAPACK | */
/* | random number generator | */
/* %-----------------------------------% */
if (inits) {
iseed[0] = 1;
iseed[1] = 3;
iseed[2] = 5;
iseed[3] = 7;
inits = FALSE_;
}
if (*ido == 0) {
/* %-------------------------------% */
/* | Initialize timing statistics | */
/* | & message level for debugging | */
/* %-------------------------------% */
arscnd_(&t0);
msglvl = debug_1.mgetv0;
*ierr = 0;
iter = 0;
first = FALSE_;
orth = FALSE_;
/* %-----------------------------------------------------% */
/* | Possibly generate a random starting vector in RESID | */
/* | Use a LAPACK random number generator used by the | */
/* | matrix generation routines. | */
/* | idist = 1: uniform (0,1) distribution; | */
/* | idist = 2: uniform (-1,1) distribution; | */
/* | idist = 3: normal (0,1) distribution; | */
/* %-----------------------------------------------------% */
if (! (*initv)) {
idist = 2;
zlarnv_(&idist, iseed, n, &resid[1]);
}
/* %----------------------------------------------------------% */
/* | Force the starting vector into the range of OP to handle | */
/* | the generalized problem when B is possibly (singular). | */
/* %----------------------------------------------------------% */
arscnd_(&t2);
if (*(unsigned char *)bmat == 'G') {
++timing_1.nopx;
ipntr[1] = 1;
ipntr[2] = *n + 1;
zcopy_(n, &resid[1], &c__1, &workd[1], &c__1);
*ido = -1;
goto L9000;
}
}
/* %----------------------------------------% */
/* | Back from computing B*(initial-vector) | */
/* %----------------------------------------% */
if (first) {
goto L20;
}
/* %-----------------------------------------------% */
/* | Back from computing B*(orthogonalized-vector) | */
/* %-----------------------------------------------% */
if (orth) {
goto L40;
}
arscnd_(&t3);
timing_1.tmvopx += t3 - t2;
/* %------------------------------------------------------% */
/* | Starting vector is now in the range of OP; r = OP*r; | */
/* | Compute B-norm of starting vector. | */
/* %------------------------------------------------------% */
arscnd_(&t2);
first = TRUE_;
if (*(unsigned char *)bmat == 'G') {
++timing_1.nbx;
zcopy_(n, &workd[*n + 1], &c__1, &resid[1], &c__1);
ipntr[1] = *n + 1;
ipntr[2] = 1;
*ido = 2;
goto L9000;
} else if (*(unsigned char *)bmat == 'I') {
zcopy_(n, &resid[1], &c__1, &workd[1], &c__1);
}
L20:
if (*(unsigned char *)bmat == 'G') {
arscnd_(&t3);
timing_1.tmvbx += t3 - t2;
}
first = FALSE_;
if (*(unsigned char *)bmat == 'G') {
zdotc_(&z__1, n, &resid[1], &c__1, &workd[1], &c__1);
cnorm.r = z__1.r, cnorm.i = z__1.i;
d__1 = cnorm.r;
d__2 = d_imag(&cnorm);
rnorm0 = sqrt(dlapy2_(&d__1, &d__2));
} else if (*(unsigned char *)bmat == 'I') {
rnorm0 = dznrm2_(n, &resid[1], &c__1);
}
*rnorm = rnorm0;
/* %---------------------------------------------% */
/* | Exit if this is the very first Arnoldi step | */
/* %---------------------------------------------% */
if (*j == 1) {
goto L50;
}
/* %---------------------------------------------------------------- */
/* | Otherwise need to B-orthogonalize the starting vector against | */
/* | the current Arnoldi basis using Gram-Schmidt with iter. ref. | */
/* | This is the case where an invariant subspace is encountered | */
/* | in the middle of the Arnoldi factorization. | */
/* | | */
/* | s = V^{T}*B*r; r = r - V*s; | */
/* | | */
/* | Stopping criteria used for iter. ref. is discussed in | */
/* | Parlett's book, page 107 and in Gragg & Reichel TOMS paper. | */
/* %---------------------------------------------------------------% */
orth = TRUE_;
L30:
i__1 = *j - 1;
zgemv_("C", n, &i__1, &c_b1, &v[v_offset], ldv, &workd[1], &c__1, &c_b2, &
workd[*n + 1], &c__1, (ftnlen)1);
i__1 = *j - 1;
z__1.r = -1., z__1.i = -0.;
zgemv_("N", n, &i__1, &z__1, &v[v_offset], ldv, &workd[*n + 1], &c__1, &
c_b1, &resid[1], &c__1, (ftnlen)1);
/* %----------------------------------------------------------% */
/* | Compute the B-norm of the orthogonalized starting vector | */
/* %----------------------------------------------------------% */
arscnd_(&t2);
if (*(unsigned char *)bmat == 'G') {
++timing_1.nbx;
zcopy_(n, &resid[1], &c__1, &workd[*n + 1], &c__1);
ipntr[1] = *n + 1;
ipntr[2] = 1;
*ido = 2;
goto L9000;
} else if (*(unsigned char *)bmat == 'I') {
zcopy_(n, &resid[1], &c__1, &workd[1], &c__1);
}
L40:
if (*(unsigned char *)bmat == 'G') {
arscnd_(&t3);
timing_1.tmvbx += t3 - t2;
}
if (*(unsigned char *)bmat == 'G') {
zdotc_(&z__1, n, &resid[1], &c__1, &workd[1], &c__1);
cnorm.r = z__1.r, cnorm.i = z__1.i;
d__1 = cnorm.r;
d__2 = d_imag(&cnorm);
*rnorm = sqrt(dlapy2_(&d__1, &d__2));
} else if (*(unsigned char *)bmat == 'I') {
*rnorm = dznrm2_(n, &resid[1], &c__1);
}
/* %--------------------------------------% */
/* | Check for further orthogonalization. | */
/* %--------------------------------------% */
if (msglvl > 2) {
dvout_(&debug_1.logfil, &c__1, &rnorm0, &debug_1.ndigit, "_getv0: re"
"-orthonalization ; rnorm0 is", (ftnlen)38);
dvout_(&debug_1.logfil, &c__1, rnorm, &debug_1.ndigit, "_getv0: re-o"
"rthonalization ; rnorm is", (ftnlen)37);
}
if (*rnorm > rnorm0 * .717f) {
goto L50;
}
++iter;
if (iter <= 1) {
/* %-----------------------------------% */
/* | Perform iterative refinement step | */
/* %-----------------------------------% */
rnorm0 = *rnorm;
goto L30;
} else {
/* %------------------------------------% */
/* | Iterative refinement step "failed" | */
/* %------------------------------------% */
i__1 = *n;
for (jj = 1; jj <= i__1; ++jj) {
i__2 = jj;
resid[i__2].r = 0., resid[i__2].i = 0.;
/* L45: */
}
*rnorm = 0.;
*ierr = -1;
}
L50:
if (msglvl > 0) {
dvout_(&debug_1.logfil, &c__1, rnorm, &debug_1.ndigit, "_getv0: B-no"
"rm of initial / restarted starting vector", (ftnlen)53);
}
if (msglvl > 2) {
zvout_(&debug_1.logfil, n, &resid[1], &debug_1.ndigit, "_getv0: init"
"ial / restarted starting vector", (ftnlen)43);
}
*ido = 99;
arscnd_(&t1);
timing_1.tgetv0 += t1 - t0;
L9000:
return 0;
/* %---------------% */
/* | End of zgetv0 | */
/* %---------------% */
} /* zgetv0_ */
/* Subroutine */ int dlasd2_(integer *nl, integer *nr, integer *sqre, integer
*k, doublereal *d__, doublereal *z__, doublereal *alpha, doublereal *
beta, doublereal *u, integer *ldu, doublereal *vt, integer *ldvt,
doublereal *dsigma, doublereal *u2, integer *ldu2, doublereal *vt2,
integer *ldvt2, integer *idxp, integer *idx, integer *idxc, integer *
idxq, integer *coltyp, integer *info)
{
/* System generated locals */
integer u_dim1, u_offset, u2_dim1, u2_offset, vt_dim1, vt_offset,
vt2_dim1, vt2_offset, i__1;
doublereal d__1, d__2;
/* Local variables */
doublereal c__;
integer i__, j, m, n;
doublereal s;
integer k2;
doublereal z1;
integer ct, jp;
doublereal eps, tau, tol;
integer psm[4], nlp1, nlp2, idxi, idxj;
extern /* Subroutine */ int drot_(integer *, doublereal *, integer *,
doublereal *, integer *, doublereal *, doublereal *);
integer ctot[4], idxjp;
extern /* Subroutine */ int dcopy_(integer *, doublereal *, integer *,
doublereal *, integer *);
integer jprev;
extern doublereal dlapy2_(doublereal *, doublereal *), dlamch_(char *);
extern /* Subroutine */ int dlamrg_(integer *, integer *, doublereal *,
integer *, integer *, integer *), dlacpy_(char *, integer *,
integer *, doublereal *, integer *, doublereal *, integer *), dlaset_(char *, integer *, integer *, doublereal *,
doublereal *, doublereal *, integer *), xerbla_(char *,
integer *);
doublereal hlftol;
/* -- LAPACK auxiliary routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* DLASD2 merges the two sets of singular values 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 */
/* singular values 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. */
/* DLASD2 is called from DLASD1. */
/* Arguments */
/* ========= */
/* NL (input) INTEGER */
/* The row dimension of the upper block. NL >= 1. */
/* NR (input) INTEGER */
/* The row dimension of the lower block. NR >= 1. */
/* SQRE (input) INTEGER */
/* = 0: the lower block is an NR-by-NR square matrix. */
/* = 1: the lower block is an NR-by-(NR+1) rectangular matrix. */
/* The bidiagonal matrix has N = NL + NR + 1 rows and */
/* M = N + SQRE >= N columns. */
/* K (output) INTEGER */
/* Contains the dimension of the non-deflated matrix, */
/* This is the order of the related secular equation. 1 <= K <=N. */
/* D (input/output) DOUBLE PRECISION array, dimension(N) */
/* On entry D contains the singular values of the two submatrices */
/* to be combined. On exit D contains the trailing (N-K) updated */
/* singular values (those which were deflated) sorted into */
/* increasing order. */
/* Z (output) DOUBLE PRECISION array, dimension(N) */
/* On exit Z contains the updating row vector in the secular */
/* equation. */
/* ALPHA (input) DOUBLE PRECISION */
/* Contains the diagonal element associated with the added row. */
/* BETA (input) DOUBLE PRECISION */
/* Contains the off-diagonal element associated with the added */
/* row. */
/* U (input/output) DOUBLE PRECISION array, dimension(LDU,N) */
/* On entry U contains the left singular vectors of two */
/* submatrices in the two square blocks with corners at (1,1), */
/* (NL, NL), and (NL+2, NL+2), (N,N). */
/* On exit U contains the trailing (N-K) updated left singular */
/* vectors (those which were deflated) in its last N-K columns. */
/* LDU (input) INTEGER */
/* The leading dimension of the array U. LDU >= N. */
/* VT (input/output) DOUBLE PRECISION array, dimension(LDVT,M) */
/* On entry VT' contains the right singular vectors of two */
/* submatrices in the two square blocks with corners at (1,1), */
/* (NL+1, NL+1), and (NL+2, NL+2), (M,M). */
/* On exit VT' contains the trailing (N-K) updated right singular */
/* vectors (those which were deflated) in its last N-K columns. */
/* In case SQRE =1, the last row of VT spans the right null */
/* space. */
/* LDVT (input) INTEGER */
/* The leading dimension of the array VT. LDVT >= M. */
/* DSIGMA (output) DOUBLE PRECISION array, dimension (N) */
/* Contains a copy of the diagonal elements (K-1 singular values */
/* and one zero) in the secular equation. */
/* U2 (output) DOUBLE PRECISION array, dimension(LDU2,N) */
/* Contains a copy of the first K-1 left singular vectors which */
/* will be used by DLASD3 in a matrix multiply (DGEMM) to solve */
/* for the new left singular vectors. U2 is arranged into four */
/* blocks. The first block contains a column with 1 at NL+1 and */
/* zero everywhere else; the second block contains non-zero */
/* entries only at and above NL; the third contains non-zero */
/* entries only below NL+1; and the fourth is dense. */
/* LDU2 (input) INTEGER */
/* The leading dimension of the array U2. LDU2 >= N. */
/* VT2 (output) DOUBLE PRECISION array, dimension(LDVT2,N) */
/* VT2' contains a copy of the first K right singular vectors */
/* which will be used by DLASD3 in a matrix multiply (DGEMM) to */
/* solve for the new right singular vectors. VT2 is arranged into */
/* three blocks. The first block contains a row that corresponds */
/* to the special 0 diagonal element in SIGMA; the second block */
/* contains non-zeros only at and before NL +1; the third block */
/* contains non-zeros only at and after NL +2. */
/* LDVT2 (input) INTEGER */
/* The leading dimension of the array VT2. LDVT2 >= M. */
/* IDXP (workspace) INTEGER array dimension(N) */
/* This will contain the permutation used to place deflated */
/* values of D at the end of the array. On output IDXP(2:K) */
/* points to the nondeflated D-values and IDXP(K+1:N) */
/* points to the deflated singular values. */
/* IDX (workspace) INTEGER array dimension(N) */
/* This will contain the permutation used to sort the contents of */
/* D into ascending order. */
/* IDXC (output) INTEGER array dimension(N) */
/* This will contain the permutation used to arrange the columns */
/* of the deflated U matrix into three groups: the first group */
/* contains non-zero entries only at and above NL, the second */
/* contains non-zero entries only below NL+2, and the third is */
/* dense. */
/* IDXQ (input/output) INTEGER array dimension(N) */
/* This contains the permutation which separately sorts the two */
/* sub-problems in D into ascending order. Note that entries in */
/* the first hlaf of this permutation must first be moved one */
/* position backward; and entries in the second half */
/* must first have NL+1 added to their values. */
/* COLTYP (workspace/output) INTEGER array dimension(N) */
/* As workspace, this will contain a label which will indicate */
/* which of the following types a column in the U2 matrix or a */
/* row in the VT2 matrix is: */
/* 1 : non-zero in the upper half only */
/* 2 : non-zero in the lower half only */
/* 3 : dense */
/* 4 : deflated */
/* On exit, it is an array of dimension 4, with COLTYP(I) being */
/* the dimension of the I-th type columns. */
/* INFO (output) INTEGER */
/* = 0: successful exit. */
/* < 0: if INFO = -i, the i-th argument had an illegal value. */
/* Further Details */
/* =============== */
/* Based on contributions by */
/* Ming Gu and Huan Ren, Computer Science Division, University of */
/* California at Berkeley, USA */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Arrays .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
--d__;
--z__;
u_dim1 = *ldu;
u_offset = 1 + u_dim1;
u -= u_offset;
vt_dim1 = *ldvt;
vt_offset = 1 + vt_dim1;
vt -= vt_offset;
--dsigma;
u2_dim1 = *ldu2;
u2_offset = 1 + u2_dim1;
u2 -= u2_offset;
vt2_dim1 = *ldvt2;
vt2_offset = 1 + vt2_dim1;
vt2 -= vt2_offset;
--idxp;
--idx;
--idxc;
--idxq;
--coltyp;
/* Function Body */
*info = 0;
if (*nl < 1) {
*info = -1;
} else if (*nr < 1) {
*info = -2;
} else if (*sqre != 1 && *sqre != 0) {
*info = -3;
}
n = *nl + *nr + 1;
m = n + *sqre;
if (*ldu < n) {
*info = -10;
} else if (*ldvt < m) {
*info = -12;
} else if (*ldu2 < n) {
*info = -15;
} else if (*ldvt2 < m) {
*info = -17;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("DLASD2", &i__1);
return 0;
}
nlp1 = *nl + 1;
nlp2 = *nl + 2;
/* Generate the first part of the vector Z; and move the singular */
/* values in the first part of D one position backward. */
z1 = *alpha * vt[nlp1 + nlp1 * vt_dim1];
z__[1] = z1;
for (i__ = *nl; i__ >= 1; --i__) {
z__[i__ + 1] = *alpha * vt[i__ + nlp1 * vt_dim1];
d__[i__ + 1] = d__[i__];
idxq[i__ + 1] = idxq[i__] + 1;
/* L10: */
}
/* Generate the second part of the vector Z. */
i__1 = m;
for (i__ = nlp2; i__ <= i__1; ++i__) {
z__[i__] = *beta * vt[i__ + nlp2 * vt_dim1];
/* L20: */
}
/* Initialize some reference arrays. */
i__1 = nlp1;
for (i__ = 2; i__ <= i__1; ++i__) {
coltyp[i__] = 1;
/* L30: */
}
i__1 = n;
for (i__ = nlp2; i__ <= i__1; ++i__) {
coltyp[i__] = 2;
/* L40: */
}
/* Sort the singular values into increasing order */
i__1 = n;
for (i__ = nlp2; i__ <= i__1; ++i__) {
idxq[i__] += nlp1;
/* L50: */
}
/* DSIGMA, IDXC, IDXC, and the first column of U2 */
/* are used as storage space. */
i__1 = n;
for (i__ = 2; i__ <= i__1; ++i__) {
dsigma[i__] = d__[idxq[i__]];
u2[i__ + u2_dim1] = z__[idxq[i__]];
idxc[i__] = coltyp[idxq[i__]];
/* L60: */
}
dlamrg_(nl, nr, &dsigma[2], &c__1, &c__1, &idx[2]);
i__1 = n;
for (i__ = 2; i__ <= i__1; ++i__) {
idxi = idx[i__] + 1;
d__[i__] = dsigma[idxi];
z__[i__] = u2[idxi + u2_dim1];
coltyp[i__] = idxc[idxi];
/* L70: */
}
/* Calculate the allowable deflation tolerance */
eps = dlamch_("Epsilon");
/* Computing MAX */
d__1 = abs(*alpha), d__2 = abs(*beta);
tol = max(d__1,d__2);
/* Computing MAX */
d__2 = (d__1 = d__[n], abs(d__1));
tol = eps * 8. * max(d__2,tol);
/* There are 2 kinds of deflation -- first a value in the z-vector */
/* is small, second two (or more) singular values are very close */
/* together (their difference is small). */
/* If the value in the z-vector is small, we simply permute the */
/* array so that the corresponding singular value is moved to the */
/* end. */
/* If two values in the D-vector are close, we perform a two-sided */
/* rotation designed to make one of the corresponding z-vector */
/* entries zero, and then permute the array so that the deflated */
/* singular value is moved to the end. */
/* If there are multiple singular values then the problem deflates. */
/* Here the number of equal singular values are found. As each equal */
/* singular value is found, an elementary reflector is computed to */
/* rotate the corresponding singular subspace so that the */
/* corresponding components of Z are zero in this new basis. */
*k = 1;
k2 = n + 1;
i__1 = n;
for (j = 2; j <= i__1; ++j) {
if ((d__1 = z__[j], abs(d__1)) <= tol) {
/* Deflate due to small z component. */
--k2;
idxp[k2] = j;
coltyp[j] = 4;
if (j == n) {
goto L120;
}
} else {
jprev = j;
goto L90;
}
/* L80: */
}
L90:
j = jprev;
L100:
++j;
if (j > n) {
goto L110;
}
if ((d__1 = z__[j], abs(d__1)) <= tol) {
/* Deflate due to small z component. */
--k2;
idxp[k2] = j;
coltyp[j] = 4;
} else {
/* Check if singular values are close enough to allow deflation. */
if ((d__1 = d__[j] - d__[jprev], abs(d__1)) <= tol) {
/* Deflation is possible. */
s = z__[jprev];
c__ = z__[j];
/* Find sqrt(a**2+b**2) without overflow or */
/* destructive underflow. */
tau = dlapy2_(&c__, &s);
c__ /= tau;
s = -s / tau;
z__[j] = tau;
z__[jprev] = 0.;
/* Apply back the Givens rotation to the left and right */
/* singular vector matrices. */
idxjp = idxq[idx[jprev] + 1];
idxj = idxq[idx[j] + 1];
if (idxjp <= nlp1) {
--idxjp;
}
if (idxj <= nlp1) {
--idxj;
}
drot_(&n, &u[idxjp * u_dim1 + 1], &c__1, &u[idxj * u_dim1 + 1], &
c__1, &c__, &s);
drot_(&m, &vt[idxjp + vt_dim1], ldvt, &vt[idxj + vt_dim1], ldvt, &
c__, &s);
if (coltyp[j] != coltyp[jprev]) {
coltyp[j] = 3;
}
coltyp[jprev] = 4;
--k2;
idxp[k2] = jprev;
jprev = j;
} else {
++(*k);
u2[*k + u2_dim1] = z__[jprev];
dsigma[*k] = d__[jprev];
idxp[*k] = jprev;
jprev = j;
}
}
goto L100;
L110:
/* Record the last singular value. */
++(*k);
u2[*k + u2_dim1] = z__[jprev];
dsigma[*k] = d__[jprev];
idxp[*k] = jprev;
L120:
/* Count up the total number of the various types of columns, then */
/* form a permutation which positions the four column types into */
/* four groups of uniform structure (although one or more of these */
/* groups may be empty). */
for (j = 1; j <= 4; ++j) {
ctot[j - 1] = 0;
/* L130: */
}
i__1 = n;
for (j = 2; j <= i__1; ++j) {
ct = coltyp[j];
++ctot[ct - 1];
/* L140: */
}
/* PSM(*) = Position in SubMatrix (of types 1 through 4) */
psm[0] = 2;
psm[1] = ctot[0] + 2;
psm[2] = psm[1] + ctot[1];
psm[3] = psm[2] + ctot[2];
/* Fill out the IDXC 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, starting from the */
/* second column. This applies similarly to the rows of VT. */
i__1 = n;
for (j = 2; j <= i__1; ++j) {
jp = idxp[j];
ct = coltyp[jp];
idxc[psm[ct - 1]] = j;
++psm[ct - 1];
/* L150: */
}
/* Sort the singular values and corresponding singular vectors into */
/* DSIGMA, U2, and VT2 respectively. The singular values/vectors */
/* which were not deflated go into the first K slots of DSIGMA, U2, */
/* and VT2 respectively, while those which were deflated go into the */
/* last N - K slots, except that the first column/row will be treated */
/* separately. */
i__1 = n;
for (j = 2; j <= i__1; ++j) {
jp = idxp[j];
dsigma[j] = d__[jp];
idxj = idxq[idx[idxp[idxc[j]]] + 1];
if (idxj <= nlp1) {
--idxj;
}
dcopy_(&n, &u[idxj * u_dim1 + 1], &c__1, &u2[j * u2_dim1 + 1], &c__1);
dcopy_(&m, &vt[idxj + vt_dim1], ldvt, &vt2[j + vt2_dim1], ldvt2);
/* L160: */
}
/* Determine DSIGMA(1), DSIGMA(2) and Z(1) */
dsigma[1] = 0.;
hlftol = tol / 2.;
if (abs(dsigma[2]) <= hlftol) {
dsigma[2] = hlftol;
}
if (m > n) {
z__[1] = dlapy2_(&z1, &z__[m]);
if (z__[1] <= tol) {
c__ = 1.;
s = 0.;
z__[1] = tol;
} else {
c__ = z1 / z__[1];
s = z__[m] / z__[1];
}
} else {
if (abs(z1) <= tol) {
z__[1] = tol;
} else {
z__[1] = z1;
}
}
/* Move the rest of the updating row to Z. */
i__1 = *k - 1;
dcopy_(&i__1, &u2[u2_dim1 + 2], &c__1, &z__[2], &c__1);
/* Determine the first column of U2, the first row of VT2 and the */
/* last row of VT. */
dlaset_("A", &n, &c__1, &c_b30, &c_b30, &u2[u2_offset], ldu2);
u2[nlp1 + u2_dim1] = 1.;
if (m > n) {
i__1 = nlp1;
for (i__ = 1; i__ <= i__1; ++i__) {
vt[m + i__ * vt_dim1] = -s * vt[nlp1 + i__ * vt_dim1];
vt2[i__ * vt2_dim1 + 1] = c__ * vt[nlp1 + i__ * vt_dim1];
/* L170: */
}
i__1 = m;
for (i__ = nlp2; i__ <= i__1; ++i__) {
vt2[i__ * vt2_dim1 + 1] = s * vt[m + i__ * vt_dim1];
vt[m + i__ * vt_dim1] = c__ * vt[m + i__ * vt_dim1];
/* L180: */
}
} else {
dcopy_(&m, &vt[nlp1 + vt_dim1], ldvt, &vt2[vt2_dim1 + 1], ldvt2);
}
if (m > n) {
dcopy_(&m, &vt[m + vt_dim1], ldvt, &vt2[m + vt2_dim1], ldvt2);
}
/* The deflated singular values and their corresponding vectors go */
/* into the back of D, U, and V respectively. */
if (n > *k) {
i__1 = n - *k;
dcopy_(&i__1, &dsigma[*k + 1], &c__1, &d__[*k + 1], &c__1);
i__1 = n - *k;
dlacpy_("A", &n, &i__1, &u2[(*k + 1) * u2_dim1 + 1], ldu2, &u[(*k + 1)
* u_dim1 + 1], ldu);
i__1 = n - *k;
dlacpy_("A", &i__1, &m, &vt2[*k + 1 + vt2_dim1], ldvt2, &vt[*k + 1 +
vt_dim1], ldvt);
}
/* Copy CTOT into COLTYP for referencing in DLASD3. */
for (j = 1; j <= 4; ++j) {
coltyp[j] = ctot[j - 1];
/* L190: */
}
return 0;
/* End of DLASD2 */
} /* dlasd2_ */
/* Subroutine */ int zsteqr_(char *compz, integer *n, doublereal *d__,
doublereal *e, doublecomplex *z__, integer *ldz, doublereal *work,
integer *info)
{
/* System generated locals */
integer z_dim1, z_offset, i__1, i__2;
doublereal d__1, d__2;
/* Builtin functions */
double sqrt(doublereal), d_sign(doublereal *, doublereal *);
/* Local variables */
doublereal b, c__, f, g;
integer i__, j, k, l, m;
doublereal p, r__, s;
integer l1, ii, mm, lm1, mm1, nm1;
doublereal rt1, rt2, eps;
integer lsv;
doublereal tst, eps2;
integer lend, jtot;
extern /* Subroutine */ int dlae2_(doublereal *, doublereal *, doublereal
*, doublereal *, doublereal *);
extern logical lsame_(char *, char *);
doublereal anorm;
extern /* Subroutine */ int zlasr_(char *, char *, char *, integer *,
integer *, doublereal *, doublereal *, doublecomplex *, integer *), zswap_(integer *, doublecomplex *,
integer *, doublecomplex *, integer *), dlaev2_(doublereal *,
doublereal *, doublereal *, doublereal *, doublereal *,
doublereal *, doublereal *);
integer lendm1, lendp1;
extern doublereal dlapy2_(doublereal *, doublereal *), dlamch_(char *);
integer iscale;
extern /* Subroutine */ int dlascl_(char *, integer *, integer *,
doublereal *, doublereal *, integer *, integer *, doublereal *,
integer *, integer *);
doublereal safmin;
extern /* Subroutine */ int dlartg_(doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *);
doublereal safmax;
extern /* Subroutine */ int xerbla_(char *, integer *);
extern doublereal dlanst_(char *, integer *, doublereal *, doublereal *);
extern /* Subroutine */ int dlasrt_(char *, integer *, doublereal *,
integer *);
integer lendsv;
doublereal ssfmin;
integer nmaxit, icompz;
doublereal ssfmax;
extern /* Subroutine */ int zlaset_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, doublecomplex *, integer *);
/* -- LAPACK routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* ZSTEQR computes all eigenvalues and, optionally, eigenvectors of a */
/* symmetric tridiagonal matrix using the implicit QL or QR method. */
/* The eigenvectors of a full or band complex Hermitian matrix can also */
/* be found if ZHETRD or ZHPTRD or ZHBTRD has been used to reduce this */
/* matrix to tridiagonal form. */
/* Arguments */
/* ========= */
/* COMPZ (input) CHARACTER*1 */
/* = 'N': Compute eigenvalues only. */
/* = 'V': Compute eigenvalues and eigenvectors of the original */
/* Hermitian matrix. On entry, Z must contain the */
/* unitary matrix used to reduce the original matrix */
/* to tridiagonal form. */
/* = 'I': Compute eigenvalues and eigenvectors of the */
/* tridiagonal matrix. Z is initialized to the identity */
/* matrix. */
/* N (input) INTEGER */
/* The order of the matrix. N >= 0. */
/* D (input/output) DOUBLE PRECISION array, dimension (N) */
/* On entry, the diagonal elements of the tridiagonal matrix. */
/* On exit, if INFO = 0, the eigenvalues in ascending order. */
/* E (input/output) DOUBLE PRECISION array, dimension (N-1) */
/* On entry, the (n-1) subdiagonal elements of the tridiagonal */
/* matrix. */
/* On exit, E has been destroyed. */
/* Z (input/output) COMPLEX*16 array, dimension (LDZ, N) */
/* On entry, if COMPZ = 'V', then Z contains the unitary */
/* matrix used in the reduction to tridiagonal form. */
/* On exit, if INFO = 0, then if COMPZ = 'V', Z contains the */
/* orthonormal eigenvectors of the original Hermitian matrix, */
/* and if COMPZ = 'I', Z contains the orthonormal eigenvectors */
/* of the symmetric tridiagonal matrix. */
/* If COMPZ = 'N', then Z is not referenced. */
/* LDZ (input) INTEGER */
/* The leading dimension of the array Z. LDZ >= 1, and if */
/* eigenvectors are desired, then LDZ >= max(1,N). */
/* WORK (workspace) DOUBLE PRECISION array, dimension (max(1,2*N-2)) */
/* If COMPZ = 'N', then WORK is not referenced. */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* > 0: the algorithm has failed to find all the eigenvalues in */
/* a total of 30*N iterations; if INFO = i, then i */
/* elements of E have not converged to zero; on exit, D */
/* and E contain the elements of a symmetric tridiagonal */
/* matrix which is unitarily similar to the original */
/* matrix. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
--d__;
--e;
z_dim1 = *ldz;
z_offset = 1 + z_dim1;
z__ -= z_offset;
--work;
/* Function Body */
*info = 0;
if (lsame_(compz, "N")) {
icompz = 0;
} else if (lsame_(compz, "V")) {
icompz = 1;
} else if (lsame_(compz, "I")) {
icompz = 2;
} else {
icompz = -1;
}
if (icompz < 0) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*ldz < 1 || icompz > 0 && *ldz < max(1,*n)) {
*info = -6;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("ZSTEQR", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
if (*n == 1) {
if (icompz == 2) {
i__1 = z_dim1 + 1;
z__[i__1].r = 1., z__[i__1].i = 0.;
}
return 0;
}
/* Determine the unit roundoff and over/underflow thresholds. */
eps = dlamch_("E");
/* Computing 2nd power */
d__1 = eps;
eps2 = d__1 * d__1;
safmin = dlamch_("S");
safmax = 1. / safmin;
ssfmax = sqrt(safmax) / 3.;
ssfmin = sqrt(safmin) / eps2;
/* Compute the eigenvalues and eigenvectors of the tridiagonal */
/* matrix. */
if (icompz == 2) {
zlaset_("Full", n, n, &c_b1, &c_b2, &z__[z_offset], ldz);
}
nmaxit = *n * 30;
jtot = 0;
/* Determine where the matrix splits and choose QL or QR iteration */
/* for each block, according to whether top or bottom diagonal */
/* element is smaller. */
l1 = 1;
nm1 = *n - 1;
L10:
if (l1 > *n) {
goto L160;
}
if (l1 > 1) {
e[l1 - 1] = 0.;
}
if (l1 <= nm1) {
i__1 = nm1;
for (m = l1; m <= i__1; ++m) {
tst = (d__1 = e[m], abs(d__1));
if (tst == 0.) {
goto L30;
}
if (tst <= sqrt((d__1 = d__[m], abs(d__1))) * sqrt((d__2 = d__[m
+ 1], abs(d__2))) * eps) {
e[m] = 0.;
goto L30;
}
/* L20: */
}
}
m = *n;
L30:
l = l1;
lsv = l;
lend = m;
lendsv = lend;
l1 = m + 1;
if (lend == l) {
goto L10;
}
/* Scale submatrix in rows and columns L to LEND */
i__1 = lend - l + 1;
anorm = dlanst_("I", &i__1, &d__[l], &e[l]);
iscale = 0;
if (anorm == 0.) {
goto L10;
}
if (anorm > ssfmax) {
iscale = 1;
i__1 = lend - l + 1;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmax, &i__1, &c__1, &d__[l], n,
info);
i__1 = lend - l;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmax, &i__1, &c__1, &e[l], n,
info);
} else if (anorm < ssfmin) {
iscale = 2;
i__1 = lend - l + 1;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &d__[l], n,
info);
i__1 = lend - l;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &e[l], n,
info);
}
/* Choose between QL and QR iteration */
if ((d__1 = d__[lend], abs(d__1)) < (d__2 = d__[l], abs(d__2))) {
lend = lsv;
l = lendsv;
}
if (lend > l) {
/* QL Iteration */
/* Look for small subdiagonal element. */
L40:
if (l != lend) {
lendm1 = lend - 1;
i__1 = lendm1;
for (m = l; m <= i__1; ++m) {
/* Computing 2nd power */
d__2 = (d__1 = e[m], abs(d__1));
tst = d__2 * d__2;
if (tst <= eps2 * (d__1 = d__[m], abs(d__1)) * (d__2 = d__[m
+ 1], abs(d__2)) + safmin) {
goto L60;
}
/* L50: */
}
}
m = lend;
L60:
if (m < lend) {
e[m] = 0.;
}
p = d__[l];
if (m == l) {
goto L80;
}
/* If remaining matrix is 2-by-2, use DLAE2 or SLAEV2 */
/* to compute its eigensystem. */
if (m == l + 1) {
if (icompz > 0) {
dlaev2_(&d__[l], &e[l], &d__[l + 1], &rt1, &rt2, &c__, &s);
work[l] = c__;
work[*n - 1 + l] = s;
zlasr_("R", "V", "B", n, &c__2, &work[l], &work[*n - 1 + l], &
z__[l * z_dim1 + 1], ldz);
} else {
dlae2_(&d__[l], &e[l], &d__[l + 1], &rt1, &rt2);
}
d__[l] = rt1;
d__[l + 1] = rt2;
e[l] = 0.;
l += 2;
if (l <= lend) {
goto L40;
}
goto L140;
}
if (jtot == nmaxit) {
goto L140;
}
++jtot;
/* Form shift. */
g = (d__[l + 1] - p) / (e[l] * 2.);
r__ = dlapy2_(&g, &c_b41);
g = d__[m] - p + e[l] / (g + d_sign(&r__, &g));
s = 1.;
c__ = 1.;
p = 0.;
/* Inner loop */
mm1 = m - 1;
i__1 = l;
for (i__ = mm1; i__ >= i__1; --i__) {
f = s * e[i__];
b = c__ * e[i__];
dlartg_(&g, &f, &c__, &s, &r__);
if (i__ != m - 1) {
e[i__ + 1] = r__;
}
g = d__[i__ + 1] - p;
r__ = (d__[i__] - g) * s + c__ * 2. * b;
p = s * r__;
d__[i__ + 1] = g + p;
g = c__ * r__ - b;
/* If eigenvectors are desired, then save rotations. */
if (icompz > 0) {
work[i__] = c__;
work[*n - 1 + i__] = -s;
}
/* L70: */
}
/* If eigenvectors are desired, then apply saved rotations. */
if (icompz > 0) {
mm = m - l + 1;
zlasr_("R", "V", "B", n, &mm, &work[l], &work[*n - 1 + l], &z__[l
* z_dim1 + 1], ldz);
}
d__[l] -= p;
e[l] = g;
goto L40;
/* Eigenvalue found. */
L80:
d__[l] = p;
++l;
if (l <= lend) {
goto L40;
}
goto L140;
} else {
/* QR Iteration */
/* Look for small superdiagonal element. */
L90:
if (l != lend) {
lendp1 = lend + 1;
i__1 = lendp1;
for (m = l; m >= i__1; --m) {
/* Computing 2nd power */
d__2 = (d__1 = e[m - 1], abs(d__1));
tst = d__2 * d__2;
if (tst <= eps2 * (d__1 = d__[m], abs(d__1)) * (d__2 = d__[m
- 1], abs(d__2)) + safmin) {
goto L110;
}
/* L100: */
}
}
m = lend;
L110:
if (m > lend) {
e[m - 1] = 0.;
}
p = d__[l];
if (m == l) {
goto L130;
}
/* If remaining matrix is 2-by-2, use DLAE2 or SLAEV2 */
/* to compute its eigensystem. */
if (m == l - 1) {
if (icompz > 0) {
dlaev2_(&d__[l - 1], &e[l - 1], &d__[l], &rt1, &rt2, &c__, &s)
;
work[m] = c__;
work[*n - 1 + m] = s;
zlasr_("R", "V", "F", n, &c__2, &work[m], &work[*n - 1 + m], &
z__[(l - 1) * z_dim1 + 1], ldz);
} else {
dlae2_(&d__[l - 1], &e[l - 1], &d__[l], &rt1, &rt2);
}
d__[l - 1] = rt1;
d__[l] = rt2;
e[l - 1] = 0.;
l += -2;
if (l >= lend) {
goto L90;
}
goto L140;
}
if (jtot == nmaxit) {
goto L140;
}
++jtot;
/* Form shift. */
g = (d__[l - 1] - p) / (e[l - 1] * 2.);
r__ = dlapy2_(&g, &c_b41);
g = d__[m] - p + e[l - 1] / (g + d_sign(&r__, &g));
s = 1.;
c__ = 1.;
p = 0.;
/* Inner loop */
lm1 = l - 1;
i__1 = lm1;
for (i__ = m; i__ <= i__1; ++i__) {
f = s * e[i__];
b = c__ * e[i__];
dlartg_(&g, &f, &c__, &s, &r__);
if (i__ != m) {
e[i__ - 1] = r__;
}
g = d__[i__] - p;
r__ = (d__[i__ + 1] - g) * s + c__ * 2. * b;
p = s * r__;
d__[i__] = g + p;
g = c__ * r__ - b;
/* If eigenvectors are desired, then save rotations. */
if (icompz > 0) {
work[i__] = c__;
work[*n - 1 + i__] = s;
}
/* L120: */
}
/* If eigenvectors are desired, then apply saved rotations. */
if (icompz > 0) {
mm = l - m + 1;
zlasr_("R", "V", "F", n, &mm, &work[m], &work[*n - 1 + m], &z__[m
* z_dim1 + 1], ldz);
}
d__[l] -= p;
e[lm1] = g;
goto L90;
/* Eigenvalue found. */
L130:
d__[l] = p;
--l;
if (l >= lend) {
goto L90;
}
goto L140;
}
/* Undo scaling if necessary */
L140:
if (iscale == 1) {
i__1 = lendsv - lsv + 1;
dlascl_("G", &c__0, &c__0, &ssfmax, &anorm, &i__1, &c__1, &d__[lsv],
n, info);
i__1 = lendsv - lsv;
dlascl_("G", &c__0, &c__0, &ssfmax, &anorm, &i__1, &c__1, &e[lsv], n,
info);
} else if (iscale == 2) {
i__1 = lendsv - lsv + 1;
dlascl_("G", &c__0, &c__0, &ssfmin, &anorm, &i__1, &c__1, &d__[lsv],
n, info);
i__1 = lendsv - lsv;
dlascl_("G", &c__0, &c__0, &ssfmin, &anorm, &i__1, &c__1, &e[lsv], n,
info);
}
/* Check for no convergence to an eigenvalue after a total */
/* of N*MAXIT iterations. */
if (jtot == nmaxit) {
i__1 = *n - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
if (e[i__] != 0.) {
++(*info);
}
/* L150: */
}
return 0;
}
goto L10;
/* Order eigenvalues and eigenvectors. */
L160:
if (icompz == 0) {
/* Use Quick Sort */
dlasrt_("I", n, &d__[1], info);
} else {
/* Use Selection Sort to minimize swaps of eigenvectors */
i__1 = *n;
for (ii = 2; ii <= i__1; ++ii) {
i__ = ii - 1;
k = i__;
p = d__[i__];
i__2 = *n;
for (j = ii; j <= i__2; ++j) {
if (d__[j] < p) {
k = j;
p = d__[j];
}
/* L170: */
}
if (k != i__) {
d__[k] = d__[i__];
d__[i__] = p;
zswap_(n, &z__[i__ * z_dim1 + 1], &c__1, &z__[k * z_dim1 + 1],
&c__1);
}
/* L180: */
}
}
return 0;
/* End of ZSTEQR */
} /* zsteqr_ */
/* Subroutine */ int dsortc_(char *which, logical *apply, integer *n,
doublereal *xreal, doublereal *ximag, doublereal *y, ftnlen which_len)
{
/* System generated locals */
integer i__1;
doublereal d__1, d__2;
/* Builtin functions */
integer s_cmp(char *, char *, ftnlen, ftnlen);
/* Local variables */
static integer i__, j, igap;
static doublereal temp, temp1, temp2;
extern doublereal dlapy2_(doublereal *, doublereal *);
/* %------------------% */
/* | Scalar Arguments | */
/* %------------------% */
/* %-----------------% */
/* | Array Arguments | */
/* %-----------------% */
/* %---------------% */
/* | Local Scalars | */
/* %---------------% */
/* %--------------------% */
/* | External Functions | */
/* %--------------------% */
/* %-----------------------% */
/* | Executable Statements | */
/* %-----------------------% */
igap = *n / 2;
if (s_cmp(which, "LM", (ftnlen)2, (ftnlen)2) == 0) {
/* %------------------------------------------------------% */
/* | Sort XREAL,XIMAG into increasing order of magnitude. | */
/* %------------------------------------------------------% */
L10:
if (igap == 0) {
goto L9000;
}
i__1 = *n - 1;
for (i__ = igap; i__ <= i__1; ++i__) {
j = i__ - igap;
L20:
if (j < 0) {
goto L30;
}
temp1 = dlapy2_(&xreal[j], &ximag[j]);
temp2 = dlapy2_(&xreal[j + igap], &ximag[j + igap]);
if (temp1 > temp2) {
temp = xreal[j];
xreal[j] = xreal[j + igap];
xreal[j + igap] = temp;
temp = ximag[j];
ximag[j] = ximag[j + igap];
ximag[j + igap] = temp;
if (*apply) {
temp = y[j];
y[j] = y[j + igap];
y[j + igap] = temp;
}
} else {
goto L30;
}
j -= igap;
goto L20;
L30:
;
}
igap /= 2;
goto L10;
} else if (s_cmp(which, "SM", (ftnlen)2, (ftnlen)2) == 0) {
/* %------------------------------------------------------% */
/* | Sort XREAL,XIMAG into decreasing order of magnitude. | */
/* %------------------------------------------------------% */
L40:
if (igap == 0) {
goto L9000;
}
i__1 = *n - 1;
for (i__ = igap; i__ <= i__1; ++i__) {
j = i__ - igap;
L50:
if (j < 0) {
goto L60;
}
temp1 = dlapy2_(&xreal[j], &ximag[j]);
temp2 = dlapy2_(&xreal[j + igap], &ximag[j + igap]);
if (temp1 < temp2) {
temp = xreal[j];
xreal[j] = xreal[j + igap];
xreal[j + igap] = temp;
temp = ximag[j];
ximag[j] = ximag[j + igap];
ximag[j + igap] = temp;
if (*apply) {
temp = y[j];
y[j] = y[j + igap];
y[j + igap] = temp;
}
} else {
goto L60;
}
j -= igap;
goto L50;
L60:
;
}
igap /= 2;
goto L40;
} else if (s_cmp(which, "LR", (ftnlen)2, (ftnlen)2) == 0) {
/* %------------------------------------------------% */
/* | Sort XREAL into increasing order of algebraic. | */
/* %------------------------------------------------% */
L70:
if (igap == 0) {
goto L9000;
}
i__1 = *n - 1;
for (i__ = igap; i__ <= i__1; ++i__) {
j = i__ - igap;
L80:
if (j < 0) {
goto L90;
}
if (xreal[j] > xreal[j + igap]) {
temp = xreal[j];
xreal[j] = xreal[j + igap];
xreal[j + igap] = temp;
temp = ximag[j];
ximag[j] = ximag[j + igap];
ximag[j + igap] = temp;
if (*apply) {
temp = y[j];
y[j] = y[j + igap];
y[j + igap] = temp;
}
} else {
goto L90;
}
j -= igap;
goto L80;
L90:
;
}
igap /= 2;
goto L70;
} else if (s_cmp(which, "SR", (ftnlen)2, (ftnlen)2) == 0) {
/* %------------------------------------------------% */
/* | Sort XREAL into decreasing order of algebraic. | */
/* %------------------------------------------------% */
L100:
if (igap == 0) {
goto L9000;
}
i__1 = *n - 1;
for (i__ = igap; i__ <= i__1; ++i__) {
j = i__ - igap;
L110:
if (j < 0) {
goto L120;
}
if (xreal[j] < xreal[j + igap]) {
temp = xreal[j];
xreal[j] = xreal[j + igap];
xreal[j + igap] = temp;
temp = ximag[j];
ximag[j] = ximag[j + igap];
ximag[j + igap] = temp;
if (*apply) {
temp = y[j];
y[j] = y[j + igap];
y[j + igap] = temp;
}
} else {
goto L120;
}
j -= igap;
goto L110;
L120:
;
}
igap /= 2;
goto L100;
} else if (s_cmp(which, "LI", (ftnlen)2, (ftnlen)2) == 0) {
/* %------------------------------------------------% */
/* | Sort XIMAG into increasing order of magnitude. | */
/* %------------------------------------------------% */
L130:
if (igap == 0) {
goto L9000;
}
i__1 = *n - 1;
for (i__ = igap; i__ <= i__1; ++i__) {
j = i__ - igap;
L140:
if (j < 0) {
goto L150;
}
if ((d__1 = ximag[j], abs(d__1)) > (d__2 = ximag[j + igap], abs(
d__2))) {
temp = xreal[j];
xreal[j] = xreal[j + igap];
xreal[j + igap] = temp;
temp = ximag[j];
ximag[j] = ximag[j + igap];
ximag[j + igap] = temp;
if (*apply) {
temp = y[j];
y[j] = y[j + igap];
y[j + igap] = temp;
}
} else {
goto L150;
}
j -= igap;
goto L140;
L150:
;
}
igap /= 2;
goto L130;
} else if (s_cmp(which, "SI", (ftnlen)2, (ftnlen)2) == 0) {
/* %------------------------------------------------% */
/* | Sort XIMAG into decreasing order of magnitude. | */
/* %------------------------------------------------% */
L160:
if (igap == 0) {
goto L9000;
}
i__1 = *n - 1;
for (i__ = igap; i__ <= i__1; ++i__) {
j = i__ - igap;
L170:
if (j < 0) {
goto L180;
}
if ((d__1 = ximag[j], abs(d__1)) < (d__2 = ximag[j + igap], abs(
d__2))) {
temp = xreal[j];
xreal[j] = xreal[j + igap];
xreal[j + igap] = temp;
temp = ximag[j];
ximag[j] = ximag[j + igap];
ximag[j + igap] = temp;
if (*apply) {
temp = y[j];
y[j] = y[j + igap];
y[j + igap] = temp;
}
} else {
goto L180;
}
j -= igap;
goto L170;
L180:
;
}
igap /= 2;
goto L160;
}
L9000:
return 0;
/* %---------------% */
/* | End of dsortc | */
/* %---------------% */
} /* dsortc_ */
/* Subroutine */ int dgeev_(char *jobvl, char *jobvr, integer *n, doublereal *
a, integer *lda, doublereal *wr, doublereal *wi, doublereal *vl,
integer *ldvl, doublereal *vr, integer *ldvr, doublereal *work,
integer *lwork, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, vl_dim1, vl_offset, vr_dim1, vr_offset, i__1,
i__2, i__3;
doublereal d__1, d__2;
/* Local variables */
integer i__, k;
doublereal r__, cs, sn;
integer ihi;
doublereal scl;
integer ilo;
doublereal dum[1], eps;
integer ibal;
char side[1];
doublereal anrm;
integer ierr, itau;
integer iwrk, nout;
logical scalea;
doublereal cscale;
logical select[1];
doublereal bignum;
integer minwrk, maxwrk;
logical wantvl;
doublereal smlnum;
integer hswork;
logical lquery, wantvr;
/* -- LAPACK driver routine (version 3.2) -- */
/* November 2006 */
/* Purpose */
/* ======= */
/* DGEEV computes for an N-by-N real nonsymmetric matrix A, the */
/* eigenvalues and, optionally, the left and/or right eigenvectors. */
/* The right eigenvector v(j) of A satisfies */
/* A * v(j) = lambda(j) * v(j) */
/* where lambda(j) is its eigenvalue. */
/* The left eigenvector u(j) of A satisfies */
/* u(j)**H * A = lambda(j) * u(j)**H */
/* where u(j)**H denotes the conjugate transpose of u(j). */
/* The computed eigenvectors are normalized to have Euclidean norm */
/* equal to 1 and largest component real. */
/* Arguments */
/* ========= */
/* JOBVL (input) CHARACTER*1 */
/* = 'N': left eigenvectors of A are not computed; */
/* = 'V': left eigenvectors of A are computed. */
/* JOBVR (input) CHARACTER*1 */
/* = 'N': right eigenvectors of A are not computed; */
/* = 'V': right eigenvectors of A are computed. */
/* N (input) INTEGER */
/* The order of the matrix A. N >= 0. */
/* A (input/output) DOUBLE PRECISION array, dimension (LDA,N) */
/* On entry, the N-by-N matrix A. */
/* On exit, A has been overwritten. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1,N). */
/* WR (output) DOUBLE PRECISION array, dimension (N) */
/* WI (output) DOUBLE PRECISION array, dimension (N) */
/* WR and WI contain the real and imaginary parts, */
/* respectively, of the computed eigenvalues. Complex */
/* conjugate pairs of eigenvalues appear consecutively */
/* with the eigenvalue having the positive imaginary part */
/* first. */
/* VL (output) DOUBLE PRECISION array, dimension (LDVL,N) */
/* If JOBVL = 'V', the left eigenvectors u(j) are stored one */
/* after another in the columns of VL, in the same order */
/* as their eigenvalues. */
/* If JOBVL = 'N', VL is not referenced. */
/* If the j-th eigenvalue is real, then u(j) = VL(:,j), */
/* the j-th column of VL. */
/* If the j-th and (j+1)-st eigenvalues form a complex */
/* conjugate pair, then u(j) = VL(:,j) + i*VL(:,j+1) and */
/* u(j+1) = VL(:,j) - i*VL(:,j+1). */
/* LDVL (input) INTEGER */
/* The leading dimension of the array VL. LDVL >= 1; if */
/* JOBVL = 'V', LDVL >= N. */
/* VR (output) DOUBLE PRECISION array, dimension (LDVR,N) */
/* If JOBVR = 'V', the right eigenvectors v(j) are stored one */
/* after another in the columns of VR, in the same order */
/* as their eigenvalues. */
/* If JOBVR = 'N', VR is not referenced. */
/* If the j-th eigenvalue is real, then v(j) = VR(:,j), */
/* the j-th column of VR. */
/* If the j-th and (j+1)-st eigenvalues form a complex */
/* conjugate pair, then v(j) = VR(:,j) + i*VR(:,j+1) and */
/* v(j+1) = VR(:,j) - i*VR(:,j+1). */
/* LDVR (input) INTEGER */
/* The leading dimension of the array VR. LDVR >= 1; if */
/* JOBVR = 'V', LDVR >= N. */
/* WORK (workspace/output) DOUBLE PRECISION array, dimension (MAX(1,LWORK)) */
/* On exit, if INFO = 0, WORK(1) returns the optimal LWORK. */
/* LWORK (input) INTEGER */
/* The dimension of the array WORK. LWORK >= max(1,3*N), and */
/* if JOBVL = 'V' or JOBVR = 'V', LWORK >= 4*N. For good */
/* performance, LWORK must generally be larger. */
/* If LWORK = -1, then a workspace query is assumed; the routine */
/* only calculates the optimal size of the WORK array, returns */
/* this value as the first entry of the WORK array, and no error */
/* message related to LWORK is issued by XERBLA. */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value. */
/* > 0: if INFO = i, the QR algorithm failed to compute all the */
/* eigenvalues, and no eigenvectors have been computed; */
/* elements i+1:N of WR and WI contain eigenvalues which */
/* have converged. */
/* ===================================================================== */
/* Test the input arguments */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
--wr;
--wi;
vl_dim1 = *ldvl;
vl_offset = 1 + vl_dim1;
vl -= vl_offset;
vr_dim1 = *ldvr;
vr_offset = 1 + vr_dim1;
vr -= vr_offset;
--work;
/* Function Body */
*info = 0;
lquery = *lwork == -1;
wantvl = lsame_(jobvl, "V");
wantvr = lsame_(jobvr, "V");
if (! wantvl && ! lsame_(jobvl, "N")) {
*info = -1;
} else if (! wantvr && ! lsame_(jobvr, "N")) {
*info = -2;
} else if (*n < 0) {
*info = -3;
} else if (*lda < max(1,*n)) {
*info = -5;
} else if (*ldvl < 1 || wantvl && *ldvl < *n) {
*info = -9;
} else if (*ldvr < 1 || wantvr && *ldvr < *n) {
*info = -11;
}
/* Compute workspace */
/* (Note: Comments in the code beginning "Workspace:" describe the */
/* minimal amount of workspace needed at that point in the code, */
/* as well as the preferred amount for good performance. */
/* NB refers to the optimal block size for the immediately */
/* following subroutine, as returned by ILAENV. */
/* HSWORK refers to the workspace preferred by DHSEQR, as */
/* calculated below. HSWORK is computed assuming ILO=1 and IHI=N, */
/* the worst case.) */
if (*info == 0) {
if (*n == 0) {
minwrk = 1;
maxwrk = 1;
} else {
maxwrk = (*n << 1) + *n * ilaenv_(&c__1, "DGEHRD", " ", n, &c__1,
n, &c__0);
if (wantvl) {
minwrk = *n << 2;
/* Computing MAX */
i__1 = maxwrk, i__2 = (*n << 1) + (*n - 1) * ilaenv_(&c__1,
"DORGHR", " ", n, &c__1, n, &c_n1);
maxwrk = max(i__1,i__2);
dhseqr_("S", "V", n, &c__1, n, &a[a_offset], lda, &wr[1], &wi[
1], &vl[vl_offset], ldvl, &work[1], &c_n1, info);
hswork = (integer) work[1];
/* Computing MAX */
i__1 = maxwrk, i__2 = *n + 1, i__1 = max(i__1,i__2), i__2 = *
n + hswork;
maxwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk, i__2 = *n << 2;
maxwrk = max(i__1,i__2);
} else if (wantvr) {
minwrk = *n << 2;
/* Computing MAX */
i__1 = maxwrk, i__2 = (*n << 1) + (*n - 1) * ilaenv_(&c__1,
"DORGHR", " ", n, &c__1, n, &c_n1);
maxwrk = max(i__1,i__2);
dhseqr_("S", "V", n, &c__1, n, &a[a_offset], lda, &wr[1], &wi[
1], &vr[vr_offset], ldvr, &work[1], &c_n1, info);
hswork = (integer) work[1];
/* Computing MAX */
i__1 = maxwrk, i__2 = *n + 1, i__1 = max(i__1,i__2), i__2 = *
n + hswork;
maxwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk, i__2 = *n << 2;
maxwrk = max(i__1,i__2);
} else {
minwrk = *n * 3;
dhseqr_("E", "N", n, &c__1, n, &a[a_offset], lda, &wr[1], &wi[
1], &vr[vr_offset], ldvr, &work[1], &c_n1, info);
hswork = (integer) work[1];
/* Computing MAX */
i__1 = maxwrk, i__2 = *n + 1, i__1 = max(i__1,i__2), i__2 = *
n + hswork;
maxwrk = max(i__1,i__2);
}
maxwrk = max(maxwrk,minwrk);
}
work[1] = (doublereal) maxwrk;
if (*lwork < minwrk && ! lquery) {
*info = -13;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("DGEEV ", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
/* Get machine constants */
eps = dlamch_("P");
smlnum = dlamch_("S");
bignum = 1. / smlnum;
dlabad_(&smlnum, &bignum);
smlnum = sqrt(smlnum) / eps;
bignum = 1. / smlnum;
/* Scale A if max element outside range [SMLNUM,BIGNUM] */
anrm = dlange_("M", n, n, &a[a_offset], lda, dum);
scalea = FALSE_;
if (anrm > 0. && anrm < smlnum) {
scalea = TRUE_;
cscale = smlnum;
} else if (anrm > bignum) {
scalea = TRUE_;
cscale = bignum;
}
if (scalea) {
dlascl_("G", &c__0, &c__0, &anrm, &cscale, n, n, &a[a_offset], lda, &
ierr);
}
/* Balance the matrix */
/* (Workspace: need N) */
ibal = 1;
dgebal_("B", n, &a[a_offset], lda, &ilo, &ihi, &work[ibal], &ierr);
/* Reduce to upper Hessenberg form */
/* (Workspace: need 3*N, prefer 2*N+N*NB) */
itau = ibal + *n;
iwrk = itau + *n;
i__1 = *lwork - iwrk + 1;
dgehrd_(n, &ilo, &ihi, &a[a_offset], lda, &work[itau], &work[iwrk], &i__1,
&ierr);
if (wantvl) {
/* Want left eigenvectors */
/* Copy Householder vectors to VL */
*(unsigned char *)side = 'L';
dlacpy_("L", n, n, &a[a_offset], lda, &vl[vl_offset], ldvl)
;
/* Generate orthogonal matrix in VL */
/* (Workspace: need 3*N-1, prefer 2*N+(N-1)*NB) */
i__1 = *lwork - iwrk + 1;
dorghr_(n, &ilo, &ihi, &vl[vl_offset], ldvl, &work[itau], &work[iwrk],
&i__1, &ierr);
/* Perform QR iteration, accumulating Schur vectors in VL */
/* (Workspace: need N+1, prefer N+HSWORK (see comments) ) */
iwrk = itau;
i__1 = *lwork - iwrk + 1;
dhseqr_("S", "V", n, &ilo, &ihi, &a[a_offset], lda, &wr[1], &wi[1], &
vl[vl_offset], ldvl, &work[iwrk], &i__1, info);
if (wantvr) {
/* Want left and right eigenvectors */
/* Copy Schur vectors to VR */
*(unsigned char *)side = 'B';
dlacpy_("F", n, n, &vl[vl_offset], ldvl, &vr[vr_offset], ldvr);
}
} else if (wantvr) {
/* Want right eigenvectors */
/* Copy Householder vectors to VR */
*(unsigned char *)side = 'R';
dlacpy_("L", n, n, &a[a_offset], lda, &vr[vr_offset], ldvr)
;
/* Generate orthogonal matrix in VR */
/* (Workspace: need 3*N-1, prefer 2*N+(N-1)*NB) */
i__1 = *lwork - iwrk + 1;
dorghr_(n, &ilo, &ihi, &vr[vr_offset], ldvr, &work[itau], &work[iwrk],
&i__1, &ierr);
/* Perform QR iteration, accumulating Schur vectors in VR */
/* (Workspace: need N+1, prefer N+HSWORK (see comments) ) */
iwrk = itau;
i__1 = *lwork - iwrk + 1;
dhseqr_("S", "V", n, &ilo, &ihi, &a[a_offset], lda, &wr[1], &wi[1], &
vr[vr_offset], ldvr, &work[iwrk], &i__1, info);
} else {
/* Compute eigenvalues only */
/* (Workspace: need N+1, prefer N+HSWORK (see comments) ) */
iwrk = itau;
i__1 = *lwork - iwrk + 1;
dhseqr_("E", "N", n, &ilo, &ihi, &a[a_offset], lda, &wr[1], &wi[1], &
vr[vr_offset], ldvr, &work[iwrk], &i__1, info);
}
/* If INFO > 0 from DHSEQR, then quit */
if (*info > 0) {
goto L50;
}
if (wantvl || wantvr) {
/* Compute left and/or right eigenvectors */
/* (Workspace: need 4*N) */
dtrevc_(side, "B", select, n, &a[a_offset], lda, &vl[vl_offset], ldvl,
&vr[vr_offset], ldvr, n, &nout, &work[iwrk], &ierr);
}
if (wantvl) {
/* Undo balancing of left eigenvectors */
/* (Workspace: need N) */
dgebak_("B", "L", n, &ilo, &ihi, &work[ibal], n, &vl[vl_offset], ldvl,
&ierr);
/* Normalize left eigenvectors and make largest component real */
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
if (wi[i__] == 0.) {
scl = 1. / dnrm2_(n, &vl[i__ * vl_dim1 + 1], &c__1);
dscal_(n, &scl, &vl[i__ * vl_dim1 + 1], &c__1);
} else if (wi[i__] > 0.) {
d__1 = dnrm2_(n, &vl[i__ * vl_dim1 + 1], &c__1);
d__2 = dnrm2_(n, &vl[(i__ + 1) * vl_dim1 + 1], &c__1);
scl = 1. / dlapy2_(&d__1, &d__2);
dscal_(n, &scl, &vl[i__ * vl_dim1 + 1], &c__1);
dscal_(n, &scl, &vl[(i__ + 1) * vl_dim1 + 1], &c__1);
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
/* Computing 2nd power */
d__1 = vl[k + i__ * vl_dim1];
/* Computing 2nd power */
d__2 = vl[k + (i__ + 1) * vl_dim1];
work[iwrk + k - 1] = d__1 * d__1 + d__2 * d__2;
}
k = idamax_(n, &work[iwrk], &c__1);
dlartg_(&vl[k + i__ * vl_dim1], &vl[k + (i__ + 1) * vl_dim1],
&cs, &sn, &r__);
drot_(n, &vl[i__ * vl_dim1 + 1], &c__1, &vl[(i__ + 1) *
vl_dim1 + 1], &c__1, &cs, &sn);
vl[k + (i__ + 1) * vl_dim1] = 0.;
}
}
}
if (wantvr) {
/* Undo balancing of right eigenvectors */
/* (Workspace: need N) */
dgebak_("B", "R", n, &ilo, &ihi, &work[ibal], n, &vr[vr_offset], ldvr,
&ierr);
/* Normalize right eigenvectors and make largest component real */
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
if (wi[i__] == 0.) {
scl = 1. / dnrm2_(n, &vr[i__ * vr_dim1 + 1], &c__1);
dscal_(n, &scl, &vr[i__ * vr_dim1 + 1], &c__1);
} else if (wi[i__] > 0.) {
d__1 = dnrm2_(n, &vr[i__ * vr_dim1 + 1], &c__1);
d__2 = dnrm2_(n, &vr[(i__ + 1) * vr_dim1 + 1], &c__1);
scl = 1. / dlapy2_(&d__1, &d__2);
dscal_(n, &scl, &vr[i__ * vr_dim1 + 1], &c__1);
dscal_(n, &scl, &vr[(i__ + 1) * vr_dim1 + 1], &c__1);
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
/* Computing 2nd power */
d__1 = vr[k + i__ * vr_dim1];
/* Computing 2nd power */
d__2 = vr[k + (i__ + 1) * vr_dim1];
work[iwrk + k - 1] = d__1 * d__1 + d__2 * d__2;
}
k = idamax_(n, &work[iwrk], &c__1);
dlartg_(&vr[k + i__ * vr_dim1], &vr[k + (i__ + 1) * vr_dim1],
&cs, &sn, &r__);
drot_(n, &vr[i__ * vr_dim1 + 1], &c__1, &vr[(i__ + 1) *
vr_dim1 + 1], &c__1, &cs, &sn);
vr[k + (i__ + 1) * vr_dim1] = 0.;
}
}
}
/* Undo scaling if necessary */
L50:
if (scalea) {
i__1 = *n - *info;
/* Computing MAX */
i__3 = *n - *info;
i__2 = max(i__3,1);
dlascl_("G", &c__0, &c__0, &cscale, &anrm, &i__1, &c__1, &wr[*info +
1], &i__2, &ierr);
i__1 = *n - *info;
/* Computing MAX */
i__3 = *n - *info;
i__2 = max(i__3,1);
dlascl_("G", &c__0, &c__0, &cscale, &anrm, &i__1, &c__1, &wi[*info +
1], &i__2, &ierr);
if (*info > 0) {
i__1 = ilo - 1;
dlascl_("G", &c__0, &c__0, &cscale, &anrm, &i__1, &c__1, &wr[1],
n, &ierr);
i__1 = ilo - 1;
dlascl_("G", &c__0, &c__0, &cscale, &anrm, &i__1, &c__1, &wi[1],
n, &ierr);
}
}
work[1] = (doublereal) maxwrk;
return 0;
/* End of DGEEV */
} /* dgeev_ */
/* Subroutine */ int dsterf_(integer *n, doublereal *d__, doublereal *e,
integer *info)
{
/* System generated locals */
integer i__1;
doublereal d__1, d__2, d__3;
/* Builtin functions */
double sqrt(doublereal), d_sign(doublereal *, doublereal *);
/* Local variables */
static doublereal oldc;
static integer lend, jtot;
extern /* Subroutine */ int dlae2_(doublereal *, doublereal *, doublereal
*, doublereal *, doublereal *);
static doublereal c__;
static integer i__, l, m;
static doublereal p, gamma, r__, s, alpha, sigma, anorm;
static integer l1;
extern doublereal dlapy2_(doublereal *, doublereal *);
static doublereal bb;
extern doublereal dlamch_(char *);
static integer iscale;
extern /* Subroutine */ int dlascl_(char *, integer *, integer *,
doublereal *, doublereal *, integer *, integer *, doublereal *,
integer *, integer *);
static doublereal oldgam, safmin;
extern /* Subroutine */ int xerbla_(char *, integer *);
static doublereal safmax;
extern doublereal dlanst_(char *, integer *, doublereal *, doublereal *);
extern /* Subroutine */ int dlasrt_(char *, integer *, doublereal *,
integer *);
static integer lendsv;
static doublereal ssfmin;
static integer nmaxit;
static doublereal ssfmax, rt1, rt2, eps, rte;
static integer lsv;
static doublereal eps2;
/* -- 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
June 30, 1999
Common block to return operation count and iteration count
ITCNT is initialized to 0, OPS is only incremented
Purpose
=======
DSTERF computes all eigenvalues of a symmetric tridiagonal matrix
using the Pal-Walker-Kahan variant of the QL or QR algorithm.
Arguments
=========
N (input) INTEGER
The order of the matrix. N >= 0.
D (input/output) DOUBLE PRECISION array, dimension (N)
On entry, the n diagonal elements of the tridiagonal matrix.
On exit, if INFO = 0, the eigenvalues in ascending order.
E (input/output) DOUBLE PRECISION array, dimension (N-1)
On entry, the (n-1) subdiagonal elements of the tridiagonal
matrix.
On exit, E has been destroyed.
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
> 0: the algorithm failed to find all of the eigenvalues in
a total of 30*N iterations; if INFO = i, then i
elements of E have not converged to zero.
=====================================================================
Test the input parameters.
Parameter adjustments */
--e;
--d__;
/* Function Body */
*info = 0;
/* Quick return if possible */
latime_1.itcnt = 0.;
if (*n < 0) {
*info = -1;
i__1 = -(*info);
xerbla_("DSTERF", &i__1);
return 0;
}
if (*n <= 1) {
return 0;
}
/* Determine the unit roundoff for this environment. */
latime_1.ops += 6;
eps = dlamch_("E");
/* Computing 2nd power */
d__1 = eps;
eps2 = d__1 * d__1;
safmin = dlamch_("S");
safmax = 1. / safmin;
ssfmax = sqrt(safmax) / 3.;
ssfmin = sqrt(safmin) / eps2;
/* Compute the eigenvalues of the tridiagonal matrix. */
nmaxit = *n * 30;
sigma = 0.;
jtot = 0;
/* Determine where the matrix splits and choose QL or QR iteration
for each block, according to whether top or bottom diagonal
element is smaller. */
l1 = 1;
L10:
if (l1 > *n) {
goto L170;
}
if (l1 > 1) {
e[l1 - 1] = 0.;
}
i__1 = *n - 1;
for (m = l1; m <= i__1; ++m) {
latime_1.ops += 4;
if ((d__3 = e[m], abs(d__3)) <= sqrt((d__1 = d__[m], abs(d__1))) *
sqrt((d__2 = d__[m + 1], abs(d__2))) * eps) {
e[m] = 0.;
goto L30;
}
/* L20: */
}
m = *n;
L30:
l = l1;
lsv = l;
lend = m;
lendsv = lend;
l1 = m + 1;
if (lend == l) {
goto L10;
}
/* Scale submatrix in rows and columns L to LEND */
latime_1.ops += lend - l + 1 << 1;
i__1 = lend - l + 1;
anorm = dlanst_("I", &i__1, &d__[l], &e[l]);
iscale = 0;
if (anorm > ssfmax) {
iscale = 1;
latime_1.ops = latime_1.ops + (lend - l << 1) + 1;
i__1 = lend - l + 1;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmax, &i__1, &c__1, &d__[l], n,
info);
i__1 = lend - l;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmax, &i__1, &c__1, &e[l], n,
info);
} else if (anorm < ssfmin) {
iscale = 2;
latime_1.ops = latime_1.ops + (lend - l << 1) + 1;
i__1 = lend - l + 1;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &d__[l], n,
info);
i__1 = lend - l;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &e[l], n,
info);
}
latime_1.ops += lend - l << 1;
i__1 = lend - 1;
for (i__ = l; i__ <= i__1; ++i__) {
/* Computing 2nd power */
d__1 = e[i__];
e[i__] = d__1 * d__1;
/* L40: */
}
/* Choose between QL and QR iteration */
if ((d__1 = d__[lend], abs(d__1)) < (d__2 = d__[l], abs(d__2))) {
lend = lsv;
l = lendsv;
}
if (lend >= l) {
/* QL Iteration
Look for small subdiagonal element. */
L50:
if (l != lend) {
i__1 = lend - 1;
for (m = l; m <= i__1; ++m) {
latime_1.ops += 3;
if ((d__2 = e[m], abs(d__2)) <= eps2 * (d__1 = d__[m] * d__[m
+ 1], abs(d__1))) {
goto L70;
}
/* L60: */
}
}
m = lend;
L70:
if (m < lend) {
e[m] = 0.;
}
p = d__[l];
if (m == l) {
goto L90;
}
/* If remaining matrix is 2 by 2, use DLAE2 to compute its
eigenvalues. */
if (m == l + 1) {
latime_1.ops += 16;
rte = sqrt(e[l]);
dlae2_(&d__[l], &rte, &d__[l + 1], &rt1, &rt2);
d__[l] = rt1;
d__[l + 1] = rt2;
e[l] = 0.;
l += 2;
if (l <= lend) {
goto L50;
}
goto L150;
}
if (jtot == nmaxit) {
goto L150;
}
++jtot;
/* Form shift. */
latime_1.ops += 14;
rte = sqrt(e[l]);
sigma = (d__[l + 1] - p) / (rte * 2.);
r__ = dlapy2_(&sigma, &c_b32);
sigma = p - rte / (sigma + d_sign(&r__, &sigma));
c__ = 1.;
s = 0.;
gamma = d__[m] - sigma;
p = gamma * gamma;
/* Inner loop */
latime_1.ops += (m - l) * 12;
i__1 = l;
for (i__ = m - 1; i__ >= i__1; --i__) {
bb = e[i__];
r__ = p + bb;
if (i__ != m - 1) {
e[i__ + 1] = s * r__;
}
oldc = c__;
c__ = p / r__;
s = bb / r__;
oldgam = gamma;
alpha = d__[i__];
gamma = c__ * (alpha - sigma) - s * oldgam;
d__[i__ + 1] = oldgam + (alpha - gamma);
if (c__ != 0.) {
p = gamma * gamma / c__;
} else {
p = oldc * bb;
}
/* L80: */
}
latime_1.ops += 2;
e[l] = s * p;
d__[l] = sigma + gamma;
goto L50;
/* Eigenvalue found. */
L90:
d__[l] = p;
++l;
if (l <= lend) {
goto L50;
}
goto L150;
} else {
/* QR Iteration
Look for small superdiagonal element. */
L100:
i__1 = lend + 1;
for (m = l; m >= i__1; --m) {
latime_1.ops += 3;
if ((d__2 = e[m - 1], abs(d__2)) <= eps2 * (d__1 = d__[m] * d__[m
- 1], abs(d__1))) {
goto L120;
}
/* L110: */
}
m = lend;
L120:
if (m > lend) {
e[m - 1] = 0.;
}
p = d__[l];
if (m == l) {
goto L140;
}
/* If remaining matrix is 2 by 2, use DLAE2 to compute its
eigenvalues. */
if (m == l - 1) {
latime_1.ops += 16;
rte = sqrt(e[l - 1]);
dlae2_(&d__[l], &rte, &d__[l - 1], &rt1, &rt2);
d__[l] = rt1;
d__[l - 1] = rt2;
e[l - 1] = 0.;
l += -2;
if (l >= lend) {
goto L100;
}
goto L150;
}
if (jtot == nmaxit) {
goto L150;
}
++jtot;
/* Form shift. */
latime_1.ops += 14;
rte = sqrt(e[l - 1]);
sigma = (d__[l - 1] - p) / (rte * 2.);
r__ = dlapy2_(&sigma, &c_b32);
sigma = p - rte / (sigma + d_sign(&r__, &sigma));
c__ = 1.;
s = 0.;
gamma = d__[m] - sigma;
p = gamma * gamma;
/* Inner loop */
latime_1.ops += (l - m) * 12;
i__1 = l - 1;
for (i__ = m; i__ <= i__1; ++i__) {
bb = e[i__];
r__ = p + bb;
if (i__ != m) {
e[i__ - 1] = s * r__;
}
oldc = c__;
c__ = p / r__;
s = bb / r__;
oldgam = gamma;
alpha = d__[i__ + 1];
gamma = c__ * (alpha - sigma) - s * oldgam;
d__[i__] = oldgam + (alpha - gamma);
if (c__ != 0.) {
p = gamma * gamma / c__;
} else {
p = oldc * bb;
}
/* L130: */
}
latime_1.ops += 2;
e[l - 1] = s * p;
d__[l] = sigma + gamma;
goto L100;
/* Eigenvalue found. */
L140:
d__[l] = p;
--l;
if (l >= lend) {
goto L100;
}
goto L150;
}
/* Undo scaling if necessary */
L150:
if (iscale == 1) {
latime_1.ops = latime_1.ops + lendsv - lsv + 1;
i__1 = lendsv - lsv + 1;
dlascl_("G", &c__0, &c__0, &ssfmax, &anorm, &i__1, &c__1, &d__[lsv],
n, info);
}
if (iscale == 2) {
latime_1.ops = latime_1.ops + lendsv - lsv + 1;
i__1 = lendsv - lsv + 1;
dlascl_("G", &c__0, &c__0, &ssfmin, &anorm, &i__1, &c__1, &d__[lsv],
n, info);
}
/* Check for no convergence to an eigenvalue after a total
of N*MAXIT iterations. */
if (jtot < nmaxit) {
goto L10;
}
i__1 = *n - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
if (e[i__] != 0.) {
++(*info);
}
/* L160: */
}
goto L180;
/* Sort eigenvalues in increasing order. */
L170:
dlasrt_("I", n, &d__[1], info);
L180:
return 0;
/* End of DSTERF */
} /* dsterf_ */
/* Subroutine */ int dneigh_(doublereal *rnorm, integer *n, doublereal *h__,
integer *ldh, doublereal *ritzr, doublereal *ritzi, doublereal *
bounds, doublereal *q, integer *ldq, doublereal *workl, integer *ierr)
{
/* System generated locals */
integer h_dim1, h_offset, q_dim1, q_offset, i__1;
doublereal d__1, d__2;
/* Local variables */
static integer i__;
static real t0, t1;
static doublereal vl[1], temp;
extern doublereal dnrm2_(integer *, doublereal *, integer *);
extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
integer *);
static integer iconj;
extern /* Subroutine */ int dgemv_(char *, integer *, integer *,
doublereal *, doublereal *, integer *, doublereal *, integer *,
doublereal *, doublereal *, integer *, ftnlen), dmout_(integer *,
integer *, integer *, doublereal *, integer *, integer *, char *,
ftnlen), dvout_(integer *, integer *, doublereal *, integer *,
char *, ftnlen);
extern doublereal dlapy2_(doublereal *, doublereal *);
extern /* Subroutine */ int dlaqrb_(logical *, integer *, integer *,
integer *, doublereal *, integer *, doublereal *, doublereal *,
doublereal *, integer *), second_(real *);
static logical select[1];
static integer msglvl;
extern /* Subroutine */ int dlacpy_(char *, integer *, integer *,
doublereal *, integer *, doublereal *, integer *, ftnlen),
dtrevc_(char *, char *, logical *, integer *, doublereal *,
integer *, doublereal *, integer *, doublereal *, integer *,
integer *, integer *, doublereal *, integer *, ftnlen, ftnlen);
/* %----------------------------------------------------% */
/* | Include files for debugging and timing information | */
/* %----------------------------------------------------% */
/* \SCCS Information: @(#) */
/* FILE: debug.h SID: 2.3 DATE OF SID: 11/16/95 RELEASE: 2 */
/* %---------------------------------% */
/* | See debug.doc for documentation | */
/* %---------------------------------% */
/* %------------------% */
/* | Scalar Arguments | */
/* %------------------% */
/* %--------------------------------% */
/* | See stat.doc for documentation | */
/* %--------------------------------% */
/* \SCCS Information: @(#) */
/* FILE: stat.h SID: 2.2 DATE OF SID: 11/16/95 RELEASE: 2 */
/* %-----------------% */
/* | Array Arguments | */
/* %-----------------% */
/* %------------% */
/* | Parameters | */
/* %------------% */
/* %------------------------% */
/* | Local Scalars & Arrays | */
/* %------------------------% */
/* %----------------------% */
/* | External Subroutines | */
/* %----------------------% */
/* %--------------------% */
/* | External Functions | */
/* %--------------------% */
/* %---------------------% */
/* | Intrinsic Functions | */
/* %---------------------% */
/* %-----------------------% */
/* | Executable Statements | */
/* %-----------------------% */
/* %-------------------------------% */
/* | Initialize timing statistics | */
/* | & message level for debugging | */
/* %-------------------------------% */
/* Parameter adjustments */
--workl;
--bounds;
--ritzi;
--ritzr;
h_dim1 = *ldh;
h_offset = 1 + h_dim1;
h__ -= h_offset;
q_dim1 = *ldq;
q_offset = 1 + q_dim1;
q -= q_offset;
/* Function Body */
second_(&t0);
msglvl = debug_1.mneigh;
if (msglvl > 2) {
dmout_(&debug_1.logfil, n, n, &h__[h_offset], ldh, &debug_1.ndigit,
"_neigh: Entering upper Hessenberg matrix H ", (ftnlen)43);
}
/* %-----------------------------------------------------------% */
/* | 1. Compute the eigenvalues, the last components of the | */
/* | corresponding Schur vectors and the full Schur form T | */
/* | of the current upper Hessenberg matrix H. | */
/* | dlaqrb returns the full Schur form of H in WORKL(1:N**2) | */
/* | and the last components of the Schur vectors in BOUNDS. | */
/* %-----------------------------------------------------------% */
dlacpy_("All", n, n, &h__[h_offset], ldh, &workl[1], n, (ftnlen)3);
dlaqrb_(&c_true, n, &c__1, n, &workl[1], n, &ritzr[1], &ritzi[1], &bounds[
1], ierr);
if (*ierr != 0) {
goto L9000;
}
if (msglvl > 1) {
dvout_(&debug_1.logfil, n, &bounds[1], &debug_1.ndigit, "_neigh: las"
"t row of the Schur matrix for H", (ftnlen)42);
}
/* %-----------------------------------------------------------% */
/* | 2. Compute the eigenvectors of the full Schur form T and | */
/* | apply the last components of the Schur vectors to get | */
/* | the last components of the corresponding eigenvectors. | */
/* | Remember that if the i-th and (i+1)-st eigenvalues are | */
/* | complex conjugate pairs, then the real & imaginary part | */
/* | of the eigenvector components are split across adjacent | */
/* | columns of Q. | */
/* %-----------------------------------------------------------% */
dtrevc_("R", "A", select, n, &workl[1], n, vl, n, &q[q_offset], ldq, n, n,
&workl[*n * *n + 1], ierr, (ftnlen)1, (ftnlen)1);
if (*ierr != 0) {
goto L9000;
}
/* %------------------------------------------------% */
/* | Scale the returning eigenvectors so that their | */
/* | euclidean norms are all one. LAPACK subroutine | */
/* | dtrevc returns each eigenvector normalized so | */
/* | that the element of largest magnitude has | */
/* | magnitude 1; here the magnitude of a complex | */
/* | number (x,y) is taken to be |x| + |y|. | */
/* %------------------------------------------------% */
iconj = 0;
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
if ((d__1 = ritzi[i__], abs(d__1)) <= 0.) {
/* %----------------------% */
/* | Real eigenvalue case | */
/* %----------------------% */
temp = dnrm2_(n, &q[i__ * q_dim1 + 1], &c__1);
d__1 = 1. / temp;
dscal_(n, &d__1, &q[i__ * q_dim1 + 1], &c__1);
} else {
/* %-------------------------------------------% */
/* | Complex conjugate pair case. Note that | */
/* | since the real and imaginary part of | */
/* | the eigenvector are stored in consecutive | */
/* | columns, we further normalize by the | */
/* | square root of two. | */
/* %-------------------------------------------% */
if (iconj == 0) {
d__1 = dnrm2_(n, &q[i__ * q_dim1 + 1], &c__1);
d__2 = dnrm2_(n, &q[(i__ + 1) * q_dim1 + 1], &c__1);
temp = dlapy2_(&d__1, &d__2);
d__1 = 1. / temp;
dscal_(n, &d__1, &q[i__ * q_dim1 + 1], &c__1);
d__1 = 1. / temp;
dscal_(n, &d__1, &q[(i__ + 1) * q_dim1 + 1], &c__1);
iconj = 1;
} else {
iconj = 0;
}
}
/* L10: */
}
dgemv_("T", n, n, &c_b18, &q[q_offset], ldq, &bounds[1], &c__1, &c_b20, &
workl[1], &c__1, (ftnlen)1);
if (msglvl > 1) {
dvout_(&debug_1.logfil, n, &workl[1], &debug_1.ndigit, "_neigh: Last"
" row of the eigenvector matrix for H", (ftnlen)48);
}
/* %----------------------------% */
/* | Compute the Ritz estimates | */
/* %----------------------------% */
iconj = 0;
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
if ((d__1 = ritzi[i__], abs(d__1)) <= 0.) {
/* %----------------------% */
/* | Real eigenvalue case | */
/* %----------------------% */
bounds[i__] = *rnorm * (d__1 = workl[i__], abs(d__1));
} else {
/* %-------------------------------------------% */
/* | Complex conjugate pair case. Note that | */
/* | since the real and imaginary part of | */
/* | the eigenvector are stored in consecutive | */
/* | columns, we need to take the magnitude | */
/* | of the last components of the two vectors | */
/* %-------------------------------------------% */
if (iconj == 0) {
bounds[i__] = *rnorm * dlapy2_(&workl[i__], &workl[i__ + 1]);
bounds[i__ + 1] = bounds[i__];
iconj = 1;
} else {
iconj = 0;
}
}
/* L20: */
}
if (msglvl > 2) {
dvout_(&debug_1.logfil, n, &ritzr[1], &debug_1.ndigit, "_neigh: Real"
" part of the eigenvalues of H", (ftnlen)41);
dvout_(&debug_1.logfil, n, &ritzi[1], &debug_1.ndigit, "_neigh: Imag"
"inary part of the eigenvalues of H", (ftnlen)46);
dvout_(&debug_1.logfil, n, &bounds[1], &debug_1.ndigit, "_neigh: Rit"
"z estimates for the eigenvalues of H", (ftnlen)47);
}
second_(&t1);
timing_1.tneigh += t1 - t0;
L9000:
return 0;
/* %---------------% */
/* | End of dneigh | */
/* %---------------% */
} /* dneigh_ */
/*< SUBROUTINE DSTEQR( COMPZ, N, D, E, Z, LDZ, WORK, INFO ) >*/
/* Subroutine */ int dsteqr_(char *compz, integer *n, doublereal *d__,
doublereal *e, doublereal *z__, integer *ldz, doublereal *work,
integer *info, ftnlen compz_len)
{
/* System generated locals */
integer z_dim1, z_offset, i__1, i__2;
doublereal d__1, d__2;
/* Builtin functions */
double sqrt(doublereal), d_sign(doublereal *, doublereal *);
/* Local variables */
doublereal b, c__, f, g;
integer i__, j, k, l, m;
doublereal p, r__, s;
integer l1, ii, mm, lm1, mm1, nm1;
doublereal rt1, rt2, eps;
integer lsv;
doublereal tst, eps2;
integer lend, jtot;
extern /* Subroutine */ int dlae2_(doublereal *, doublereal *, doublereal
*, doublereal *, doublereal *);
extern logical lsame_(const char *, const char *, ftnlen, ftnlen);
extern /* Subroutine */ int dlasr_(char *, char *, char *, integer *,
integer *, doublereal *, doublereal *, doublereal *, integer *,
ftnlen, ftnlen, ftnlen);
doublereal anorm;
extern /* Subroutine */ int dswap_(integer *, doublereal *, integer *,
doublereal *, integer *), dlaev2_(doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *, doublereal *,
doublereal *);
integer lendm1, lendp1;
extern doublereal dlapy2_(doublereal *, doublereal *), dlamch_(char *,
ftnlen);
integer iscale;
extern /* Subroutine */ int dlascl_(char *, integer *, integer *,
doublereal *, doublereal *, integer *, integer *, doublereal *,
integer *, integer *, ftnlen), dlaset_(char *, integer *, integer
*, doublereal *, doublereal *, doublereal *, integer *, ftnlen);
doublereal safmin;
extern /* Subroutine */ int dlartg_(doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *);
doublereal safmax;
extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
extern doublereal dlanst_(char *, integer *, doublereal *, doublereal *,
ftnlen);
extern /* Subroutine */ int dlasrt_(char *, integer *, doublereal *,
integer *, ftnlen);
integer lendsv;
doublereal ssfmin;
integer nmaxit, icompz;
doublereal ssfmax;
/* -- LAPACK routine (version 3.2) -- */
/* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
/* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
/* November 2006 */
/* .. Scalar Arguments .. */
/*< CHARACTER COMPZ >*/
/*< INTEGER INFO, LDZ, N >*/
/* .. */
/* .. Array Arguments .. */
/*< DOUBLE PRECISION D( * ), E( * ), WORK( * ), Z( LDZ, * ) >*/
/* .. */
/* Purpose */
/* ======= */
/* DSTEQR computes all eigenvalues and, optionally, eigenvectors of a */
/* symmetric tridiagonal matrix using the implicit QL or QR method. */
/* The eigenvectors of a full or band symmetric matrix can also be found */
/* if DSYTRD or DSPTRD or DSBTRD has been used to reduce this matrix to */
/* tridiagonal form. */
/* Arguments */
/* ========= */
/* COMPZ (input) CHARACTER*1 */
/* = 'N': Compute eigenvalues only. */
/* = 'V': Compute eigenvalues and eigenvectors of the original */
/* symmetric matrix. On entry, Z must contain the */
/* orthogonal matrix used to reduce the original matrix */
/* to tridiagonal form. */
/* = 'I': Compute eigenvalues and eigenvectors of the */
/* tridiagonal matrix. Z is initialized to the identity */
/* matrix. */
/* N (input) INTEGER */
/* The order of the matrix. N >= 0. */
/* D (input/output) DOUBLE PRECISION array, dimension (N) */
/* On entry, the diagonal elements of the tridiagonal matrix. */
/* On exit, if INFO = 0, the eigenvalues in ascending order. */
/* E (input/output) DOUBLE PRECISION array, dimension (N-1) */
/* On entry, the (n-1) subdiagonal elements of the tridiagonal */
/* matrix. */
/* On exit, E has been destroyed. */
/* Z (input/output) DOUBLE PRECISION array, dimension (LDZ, N) */
/* On entry, if COMPZ = 'V', then Z contains the orthogonal */
/* matrix used in the reduction to tridiagonal form. */
/* On exit, if INFO = 0, then if COMPZ = 'V', Z contains the */
/* orthonormal eigenvectors of the original symmetric matrix, */
/* and if COMPZ = 'I', Z contains the orthonormal eigenvectors */
/* of the symmetric tridiagonal matrix. */
/* If COMPZ = 'N', then Z is not referenced. */
/* LDZ (input) INTEGER */
/* The leading dimension of the array Z. LDZ >= 1, and if */
/* eigenvectors are desired, then LDZ >= max(1,N). */
/* WORK (workspace) DOUBLE PRECISION array, dimension (max(1,2*N-2)) */
/* If COMPZ = 'N', then WORK is not referenced. */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* > 0: the algorithm has failed to find all the eigenvalues in */
/* a total of 30*N iterations; if INFO = i, then i */
/* elements of E have not converged to zero; on exit, D */
/* and E contain the elements of a symmetric tridiagonal */
/* matrix which is orthogonally similar to the original */
/* matrix. */
/* ===================================================================== */
/* .. Parameters .. */
/*< DOUBLE PRECISION ZERO, ONE, TWO, THREE >*/
/*< >*/
/*< INTEGER MAXIT >*/
/*< PARAMETER ( MAXIT = 30 ) >*/
/* .. */
/* .. Local Scalars .. */
/*< >*/
/*< >*/
/* .. */
/* .. External Functions .. */
/*< LOGICAL LSAME >*/
/*< DOUBLE PRECISION DLAMCH, DLANST, DLAPY2 >*/
/*< EXTERNAL LSAME, DLAMCH, DLANST, DLAPY2 >*/
/* .. */
/* .. External Subroutines .. */
/*< >*/
/* .. */
/* .. Intrinsic Functions .. */
/*< INTRINSIC ABS, MAX, SIGN, SQRT >*/
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/*< INFO = 0 >*/
/* Parameter adjustments */
--d__;
--e;
z_dim1 = *ldz;
z_offset = 1 + z_dim1;
z__ -= z_offset;
--work;
/* Function Body */
*info = 0;
/*< IF( LSAME( COMPZ, 'N' ) ) THEN >*/
if (lsame_(compz, "N", (ftnlen)1, (ftnlen)1)) {
/*< ICOMPZ = 0 >*/
icompz = 0;
/*< ELSE IF( LSAME( COMPZ, 'V' ) ) THEN >*/
} else if (lsame_(compz, "V", (ftnlen)1, (ftnlen)1)) {
/*< ICOMPZ = 1 >*/
icompz = 1;
/*< ELSE IF( LSAME( COMPZ, 'I' ) ) THEN >*/
} else if (lsame_(compz, "I", (ftnlen)1, (ftnlen)1)) {
/*< ICOMPZ = 2 >*/
icompz = 2;
/*< ELSE >*/
} else {
/*< ICOMPZ = -1 >*/
icompz = -1;
/*< END IF >*/
}
/*< IF( ICOMPZ.LT.0 ) THEN >*/
if (icompz < 0) {
/*< INFO = -1 >*/
*info = -1;
/*< ELSE IF( N.LT.0 ) THEN >*/
} else if (*n < 0) {
/*< INFO = -2 >*/
*info = -2;
/*< >*/
} else if (*ldz < 1 || (icompz > 0 && *ldz < max(1,*n))) {
/*< INFO = -6 >*/
*info = -6;
/*< END IF >*/
}
/*< IF( INFO.NE.0 ) THEN >*/
if (*info != 0) {
/*< CALL XERBLA( 'DSTEQR', -INFO ) >*/
i__1 = -(*info);
xerbla_("DSTEQR", &i__1, (ftnlen)6);
/*< RETURN >*/
return 0;
/*< END IF >*/
}
/* Quick return if possible */
/*< >*/
if (*n == 0) {
return 0;
}
/*< IF( N.EQ.1 ) THEN >*/
if (*n == 1) {
/*< >*/
if (icompz == 2) {
z__[z_dim1 + 1] = 1.;
}
/*< RETURN >*/
return 0;
/*< END IF >*/
}
/* Determine the unit roundoff and over/underflow thresholds. */
/*< EPS = DLAMCH( 'E' ) >*/
eps = dlamch_("E", (ftnlen)1);
/*< EPS2 = EPS**2 >*/
/* Computing 2nd power */
d__1 = eps;
eps2 = d__1 * d__1;
/*< SAFMIN = DLAMCH( 'S' ) >*/
safmin = dlamch_("S", (ftnlen)1);
/*< SAFMAX = ONE / SAFMIN >*/
safmax = 1. / safmin;
/*< SSFMAX = SQRT( SAFMAX ) / THREE >*/
ssfmax = sqrt(safmax) / 3.;
/*< SSFMIN = SQRT( SAFMIN ) / EPS2 >*/
ssfmin = sqrt(safmin) / eps2;
/* Compute the eigenvalues and eigenvectors of the tridiagonal */
/* matrix. */
/*< >*/
if (icompz == 2) {
dlaset_("Full", n, n, &c_b9, &c_b10, &z__[z_offset], ldz, (ftnlen)4);
}
/*< NMAXIT = N*MAXIT >*/
nmaxit = *n * 30;
/*< JTOT = 0 >*/
jtot = 0;
/* Determine where the matrix splits and choose QL or QR iteration */
/* for each block, according to whether top or bottom diagonal */
/* element is smaller. */
/*< L1 = 1 >*/
l1 = 1;
/*< NM1 = N - 1 >*/
nm1 = *n - 1;
/*< 10 CONTINUE >*/
L10:
/*< >*/
if (l1 > *n) {
goto L160;
}
/*< >*/
if (l1 > 1) {
e[l1 - 1] = 0.;
}
/*< IF( L1.LE.NM1 ) THEN >*/
if (l1 <= nm1) {
/*< DO 20 M = L1, NM1 >*/
i__1 = nm1;
for (m = l1; m <= i__1; ++m) {
/*< TST = ABS( E( M ) ) >*/
tst = (d__1 = e[m], abs(d__1));
/*< >*/
if (tst == 0.) {
goto L30;
}
/*< >*/
if (tst <= sqrt((d__1 = d__[m], abs(d__1))) * sqrt((d__2 = d__[m
+ 1], abs(d__2))) * eps) {
/*< E( M ) = ZERO >*/
e[m] = 0.;
/*< GO TO 30 >*/
goto L30;
/*< END IF >*/
}
/*< 20 CONTINUE >*/
/* L20: */
}
/*< END IF >*/
}
/*< M = N >*/
m = *n;
/*< 30 CONTINUE >*/
L30:
/*< L = L1 >*/
l = l1;
/*< LSV = L >*/
lsv = l;
/*< LEND = M >*/
lend = m;
/*< LENDSV = LEND >*/
lendsv = lend;
/*< L1 = M + 1 >*/
l1 = m + 1;
/*< >*/
if (lend == l) {
goto L10;
}
/* Scale submatrix in rows and columns L to LEND */
/*< ANORM = DLANST( 'I', LEND-L+1, D( L ), E( L ) ) >*/
i__1 = lend - l + 1;
anorm = dlanst_("I", &i__1, &d__[l], &e[l], (ftnlen)1);
/*< ISCALE = 0 >*/
iscale = 0;
/*< >*/
if (anorm == 0.) {
goto L10;
}
/*< IF( ANORM.GT.SSFMAX ) THEN >*/
if (anorm > ssfmax) {
/*< ISCALE = 1 >*/
iscale = 1;
/*< >*/
i__1 = lend - l + 1;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmax, &i__1, &c__1, &d__[l], n,
info, (ftnlen)1);
/*< >*/
i__1 = lend - l;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmax, &i__1, &c__1, &e[l], n,
info, (ftnlen)1);
/*< ELSE IF( ANORM.LT.SSFMIN ) THEN >*/
} else if (anorm < ssfmin) {
/*< ISCALE = 2 >*/
iscale = 2;
/*< >*/
i__1 = lend - l + 1;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &d__[l], n,
info, (ftnlen)1);
/*< >*/
i__1 = lend - l;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &e[l], n,
info, (ftnlen)1);
/*< END IF >*/
}
/* Choose between QL and QR iteration */
/*< IF( ABS( D( LEND ) ).LT.ABS( D( L ) ) ) THEN >*/
if ((d__1 = d__[lend], abs(d__1)) < (d__2 = d__[l], abs(d__2))) {
/*< LEND = LSV >*/
lend = lsv;
/*< L = LENDSV >*/
l = lendsv;
/*< END IF >*/
}
/*< IF( LEND.GT.L ) THEN >*/
if (lend > l) {
/* QL Iteration */
/* Look for small subdiagonal element. */
/*< 40 CONTINUE >*/
L40:
/*< IF( L.NE.LEND ) THEN >*/
if (l != lend) {
/*< LENDM1 = LEND - 1 >*/
lendm1 = lend - 1;
/*< DO 50 M = L, LENDM1 >*/
i__1 = lendm1;
for (m = l; m <= i__1; ++m) {
/*< TST = ABS( E( M ) )**2 >*/
/* Computing 2nd power */
d__2 = (d__1 = e[m], abs(d__1));
tst = d__2 * d__2;
/*< >*/
if (tst <= eps2 * (d__1 = d__[m], abs(d__1)) * (d__2 = d__[m
+ 1], abs(d__2)) + safmin) {
goto L60;
}
/*< 50 CONTINUE >*/
/* L50: */
}
/*< END IF >*/
}
/*< M = LEND >*/
m = lend;
/*< 60 CONTINUE >*/
L60:
/*< >*/
if (m < lend) {
e[m] = 0.;
}
/*< P = D( L ) >*/
p = d__[l];
/*< >*/
if (m == l) {
goto L80;
}
/* If remaining matrix is 2-by-2, use DLAE2 or SLAEV2 */
/* to compute its eigensystem. */
/*< IF( M.EQ.L+1 ) THEN >*/
if (m == l + 1) {
/*< IF( ICOMPZ.GT.0 ) THEN >*/
if (icompz > 0) {
/*< CALL DLAEV2( D( L ), E( L ), D( L+1 ), RT1, RT2, C, S ) >*/
dlaev2_(&d__[l], &e[l], &d__[l + 1], &rt1, &rt2, &c__, &s);
/*< WORK( L ) = C >*/
work[l] = c__;
/*< WORK( N-1+L ) = S >*/
work[*n - 1 + l] = s;
/*< >*/
dlasr_("R", "V", "B", n, &c__2, &work[l], &work[*n - 1 + l], &
z__[l * z_dim1 + 1], ldz, (ftnlen)1, (ftnlen)1, (
ftnlen)1);
/*< ELSE >*/
} else {
/*< CALL DLAE2( D( L ), E( L ), D( L+1 ), RT1, RT2 ) >*/
dlae2_(&d__[l], &e[l], &d__[l + 1], &rt1, &rt2);
/*< END IF >*/
}
/*< D( L ) = RT1 >*/
d__[l] = rt1;
/*< D( L+1 ) = RT2 >*/
d__[l + 1] = rt2;
/*< E( L ) = ZERO >*/
e[l] = 0.;
/*< L = L + 2 >*/
l += 2;
/*< >*/
if (l <= lend) {
goto L40;
}
/*< GO TO 140 >*/
goto L140;
/*< END IF >*/
}
/*< >*/
if (jtot == nmaxit) {
goto L140;
}
/*< JTOT = JTOT + 1 >*/
++jtot;
/* Form shift. */
/*< G = ( D( L+1 )-P ) / ( TWO*E( L ) ) >*/
g = (d__[l + 1] - p) / (e[l] * 2.);
/*< R = DLAPY2( G, ONE ) >*/
r__ = dlapy2_(&g, &c_b10);
/*< G = D( M ) - P + ( E( L ) / ( G+SIGN( R, G ) ) ) >*/
g = d__[m] - p + e[l] / (g + d_sign(&r__, &g));
/*< S = ONE >*/
s = 1.;
/*< C = ONE >*/
c__ = 1.;
/*< P = ZERO >*/
p = 0.;
/* Inner loop */
/*< MM1 = M - 1 >*/
mm1 = m - 1;
/*< DO 70 I = MM1, L, -1 >*/
i__1 = l;
for (i__ = mm1; i__ >= i__1; --i__) {
/*< F = S*E( I ) >*/
f = s * e[i__];
/*< B = C*E( I ) >*/
b = c__ * e[i__];
/*< CALL DLARTG( G, F, C, S, R ) >*/
dlartg_(&g, &f, &c__, &s, &r__);
/*< >*/
if (i__ != m - 1) {
e[i__ + 1] = r__;
}
/*< G = D( I+1 ) - P >*/
g = d__[i__ + 1] - p;
/*< R = ( D( I )-G )*S + TWO*C*B >*/
r__ = (d__[i__] - g) * s + c__ * 2. * b;
/*< P = S*R >*/
p = s * r__;
/*< D( I+1 ) = G + P >*/
d__[i__ + 1] = g + p;
/*< G = C*R - B >*/
g = c__ * r__ - b;
/* If eigenvectors are desired, then save rotations. */
/*< IF( ICOMPZ.GT.0 ) THEN >*/
if (icompz > 0) {
/*< WORK( I ) = C >*/
work[i__] = c__;
/*< WORK( N-1+I ) = -S >*/
work[*n - 1 + i__] = -s;
/*< END IF >*/
}
/*< 70 CONTINUE >*/
/* L70: */
}
/* If eigenvectors are desired, then apply saved rotations. */
/*< IF( ICOMPZ.GT.0 ) THEN >*/
if (icompz > 0) {
/*< MM = M - L + 1 >*/
mm = m - l + 1;
/*< >*/
dlasr_("R", "V", "B", n, &mm, &work[l], &work[*n - 1 + l], &z__[l
* z_dim1 + 1], ldz, (ftnlen)1, (ftnlen)1, (ftnlen)1);
/*< END IF >*/
}
/*< D( L ) = D( L ) - P >*/
d__[l] -= p;
/*< E( L ) = G >*/
e[l] = g;
/*< GO TO 40 >*/
goto L40;
/* Eigenvalue found. */
/*< 80 CONTINUE >*/
L80:
/*< D( L ) = P >*/
d__[l] = p;
/*< L = L + 1 >*/
++l;
/*< >*/
if (l <= lend) {
goto L40;
}
/*< GO TO 140 >*/
goto L140;
/*< ELSE >*/
} else {
/* QR Iteration */
/* Look for small superdiagonal element. */
/*< 90 CONTINUE >*/
L90:
/*< IF( L.NE.LEND ) THEN >*/
if (l != lend) {
/*< LENDP1 = LEND + 1 >*/
lendp1 = lend + 1;
/*< DO 100 M = L, LENDP1, -1 >*/
i__1 = lendp1;
for (m = l; m >= i__1; --m) {
/*< TST = ABS( E( M-1 ) )**2 >*/
/* Computing 2nd power */
d__2 = (d__1 = e[m - 1], abs(d__1));
tst = d__2 * d__2;
/*< >*/
if (tst <= eps2 * (d__1 = d__[m], abs(d__1)) * (d__2 = d__[m
- 1], abs(d__2)) + safmin) {
goto L110;
}
/*< 100 CONTINUE >*/
/* L100: */
}
/*< END IF >*/
}
/*< M = LEND >*/
m = lend;
/*< 110 CONTINUE >*/
L110:
/*< >*/
if (m > lend) {
e[m - 1] = 0.;
}
/*< P = D( L ) >*/
p = d__[l];
/*< >*/
if (m == l) {
goto L130;
}
/* If remaining matrix is 2-by-2, use DLAE2 or SLAEV2 */
/* to compute its eigensystem. */
/*< IF( M.EQ.L-1 ) THEN >*/
if (m == l - 1) {
/*< IF( ICOMPZ.GT.0 ) THEN >*/
if (icompz > 0) {
/*< CALL DLAEV2( D( L-1 ), E( L-1 ), D( L ), RT1, RT2, C, S ) >*/
dlaev2_(&d__[l - 1], &e[l - 1], &d__[l], &rt1, &rt2, &c__, &s)
;
/*< WORK( M ) = C >*/
work[m] = c__;
/*< WORK( N-1+M ) = S >*/
work[*n - 1 + m] = s;
/*< >*/
dlasr_("R", "V", "F", n, &c__2, &work[m], &work[*n - 1 + m], &
z__[(l - 1) * z_dim1 + 1], ldz, (ftnlen)1, (ftnlen)1,
(ftnlen)1);
/*< ELSE >*/
} else {
/*< CALL DLAE2( D( L-1 ), E( L-1 ), D( L ), RT1, RT2 ) >*/
dlae2_(&d__[l - 1], &e[l - 1], &d__[l], &rt1, &rt2);
/*< END IF >*/
}
/*< D( L-1 ) = RT1 >*/
d__[l - 1] = rt1;
/*< D( L ) = RT2 >*/
d__[l] = rt2;
/*< E( L-1 ) = ZERO >*/
e[l - 1] = 0.;
/*< L = L - 2 >*/
l += -2;
/*< >*/
if (l >= lend) {
goto L90;
}
/*< GO TO 140 >*/
goto L140;
/*< END IF >*/
}
/*< >*/
if (jtot == nmaxit) {
goto L140;
}
/*< JTOT = JTOT + 1 >*/
++jtot;
/* Form shift. */
/*< G = ( D( L-1 )-P ) / ( TWO*E( L-1 ) ) >*/
g = (d__[l - 1] - p) / (e[l - 1] * 2.);
/*< R = DLAPY2( G, ONE ) >*/
r__ = dlapy2_(&g, &c_b10);
/*< G = D( M ) - P + ( E( L-1 ) / ( G+SIGN( R, G ) ) ) >*/
g = d__[m] - p + e[l - 1] / (g + d_sign(&r__, &g));
/*< S = ONE >*/
s = 1.;
/*< C = ONE >*/
c__ = 1.;
/*< P = ZERO >*/
p = 0.;
/* Inner loop */
/*< LM1 = L - 1 >*/
lm1 = l - 1;
/*< DO 120 I = M, LM1 >*/
i__1 = lm1;
for (i__ = m; i__ <= i__1; ++i__) {
/*< F = S*E( I ) >*/
f = s * e[i__];
/*< B = C*E( I ) >*/
b = c__ * e[i__];
/*< CALL DLARTG( G, F, C, S, R ) >*/
dlartg_(&g, &f, &c__, &s, &r__);
/*< >*/
if (i__ != m) {
e[i__ - 1] = r__;
}
/*< G = D( I ) - P >*/
g = d__[i__] - p;
/*< R = ( D( I+1 )-G )*S + TWO*C*B >*/
r__ = (d__[i__ + 1] - g) * s + c__ * 2. * b;
/*< P = S*R >*/
p = s * r__;
/*< D( I ) = G + P >*/
d__[i__] = g + p;
/*< G = C*R - B >*/
g = c__ * r__ - b;
/* If eigenvectors are desired, then save rotations. */
/*< IF( ICOMPZ.GT.0 ) THEN >*/
if (icompz > 0) {
/*< WORK( I ) = C >*/
work[i__] = c__;
/*< WORK( N-1+I ) = S >*/
work[*n - 1 + i__] = s;
/*< END IF >*/
}
/*< 120 CONTINUE >*/
/* L120: */
}
/* If eigenvectors are desired, then apply saved rotations. */
/*< IF( ICOMPZ.GT.0 ) THEN >*/
if (icompz > 0) {
/*< MM = L - M + 1 >*/
mm = l - m + 1;
/*< >*/
dlasr_("R", "V", "F", n, &mm, &work[m], &work[*n - 1 + m], &z__[m
* z_dim1 + 1], ldz, (ftnlen)1, (ftnlen)1, (ftnlen)1);
/*< END IF >*/
}
/*< D( L ) = D( L ) - P >*/
d__[l] -= p;
/*< E( LM1 ) = G >*/
e[lm1] = g;
/*< GO TO 90 >*/
goto L90;
/* Eigenvalue found. */
/*< 130 CONTINUE >*/
L130:
/*< D( L ) = P >*/
d__[l] = p;
/*< L = L - 1 >*/
--l;
/*< >*/
if (l >= lend) {
goto L90;
}
/*< GO TO 140 >*/
goto L140;
/*< END IF >*/
}
/* Undo scaling if necessary */
/*< 140 CONTINUE >*/
L140:
/*< IF( ISCALE.EQ.1 ) THEN >*/
if (iscale == 1) {
/*< >*/
i__1 = lendsv - lsv + 1;
dlascl_("G", &c__0, &c__0, &ssfmax, &anorm, &i__1, &c__1, &d__[lsv],
n, info, (ftnlen)1);
/*< >*/
i__1 = lendsv - lsv;
dlascl_("G", &c__0, &c__0, &ssfmax, &anorm, &i__1, &c__1, &e[lsv], n,
info, (ftnlen)1);
/*< ELSE IF( ISCALE.EQ.2 ) THEN >*/
} else if (iscale == 2) {
/*< >*/
i__1 = lendsv - lsv + 1;
dlascl_("G", &c__0, &c__0, &ssfmin, &anorm, &i__1, &c__1, &d__[lsv],
n, info, (ftnlen)1);
/*< >*/
i__1 = lendsv - lsv;
dlascl_("G", &c__0, &c__0, &ssfmin, &anorm, &i__1, &c__1, &e[lsv], n,
info, (ftnlen)1);
/*< END IF >*/
}
/* Check for no convergence to an eigenvalue after a total */
/* of N*MAXIT iterations. */
/*< >*/
if (jtot < nmaxit) {
goto L10;
}
/*< DO 150 I = 1, N - 1 >*/
i__1 = *n - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
/*< >*/
if (e[i__] != 0.) {
++(*info);
}
/*< 150 CONTINUE >*/
/* L150: */
}
/*< GO TO 190 >*/
goto L190;
/* Order eigenvalues and eigenvectors. */
/*< 160 CONTINUE >*/
L160:
/*< IF( ICOMPZ.EQ.0 ) THEN >*/
if (icompz == 0) {
/* Use Quick Sort */
/*< CALL DLASRT( 'I', N, D, INFO ) >*/
dlasrt_("I", n, &d__[1], info, (ftnlen)1);
/*< ELSE >*/
} else {
/* Use Selection Sort to minimize swaps of eigenvectors */
/*< DO 180 II = 2, N >*/
i__1 = *n;
for (ii = 2; ii <= i__1; ++ii) {
/*< I = II - 1 >*/
i__ = ii - 1;
/*< K = I >*/
k = i__;
/*< P = D( I ) >*/
p = d__[i__];
/*< DO 170 J = II, N >*/
i__2 = *n;
for (j = ii; j <= i__2; ++j) {
/*< IF( D( J ).LT.P ) THEN >*/
if (d__[j] < p) {
/*< K = J >*/
k = j;
/*< P = D( J ) >*/
p = d__[j];
/*< END IF >*/
}
/*< 170 CONTINUE >*/
/* L170: */
}
/*< IF( K.NE.I ) THEN >*/
if (k != i__) {
/*< D( K ) = D( I ) >*/
d__[k] = d__[i__];
/*< D( I ) = P >*/
d__[i__] = p;
/*< CALL DSWAP( N, Z( 1, I ), 1, Z( 1, K ), 1 ) >*/
dswap_(n, &z__[i__ * z_dim1 + 1], &c__1, &z__[k * z_dim1 + 1],
&c__1);
/*< END IF >*/
}
/*< 180 CONTINUE >*/
/* L180: */
}
/*< END IF >*/
}
/*< 190 CONTINUE >*/
L190:
/*< RETURN >*/
return 0;
/* End of DSTEQR */
/*< END >*/
} /* dsteqr_ */
/* Subroutine */ int zhseqr_(char *job, char *compz, integer *n, integer *ilo,
integer *ihi, doublecomplex *h__, integer *ldh, doublecomplex *w,
doublecomplex *z__, integer *ldz, doublecomplex *work, integer *lwork,
integer *info)
{
/* -- LAPACK routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
June 30, 1999
Purpose
=======
ZHSEQR computes the eigenvalues of a complex upper Hessenberg
matrix H, and, optionally, the matrices T and Z from the Schur
decomposition H = Z T Z**H, where T is an upper triangular matrix
(the Schur form), and Z is the unitary matrix of Schur vectors.
Optionally Z may be postmultiplied into an input unitary matrix Q,
so that this routine can give the Schur factorization of a matrix A
which has been reduced to the Hessenberg form H by the unitary
matrix Q: A = Q*H*Q**H = (QZ)*T*(QZ)**H.
Arguments
=========
JOB (input) CHARACTER*1
= 'E': compute eigenvalues only;
= 'S': compute eigenvalues and the Schur form T.
COMPZ (input) CHARACTER*1
= 'N': no Schur vectors are computed;
= 'I': Z is initialized to the unit matrix and the matrix Z
of Schur vectors of H is returned;
= 'V': Z must contain an unitary matrix Q on entry, and
the product Q*Z is returned.
N (input) INTEGER
The order of the matrix H. N >= 0.
ILO (input) INTEGER
IHI (input) INTEGER
It is assumed that H is already upper triangular in rows
and columns 1:ILO-1 and IHI+1:N. ILO and IHI are normally
set by a previous call to ZGEBAL, and then passed to CGEHRD
when the matrix output by ZGEBAL is reduced to Hessenberg
form. Otherwise ILO and IHI should be set to 1 and N
respectively.
1 <= ILO <= IHI <= N, if N > 0; ILO=1 and IHI=0, if N=0.
H (input/output) COMPLEX*16 array, dimension (LDH,N)
On entry, the upper Hessenberg matrix H.
On exit, if JOB = 'S', H contains the upper triangular matrix
T from the Schur decomposition (the Schur form). If
JOB = 'E', the contents of H are unspecified on exit.
LDH (input) INTEGER
The leading dimension of the array H. LDH >= max(1,N).
W (output) COMPLEX*16 array, dimension (N)
The computed eigenvalues. If JOB = 'S', the eigenvalues are
stored in the same order as on the diagonal of the Schur form
returned in H, with W(i) = H(i,i).
Z (input/output) COMPLEX*16 array, dimension (LDZ,N)
If COMPZ = 'N': Z is not referenced.
If COMPZ = 'I': on entry, Z need not be set, and on exit, Z
contains the unitary matrix Z of the Schur vectors of H.
If COMPZ = 'V': on entry Z must contain an N-by-N matrix Q,
which is assumed to be equal to the unit matrix except for
the submatrix Z(ILO:IHI,ILO:IHI); on exit Z contains Q*Z.
Normally Q is the unitary matrix generated by ZUNGHR after
the call to ZGEHRD which formed the Hessenberg matrix H.
LDZ (input) INTEGER
The leading dimension of the array Z.
LDZ >= max(1,N) if COMPZ = 'I' or 'V'; LDZ >= 1 otherwise.
WORK (workspace/output) COMPLEX*16 array, dimension (LWORK)
On exit, if INFO = 0, WORK(1) returns the optimal LWORK.
LWORK (input) INTEGER
The dimension of the array WORK. LWORK >= max(1,N).
If LWORK = -1, then a workspace query is assumed; the routine
only calculates the optimal size of the WORK array, returns
this value as the first entry of the WORK array, and no error
message related to LWORK is issued by XERBLA.
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
> 0: if INFO = i, ZHSEQR failed to compute all the
eigenvalues in a total of 30*(IHI-ILO+1) iterations;
elements 1:ilo-1 and i+1:n of W contain those
eigenvalues which have been successfully computed.
=====================================================================
Decode and test the input parameters
Parameter adjustments */
/* Table of constant values */
static doublecomplex c_b1 = {0.,0.};
static doublecomplex c_b2 = {1.,0.};
static integer c__1 = 1;
static integer c__4 = 4;
static integer c_n1 = -1;
static integer c__2 = 2;
static integer c__8 = 8;
static integer c__15 = 15;
static logical c_false = FALSE_;
/* System generated locals */
address a__1[2];
integer h_dim1, h_offset, z_dim1, z_offset, i__1, i__2, i__3, i__4[2],
i__5, i__6;
doublereal d__1, d__2, d__3, d__4;
doublecomplex z__1;
char ch__1[2];
/* Builtin functions */
double d_imag(doublecomplex *);
void d_cnjg(doublecomplex *, doublecomplex *);
/* Subroutine */ int s_cat(char *, char **, integer *, integer *, ftnlen);
/* Local variables */
static integer maxb, ierr;
static doublereal unfl;
static doublecomplex temp;
static doublereal ovfl;
static integer i__, j, k, l;
static doublecomplex s[225] /* was [15][15] */, v[16];
extern logical lsame_(char *, char *);
extern /* Subroutine */ int zscal_(integer *, doublecomplex *,
doublecomplex *, integer *);
static integer itemp;
static doublereal rtemp;
static integer i1, i2;
extern /* Subroutine */ int zgemv_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublecomplex *, doublecomplex *, integer *);
static logical initz, wantt, wantz;
static doublereal rwork[1];
extern /* Subroutine */ int zcopy_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *);
extern doublereal dlapy2_(doublereal *, doublereal *);
extern /* Subroutine */ int dlabad_(doublereal *, doublereal *);
static integer ii, nh;
extern doublereal dlamch_(char *);
static integer nr, ns, nv;
static doublecomplex vv[16];
extern /* Subroutine */ int xerbla_(char *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *, ftnlen, ftnlen);
extern /* Subroutine */ int zdscal_(integer *, doublereal *,
doublecomplex *, integer *), zlarfg_(integer *, doublecomplex *,
doublecomplex *, integer *, doublecomplex *);
extern integer izamax_(integer *, doublecomplex *, integer *);
extern doublereal zlanhs_(char *, integer *, doublecomplex *, integer *,
doublereal *);
extern /* Subroutine */ int zlahqr_(logical *, logical *, integer *,
integer *, integer *, doublecomplex *, integer *, doublecomplex *,
integer *, integer *, doublecomplex *, integer *, integer *),
zlacpy_(char *, integer *, integer *, doublecomplex *, integer *,
doublecomplex *, integer *), zlaset_(char *, integer *,
integer *, doublecomplex *, doublecomplex *, doublecomplex *,
integer *), zlarfx_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, doublecomplex *, integer *,
doublecomplex *);
static doublereal smlnum;
static logical lquery;
static integer itn;
static doublecomplex tau;
static integer its;
static doublereal ulp, tst1;
#define h___subscr(a_1,a_2) (a_2)*h_dim1 + a_1
#define h___ref(a_1,a_2) h__[h___subscr(a_1,a_2)]
#define s_subscr(a_1,a_2) (a_2)*15 + a_1 - 16
#define s_ref(a_1,a_2) s[s_subscr(a_1,a_2)]
#define z___subscr(a_1,a_2) (a_2)*z_dim1 + a_1
#define z___ref(a_1,a_2) z__[z___subscr(a_1,a_2)]
h_dim1 = *ldh;
h_offset = 1 + h_dim1 * 1;
h__ -= h_offset;
--w;
z_dim1 = *ldz;
z_offset = 1 + z_dim1 * 1;
z__ -= z_offset;
--work;
/* Function Body */
wantt = lsame_(job, "S");
initz = lsame_(compz, "I");
wantz = initz || lsame_(compz, "V");
*info = 0;
i__1 = max(1,*n);
work[1].r = (doublereal) i__1, work[1].i = 0.;
lquery = *lwork == -1;
if (! lsame_(job, "E") && ! wantt) {
*info = -1;
} else if (! lsame_(compz, "N") && ! wantz) {
*info = -2;
} else if (*n < 0) {
*info = -3;
} else if (*ilo < 1 || *ilo > max(1,*n)) {
*info = -4;
} else if (*ihi < min(*ilo,*n) || *ihi > *n) {
*info = -5;
} else if (*ldh < max(1,*n)) {
*info = -7;
} else if (*ldz < 1 || wantz && *ldz < max(1,*n)) {
*info = -10;
} else if (*lwork < max(1,*n) && ! lquery) {
*info = -12;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("ZHSEQR", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Initialize Z, if necessary */
if (initz) {
zlaset_("Full", n, n, &c_b1, &c_b2, &z__[z_offset], ldz);
}
/* Store the eigenvalues isolated by ZGEBAL. */
i__1 = *ilo - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = i__;
i__3 = h___subscr(i__, i__);
w[i__2].r = h__[i__3].r, w[i__2].i = h__[i__3].i;
/* L10: */
}
i__1 = *n;
for (i__ = *ihi + 1; i__ <= i__1; ++i__) {
i__2 = i__;
i__3 = h___subscr(i__, i__);
w[i__2].r = h__[i__3].r, w[i__2].i = h__[i__3].i;
/* L20: */
}
/* Quick return if possible. */
if (*n == 0) {
return 0;
}
if (*ilo == *ihi) {
i__1 = *ilo;
i__2 = h___subscr(*ilo, *ilo);
w[i__1].r = h__[i__2].r, w[i__1].i = h__[i__2].i;
return 0;
}
/* Set rows and columns ILO to IHI to zero below the first
subdiagonal. */
i__1 = *ihi - 2;
for (j = *ilo; j <= i__1; ++j) {
i__2 = *n;
for (i__ = j + 2; i__ <= i__2; ++i__) {
i__3 = h___subscr(i__, j);
h__[i__3].r = 0., h__[i__3].i = 0.;
/* L30: */
}
/* L40: */
}
nh = *ihi - *ilo + 1;
/* I1 and I2 are the indices of the first row and last column of H
to which transformations must be applied. If eigenvalues only are
being computed, I1 and I2 are re-set inside the main loop. */
if (wantt) {
i1 = 1;
i2 = *n;
} else {
i1 = *ilo;
i2 = *ihi;
}
/* Ensure that the subdiagonal elements are real. */
i__1 = *ihi;
for (i__ = *ilo + 1; i__ <= i__1; ++i__) {
i__2 = h___subscr(i__, i__ - 1);
temp.r = h__[i__2].r, temp.i = h__[i__2].i;
if (d_imag(&temp) != 0.) {
d__1 = temp.r;
d__2 = d_imag(&temp);
rtemp = dlapy2_(&d__1, &d__2);
i__2 = h___subscr(i__, i__ - 1);
h__[i__2].r = rtemp, h__[i__2].i = 0.;
z__1.r = temp.r / rtemp, z__1.i = temp.i / rtemp;
temp.r = z__1.r, temp.i = z__1.i;
if (i2 > i__) {
i__2 = i2 - i__;
d_cnjg(&z__1, &temp);
zscal_(&i__2, &z__1, &h___ref(i__, i__ + 1), ldh);
}
i__2 = i__ - i1;
zscal_(&i__2, &temp, &h___ref(i1, i__), &c__1);
if (i__ < *ihi) {
i__2 = h___subscr(i__ + 1, i__);
i__3 = h___subscr(i__ + 1, i__);
z__1.r = temp.r * h__[i__3].r - temp.i * h__[i__3].i, z__1.i =
temp.r * h__[i__3].i + temp.i * h__[i__3].r;
h__[i__2].r = z__1.r, h__[i__2].i = z__1.i;
}
if (wantz) {
zscal_(&nh, &temp, &z___ref(*ilo, i__), &c__1);
}
}
/* L50: */
}
/* Determine the order of the multi-shift QR algorithm to be used.
Writing concatenation */
i__4[0] = 1, a__1[0] = job;
i__4[1] = 1, a__1[1] = compz;
s_cat(ch__1, a__1, i__4, &c__2, (ftnlen)2);
ns = ilaenv_(&c__4, "ZHSEQR", ch__1, n, ilo, ihi, &c_n1, (ftnlen)6, (
ftnlen)2);
/* Writing concatenation */
i__4[0] = 1, a__1[0] = job;
i__4[1] = 1, a__1[1] = compz;
s_cat(ch__1, a__1, i__4, &c__2, (ftnlen)2);
maxb = ilaenv_(&c__8, "ZHSEQR", ch__1, n, ilo, ihi, &c_n1, (ftnlen)6, (
ftnlen)2);
if (ns <= 1 || ns > nh || maxb >= nh) {
/* Use the standard double-shift algorithm */
zlahqr_(&wantt, &wantz, n, ilo, ihi, &h__[h_offset], ldh, &w[1], ilo,
ihi, &z__[z_offset], ldz, info);
return 0;
}
maxb = max(2,maxb);
/* Computing MIN */
i__1 = min(ns,maxb);
ns = min(i__1,15);
/* Now 1 < NS <= MAXB < NH.
Set machine-dependent constants for the stopping criterion.
If norm(H) <= sqrt(OVFL), overflow should not occur. */
unfl = dlamch_("Safe minimum");
ovfl = 1. / unfl;
dlabad_(&unfl, &ovfl);
ulp = dlamch_("Precision");
smlnum = unfl * (nh / ulp);
/* ITN is the total number of multiple-shift QR iterations allowed. */
itn = nh * 30;
/* The main loop begins here. I is the loop index and decreases from
IHI to ILO in steps of at most MAXB. Each iteration of the loop
works with the active submatrix in rows and columns L to I.
Eigenvalues I+1 to IHI have already converged. Either L = ILO, or
H(L,L-1) is negligible so that the matrix splits. */
i__ = *ihi;
L60:
if (i__ < *ilo) {
goto L180;
}
/* Perform multiple-shift QR iterations on rows and columns ILO to I
until a submatrix of order at most MAXB splits off at the bottom
because a subdiagonal element has become negligible. */
l = *ilo;
i__1 = itn;
for (its = 0; its <= i__1; ++its) {
/* Look for a single small subdiagonal element. */
i__2 = l + 1;
for (k = i__; k >= i__2; --k) {
i__3 = h___subscr(k - 1, k - 1);
i__5 = h___subscr(k, k);
tst1 = (d__1 = h__[i__3].r, abs(d__1)) + (d__2 = d_imag(&h___ref(
k - 1, k - 1)), abs(d__2)) + ((d__3 = h__[i__5].r, abs(
d__3)) + (d__4 = d_imag(&h___ref(k, k)), abs(d__4)));
if (tst1 == 0.) {
i__3 = i__ - l + 1;
tst1 = zlanhs_("1", &i__3, &h___ref(l, l), ldh, rwork);
}
i__3 = h___subscr(k, k - 1);
/* Computing MAX */
d__2 = ulp * tst1;
if ((d__1 = h__[i__3].r, abs(d__1)) <= max(d__2,smlnum)) {
goto L80;
}
/* L70: */
}
L80:
l = k;
if (l > *ilo) {
/* H(L,L-1) is negligible. */
i__2 = h___subscr(l, l - 1);
h__[i__2].r = 0., h__[i__2].i = 0.;
}
/* Exit from loop if a submatrix of order <= MAXB has split off. */
if (l >= i__ - maxb + 1) {
goto L170;
}
/* Now the active submatrix is in rows and columns L to I. If
eigenvalues only are being computed, only the active submatrix
need be transformed. */
if (! wantt) {
i1 = l;
i2 = i__;
}
if (its == 20 || its == 30) {
/* Exceptional shifts. */
i__2 = i__;
for (ii = i__ - ns + 1; ii <= i__2; ++ii) {
i__3 = ii;
i__5 = h___subscr(ii, ii - 1);
i__6 = h___subscr(ii, ii);
d__3 = ((d__1 = h__[i__5].r, abs(d__1)) + (d__2 = h__[i__6].r,
abs(d__2))) * 1.5;
w[i__3].r = d__3, w[i__3].i = 0.;
/* L90: */
}
} else {
/* Use eigenvalues of trailing submatrix of order NS as shifts. */
zlacpy_("Full", &ns, &ns, &h___ref(i__ - ns + 1, i__ - ns + 1),
ldh, s, &c__15);
zlahqr_(&c_false, &c_false, &ns, &c__1, &ns, s, &c__15, &w[i__ -
ns + 1], &c__1, &ns, &z__[z_offset], ldz, &ierr);
if (ierr > 0) {
/* If ZLAHQR failed to compute all NS eigenvalues, use the
unconverged diagonal elements as the remaining shifts. */
i__2 = ierr;
for (ii = 1; ii <= i__2; ++ii) {
i__3 = i__ - ns + ii;
i__5 = s_subscr(ii, ii);
w[i__3].r = s[i__5].r, w[i__3].i = s[i__5].i;
/* L100: */
}
}
}
/* Form the first column of (G-w(1)) (G-w(2)) . . . (G-w(ns))
where G is the Hessenberg submatrix H(L:I,L:I) and w is
the vector of shifts (stored in W). The result is
stored in the local array V. */
v[0].r = 1., v[0].i = 0.;
i__2 = ns + 1;
for (ii = 2; ii <= i__2; ++ii) {
i__3 = ii - 1;
v[i__3].r = 0., v[i__3].i = 0.;
/* L110: */
}
nv = 1;
i__2 = i__;
for (j = i__ - ns + 1; j <= i__2; ++j) {
i__3 = nv + 1;
zcopy_(&i__3, v, &c__1, vv, &c__1);
i__3 = nv + 1;
i__5 = j;
z__1.r = -w[i__5].r, z__1.i = -w[i__5].i;
zgemv_("No transpose", &i__3, &nv, &c_b2, &h___ref(l, l), ldh, vv,
&c__1, &z__1, v, &c__1);
++nv;
/* Scale V(1:NV) so that max(abs(V(i))) = 1. If V is zero,
reset it to the unit vector. */
itemp = izamax_(&nv, v, &c__1);
i__3 = itemp - 1;
rtemp = (d__1 = v[i__3].r, abs(d__1)) + (d__2 = d_imag(&v[itemp -
1]), abs(d__2));
if (rtemp == 0.) {
v[0].r = 1., v[0].i = 0.;
i__3 = nv;
for (ii = 2; ii <= i__3; ++ii) {
i__5 = ii - 1;
v[i__5].r = 0., v[i__5].i = 0.;
/* L120: */
}
} else {
rtemp = max(rtemp,smlnum);
d__1 = 1. / rtemp;
zdscal_(&nv, &d__1, v, &c__1);
}
/* L130: */
}
/* Multiple-shift QR step */
i__2 = i__ - 1;
for (k = l; k <= i__2; ++k) {
/* The first iteration of this loop determines a reflection G
from the vector V and applies it from left and right to H,
thus creating a nonzero bulge below the subdiagonal.
Each subsequent iteration determines a reflection G to
restore the Hessenberg form in the (K-1)th column, and thus
chases the bulge one step toward the bottom of the active
submatrix. NR is the order of G.
Computing MIN */
i__3 = ns + 1, i__5 = i__ - k + 1;
nr = min(i__3,i__5);
if (k > l) {
zcopy_(&nr, &h___ref(k, k - 1), &c__1, v, &c__1);
}
zlarfg_(&nr, v, &v[1], &c__1, &tau);
if (k > l) {
i__3 = h___subscr(k, k - 1);
h__[i__3].r = v[0].r, h__[i__3].i = v[0].i;
i__3 = i__;
for (ii = k + 1; ii <= i__3; ++ii) {
i__5 = h___subscr(ii, k - 1);
h__[i__5].r = 0., h__[i__5].i = 0.;
/* L140: */
}
}
v[0].r = 1., v[0].i = 0.;
/* Apply G' from the left to transform the rows of the matrix
in columns K to I2. */
i__3 = i2 - k + 1;
d_cnjg(&z__1, &tau);
zlarfx_("Left", &nr, &i__3, v, &z__1, &h___ref(k, k), ldh, &work[
1]);
/* Apply G from the right to transform the columns of the
matrix in rows I1 to min(K+NR,I).
Computing MIN */
i__5 = k + nr;
i__3 = min(i__5,i__) - i1 + 1;
zlarfx_("Right", &i__3, &nr, v, &tau, &h___ref(i1, k), ldh, &work[
1]);
if (wantz) {
/* Accumulate transformations in the matrix Z */
zlarfx_("Right", &nh, &nr, v, &tau, &z___ref(*ilo, k), ldz, &
work[1]);
}
/* L150: */
}
/* Ensure that H(I,I-1) is real. */
i__2 = h___subscr(i__, i__ - 1);
temp.r = h__[i__2].r, temp.i = h__[i__2].i;
if (d_imag(&temp) != 0.) {
d__1 = temp.r;
d__2 = d_imag(&temp);
rtemp = dlapy2_(&d__1, &d__2);
i__2 = h___subscr(i__, i__ - 1);
h__[i__2].r = rtemp, h__[i__2].i = 0.;
z__1.r = temp.r / rtemp, z__1.i = temp.i / rtemp;
temp.r = z__1.r, temp.i = z__1.i;
if (i2 > i__) {
i__2 = i2 - i__;
d_cnjg(&z__1, &temp);
zscal_(&i__2, &z__1, &h___ref(i__, i__ + 1), ldh);
}
i__2 = i__ - i1;
zscal_(&i__2, &temp, &h___ref(i1, i__), &c__1);
if (wantz) {
zscal_(&nh, &temp, &z___ref(*ilo, i__), &c__1);
}
}
/* L160: */
}
/* Failure to converge in remaining number of iterations */
*info = i__;
return 0;
L170:
/* A submatrix of order <= MAXB in rows and columns L to I has split
off. Use the double-shift QR algorithm to handle it. */
zlahqr_(&wantt, &wantz, n, &l, &i__, &h__[h_offset], ldh, &w[1], ilo, ihi,
&z__[z_offset], ldz, info);
if (*info > 0) {
return 0;
}
/* Decrement number of remaining iterations, and return to start of
the main loop with a new value of I. */
itn -= its;
i__ = l - 1;
goto L60;
L180:
i__1 = max(1,*n);
work[1].r = (doublereal) i__1, work[1].i = 0.;
return 0;
/* End of ZHSEQR */
} /* zhseqr_ */
/* Subroutine */
int dlasd2_(integer *nl, integer *nr, integer *sqre, integer *k, doublereal *d__, doublereal *z__, doublereal *alpha, doublereal * beta, doublereal *u, integer *ldu, doublereal *vt, integer *ldvt, doublereal *dsigma, doublereal *u2, integer *ldu2, doublereal *vt2, integer *ldvt2, integer *idxp, integer *idx, integer *idxc, integer * idxq, integer *coltyp, integer *info)
{
/* System generated locals */
integer u_dim1, u_offset, u2_dim1, u2_offset, vt_dim1, vt_offset, vt2_dim1, vt2_offset, i__1;
doublereal d__1, d__2;
/* Local variables */
doublereal c__;
integer i__, j, m, n;
doublereal s;
integer k2;
doublereal z1;
integer ct, jp;
doublereal eps, tau, tol;
integer psm[4], nlp1, nlp2, idxi, idxj;
extern /* Subroutine */
int drot_(integer *, doublereal *, integer *, doublereal *, integer *, doublereal *, doublereal *);
integer ctot[4], idxjp;
extern /* Subroutine */
int dcopy_(integer *, doublereal *, integer *, doublereal *, integer *);
integer jprev;
extern doublereal dlapy2_(doublereal *, doublereal *), dlamch_(char *);
extern /* Subroutine */
int dlamrg_(integer *, integer *, doublereal *, integer *, integer *, integer *), dlacpy_(char *, integer *, integer *, doublereal *, integer *, doublereal *, integer *), dlaset_(char *, integer *, integer *, doublereal *, doublereal *, doublereal *, integer *), xerbla_(char *, integer *);
doublereal hlftol;
/* -- 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 Arrays .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
--d__;
--z__;
u_dim1 = *ldu;
u_offset = 1 + u_dim1;
u -= u_offset;
vt_dim1 = *ldvt;
vt_offset = 1 + vt_dim1;
vt -= vt_offset;
--dsigma;
u2_dim1 = *ldu2;
u2_offset = 1 + u2_dim1;
u2 -= u2_offset;
vt2_dim1 = *ldvt2;
vt2_offset = 1 + vt2_dim1;
vt2 -= vt2_offset;
--idxp;
--idx;
--idxc;
--idxq;
--coltyp;
/* Function Body */
*info = 0;
if (*nl < 1)
{
*info = -1;
}
else if (*nr < 1)
{
*info = -2;
}
else if (*sqre != 1 && *sqre != 0)
{
*info = -3;
}
n = *nl + *nr + 1;
m = n + *sqre;
if (*ldu < n)
{
*info = -10;
}
else if (*ldvt < m)
{
*info = -12;
}
else if (*ldu2 < n)
{
*info = -15;
}
else if (*ldvt2 < m)
{
*info = -17;
}
if (*info != 0)
{
i__1 = -(*info);
xerbla_("DLASD2", &i__1);
return 0;
}
nlp1 = *nl + 1;
nlp2 = *nl + 2;
/* Generate the first part of the vector Z;
and move the singular */
/* values in the first part of D one position backward. */
z1 = *alpha * vt[nlp1 + nlp1 * vt_dim1];
z__[1] = z1;
for (i__ = *nl;
i__ >= 1;
--i__)
{
z__[i__ + 1] = *alpha * vt[i__ + nlp1 * vt_dim1];
d__[i__ + 1] = d__[i__];
idxq[i__ + 1] = idxq[i__] + 1;
/* L10: */
}
/* Generate the second part of the vector Z. */
i__1 = m;
for (i__ = nlp2;
i__ <= i__1;
++i__)
{
z__[i__] = *beta * vt[i__ + nlp2 * vt_dim1];
/* L20: */
}
/* Initialize some reference arrays. */
i__1 = nlp1;
for (i__ = 2;
i__ <= i__1;
++i__)
{
coltyp[i__] = 1;
/* L30: */
}
i__1 = n;
for (i__ = nlp2;
i__ <= i__1;
++i__)
{
coltyp[i__] = 2;
/* L40: */
}
/* Sort the singular values into increasing order */
i__1 = n;
for (i__ = nlp2;
i__ <= i__1;
++i__)
{
idxq[i__] += nlp1;
/* L50: */
}
/* DSIGMA, IDXC, IDXC, and the first column of U2 */
/* are used as storage space. */
i__1 = n;
for (i__ = 2;
i__ <= i__1;
++i__)
{
dsigma[i__] = d__[idxq[i__]];
u2[i__ + u2_dim1] = z__[idxq[i__]];
idxc[i__] = coltyp[idxq[i__]];
/* L60: */
}
dlamrg_(nl, nr, &dsigma[2], &c__1, &c__1, &idx[2]);
i__1 = n;
for (i__ = 2;
i__ <= i__1;
++i__)
{
idxi = idx[i__] + 1;
d__[i__] = dsigma[idxi];
z__[i__] = u2[idxi + u2_dim1];
coltyp[i__] = idxc[idxi];
/* L70: */
}
/* Calculate the allowable deflation tolerance */
eps = dlamch_("Epsilon");
/* Computing MAX */
d__1 = f2c_abs(*alpha);
d__2 = f2c_abs(*beta); // , expr subst
tol = max(d__1,d__2);
/* Computing MAX */
d__2 = (d__1 = d__[n], f2c_abs(d__1));
tol = eps * 8. * max(d__2,tol);
/* There are 2 kinds of deflation -- first a value in the z-vector */
/* is small, second two (or more) singular values are very close */
/* together (their difference is small). */
/* If the value in the z-vector is small, we simply permute the */
/* array so that the corresponding singular value is moved to the */
/* end. */
/* If two values in the D-vector are close, we perform a two-sided */
/* rotation designed to make one of the corresponding z-vector */
/* entries zero, and then permute the array so that the deflated */
/* singular value is moved to the end. */
/* If there are multiple singular values then the problem deflates. */
/* Here the number of equal singular values are found. As each equal */
/* singular value is found, an elementary reflector is computed to */
/* rotate the corresponding singular subspace so that the */
/* corresponding components of Z are zero in this new basis. */
*k = 1;
k2 = n + 1;
i__1 = n;
for (j = 2;
j <= i__1;
++j)
{
if ((d__1 = z__[j], f2c_abs(d__1)) <= tol)
{
/* Deflate due to small z component. */
--k2;
idxp[k2] = j;
coltyp[j] = 4;
if (j == n)
{
goto L120;
}
}
else
{
jprev = j;
goto L90;
}
/* L80: */
}
L90:
j = jprev;
L100:
++j;
if (j > n)
{
goto L110;
}
if ((d__1 = z__[j], f2c_abs(d__1)) <= tol)
{
/* Deflate due to small z component. */
--k2;
idxp[k2] = j;
coltyp[j] = 4;
}
else
{
/* Check if singular values are close enough to allow deflation. */
if ((d__1 = d__[j] - d__[jprev], f2c_abs(d__1)) <= tol)
{
/* Deflation is possible. */
s = z__[jprev];
c__ = z__[j];
/* Find sqrt(a**2+b**2) without overflow or */
/* destructive underflow. */
tau = dlapy2_(&c__, &s);
c__ /= tau;
s = -s / tau;
z__[j] = tau;
z__[jprev] = 0.;
/* Apply back the Givens rotation to the left and right */
/* singular vector matrices. */
idxjp = idxq[idx[jprev] + 1];
idxj = idxq[idx[j] + 1];
if (idxjp <= nlp1)
{
--idxjp;
}
if (idxj <= nlp1)
{
--idxj;
}
drot_(&n, &u[idxjp * u_dim1 + 1], &c__1, &u[idxj * u_dim1 + 1], & c__1, &c__, &s);
drot_(&m, &vt[idxjp + vt_dim1], ldvt, &vt[idxj + vt_dim1], ldvt, & c__, &s);
if (coltyp[j] != coltyp[jprev])
{
coltyp[j] = 3;
}
coltyp[jprev] = 4;
--k2;
idxp[k2] = jprev;
jprev = j;
}
else
{
++(*k);
u2[*k + u2_dim1] = z__[jprev];
dsigma[*k] = d__[jprev];
idxp[*k] = jprev;
jprev = j;
}
}
goto L100;
L110: /* Record the last singular value. */
++(*k);
u2[*k + u2_dim1] = z__[jprev];
dsigma[*k] = d__[jprev];
idxp[*k] = jprev;
L120: /* Count up the total number of the various types of columns, then */
/* form a permutation which positions the four column types into */
/* four groups of uniform structure (although one or more of these */
/* groups may be empty). */
for (j = 1;
j <= 4;
++j)
{
ctot[j - 1] = 0;
/* L130: */
}
i__1 = n;
for (j = 2;
j <= i__1;
++j)
{
ct = coltyp[j];
++ctot[ct - 1];
/* L140: */
}
/* PSM(*) = Position in SubMatrix (of types 1 through 4) */
psm[0] = 2;
psm[1] = ctot[0] + 2;
psm[2] = psm[1] + ctot[1];
psm[3] = psm[2] + ctot[2];
/* Fill out the IDXC 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, starting from the */
/* second column. This applies similarly to the rows of VT. */
i__1 = n;
for (j = 2;
j <= i__1;
++j)
{
jp = idxp[j];
ct = coltyp[jp];
idxc[psm[ct - 1]] = j;
++psm[ct - 1];
/* L150: */
}
/* Sort the singular values and corresponding singular vectors into */
/* DSIGMA, U2, and VT2 respectively. The singular values/vectors */
/* which were not deflated go into the first K slots of DSIGMA, U2, */
/* and VT2 respectively, while those which were deflated go into the */
/* last N - K slots, except that the first column/row will be treated */
/* separately. */
i__1 = n;
for (j = 2;
j <= i__1;
++j)
{
jp = idxp[j];
dsigma[j] = d__[jp];
idxj = idxq[idx[idxp[idxc[j]]] + 1];
if (idxj <= nlp1)
{
--idxj;
}
dcopy_(&n, &u[idxj * u_dim1 + 1], &c__1, &u2[j * u2_dim1 + 1], &c__1);
dcopy_(&m, &vt[idxj + vt_dim1], ldvt, &vt2[j + vt2_dim1], ldvt2);
/* L160: */
}
/* Determine DSIGMA(1), DSIGMA(2) and Z(1) */
dsigma[1] = 0.;
hlftol = tol / 2.;
if (f2c_abs(dsigma[2]) <= hlftol)
{
dsigma[2] = hlftol;
}
if (m > n)
{
z__[1] = dlapy2_(&z1, &z__[m]);
if (z__[1] <= tol)
{
c__ = 1.;
s = 0.;
z__[1] = tol;
}
else
{
c__ = z1 / z__[1];
s = z__[m] / z__[1];
}
}
else
{
if (f2c_abs(z1) <= tol)
{
z__[1] = tol;
}
else
{
z__[1] = z1;
}
}
/* Move the rest of the updating row to Z. */
i__1 = *k - 1;
dcopy_(&i__1, &u2[u2_dim1 + 2], &c__1, &z__[2], &c__1);
/* Determine the first column of U2, the first row of VT2 and the */
/* last row of VT. */
dlaset_("A", &n, &c__1, &c_b30, &c_b30, &u2[u2_offset], ldu2);
u2[nlp1 + u2_dim1] = 1.;
if (m > n)
{
i__1 = nlp1;
for (i__ = 1;
i__ <= i__1;
++i__)
{
vt[m + i__ * vt_dim1] = -s * vt[nlp1 + i__ * vt_dim1];
vt2[i__ * vt2_dim1 + 1] = c__ * vt[nlp1 + i__ * vt_dim1];
/* L170: */
}
i__1 = m;
for (i__ = nlp2;
i__ <= i__1;
++i__)
{
vt2[i__ * vt2_dim1 + 1] = s * vt[m + i__ * vt_dim1];
vt[m + i__ * vt_dim1] = c__ * vt[m + i__ * vt_dim1];
/* L180: */
}
}
else
{
dcopy_(&m, &vt[nlp1 + vt_dim1], ldvt, &vt2[vt2_dim1 + 1], ldvt2);
}
if (m > n)
{
dcopy_(&m, &vt[m + vt_dim1], ldvt, &vt2[m + vt2_dim1], ldvt2);
}
/* The deflated singular values and their corresponding vectors go */
/* into the back of D, U, and V respectively. */
if (n > *k)
{
i__1 = n - *k;
dcopy_(&i__1, &dsigma[*k + 1], &c__1, &d__[*k + 1], &c__1);
i__1 = n - *k;
dlacpy_("A", &n, &i__1, &u2[(*k + 1) * u2_dim1 + 1], ldu2, &u[(*k + 1) * u_dim1 + 1], ldu);
i__1 = n - *k;
dlacpy_("A", &i__1, &m, &vt2[*k + 1 + vt2_dim1], ldvt2, &vt[*k + 1 + vt_dim1], ldvt);
}
/* Copy CTOT into COLTYP for referencing in DLASD3. */
for (j = 1;
j <= 4;
++j)
{
coltyp[j] = ctot[j - 1];
/* L190: */
}
return 0;
/* End of DLASD2 */
}
/* Subroutine */ int dlagv2_(doublereal *a, integer *lda, doublereal *b,
integer *ldb, doublereal *alphar, doublereal *alphai, doublereal *
beta, doublereal *csl, doublereal *snl, doublereal *csr, doublereal *
snr)
{
/* -- 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
=======
DLAGV2 computes the Generalized Schur factorization of a real 2-by-2
matrix pencil (A,B) where B is upper triangular. This routine
computes orthogonal (rotation) matrices given by CSL, SNL and CSR,
SNR such that
1) if the pencil (A,B) has two real eigenvalues (include 0/0 or 1/0
types), then
[ a11 a12 ] := [ CSL SNL ] [ a11 a12 ] [ CSR -SNR ]
[ 0 a22 ] [ -SNL CSL ] [ a21 a22 ] [ SNR CSR ]
[ b11 b12 ] := [ CSL SNL ] [ b11 b12 ] [ CSR -SNR ]
[ 0 b22 ] [ -SNL CSL ] [ 0 b22 ] [ SNR CSR ],
2) if the pencil (A,B) has a pair of complex conjugate eigenvalues,
then
[ a11 a12 ] := [ CSL SNL ] [ a11 a12 ] [ CSR -SNR ]
[ a21 a22 ] [ -SNL CSL ] [ a21 a22 ] [ SNR CSR ]
[ b11 0 ] := [ CSL SNL ] [ b11 b12 ] [ CSR -SNR ]
[ 0 b22 ] [ -SNL CSL ] [ 0 b22 ] [ SNR CSR ]
where b11 >= b22 > 0.
Arguments
=========
A (input/output) DOUBLE PRECISION array, dimension (LDA, 2)
On entry, the 2 x 2 matrix A.
On exit, A is overwritten by the ``A-part'' of the
generalized Schur form.
LDA (input) INTEGER
THe leading dimension of the array A. LDA >= 2.
B (input/output) DOUBLE PRECISION array, dimension (LDB, 2)
On entry, the upper triangular 2 x 2 matrix B.
On exit, B is overwritten by the ``B-part'' of the
generalized Schur form.
LDB (input) INTEGER
THe leading dimension of the array B. LDB >= 2.
ALPHAR (output) DOUBLE PRECISION array, dimension (2)
ALPHAI (output) DOUBLE PRECISION array, dimension (2)
BETA (output) DOUBLE PRECISION array, dimension (2)
(ALPHAR(k)+i*ALPHAI(k))/BETA(k) are the eigenvalues of the
pencil (A,B), k=1,2, i = sqrt(-1). Note that BETA(k) may
be zero.
CSL (output) DOUBLE PRECISION
The cosine of the left rotation matrix.
SNL (output) DOUBLE PRECISION
The sine of the left rotation matrix.
CSR (output) DOUBLE PRECISION
The cosine of the right rotation matrix.
SNR (output) DOUBLE PRECISION
The sine of the right rotation matrix.
Further Details
===============
Based on contributions by
Mark Fahey, Department of Mathematics, Univ. of Kentucky, USA
=====================================================================
Parameter adjustments */
/* Table of constant values */
static integer c__2 = 2;
static integer c__1 = 1;
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset;
doublereal d__1, d__2, d__3, d__4, d__5, d__6;
/* Local variables */
extern /* Subroutine */ int drot_(integer *, doublereal *, integer *,
doublereal *, integer *, doublereal *, doublereal *), dlag2_(
doublereal *, integer *, doublereal *, integer *, doublereal *,
doublereal *, doublereal *, doublereal *, doublereal *,
doublereal *);
static doublereal r__, t, anorm, bnorm, h1, h2, h3, scale1, scale2;
extern /* Subroutine */ int dlasv2_(doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *);
extern doublereal dlapy2_(doublereal *, doublereal *);
static doublereal ascale, bscale;
extern doublereal dlamch_(char *);
static doublereal wi, qq, rr, safmin;
extern /* Subroutine */ int dlartg_(doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *);
static doublereal wr1, wr2, ulp;
#define a_ref(a_1,a_2) a[(a_2)*a_dim1 + a_1]
#define b_ref(a_1,a_2) b[(a_2)*b_dim1 + a_1]
a_dim1 = *lda;
a_offset = 1 + a_dim1 * 1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1 * 1;
b -= b_offset;
--alphar;
--alphai;
--beta;
/* Function Body */
safmin = dlamch_("S");
ulp = dlamch_("P");
/* Scale A
Computing MAX */
d__5 = (d__1 = a_ref(1, 1), abs(d__1)) + (d__2 = a_ref(2, 1), abs(d__2)),
d__6 = (d__3 = a_ref(1, 2), abs(d__3)) + (d__4 = a_ref(2, 2), abs(
d__4)), d__5 = max(d__5,d__6);
anorm = max(d__5,safmin);
ascale = 1. / anorm;
a_ref(1, 1) = ascale * a_ref(1, 1);
a_ref(1, 2) = ascale * a_ref(1, 2);
a_ref(2, 1) = ascale * a_ref(2, 1);
a_ref(2, 2) = ascale * a_ref(2, 2);
/* Scale B
Computing MAX */
d__4 = (d__3 = b_ref(1, 1), abs(d__3)), d__5 = (d__1 = b_ref(1, 2), abs(
d__1)) + (d__2 = b_ref(2, 2), abs(d__2)), d__4 = max(d__4,d__5);
bnorm = max(d__4,safmin);
bscale = 1. / bnorm;
b_ref(1, 1) = bscale * b_ref(1, 1);
b_ref(1, 2) = bscale * b_ref(1, 2);
b_ref(2, 2) = bscale * b_ref(2, 2);
/* Check if A can be deflated */
if ((d__1 = a_ref(2, 1), abs(d__1)) <= ulp) {
*csl = 1.;
*snl = 0.;
*csr = 1.;
*snr = 0.;
a_ref(2, 1) = 0.;
b_ref(2, 1) = 0.;
/* Check if B is singular */
} else if ((d__1 = b_ref(1, 1), abs(d__1)) <= ulp) {
dlartg_(&a_ref(1, 1), &a_ref(2, 1), csl, snl, &r__);
*csr = 1.;
*snr = 0.;
drot_(&c__2, &a_ref(1, 1), lda, &a_ref(2, 1), lda, csl, snl);
drot_(&c__2, &b_ref(1, 1), ldb, &b_ref(2, 1), ldb, csl, snl);
a_ref(2, 1) = 0.;
b_ref(1, 1) = 0.;
b_ref(2, 1) = 0.;
} else if ((d__1 = b_ref(2, 2), abs(d__1)) <= ulp) {
dlartg_(&a_ref(2, 2), &a_ref(2, 1), csr, snr, &t);
*snr = -(*snr);
drot_(&c__2, &a_ref(1, 1), &c__1, &a_ref(1, 2), &c__1, csr, snr);
drot_(&c__2, &b_ref(1, 1), &c__1, &b_ref(1, 2), &c__1, csr, snr);
*csl = 1.;
*snl = 0.;
a_ref(2, 1) = 0.;
b_ref(2, 1) = 0.;
b_ref(2, 2) = 0.;
} else {
/* B is nonsingular, first compute the eigenvalues of (A,B) */
dlag2_(&a[a_offset], lda, &b[b_offset], ldb, &safmin, &scale1, &
scale2, &wr1, &wr2, &wi);
if (wi == 0.) {
/* two real eigenvalues, compute s*A-w*B */
h1 = scale1 * a_ref(1, 1) - wr1 * b_ref(1, 1);
h2 = scale1 * a_ref(1, 2) - wr1 * b_ref(1, 2);
h3 = scale1 * a_ref(2, 2) - wr1 * b_ref(2, 2);
rr = dlapy2_(&h1, &h2);
d__1 = scale1 * a_ref(2, 1);
qq = dlapy2_(&d__1, &h3);
if (rr > qq) {
/* find right rotation matrix to zero 1,1 element of
(sA - wB) */
dlartg_(&h2, &h1, csr, snr, &t);
} else {
/* find right rotation matrix to zero 2,1 element of
(sA - wB) */
d__1 = scale1 * a_ref(2, 1);
dlartg_(&h3, &d__1, csr, snr, &t);
}
*snr = -(*snr);
drot_(&c__2, &a_ref(1, 1), &c__1, &a_ref(1, 2), &c__1, csr, snr);
drot_(&c__2, &b_ref(1, 1), &c__1, &b_ref(1, 2), &c__1, csr, snr);
/* compute inf norms of A and B
Computing MAX */
d__5 = (d__1 = a_ref(1, 1), abs(d__1)) + (d__2 = a_ref(1, 2), abs(
d__2)), d__6 = (d__3 = a_ref(2, 1), abs(d__3)) + (d__4 =
a_ref(2, 2), abs(d__4));
h1 = max(d__5,d__6);
/* Computing MAX */
d__5 = (d__1 = b_ref(1, 1), abs(d__1)) + (d__2 = b_ref(1, 2), abs(
d__2)), d__6 = (d__3 = b_ref(2, 1), abs(d__3)) + (d__4 =
b_ref(2, 2), abs(d__4));
h2 = max(d__5,d__6);
if (scale1 * h1 >= abs(wr1) * h2) {
/* find left rotation matrix Q to zero out B(2,1) */
dlartg_(&b_ref(1, 1), &b_ref(2, 1), csl, snl, &r__);
} else {
/* find left rotation matrix Q to zero out A(2,1) */
dlartg_(&a_ref(1, 1), &a_ref(2, 1), csl, snl, &r__);
}
drot_(&c__2, &a_ref(1, 1), lda, &a_ref(2, 1), lda, csl, snl);
drot_(&c__2, &b_ref(1, 1), ldb, &b_ref(2, 1), ldb, csl, snl);
a_ref(2, 1) = 0.;
b_ref(2, 1) = 0.;
} else {
/* a pair of complex conjugate eigenvalues
first compute the SVD of the matrix B */
dlasv2_(&b_ref(1, 1), &b_ref(1, 2), &b_ref(2, 2), &r__, &t, snr,
csr, snl, csl);
/* Form (A,B) := Q(A,B)Z' where Q is left rotation matrix and
Z is right rotation matrix computed from DLASV2 */
drot_(&c__2, &a_ref(1, 1), lda, &a_ref(2, 1), lda, csl, snl);
drot_(&c__2, &b_ref(1, 1), ldb, &b_ref(2, 1), ldb, csl, snl);
drot_(&c__2, &a_ref(1, 1), &c__1, &a_ref(1, 2), &c__1, csr, snr);
drot_(&c__2, &b_ref(1, 1), &c__1, &b_ref(1, 2), &c__1, csr, snr);
b_ref(2, 1) = 0.;
b_ref(1, 2) = 0.;
}
}
/* Unscaling */
a_ref(1, 1) = anorm * a_ref(1, 1);
a_ref(2, 1) = anorm * a_ref(2, 1);
a_ref(1, 2) = anorm * a_ref(1, 2);
a_ref(2, 2) = anorm * a_ref(2, 2);
b_ref(1, 1) = bnorm * b_ref(1, 1);
b_ref(2, 1) = bnorm * b_ref(2, 1);
b_ref(1, 2) = bnorm * b_ref(1, 2);
b_ref(2, 2) = bnorm * b_ref(2, 2);
if (wi == 0.) {
alphar[1] = a_ref(1, 1);
alphar[2] = a_ref(2, 2);
alphai[1] = 0.;
alphai[2] = 0.;
beta[1] = b_ref(1, 1);
beta[2] = b_ref(2, 2);
} else {
alphar[1] = anorm * wr1 / scale1 / bnorm;
alphai[1] = anorm * wi / scale1 / bnorm;
alphar[2] = alphar[1];
alphai[2] = -alphai[1];
beta[1] = 1.;
beta[2] = 1.;
}
/* L10: */
return 0;
/* End of DLAGV2 */
} /* dlagv2_ */
int TRL::dstqrb_(integer_ * n, doublereal_ * d__, doublereal_ * e,
doublereal_ * z__, doublereal_ * work, integer_ * info)
{
/*
Purpose
=======
DSTQRB computes all eigenvalues and the last component of the eigenvectors of a
symmetric tridiagonal matrix using the implicit QL or QR method.
This is mainly a modification of the CLAPACK subroutine dsteqr.c
Arguments
=========
N (input) INTEGER
The order of the matrix. N >= 0.
D (input/output) DOUBLE PRECISION array, dimension (N)
On entry, the diagonal elements of the tridiagonal matrix.
On exit, if INFO = 0, the eigenvalues in ascending order.
E (input/output) DOUBLE PRECISION array, dimension (N-1)
On entry, the (n-1) subdiagonal elements of the tridiagonal
matrix.
On exit, E has been destroyed.
Z (input/output) DOUBLE PRECISION array, dimension (LDZ, N)
On entry, if COMPZ = 'V', then Z contains the orthogonal
matrix used in the reduction to tridiagonal form.
On exit, if INFO = 0, then if COMPZ = 'V', Z contains the
orthonormal eigenvectors of the original symmetric matrix,
and if COMPZ = 'I', Z contains the orthonormal eigenvectors
of the symmetric tridiagonal matrix.
If COMPZ = 'N', then Z is not referenced.
WORK (workspace) DOUBLE PRECISION array, dimension (max(1,2*N-2))
If COMPZ = 'N', then WORK is not referenced.
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
> 0: the algorithm has failed to find all the eigenvalues in
a total of 30*N iterations; if INFO = i, then i
elements of E have not converged to zero; on exit, D
and E contain the elements of a symmetric tridiagonal
matrix which is orthogonally similar to the original
matrix.
=====================================================================
*/
/* Table of constant values */
doublereal_ c_b10 = 1.;
integer_ c__0 = 0;
integer_ c__1 = 1;
/* System generated locals */
integer_ i__1, i__2;
doublereal_ d__1, d__2;
/* Builtin functions */
//double sqrt_(doublereal_), d_sign(doublereal_ *, doublereal_ *);
//extern /* Subroutine */ int dlae2_(doublereal_ *, doublereal_ *, doublereal_
// *, doublereal_ *, doublereal_ *);
doublereal_ b, c__, f, g;
integer_ i__, j, k, l, m;
doublereal_ p, r__, s;
//extern logical_ lsame_(char *, char *);
//extern /* Subroutine */ int dlasr_(char *, char *, char *, integer_ *,
// integer_ *, doublereal_ *,
// doublereal_ *, doublereal_ *,
// integer_ *);
doublereal_ anorm;
//extern /* Subroutine */ int dswap_(integer_ *, doublereal_ *, integer_ *,
// doublereal_ *, integer_ *);
integer_ l1;
//extern /* Subroutine */ int dlaev2_(doublereal_ *, doublereal_ *,
// doublereal_ *, doublereal_ *,
// doublereal_ *, doublereal_ *,
// doublereal_ *);
integer_ lendm1, lendp1;
//extern doublereal_ dlapy2_(doublereal_ *, doublereal_ *);
integer_ ii;
//extern doublereal_ dlamch_(char *);
integer_ mm, iscale;
//extern /* Subroutine */ int dlascl_(char *, integer_ *, integer_ *,
// doublereal_ *, doublereal_ *,
// integer_ *, integer_ *, doublereal_ *,
// integer_ *, integer_ *),
// dlaset_(char *, integer_ *, integer_ *, doublereal_ *, doublereal_ *,
// doublereal_ *, integer_ *);
doublereal_ safmin;
//extern /* Subroutine */ int dlartg_(doublereal_ *, doublereal_ *,
// doublereal_ *, doublereal_ *,
// doublereal_ *);
doublereal_ safmax;
//extern /* Subroutine */ int xerbla_(char *, integer_ *);
//extern doublereal_ dlanst_(char *, integer_ *, doublereal_ *,
// doublereal_ *);
//extern /* Subroutine */ int dlasrt_(char *, integer_ *, doublereal_ *,
// integer_ *);
/* Local variables */
integer_ lend, jtot;
integer_ lendsv;
doublereal_ ssfmin;
integer_ nmaxit;
doublereal_ ssfmax;
integer_ lm1, mm1, nm1;
doublereal_ rt1, rt2, eps;
integer_ lsv;
doublereal_ tst, eps2;
--d__;
--e;
--z__;
/* z_dim1 = *ldz; */
/* z_offset = 1 + z_dim1 * 1; */
/* z__ -= z_offset; */
--work;
/* Function Body */
*info = 0;
/* Taken out for TRLan
if (lsame_(compz, "N")) {
icompz = 0;
} else if (lsame_(compz, "V")) {
icompz = 1;
} else if (lsame_(compz, "I")) {
icompz = 2;
} else {
icompz = -1;
}
if (icompz < 0) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*ldz < 1 || icompz > 0 && *ldz < max(1,*n)) {
*info = -6;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("DSTEQR", &i__1);
return 0;
}
*/
/* icompz = 2; */
/* Quick return if possible */
if (*n == 0) {
return 0;
}
if (*n == 1) {
z__[1] = 1;
return 0;
}
/* Determine the unit roundoff and over/underflow thresholds. */
eps = dlamch_("E");
/* Computing 2nd power */
d__1 = eps;
eps2 = d__1 * d__1;
safmin = dlamch_("S");
safmax = 1. / safmin;
ssfmax = sqrt_(safmax) / 3.;
ssfmin = sqrt_(safmin) / eps2;
/* Compute the eigenvalues and eigenvectors of the tridiagonal
matrix. */
/* Taken out for TRLan
if (icompz == 2) {
dlaset_("Full", n, n, &c_b9, &c_b10, &z__[z_offset], ldz);
}
*/
for (j = 1; j < *n; j++) {
z__[j] = 0.0;
}
z__[*n] = 1.0;
nmaxit = *n * 30;
jtot = 0;
/* Determine where the matrix splits and choose QL or QR iteration
for each block, according to whether top or bottom diagonal
element is smaller. */
l1 = 1;
nm1 = *n - 1;
L10:
if (l1 > *n) {
goto L160;
}
if (l1 > 1) {
e[l1 - 1] = 0.;
}
if (l1 <= nm1) {
i__1 = nm1;
for (m = l1; m <= i__1; ++m) {
tst = (d__1 = e[m], fabs(d__1));
if (tst == 0.) {
goto L30;
}
if (tst <=
sqrt_((d__1 = d__[m], fabs(d__1))) * sqrt_((d__2 =
d__[m + 1],
fabs(d__2))) *
eps) {
e[m] = 0.;
goto L30;
}
/* L20: */
}
}
m = *n;
L30:
l = l1;
lsv = l;
lend = m;
lendsv = lend;
l1 = m + 1;
if (lend == l) {
goto L10;
}
/* Scale submatrix in rows and columns L to LEND */
i__1 = lend - l + 1;
anorm = dlanst_("I", &i__1, &d__[l], &e[l]);
iscale = 0;
if (anorm == 0.) {
goto L10;
}
if (anorm > ssfmax) {
iscale = 1;
i__1 = lend - l + 1;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmax, &i__1, &c__1, &d__[l],
n, info);
i__1 = lend - l;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmax, &i__1, &c__1, &e[l], n,
info);
} else if (anorm < ssfmin) {
iscale = 2;
i__1 = lend - l + 1;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &d__[l],
n, info);
i__1 = lend - l;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &e[l], n,
info);
}
/* Choose between QL and QR iteration */
if ((d__1 = d__[lend], fabs(d__1)) < (d__2 = d__[l], fabs(d__2))) {
lend = lsv;
l = lendsv;
}
if (lend > l) {
/* QL Iteration
Look for small subdiagonal element. */
L40:
if (l != lend) {
lendm1 = lend - 1;
i__1 = lendm1;
for (m = l; m <= i__1; ++m) {
/* Computing 2nd power */
d__2 = (d__1 = e[m], fabs(d__1));
tst = d__2 * d__2;
if (tst <=
eps2 * (d__1 = d__[m], fabs(d__1)) * (d__2 =
d__[m + 1],
fabs(d__2)) +
safmin) {
goto L60;
}
/* L50: */
}
}
m = lend;
L60:
if (m < lend) {
e[m] = 0.;
}
p = d__[l];
if (m == l) {
goto L80;
}
/* If remaining matrix is 2-by-2, use DLAE2 or SLAEV2
to compute its eigensystem. */
if (m == l + 1) {
dlaev2_(&d__[l], &e[l], &d__[l + 1], &rt1, &rt2, &c__, &s);
work[l] = c__;
work[*n - 1 + l] = s;
/* Taken out for TRLan
dlasr_("R", "V", "B", n, &c__2, &work[l], &work[*n - 1 + l], &
z___ref(1, l), ldz);
*/
tst = z__[l + 1];
z__[l + 1] = c__ * tst - s * z__[l];
z__[l] = s * tst + c__ * z__[l];
d__[l] = rt1;
d__[l + 1] = rt2;
e[l] = 0.;
l += 2;
if (l <= lend) {
goto L40;
}
goto L140;
}
if (jtot == nmaxit) {
goto L140;
}
++jtot;
/* Form shift. */
g = (d__[l + 1] - p) / (e[l] * 2.);
r__ = dlapy2_(&g, &c_b10);
g = d__[m] - p + e[l] / (g + d_sign(&r__, &g));
s = 1.;
c__ = 1.;
p = 0.;
/* Inner loop */
mm1 = m - 1;
i__1 = l;
for (i__ = mm1; i__ >= i__1; --i__) {
f = s * e[i__];
b = c__ * e[i__];
dlartg_(&g, &f, &c__, &s, &r__);
if (i__ != m - 1) {
e[i__ + 1] = r__;
}
g = d__[i__ + 1] - p;
r__ = (d__[i__] - g) * s + c__ * 2. * b;
p = s * r__;
d__[i__ + 1] = g + p;
g = c__ * r__ - b;
/* If eigenvectors are desired, then save rotations. */
work[i__] = c__;
work[*n - 1 + i__] = -s;
/* L70: */
}
/* If eigenvectors are desired, then apply saved rotations. */
mm = m - l + 1;
/* Taken out for TRLan
dlasr_("R", "V", "B", n, &mm, &work[l], &work[*n - 1 + l], &
z___ref(1, l), ldz);
*/
dlasr_("R", "V", "B", &c__1, &mm, &work[l], &work[*n - 1 + l],
&z__[l], &c__1);
d__[l] -= p;
e[l] = g;
goto L40;
/* Eigenvalue found. */
L80:
d__[l] = p;
++l;
if (l <= lend) {
goto L40;
}
goto L140;
} else {
/* QR Iteration
Look for small superdiagonal element. */
L90:
if (l != lend) {
lendp1 = lend + 1;
i__1 = lendp1;
for (m = l; m >= i__1; --m) {
/* Computing 2nd power */
d__2 = (d__1 = e[m - 1], fabs(d__1));
tst = d__2 * d__2;
if (tst <=
eps2 * (d__1 = d__[m], fabs(d__1)) * (d__2 =
d__[m - 1],
fabs(d__2)) +
safmin) {
goto L110;
}
/* L100: */
}
}
m = lend;
L110:
if (m > lend) {
e[m - 1] = 0.;
}
p = d__[l];
if (m == l) {
goto L130;
}
/* If remaining matrix is 2-by-2, use DLAE2 or SLAEV2
to compute its eigensystem. */
if (m == l - 1) {
dlaev2_(&d__[l - 1], &e[l - 1], &d__[l], &rt1, &rt2, &c__, &s);
/* Taken out for TRLan
work[m] = c__;
work[*n - 1 + m] = s;
dlasr_("R", "V", "F", n, &c__2, &work[m], &work[*n - 1 + m], &
z___ref(1, l - 1), ldz);
*/
tst = z__[l];
z__[l] = c__ * tst - s * z__[l - 1];
z__[l - 1] = s * tst + c__ * z__[l - 1];
d__[l - 1] = rt1;
d__[l] = rt2;
e[l - 1] = 0.;
l += -2;
if (l >= lend) {
goto L90;
}
goto L140;
}
if (jtot == nmaxit) {
goto L140;
}
++jtot;
/* Form shift. */
g = (d__[l - 1] - p) / (e[l - 1] * 2.);
r__ = dlapy2_(&g, &c_b10);
g = d__[m] - p + e[l - 1] / (g + d_sign(&r__, &g));
s = 1.;
c__ = 1.;
p = 0.;
/* Inner loop */
lm1 = l - 1;
i__1 = lm1;
for (i__ = m; i__ <= i__1; ++i__) {
f = s * e[i__];
b = c__ * e[i__];
dlartg_(&g, &f, &c__, &s, &r__);
if (i__ != m) {
e[i__ - 1] = r__;
}
g = d__[i__] - p;
r__ = (d__[i__ + 1] - g) * s + c__ * 2. * b;
p = s * r__;
d__[i__] = g + p;
g = c__ * r__ - b;
/* If eigenvectors are desired, then save rotations. */
work[i__] = c__;
work[*n - 1 + i__] = s;
/* L120: */
}
/* If eigenvectors are desired, then apply saved rotations. */
mm = l - m + 1;
/*
dlasr_("R", "V", "F", n, &mm, &work[m], &work[*n - 1 + m], &
z___ref(1, m), ldz);
*/
dlasr_("R", "V", "F", &c__1, &mm, &work[m], &work[*n - 1 + m],
&z__[m], &c__1);
d__[l] -= p;
e[lm1] = g;
goto L90;
/* Eigenvalue found. */
L130:
d__[l] = p;
--l;
if (l >= lend) {
goto L90;
}
goto L140;
}
/* Undo scaling if necessary */
L140:
if (iscale == 1) {
i__1 = lendsv - lsv + 1;
dlascl_("G", &c__0, &c__0, &ssfmax, &anorm, &i__1, &c__1,
&d__[lsv], n, info);
i__1 = lendsv - lsv;
dlascl_("G", &c__0, &c__0, &ssfmax, &anorm, &i__1, &c__1, &e[lsv],
n, info);
} else if (iscale == 2) {
i__1 = lendsv - lsv + 1;
dlascl_("G", &c__0, &c__0, &ssfmin, &anorm, &i__1, &c__1,
&d__[lsv], n, info);
i__1 = lendsv - lsv;
dlascl_("G", &c__0, &c__0, &ssfmin, &anorm, &i__1, &c__1, &e[lsv],
n, info);
}
/* Check for no convergence to an eigenvalue after a total
of N*MAXIT iterations. */
if (jtot < nmaxit) {
goto L10;
}
i__1 = *n - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
if (e[i__] != 0.) {
++(*info);
}
/* L150: */
}
goto L190;
/* Order eigenvalues and eigenvectors. */
L160:
/* Use Selection Sort to minimize swaps of eigenvectors */
i__1 = *n;
for (ii = 2; ii <= i__1; ++ii) {
i__ = ii - 1;
k = i__;
p = d__[i__];
i__2 = *n;
for (j = ii; j <= i__2; ++j) {
if (d__[j] < p) {
k = j;
p = d__[j];
}
/* L170: */
}
if (k != i__) {
d__[k] = d__[i__];
d__[i__] = p;
/* Taken out for TRLan
dswap_(n, &z___ref(1, i__), &c__1, &z___ref(1, k), &c__1);
*/
p = z__[k];
z__[k] = z__[i__];
z__[i__] = p;
}
/* L180: */
}
L190:
return 0;
}
/* Subroutine */ int dtgsna_(char *job, char *howmny, logical *select,
integer *n, doublereal *a, integer *lda, doublereal *b, integer *ldb,
doublereal *vl, integer *ldvl, doublereal *vr, integer *ldvr,
doublereal *s, doublereal *dif, integer *mm, integer *m, doublereal *
work, integer *lwork, integer *iwork, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, vl_dim1, vl_offset, vr_dim1,
vr_offset, i__1, i__2;
doublereal d__1, d__2;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
integer i__, k;
doublereal c1, c2;
integer n1, n2, ks, iz;
doublereal eps, beta, cond;
extern doublereal ddot_(integer *, doublereal *, integer *, doublereal *,
integer *);
logical pair;
integer ierr;
doublereal uhav, uhbv;
integer ifst;
doublereal lnrm;
integer ilst;
doublereal rnrm;
extern /* Subroutine */ int dlag2_(doublereal *, integer *, doublereal *,
integer *, doublereal *, doublereal *, doublereal *, doublereal *,
doublereal *, doublereal *);
extern doublereal dnrm2_(integer *, doublereal *, integer *);
doublereal root1, root2, scale;
extern logical lsame_(char *, char *);
extern /* Subroutine */ int dgemv_(char *, integer *, integer *,
doublereal *, doublereal *, integer *, doublereal *, integer *,
doublereal *, doublereal *, integer *);
doublereal uhavi, uhbvi, tmpii;
integer lwmin;
logical wants;
doublereal tmpir, tmpri, dummy[1], tmprr;
extern doublereal dlapy2_(doublereal *, doublereal *);
doublereal dummy1[1];
extern doublereal dlamch_(char *);
doublereal alphai, alphar;
extern /* Subroutine */ int dlacpy_(char *, integer *, integer *,
doublereal *, integer *, doublereal *, integer *),
xerbla_(char *, integer *), dtgexc_(logical *, logical *,
integer *, doublereal *, integer *, doublereal *, integer *,
doublereal *, integer *, doublereal *, integer *, integer *,
integer *, doublereal *, integer *, integer *);
logical wantbh, wantdf, somcon;
doublereal alprqt;
extern /* Subroutine */ int dtgsyl_(char *, integer *, integer *, integer
*, doublereal *, integer *, doublereal *, integer *, doublereal *,
integer *, doublereal *, integer *, doublereal *, integer *,
doublereal *, integer *, doublereal *, doublereal *, doublereal *,
integer *, integer *, integer *);
doublereal smlnum;
logical lquery;
/* -- LAPACK routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* DTGSNA estimates reciprocal condition numbers for specified */
/* eigenvalues and/or eigenvectors of a matrix pair (A, B) in */
/* generalized real Schur canonical form (or of any matrix pair */
/* (Q*A*Z', Q*B*Z') with orthogonal matrices Q and Z, where */
/* Z' denotes the transpose of Z. */
/* (A, B) must be in generalized real Schur form (as returned by DGGES), */
/* i.e. A is block upper triangular with 1-by-1 and 2-by-2 diagonal */
/* blocks. B is upper triangular. */
/* Arguments */
/* ========= */
/* JOB (input) CHARACTER*1 */
/* Specifies whether condition numbers are required for */
/* eigenvalues (S) or eigenvectors (DIF): */
/* = 'E': for eigenvalues only (S); */
/* = 'V': for eigenvectors only (DIF); */
/* = 'B': for both eigenvalues and eigenvectors (S and DIF). */
/* HOWMNY (input) CHARACTER*1 */
/* = 'A': compute condition numbers for all eigenpairs; */
/* = 'S': compute condition numbers for selected eigenpairs */
/* specified by the array SELECT. */
/* SELECT (input) LOGICAL array, dimension (N) */
/* If HOWMNY = 'S', SELECT specifies the eigenpairs for which */
/* condition numbers are required. To select condition numbers */
/* for the eigenpair corresponding to a real eigenvalue w(j), */
/* SELECT(j) must be set to .TRUE.. To select condition numbers */
/* corresponding to a complex conjugate pair of eigenvalues w(j) */
/* and w(j+1), either SELECT(j) or SELECT(j+1) or both, must be */
/* set to .TRUE.. */
/* If HOWMNY = 'A', SELECT is not referenced. */
/* N (input) INTEGER */
/* The order of the square matrix pair (A, B). N >= 0. */
/* A (input) DOUBLE PRECISION array, dimension (LDA,N) */
/* The upper quasi-triangular matrix A in the pair (A,B). */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1,N). */
/* B (input) DOUBLE PRECISION array, dimension (LDB,N) */
/* The upper triangular matrix B in the pair (A,B). */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= max(1,N). */
/* VL (input) DOUBLE PRECISION array, dimension (LDVL,M) */
/* If JOB = 'E' or 'B', VL must contain left eigenvectors of */
/* (A, B), corresponding to the eigenpairs specified by HOWMNY */
/* and SELECT. The eigenvectors must be stored in consecutive */
/* columns of VL, as returned by DTGEVC. */
/* If JOB = 'V', VL is not referenced. */
/* LDVL (input) INTEGER */
/* The leading dimension of the array VL. LDVL >= 1. */
/* If JOB = 'E' or 'B', LDVL >= N. */
/* VR (input) DOUBLE PRECISION array, dimension (LDVR,M) */
/* If JOB = 'E' or 'B', VR must contain right eigenvectors of */
/* (A, B), corresponding to the eigenpairs specified by HOWMNY */
/* and SELECT. The eigenvectors must be stored in consecutive */
/* columns ov VR, as returned by DTGEVC. */
/* If JOB = 'V', VR is not referenced. */
/* LDVR (input) INTEGER */
/* The leading dimension of the array VR. LDVR >= 1. */
/* If JOB = 'E' or 'B', LDVR >= N. */
/* S (output) DOUBLE PRECISION array, dimension (MM) */
/* If JOB = 'E' or 'B', the reciprocal condition numbers of the */
/* selected eigenvalues, stored in consecutive elements of the */
/* array. For a complex conjugate pair of eigenvalues two */
/* consecutive elements of S are set to the same value. Thus */
/* S(j), DIF(j), and the j-th columns of VL and VR all */
/* correspond to the same eigenpair (but not in general the */
/* j-th eigenpair, unless all eigenpairs are selected). */
/* If JOB = 'V', S is not referenced. */
/* DIF (output) DOUBLE PRECISION array, dimension (MM) */
/* If JOB = 'V' or 'B', the estimated reciprocal condition */
/* numbers of the selected eigenvectors, stored in consecutive */
/* elements of the array. For a complex eigenvector two */
/* consecutive elements of DIF are set to the same value. If */
/* the eigenvalues cannot be reordered to compute DIF(j), DIF(j) */
/* is set to 0; this can only occur when the true value would be */
/* very small anyway. */
/* If JOB = 'E', DIF is not referenced. */
/* MM (input) INTEGER */
/* The number of elements in the arrays S and DIF. MM >= M. */
/* M (output) INTEGER */
/* The number of elements of the arrays S and DIF used to store */
/* the specified condition numbers; for each selected real */
/* eigenvalue one element is used, and for each selected complex */
/* conjugate pair of eigenvalues, two elements are used. */
/* If HOWMNY = 'A', M is set to N. */
/* WORK (workspace/output) DOUBLE PRECISION array, dimension (MAX(1,LWORK)) */
/* On exit, if INFO = 0, WORK(1) returns the optimal LWORK. */
/* LWORK (input) INTEGER */
/* The dimension of the array WORK. LWORK >= max(1,N). */
/* If JOB = 'V' or 'B' LWORK >= 2*N*(N+2)+16. */
/* If LWORK = -1, then a workspace query is assumed; the routine */
/* only calculates the optimal size of the WORK array, returns */
/* this value as the first entry of the WORK array, and no error */
/* message related to LWORK is issued by XERBLA. */
/* IWORK (workspace) INTEGER array, dimension (N + 6) */
/* If JOB = 'E', IWORK is not referenced. */
/* INFO (output) INTEGER */
/* =0: Successful exit */
/* <0: If INFO = -i, the i-th argument had an illegal value */
/* Further Details */
/* =============== */
/* The reciprocal of the condition number of a generalized eigenvalue */
/* w = (a, b) is defined as */
/* S(w) = (|u'Av|**2 + |u'Bv|**2)**(1/2) / (norm(u)*norm(v)) */
/* where u and v are the left and right eigenvectors of (A, B) */
/* corresponding to w; |z| denotes the absolute value of the complex */
/* number, and norm(u) denotes the 2-norm of the vector u. */
/* The pair (a, b) corresponds to an eigenvalue w = a/b (= u'Av/u'Bv) */
/* of the matrix pair (A, B). If both a and b equal zero, then (A B) is */
/* singular and S(I) = -1 is returned. */
/* An approximate error bound on the chordal distance between the i-th */
/* computed generalized eigenvalue w and the corresponding exact */
/* eigenvalue lambda is */
/* chord(w, lambda) <= EPS * norm(A, B) / S(I) */
/* where EPS is the machine precision. */
/* The reciprocal of the condition number DIF(i) of right eigenvector u */
/* and left eigenvector v corresponding to the generalized eigenvalue w */
/* is defined as follows: */
/* a) If the i-th eigenvalue w = (a,b) is real */
/* Suppose U and V are orthogonal transformations such that */
/* U'*(A, B)*V = (S, T) = ( a * ) ( b * ) 1 */
/* ( 0 S22 ),( 0 T22 ) n-1 */
/* 1 n-1 1 n-1 */
/* Then the reciprocal condition number DIF(i) is */
/* Difl((a, b), (S22, T22)) = sigma-min( Zl ), */
/* where sigma-min(Zl) denotes the smallest singular value of the */
/* 2(n-1)-by-2(n-1) matrix */
/* Zl = [ kron(a, In-1) -kron(1, S22) ] */
/* [ kron(b, In-1) -kron(1, T22) ] . */
/* Here In-1 is the identity matrix of size n-1. kron(X, Y) is the */
/* Kronecker product between the matrices X and Y. */
/* Note that if the default method for computing DIF(i) is wanted */
/* (see DLATDF), then the parameter DIFDRI (see below) should be */
/* changed from 3 to 4 (routine DLATDF(IJOB = 2 will be used)). */
/* See DTGSYL for more details. */
/* b) If the i-th and (i+1)-th eigenvalues are complex conjugate pair, */
/* Suppose U and V are orthogonal transformations such that */
/* U'*(A, B)*V = (S, T) = ( S11 * ) ( T11 * ) 2 */
/* ( 0 S22 ),( 0 T22) n-2 */
/* 2 n-2 2 n-2 */
/* and (S11, T11) corresponds to the complex conjugate eigenvalue */
/* pair (w, conjg(w)). There exist unitary matrices U1 and V1 such */
/* that */
/* U1'*S11*V1 = ( s11 s12 ) and U1'*T11*V1 = ( t11 t12 ) */
/* ( 0 s22 ) ( 0 t22 ) */
/* where the generalized eigenvalues w = s11/t11 and */
/* conjg(w) = s22/t22. */
/* Then the reciprocal condition number DIF(i) is bounded by */
/* min( d1, max( 1, |real(s11)/real(s22)| )*d2 ) */
/* where, d1 = Difl((s11, t11), (s22, t22)) = sigma-min(Z1), where */
/* Z1 is the complex 2-by-2 matrix */
/* Z1 = [ s11 -s22 ] */
/* [ t11 -t22 ], */
/* This is done by computing (using real arithmetic) the */
/* roots of the characteristical polynomial det(Z1' * Z1 - lambda I), */
/* where Z1' denotes the conjugate transpose of Z1 and det(X) denotes */
/* the determinant of X. */
/* and d2 is an upper bound on Difl((S11, T11), (S22, T22)), i.e. an */
/* upper bound on sigma-min(Z2), where Z2 is (2n-2)-by-(2n-2) */
/* Z2 = [ kron(S11', In-2) -kron(I2, S22) ] */
/* [ kron(T11', In-2) -kron(I2, T22) ] */
/* Note that if the default method for computing DIF is wanted (see */
/* DLATDF), then the parameter DIFDRI (see below) should be changed */
/* from 3 to 4 (routine DLATDF(IJOB = 2 will be used)). See DTGSYL */
/* for more details. */
/* For each eigenvalue/vector specified by SELECT, DIF stores a */
/* Frobenius norm-based estimate of Difl. */
/* An approximate error bound for the i-th computed eigenvector VL(i) or */
/* VR(i) is given by */
/* EPS * norm(A, B) / DIF(i). */
/* See ref. [2-3] for more details and further references. */
/* Based on contributions by */
/* Bo Kagstrom and Peter Poromaa, Department of Computing Science, */
/* Umea University, S-901 87 Umea, Sweden. */
/* References */
/* ========== */
/* [1] B. Kagstrom; A Direct Method for Reordering Eigenvalues in the */
/* Generalized Real Schur Form of a Regular Matrix Pair (A, B), in */
/* M.S. Moonen et al (eds), Linear Algebra for Large Scale and */
/* Real-Time Applications, Kluwer Academic Publ. 1993, pp 195-218. */
/* [2] B. Kagstrom and P. Poromaa; Computing Eigenspaces with Specified */
/* Eigenvalues of a Regular Matrix Pair (A, B) and Condition */
/* Estimation: Theory, Algorithms and Software, */
/* Report UMINF - 94.04, Department of Computing Science, Umea */
/* University, S-901 87 Umea, Sweden, 1994. Also as LAPACK Working */
/* Note 87. To appear in Numerical Algorithms, 1996. */
/* [3] B. Kagstrom and P. Poromaa, LAPACK-Style Algorithms and Software */
/* for Solving the Generalized Sylvester Equation and Estimating the */
/* Separation between Regular Matrix Pairs, Report UMINF - 93.23, */
/* Department of Computing Science, Umea University, S-901 87 Umea, */
/* Sweden, December 1993, Revised April 1994, Also as LAPACK Working */
/* Note 75. To appear in ACM Trans. on Math. Software, Vol 22, */
/* No 1, 1996. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. Local Arrays .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Decode and test the input parameters */
/* Parameter adjustments */
--select;
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
vl_dim1 = *ldvl;
vl_offset = 1 + vl_dim1;
vl -= vl_offset;
vr_dim1 = *ldvr;
vr_offset = 1 + vr_dim1;
vr -= vr_offset;
--s;
--dif;
--work;
--iwork;
/* Function Body */
wantbh = lsame_(job, "B");
wants = lsame_(job, "E") || wantbh;
wantdf = lsame_(job, "V") || wantbh;
somcon = lsame_(howmny, "S");
*info = 0;
lquery = *lwork == -1;
if (! wants && ! wantdf) {
*info = -1;
} else if (! lsame_(howmny, "A") && ! somcon) {
*info = -2;
} else if (*n < 0) {
*info = -4;
} else if (*lda < max(1,*n)) {
*info = -6;
} else if (*ldb < max(1,*n)) {
*info = -8;
} else if (wants && *ldvl < *n) {
*info = -10;
} else if (wants && *ldvr < *n) {
*info = -12;
} else {
/* Set M to the number of eigenpairs for which condition numbers */
/* are required, and test MM. */
if (somcon) {
*m = 0;
pair = FALSE_;
i__1 = *n;
for (k = 1; k <= i__1; ++k) {
if (pair) {
pair = FALSE_;
} else {
if (k < *n) {
if (a[k + 1 + k * a_dim1] == 0.) {
if (select[k]) {
++(*m);
}
} else {
pair = TRUE_;
if (select[k] || select[k + 1]) {
*m += 2;
}
}
} else {
if (select[*n]) {
++(*m);
}
}
}
/* L10: */
}
} else {
*m = *n;
}
if (*n == 0) {
lwmin = 1;
} else if (lsame_(job, "V") || lsame_(job,
"B")) {
lwmin = (*n << 1) * (*n + 2) + 16;
} else {
lwmin = *n;
}
work[1] = (doublereal) lwmin;
if (*mm < *m) {
*info = -15;
} else if (*lwork < lwmin && ! lquery) {
*info = -18;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("DTGSNA", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
/* Get machine constants */
eps = dlamch_("P");
smlnum = dlamch_("S") / eps;
ks = 0;
pair = FALSE_;
i__1 = *n;
for (k = 1; k <= i__1; ++k) {
/* Determine whether A(k,k) begins a 1-by-1 or 2-by-2 block. */
if (pair) {
pair = FALSE_;
goto L20;
} else {
if (k < *n) {
pair = a[k + 1 + k * a_dim1] != 0.;
}
}
/* Determine whether condition numbers are required for the k-th */
/* eigenpair. */
if (somcon) {
if (pair) {
if (! select[k] && ! select[k + 1]) {
goto L20;
}
} else {
if (! select[k]) {
goto L20;
}
}
}
++ks;
if (wants) {
/* Compute the reciprocal condition number of the k-th */
/* eigenvalue. */
if (pair) {
/* Complex eigenvalue pair. */
d__1 = dnrm2_(n, &vr[ks * vr_dim1 + 1], &c__1);
d__2 = dnrm2_(n, &vr[(ks + 1) * vr_dim1 + 1], &c__1);
rnrm = dlapy2_(&d__1, &d__2);
d__1 = dnrm2_(n, &vl[ks * vl_dim1 + 1], &c__1);
d__2 = dnrm2_(n, &vl[(ks + 1) * vl_dim1 + 1], &c__1);
lnrm = dlapy2_(&d__1, &d__2);
dgemv_("N", n, n, &c_b19, &a[a_offset], lda, &vr[ks * vr_dim1
+ 1], &c__1, &c_b21, &work[1], &c__1);
tmprr = ddot_(n, &work[1], &c__1, &vl[ks * vl_dim1 + 1], &
c__1);
tmpri = ddot_(n, &work[1], &c__1, &vl[(ks + 1) * vl_dim1 + 1],
&c__1);
dgemv_("N", n, n, &c_b19, &a[a_offset], lda, &vr[(ks + 1) *
vr_dim1 + 1], &c__1, &c_b21, &work[1], &c__1);
tmpii = ddot_(n, &work[1], &c__1, &vl[(ks + 1) * vl_dim1 + 1],
&c__1);
tmpir = ddot_(n, &work[1], &c__1, &vl[ks * vl_dim1 + 1], &
c__1);
uhav = tmprr + tmpii;
uhavi = tmpir - tmpri;
dgemv_("N", n, n, &c_b19, &b[b_offset], ldb, &vr[ks * vr_dim1
+ 1], &c__1, &c_b21, &work[1], &c__1);
tmprr = ddot_(n, &work[1], &c__1, &vl[ks * vl_dim1 + 1], &
c__1);
tmpri = ddot_(n, &work[1], &c__1, &vl[(ks + 1) * vl_dim1 + 1],
&c__1);
dgemv_("N", n, n, &c_b19, &b[b_offset], ldb, &vr[(ks + 1) *
vr_dim1 + 1], &c__1, &c_b21, &work[1], &c__1);
tmpii = ddot_(n, &work[1], &c__1, &vl[(ks + 1) * vl_dim1 + 1],
&c__1);
tmpir = ddot_(n, &work[1], &c__1, &vl[ks * vl_dim1 + 1], &
c__1);
uhbv = tmprr + tmpii;
uhbvi = tmpir - tmpri;
uhav = dlapy2_(&uhav, &uhavi);
uhbv = dlapy2_(&uhbv, &uhbvi);
cond = dlapy2_(&uhav, &uhbv);
s[ks] = cond / (rnrm * lnrm);
s[ks + 1] = s[ks];
} else {
/* Real eigenvalue. */
rnrm = dnrm2_(n, &vr[ks * vr_dim1 + 1], &c__1);
lnrm = dnrm2_(n, &vl[ks * vl_dim1 + 1], &c__1);
dgemv_("N", n, n, &c_b19, &a[a_offset], lda, &vr[ks * vr_dim1
+ 1], &c__1, &c_b21, &work[1], &c__1);
uhav = ddot_(n, &work[1], &c__1, &vl[ks * vl_dim1 + 1], &c__1)
;
dgemv_("N", n, n, &c_b19, &b[b_offset], ldb, &vr[ks * vr_dim1
+ 1], &c__1, &c_b21, &work[1], &c__1);
uhbv = ddot_(n, &work[1], &c__1, &vl[ks * vl_dim1 + 1], &c__1)
;
cond = dlapy2_(&uhav, &uhbv);
if (cond == 0.) {
s[ks] = -1.;
} else {
s[ks] = cond / (rnrm * lnrm);
}
}
}
if (wantdf) {
if (*n == 1) {
dif[ks] = dlapy2_(&a[a_dim1 + 1], &b[b_dim1 + 1]);
goto L20;
}
/* Estimate the reciprocal condition number of the k-th */
/* eigenvectors. */
if (pair) {
/* Copy the 2-by 2 pencil beginning at (A(k,k), B(k, k)). */
/* Compute the eigenvalue(s) at position K. */
work[1] = a[k + k * a_dim1];
work[2] = a[k + 1 + k * a_dim1];
work[3] = a[k + (k + 1) * a_dim1];
work[4] = a[k + 1 + (k + 1) * a_dim1];
work[5] = b[k + k * b_dim1];
work[6] = b[k + 1 + k * b_dim1];
work[7] = b[k + (k + 1) * b_dim1];
work[8] = b[k + 1 + (k + 1) * b_dim1];
d__1 = smlnum * eps;
dlag2_(&work[1], &c__2, &work[5], &c__2, &d__1, &beta, dummy1,
&alphar, dummy, &alphai);
alprqt = 1.;
c1 = (alphar * alphar + alphai * alphai + beta * beta) * 2.;
c2 = beta * 4. * beta * alphai * alphai;
root1 = c1 + sqrt(c1 * c1 - c2 * 4.);
root2 = c2 / root1;
root1 /= 2.;
/* Computing MIN */
d__1 = sqrt(root1), d__2 = sqrt(root2);
cond = min(d__1,d__2);
}
/* Copy the matrix (A, B) to the array WORK and swap the */
/* diagonal block beginning at A(k,k) to the (1,1) position. */
dlacpy_("Full", n, n, &a[a_offset], lda, &work[1], n);
dlacpy_("Full", n, n, &b[b_offset], ldb, &work[*n * *n + 1], n);
ifst = k;
ilst = 1;
i__2 = *lwork - (*n << 1) * *n;
dtgexc_(&c_false, &c_false, n, &work[1], n, &work[*n * *n + 1], n,
dummy, &c__1, dummy1, &c__1, &ifst, &ilst, &work[(*n * *
n << 1) + 1], &i__2, &ierr);
if (ierr > 0) {
/* Ill-conditioned problem - swap rejected. */
dif[ks] = 0.;
} else {
/* Reordering successful, solve generalized Sylvester */
/* equation for R and L, */
/* A22 * R - L * A11 = A12 */
/* B22 * R - L * B11 = B12, */
/* and compute estimate of Difl((A11,B11), (A22, B22)). */
n1 = 1;
if (work[2] != 0.) {
n1 = 2;
}
n2 = *n - n1;
if (n2 == 0) {
dif[ks] = cond;
} else {
i__ = *n * *n + 1;
iz = (*n << 1) * *n + 1;
i__2 = *lwork - (*n << 1) * *n;
dtgsyl_("N", &c__3, &n2, &n1, &work[*n * n1 + n1 + 1], n,
&work[1], n, &work[n1 + 1], n, &work[*n * n1 + n1
+ i__], n, &work[i__], n, &work[n1 + i__], n, &
scale, &dif[ks], &work[iz + 1], &i__2, &iwork[1],
&ierr);
if (pair) {
/* Computing MIN */
d__1 = max(1.,alprqt) * dif[ks];
dif[ks] = min(d__1,cond);
}
}
}
if (pair) {
dif[ks + 1] = dif[ks];
}
}
if (pair) {
++ks;
}
L20:
;
}
work[1] = (doublereal) lwmin;
return 0;
/* End of DTGSNA */
} /* dtgsna_ */
/* Subroutine */ int dgeev_(char *jobvl, char *jobvr, integer *n, doublereal *
a, integer *lda, doublereal *wr, doublereal *wi, doublereal *vl,
integer *ldvl, doublereal *vr, integer *ldvr, doublereal *work,
integer *lwork, integer *info)
{
/* -- LAPACK driver routine (version 2.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
September 30, 1994
Purpose
=======
DGEEV computes for an N-by-N real nonsymmetric matrix A, the
eigenvalues and, optionally, the left and/or right eigenvectors.
The right eigenvector v(j) of A satisfies
A * v(j) = lambda(j) * v(j)
where lambda(j) is its eigenvalue.
The left eigenvector u(j) of A satisfies
u(j)**H * A = lambda(j) * u(j)**H
where u(j)**H denotes the conjugate transpose of u(j).
The computed eigenvectors are normalized to have Euclidean norm
equal to 1 and largest component real.
Arguments
=========
JOBVL (input) CHARACTER*1
= 'N': left eigenvectors of A are not computed;
= 'V': left eigenvectors of A are computed.
JOBVR (input) CHARACTER*1
= 'N': right eigenvectors of A are not computed;
= 'V': right eigenvectors of A are computed.
N (input) INTEGER
The order of the matrix A. N >= 0.
A (input/output) DOUBLE PRECISION array, dimension (LDA,N)
On entry, the N-by-N matrix A.
On exit, A has been overwritten.
LDA (input) INTEGER
The leading dimension of the array A. LDA >= max(1,N).
WR (output) DOUBLE PRECISION array, dimension (N)
WI (output) DOUBLE PRECISION array, dimension (N)
WR and WI contain the real and imaginary parts,
respectively, of the computed eigenvalues. Complex
conjugate pairs of eigenvalues appear consecutively
with the eigenvalue having the positive imaginary part
first.
VL (output) DOUBLE PRECISION array, dimension (LDVL,N)
If JOBVL = 'V', the left eigenvectors u(j) are stored one
after another in the columns of VL, in the same order
as their eigenvalues.
If JOBVL = 'N', VL is not referenced.
If the j-th eigenvalue is real, then u(j) = VL(:,j),
the j-th column of VL.
If the j-th and (j+1)-st eigenvalues form a complex
conjugate pair, then u(j) = VL(:,j) + i*VL(:,j+1) and
u(j+1) = VL(:,j) - i*VL(:,j+1).
LDVL (input) INTEGER
The leading dimension of the array VL. LDVL >= 1; if
JOBVL = 'V', LDVL >= N.
VR (output) DOUBLE PRECISION array, dimension (LDVR,N)
If JOBVR = 'V', the right eigenvectors v(j) are stored one
after another in the columns of VR, in the same order
as their eigenvalues.
If JOBVR = 'N', VR is not referenced.
If the j-th eigenvalue is real, then v(j) = VR(:,j),
the j-th column of VR.
If the j-th and (j+1)-st eigenvalues form a complex
conjugate pair, then v(j) = VR(:,j) + i*VR(:,j+1) and
v(j+1) = VR(:,j) - i*VR(:,j+1).
LDVR (input) INTEGER
The leading dimension of the array VR. LDVR >= 1; if
JOBVR = 'V', LDVR >= N.
WORK (workspace/output) DOUBLE PRECISION array, dimension (LWORK)
On exit, if INFO = 0, WORK(1) returns the optimal LWORK.
LWORK (input) INTEGER
The dimension of the array WORK. LWORK >= max(1,3*N), and
if JOBVL = 'V' or JOBVR = 'V', LWORK >= 4*N. For good
performance, LWORK must generally be larger.
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value.
> 0: if INFO = i, the QR algorithm failed to compute all the
eigenvalues, and no eigenvectors have been computed;
elements i+1:N of WR and WI contain eigenvalues which
have converged.
=====================================================================
Test the input arguments
Parameter adjustments
Function Body */
/* Table of constant values */
static integer c__1 = 1;
static integer c__0 = 0;
static integer c__8 = 8;
static integer c_n1 = -1;
static integer c__4 = 4;
/* System generated locals */
integer a_dim1, a_offset, vl_dim1, vl_offset, vr_dim1, vr_offset, i__1,
i__2, i__3, i__4;
doublereal d__1, d__2;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
static integer ibal;
static char side[1];
static integer maxb;
static doublereal anrm;
static integer ierr, itau;
extern /* Subroutine */ int drot_(integer *, doublereal *, integer *,
doublereal *, integer *, doublereal *, doublereal *);
static integer iwrk, nout;
extern doublereal dnrm2_(integer *, doublereal *, integer *);
static integer i, k;
static doublereal r;
extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
integer *);
extern logical lsame_(char *, char *);
extern doublereal dlapy2_(doublereal *, doublereal *);
extern /* Subroutine */ int dlabad_(doublereal *, doublereal *), dgebak_(
char *, char *, integer *, integer *, integer *, doublereal *,
integer *, doublereal *, integer *, integer *),
dgebal_(char *, integer *, doublereal *, integer *, integer *,
integer *, doublereal *, integer *);
static doublereal cs;
static logical scalea;
extern doublereal dlamch_(char *);
static doublereal cscale;
extern doublereal dlange_(char *, integer *, integer *, doublereal *,
integer *, doublereal *);
extern /* Subroutine */ int dgehrd_(integer *, integer *, integer *,
doublereal *, integer *, doublereal *, doublereal *, integer *,
integer *);
static doublereal sn;
extern /* Subroutine */ int dlascl_(char *, integer *, integer *,
doublereal *, doublereal *, integer *, integer *, doublereal *,
integer *, integer *);
extern integer idamax_(integer *, doublereal *, integer *);
extern /* Subroutine */ int dlacpy_(char *, integer *, integer *,
doublereal *, integer *, doublereal *, integer *),
dlartg_(doublereal *, doublereal *, doublereal *, doublereal *,
doublereal *), xerbla_(char *, integer *);
static logical select[1];
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *, ftnlen, ftnlen);
static doublereal bignum;
extern /* Subroutine */ int dorghr_(integer *, integer *, integer *,
doublereal *, integer *, doublereal *, doublereal *, integer *,
integer *), dhseqr_(char *, char *, integer *, integer *, integer
*, doublereal *, integer *, doublereal *, doublereal *,
doublereal *, integer *, doublereal *, integer *, integer *), dtrevc_(char *, char *, logical *, integer *,
doublereal *, integer *, doublereal *, integer *, doublereal *,
integer *, integer *, integer *, doublereal *, integer *);
static integer minwrk, maxwrk;
static logical wantvl;
static doublereal smlnum;
static integer hswork;
static logical wantvr;
static integer ihi;
static doublereal scl;
static integer ilo;
static doublereal dum[1], eps;
#define DUM(I) dum[(I)]
#define WR(I) wr[(I)-1]
#define WI(I) wi[(I)-1]
#define WORK(I) work[(I)-1]
#define A(I,J) a[(I)-1 + ((J)-1)* ( *lda)]
#define VL(I,J) vl[(I)-1 + ((J)-1)* ( *ldvl)]
#define VR(I,J) vr[(I)-1 + ((J)-1)* ( *ldvr)]
*info = 0;
wantvl = lsame_(jobvl, "V");
wantvr = lsame_(jobvr, "V");
if (! wantvl && ! lsame_(jobvl, "N")) {
*info = -1;
} else if (! wantvr && ! lsame_(jobvr, "N")) {
*info = -2;
} else if (*n < 0) {
*info = -3;
} else if (*lda < max(1,*n)) {
*info = -5;
} else if (*ldvl < 1 || wantvl && *ldvl < *n) {
*info = -9;
} else if (*ldvr < 1 || wantvr && *ldvr < *n) {
*info = -11;
}
/* Compute workspace
(Note: Comments in the code beginning "Workspace:" describe the
minimal amount of workspace needed at that point in the code,
as well as the preferred amount for good performance.
NB refers to the optimal block size for the immediately
following subroutine, as returned by ILAENV.
HSWORK refers to the workspace preferred by DHSEQR, as
calculated below. HSWORK is computed assuming ILO=1 and IHI=N,
the worst case.) */
minwrk = 1;
if (*info == 0 && *lwork >= 1) {
maxwrk = (*n << 1) + *n * ilaenv_(&c__1, "DGEHRD", " ", n, &c__1, n, &
c__0, 6L, 1L);
if (! wantvl && ! wantvr) {
/* Computing MAX */
i__1 = 1, i__2 = *n * 3;
minwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = ilaenv_(&c__8, "DHSEQR", "EN", n, &c__1, n, &c_n1, 6L, 2L);
maxb = max(i__1,2);
/* Computing MIN
Computing MAX */
i__3 = 2, i__4 = ilaenv_(&c__4, "DHSEQR", "EN", n, &c__1, n, &
c_n1, 6L, 2L);
i__1 = min(maxb,*n), i__2 = max(i__3,i__4);
k = min(i__1,i__2);
/* Computing MAX */
i__1 = k * (k + 2), i__2 = *n << 1;
hswork = max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk, i__2 = *n + 1, i__1 = max(i__1,i__2), i__2 = *n +
hswork;
maxwrk = max(i__1,i__2);
} else {
/* Computing MAX */
i__1 = 1, i__2 = *n << 2;
minwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk, i__2 = (*n << 1) + (*n - 1) * ilaenv_(&c__1, "DOR"
"GHR", " ", n, &c__1, n, &c_n1, 6L, 1L);
maxwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = ilaenv_(&c__8, "DHSEQR", "SV", n, &c__1, n, &c_n1, 6L, 2L);
maxb = max(i__1,2);
/* Computing MIN
Computing MAX */
i__3 = 2, i__4 = ilaenv_(&c__4, "DHSEQR", "SV", n, &c__1, n, &
c_n1, 6L, 2L);
i__1 = min(maxb,*n), i__2 = max(i__3,i__4);
k = min(i__1,i__2);
/* Computing MAX */
i__1 = k * (k + 2), i__2 = *n << 1;
hswork = max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk, i__2 = *n + 1, i__1 = max(i__1,i__2), i__2 = *n +
hswork;
maxwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk, i__2 = *n << 2;
maxwrk = max(i__1,i__2);
}
WORK(1) = (doublereal) maxwrk;
}
if (*lwork < minwrk) {
*info = -13;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("DGEEV ", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
/* Get machine constants */
eps = dlamch_("P");
smlnum = dlamch_("S");
bignum = 1. / smlnum;
dlabad_(&smlnum, &bignum);
smlnum = sqrt(smlnum) / eps;
bignum = 1. / smlnum;
/* Scale A if max element outside range [SMLNUM,BIGNUM] */
anrm = dlange_("M", n, n, &A(1,1), lda, dum);
scalea = FALSE_;
if (anrm > 0. && anrm < smlnum) {
scalea = TRUE_;
cscale = smlnum;
} else if (anrm > bignum) {
scalea = TRUE_;
cscale = bignum;
}
if (scalea) {
dlascl_("G", &c__0, &c__0, &anrm, &cscale, n, n, &A(1,1), lda, &
ierr);
}
/* Balance the matrix
(Workspace: need N) */
ibal = 1;
dgebal_("B", n, &A(1,1), lda, &ilo, &ihi, &WORK(ibal), &ierr);
/* Reduce to upper Hessenberg form
(Workspace: need 3*N, prefer 2*N+N*NB) */
itau = ibal + *n;
iwrk = itau + *n;
i__1 = *lwork - iwrk + 1;
dgehrd_(n, &ilo, &ihi, &A(1,1), lda, &WORK(itau), &WORK(iwrk), &i__1,
&ierr);
if (wantvl) {
/* Want left eigenvectors
Copy Householder vectors to VL */
*(unsigned char *)side = 'L';
dlacpy_("L", n, n, &A(1,1), lda, &VL(1,1), ldvl);
/* Generate orthogonal matrix in VL
(Workspace: need 3*N-1, prefer 2*N+(N-1)*NB) */
i__1 = *lwork - iwrk + 1;
dorghr_(n, &ilo, &ihi, &VL(1,1), ldvl, &WORK(itau), &WORK(iwrk),
&i__1, &ierr);
/* Perform QR iteration, accumulating Schur vectors in VL
(Workspace: need N+1, prefer N+HSWORK (see comments) ) */
iwrk = itau;
i__1 = *lwork - iwrk + 1;
dhseqr_("S", "V", n, &ilo, &ihi, &A(1,1), lda, &WR(1), &WI(1), &
VL(1,1), ldvl, &WORK(iwrk), &i__1, info);
if (wantvr) {
/* Want left and right eigenvectors
Copy Schur vectors to VR */
*(unsigned char *)side = 'B';
dlacpy_("F", n, n, &VL(1,1), ldvl, &VR(1,1), ldvr)
;
}
} else if (wantvr) {
/* Want right eigenvectors
Copy Householder vectors to VR */
*(unsigned char *)side = 'R';
dlacpy_("L", n, n, &A(1,1), lda, &VR(1,1), ldvr);
/* Generate orthogonal matrix in VR
(Workspace: need 3*N-1, prefer 2*N+(N-1)*NB) */
i__1 = *lwork - iwrk + 1;
dorghr_(n, &ilo, &ihi, &VR(1,1), ldvr, &WORK(itau), &WORK(iwrk),
&i__1, &ierr);
/* Perform QR iteration, accumulating Schur vectors in VR
(Workspace: need N+1, prefer N+HSWORK (see comments) ) */
iwrk = itau;
i__1 = *lwork - iwrk + 1;
dhseqr_("S", "V", n, &ilo, &ihi, &A(1,1), lda, &WR(1), &WI(1), &
VR(1,1), ldvr, &WORK(iwrk), &i__1, info);
} else {
/* Compute eigenvalues only
(Workspace: need N+1, prefer N+HSWORK (see comments) ) */
iwrk = itau;
i__1 = *lwork - iwrk + 1;
dhseqr_("E", "N", n, &ilo, &ihi, &A(1,1), lda, &WR(1), &WI(1), &
VR(1,1), ldvr, &WORK(iwrk), &i__1, info);
}
/* If INFO > 0 from DHSEQR, then quit */
if (*info > 0) {
goto L50;
}
if (wantvl || wantvr) {
/* Compute left and/or right eigenvectors
(Workspace: need 4*N) */
dtrevc_(side, "B", select, n, &A(1,1), lda, &VL(1,1), ldvl,
&VR(1,1), ldvr, n, &nout, &WORK(iwrk), &ierr);
}
if (wantvl) {
/* Undo balancing of left eigenvectors
(Workspace: need N) */
dgebak_("B", "L", n, &ilo, &ihi, &WORK(ibal), n, &VL(1,1), ldvl,
&ierr);
/* Normalize left eigenvectors and make largest component real
*/
i__1 = *n;
for (i = 1; i <= *n; ++i) {
if (WI(i) == 0.) {
scl = 1. / dnrm2_(n, &VL(1,i), &c__1);
dscal_(n, &scl, &VL(1,i), &c__1);
} else if (WI(i) > 0.) {
d__1 = dnrm2_(n, &VL(1,i), &c__1);
d__2 = dnrm2_(n, &VL(1,i+1), &c__1);
scl = 1. / dlapy2_(&d__1, &d__2);
dscal_(n, &scl, &VL(1,i), &c__1);
dscal_(n, &scl, &VL(1,i+1), &c__1);
i__2 = *n;
for (k = 1; k <= *n; ++k) {
/* Computing 2nd power */
d__1 = VL(k,i);
/* Computing 2nd power */
d__2 = VL(k,i+1);
WORK(iwrk + k - 1) = d__1 * d__1 + d__2 * d__2;
/* L10: */
}
k = idamax_(n, &WORK(iwrk), &c__1);
dlartg_(&VL(k,i), &VL(k,i+1), &cs,
&sn, &r);
drot_(n, &VL(1,i), &c__1, &VL(1,i+1), &c__1, &cs, &sn);
VL(k,i+1) = 0.;
}
/* L20: */
}
}
if (wantvr) {
/* Undo balancing of right eigenvectors
(Workspace: need N) */
dgebak_("B", "R", n, &ilo, &ihi, &WORK(ibal), n, &VR(1,1), ldvr,
&ierr);
/* Normalize right eigenvectors and make largest component real
*/
i__1 = *n;
for (i = 1; i <= *n; ++i) {
if (WI(i) == 0.) {
scl = 1. / dnrm2_(n, &VR(1,i), &c__1);
dscal_(n, &scl, &VR(1,i), &c__1);
} else if (WI(i) > 0.) {
d__1 = dnrm2_(n, &VR(1,i), &c__1);
d__2 = dnrm2_(n, &VR(1,i+1), &c__1);
scl = 1. / dlapy2_(&d__1, &d__2);
dscal_(n, &scl, &VR(1,i), &c__1);
dscal_(n, &scl, &VR(1,i+1), &c__1);
i__2 = *n;
for (k = 1; k <= *n; ++k) {
/* Computing 2nd power */
d__1 = VR(k,i);
/* Computing 2nd power */
d__2 = VR(k,i+1);
WORK(iwrk + k - 1) = d__1 * d__1 + d__2 * d__2;
/* L30: */
}
k = idamax_(n, &WORK(iwrk), &c__1);
dlartg_(&VR(k,i), &VR(k,i+1), &cs,
&sn, &r);
drot_(n, &VR(1,i), &c__1, &VR(1,i+1), &c__1, &cs, &sn);
VR(k,i+1) = 0.;
}
/* L40: */
}
}
/* Undo scaling if necessary */
L50:
if (scalea) {
i__1 = *n - *info;
/* Computing MAX */
i__3 = *n - *info;
i__2 = max(i__3,1);
dlascl_("G", &c__0, &c__0, &cscale, &anrm, &i__1, &c__1, &WR(*info +
1), &i__2, &ierr);
i__1 = *n - *info;
/* Computing MAX */
i__3 = *n - *info;
i__2 = max(i__3,1);
dlascl_("G", &c__0, &c__0, &cscale, &anrm, &i__1, &c__1, &WI(*info +
1), &i__2, &ierr);
if (*info > 0) {
i__1 = ilo - 1;
dlascl_("G", &c__0, &c__0, &cscale, &anrm, &i__1, &c__1, &WR(1),
n, &ierr);
i__1 = ilo - 1;
dlascl_("G", &c__0, &c__0, &cscale, &anrm, &i__1, &c__1, &WI(1),
n, &ierr);
}
}
WORK(1) = (doublereal) maxwrk;
return 0;
/* End of DGEEV */
} /* dgeev_ */
/* Subroutine */ int zhetf2_(char *uplo, integer *n, doublecomplex *a,
integer *lda, integer *ipiv, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, i__1, i__2, i__3, i__4, i__5, i__6;
doublereal d__1, d__2, d__3, d__4;
doublecomplex z__1, z__2, z__3, z__4, z__5, z__6;
/* Builtin functions */
double sqrt(doublereal), d_imag(doublecomplex *);
void d_cnjg(doublecomplex *, doublecomplex *);
/* Local variables */
doublereal d__;
integer i__, j, k;
doublecomplex t;
doublereal r1, d11;
doublecomplex d12;
doublereal d22;
doublecomplex d21;
integer kk, kp;
doublecomplex wk;
doublereal tt;
doublecomplex wkm1, wkp1;
integer imax, jmax;
extern /* Subroutine */ int zher_(char *, integer *, doublereal *,
doublecomplex *, integer *, doublecomplex *, integer *);
doublereal alpha;
extern logical lsame_(char *, char *);
integer kstep;
logical upper;
extern /* Subroutine */ int zswap_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *);
extern doublereal dlapy2_(doublereal *, doublereal *);
doublereal absakk;
extern logical disnan_(doublereal *);
extern /* Subroutine */ int xerbla_(char *, integer *), zdscal_(
integer *, doublereal *, doublecomplex *, integer *);
doublereal colmax;
extern integer izamax_(integer *, doublecomplex *, integer *);
doublereal rowmax;
/* -- LAPACK routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* ZHETF2 computes the factorization of a complex Hermitian 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 conjugate transpose of U, and D is */
/* Hermitian 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 */
/* Hermitian matrix A is stored: */
/* = 'U': Upper triangular */
/* = 'L': Lower triangular */
/* N (input) INTEGER */
/* The order of the matrix A. N >= 0. */
/* A (input/output) COMPLEX*16 array, dimension (LDA,N) */
/* On entry, the Hermitian matrix A. If UPLO = 'U', the leading */
/* n-by-n upper triangular part of A contains the upper */
/* triangular part of the matrix A, and the strictly lower */
/* triangular part of A is not referenced. If UPLO = 'L', the */
/* leading n-by-n lower triangular part of A contains the lower */
/* triangular part of the matrix A, and the strictly upper */
/* triangular part of A is not referenced. */
/* 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.210 and l.393 */
/* 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 */
/* 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 .. */
/* .. */
/* .. Statement Functions .. */
/* .. */
/* .. Statement Function definitions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
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_("ZHETF2", &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 L90;
}
kstep = 1;
/* Determine rows and columns to be interchanged and whether */
/* a 1-by-1 or 2-by-2 pivot block will be used */
i__1 = k + k * a_dim1;
absakk = (d__1 = a[i__1].r, 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 = izamax_(&i__1, &a[k * a_dim1 + 1], &c__1);
i__1 = imax + k * a_dim1;
colmax = (d__1 = a[i__1].r, abs(d__1)) + (d__2 = d_imag(&a[imax +
k * a_dim1]), abs(d__2));
} 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;
i__1 = k + k * a_dim1;
i__2 = k + k * a_dim1;
d__1 = a[i__2].r;
a[i__1].r = d__1, a[i__1].i = 0.;
} 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 + izamax_(&i__1, &a[imax + (imax + 1) * a_dim1],
lda);
i__1 = imax + jmax * a_dim1;
rowmax = (d__1 = a[i__1].r, abs(d__1)) + (d__2 = d_imag(&a[
imax + jmax * a_dim1]), abs(d__2));
if (imax > 1) {
i__1 = imax - 1;
jmax = izamax_(&i__1, &a[imax * a_dim1 + 1], &c__1);
/* Computing MAX */
i__1 = jmax + imax * a_dim1;
d__3 = rowmax, d__4 = (d__1 = a[i__1].r, abs(d__1)) + (
d__2 = d_imag(&a[jmax + imax * a_dim1]), abs(d__2)
);
rowmax = max(d__3,d__4);
}
if (absakk >= alpha * colmax * (colmax / rowmax)) {
/* no interchange, use 1-by-1 pivot block */
kp = k;
} else /* if(complicated condition) */ {
i__1 = imax + imax * a_dim1;
if ((d__1 = a[i__1].r, 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;
zswap_(&i__1, &a[kk * a_dim1 + 1], &c__1, &a[kp * a_dim1 + 1],
&c__1);
i__1 = kk - 1;
for (j = kp + 1; j <= i__1; ++j) {
d_cnjg(&z__1, &a[j + kk * a_dim1]);
t.r = z__1.r, t.i = z__1.i;
i__2 = j + kk * a_dim1;
d_cnjg(&z__1, &a[kp + j * a_dim1]);
a[i__2].r = z__1.r, a[i__2].i = z__1.i;
i__2 = kp + j * a_dim1;
a[i__2].r = t.r, a[i__2].i = t.i;
/* L20: */
}
i__1 = kp + kk * a_dim1;
d_cnjg(&z__1, &a[kp + kk * a_dim1]);
a[i__1].r = z__1.r, a[i__1].i = z__1.i;
i__1 = kk + kk * a_dim1;
r1 = a[i__1].r;
i__1 = kk + kk * a_dim1;
i__2 = kp + kp * a_dim1;
d__1 = a[i__2].r;
a[i__1].r = d__1, a[i__1].i = 0.;
i__1 = kp + kp * a_dim1;
a[i__1].r = r1, a[i__1].i = 0.;
if (kstep == 2) {
i__1 = k + k * a_dim1;
i__2 = k + k * a_dim1;
d__1 = a[i__2].r;
a[i__1].r = d__1, a[i__1].i = 0.;
i__1 = k - 1 + k * a_dim1;
t.r = a[i__1].r, t.i = a[i__1].i;
i__1 = k - 1 + k * a_dim1;
i__2 = kp + k * a_dim1;
a[i__1].r = a[i__2].r, a[i__1].i = a[i__2].i;
i__1 = kp + k * a_dim1;
a[i__1].r = t.r, a[i__1].i = t.i;
}
} else {
i__1 = k + k * a_dim1;
i__2 = k + k * a_dim1;
d__1 = a[i__2].r;
a[i__1].r = d__1, a[i__1].i = 0.;
if (kstep == 2) {
i__1 = k - 1 + (k - 1) * a_dim1;
i__2 = k - 1 + (k - 1) * a_dim1;
d__1 = a[i__2].r;
a[i__1].r = d__1, a[i__1].i = 0.;
}
}
/* 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)' */
i__1 = k + k * a_dim1;
r1 = 1. / a[i__1].r;
i__1 = k - 1;
d__1 = -r1;
zher_(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;
zdscal_(&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) {
i__1 = k - 1 + k * a_dim1;
d__1 = a[i__1].r;
d__2 = d_imag(&a[k - 1 + k * a_dim1]);
d__ = dlapy2_(&d__1, &d__2);
i__1 = k - 1 + (k - 1) * a_dim1;
d22 = a[i__1].r / d__;
i__1 = k + k * a_dim1;
d11 = a[i__1].r / d__;
tt = 1. / (d11 * d22 - 1.);
i__1 = k - 1 + k * a_dim1;
z__1.r = a[i__1].r / d__, z__1.i = a[i__1].i / d__;
d12.r = z__1.r, d12.i = z__1.i;
d__ = tt / d__;
for (j = k - 2; j >= 1; --j) {
i__1 = j + (k - 1) * a_dim1;
z__3.r = d11 * a[i__1].r, z__3.i = d11 * a[i__1].i;
d_cnjg(&z__5, &d12);
i__2 = j + k * a_dim1;
z__4.r = z__5.r * a[i__2].r - z__5.i * a[i__2].i,
z__4.i = z__5.r * a[i__2].i + z__5.i * a[i__2]
.r;
z__2.r = z__3.r - z__4.r, z__2.i = z__3.i - z__4.i;
z__1.r = d__ * z__2.r, z__1.i = d__ * z__2.i;
wkm1.r = z__1.r, wkm1.i = z__1.i;
i__1 = j + k * a_dim1;
z__3.r = d22 * a[i__1].r, z__3.i = d22 * a[i__1].i;
i__2 = j + (k - 1) * a_dim1;
z__4.r = d12.r * a[i__2].r - d12.i * a[i__2].i,
z__4.i = d12.r * a[i__2].i + d12.i * a[i__2]
.r;
z__2.r = z__3.r - z__4.r, z__2.i = z__3.i - z__4.i;
z__1.r = d__ * z__2.r, z__1.i = d__ * z__2.i;
wk.r = z__1.r, wk.i = z__1.i;
for (i__ = j; i__ >= 1; --i__) {
i__1 = i__ + j * a_dim1;
i__2 = i__ + j * a_dim1;
i__3 = i__ + k * a_dim1;
d_cnjg(&z__4, &wk);
z__3.r = a[i__3].r * z__4.r - a[i__3].i * z__4.i,
z__3.i = a[i__3].r * z__4.i + a[i__3].i *
z__4.r;
z__2.r = a[i__2].r - z__3.r, z__2.i = a[i__2].i -
z__3.i;
i__4 = i__ + (k - 1) * a_dim1;
d_cnjg(&z__6, &wkm1);
z__5.r = a[i__4].r * z__6.r - a[i__4].i * z__6.i,
z__5.i = a[i__4].r * z__6.i + a[i__4].i *
z__6.r;
z__1.r = z__2.r - z__5.r, z__1.i = z__2.i -
z__5.i;
a[i__1].r = z__1.r, a[i__1].i = z__1.i;
/* L30: */
}
i__1 = j + k * a_dim1;
a[i__1].r = wk.r, a[i__1].i = wk.i;
i__1 = j + (k - 1) * a_dim1;
a[i__1].r = wkm1.r, a[i__1].i = wkm1.i;
i__1 = j + j * a_dim1;
i__2 = j + j * a_dim1;
d__1 = a[i__2].r;
z__1.r = d__1, z__1.i = 0.;
a[i__1].r = z__1.r, a[i__1].i = z__1.i;
/* L40: */
}
}
}
}
/* 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;
L50:
/* If K > N, exit from loop */
if (k > *n) {
goto L90;
}
kstep = 1;
/* Determine rows and columns to be interchanged and whether */
/* a 1-by-1 or 2-by-2 pivot block will be used */
i__1 = k + k * a_dim1;
absakk = (d__1 = a[i__1].r, 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 + izamax_(&i__1, &a[k + 1 + k * a_dim1], &c__1);
i__1 = imax + k * a_dim1;
colmax = (d__1 = a[i__1].r, abs(d__1)) + (d__2 = d_imag(&a[imax +
k * a_dim1]), abs(d__2));
} 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;
i__1 = k + k * a_dim1;
i__2 = k + k * a_dim1;
d__1 = a[i__2].r;
a[i__1].r = d__1, a[i__1].i = 0.;
} 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 + izamax_(&i__1, &a[imax + k * a_dim1], lda);
i__1 = imax + jmax * a_dim1;
rowmax = (d__1 = a[i__1].r, abs(d__1)) + (d__2 = d_imag(&a[
imax + jmax * a_dim1]), abs(d__2));
if (imax < *n) {
i__1 = *n - imax;
jmax = imax + izamax_(&i__1, &a[imax + 1 + imax * a_dim1],
&c__1);
/* Computing MAX */
i__1 = jmax + imax * a_dim1;
d__3 = rowmax, d__4 = (d__1 = a[i__1].r, abs(d__1)) + (
d__2 = d_imag(&a[jmax + imax * a_dim1]), abs(d__2)
);
rowmax = max(d__3,d__4);
}
if (absakk >= alpha * colmax * (colmax / rowmax)) {
/* no interchange, use 1-by-1 pivot block */
kp = k;
} else /* if(complicated condition) */ {
i__1 = imax + imax * a_dim1;
if ((d__1 = a[i__1].r, 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;
zswap_(&i__1, &a[kp + 1 + kk * a_dim1], &c__1, &a[kp + 1
+ kp * a_dim1], &c__1);
}
i__1 = kp - 1;
for (j = kk + 1; j <= i__1; ++j) {
d_cnjg(&z__1, &a[j + kk * a_dim1]);
t.r = z__1.r, t.i = z__1.i;
i__2 = j + kk * a_dim1;
d_cnjg(&z__1, &a[kp + j * a_dim1]);
a[i__2].r = z__1.r, a[i__2].i = z__1.i;
i__2 = kp + j * a_dim1;
a[i__2].r = t.r, a[i__2].i = t.i;
/* L60: */
}
i__1 = kp + kk * a_dim1;
d_cnjg(&z__1, &a[kp + kk * a_dim1]);
a[i__1].r = z__1.r, a[i__1].i = z__1.i;
i__1 = kk + kk * a_dim1;
r1 = a[i__1].r;
i__1 = kk + kk * a_dim1;
i__2 = kp + kp * a_dim1;
d__1 = a[i__2].r;
a[i__1].r = d__1, a[i__1].i = 0.;
i__1 = kp + kp * a_dim1;
a[i__1].r = r1, a[i__1].i = 0.;
if (kstep == 2) {
i__1 = k + k * a_dim1;
i__2 = k + k * a_dim1;
d__1 = a[i__2].r;
a[i__1].r = d__1, a[i__1].i = 0.;
i__1 = k + 1 + k * a_dim1;
t.r = a[i__1].r, t.i = a[i__1].i;
i__1 = k + 1 + k * a_dim1;
i__2 = kp + k * a_dim1;
a[i__1].r = a[i__2].r, a[i__1].i = a[i__2].i;
i__1 = kp + k * a_dim1;
a[i__1].r = t.r, a[i__1].i = t.i;
}
} else {
i__1 = k + k * a_dim1;
i__2 = k + k * a_dim1;
d__1 = a[i__2].r;
a[i__1].r = d__1, a[i__1].i = 0.;
if (kstep == 2) {
i__1 = k + 1 + (k + 1) * a_dim1;
i__2 = k + 1 + (k + 1) * a_dim1;
d__1 = a[i__2].r;
a[i__1].r = d__1, a[i__1].i = 0.;
}
}
/* 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)' */
i__1 = k + k * a_dim1;
r1 = 1. / a[i__1].r;
i__1 = *n - k;
d__1 = -r1;
zher_(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;
zdscal_(&i__1, &r1, &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 - ( 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) )' */
/* where L(k) and L(k+1) are the k-th and (k+1)-th */
/* columns of L */
i__1 = k + 1 + k * a_dim1;
d__1 = a[i__1].r;
d__2 = d_imag(&a[k + 1 + k * a_dim1]);
d__ = dlapy2_(&d__1, &d__2);
i__1 = k + 1 + (k + 1) * a_dim1;
d11 = a[i__1].r / d__;
i__1 = k + k * a_dim1;
d22 = a[i__1].r / d__;
tt = 1. / (d11 * d22 - 1.);
i__1 = k + 1 + k * a_dim1;
z__1.r = a[i__1].r / d__, z__1.i = a[i__1].i / d__;
d21.r = z__1.r, d21.i = z__1.i;
d__ = tt / d__;
i__1 = *n;
for (j = k + 2; j <= i__1; ++j) {
i__2 = j + k * a_dim1;
z__3.r = d11 * a[i__2].r, z__3.i = d11 * a[i__2].i;
i__3 = j + (k + 1) * a_dim1;
z__4.r = d21.r * a[i__3].r - d21.i * a[i__3].i,
z__4.i = d21.r * a[i__3].i + d21.i * a[i__3]
.r;
z__2.r = z__3.r - z__4.r, z__2.i = z__3.i - z__4.i;
z__1.r = d__ * z__2.r, z__1.i = d__ * z__2.i;
wk.r = z__1.r, wk.i = z__1.i;
i__2 = j + (k + 1) * a_dim1;
z__3.r = d22 * a[i__2].r, z__3.i = d22 * a[i__2].i;
d_cnjg(&z__5, &d21);
i__3 = j + k * a_dim1;
z__4.r = z__5.r * a[i__3].r - z__5.i * a[i__3].i,
z__4.i = z__5.r * a[i__3].i + z__5.i * a[i__3]
.r;
z__2.r = z__3.r - z__4.r, z__2.i = z__3.i - z__4.i;
z__1.r = d__ * z__2.r, z__1.i = d__ * z__2.i;
wkp1.r = z__1.r, wkp1.i = z__1.i;
i__2 = *n;
for (i__ = j; i__ <= i__2; ++i__) {
i__3 = i__ + j * a_dim1;
i__4 = i__ + j * a_dim1;
i__5 = i__ + k * a_dim1;
d_cnjg(&z__4, &wk);
z__3.r = a[i__5].r * z__4.r - a[i__5].i * z__4.i,
z__3.i = a[i__5].r * z__4.i + a[i__5].i *
z__4.r;
z__2.r = a[i__4].r - z__3.r, z__2.i = a[i__4].i -
z__3.i;
i__6 = i__ + (k + 1) * a_dim1;
d_cnjg(&z__6, &wkp1);
z__5.r = a[i__6].r * z__6.r - a[i__6].i * z__6.i,
z__5.i = a[i__6].r * z__6.i + a[i__6].i *
z__6.r;
z__1.r = z__2.r - z__5.r, z__1.i = z__2.i -
z__5.i;
a[i__3].r = z__1.r, a[i__3].i = z__1.i;
/* L70: */
}
i__2 = j + k * a_dim1;
a[i__2].r = wk.r, a[i__2].i = wk.i;
i__2 = j + (k + 1) * a_dim1;
a[i__2].r = wkp1.r, a[i__2].i = wkp1.i;
i__2 = j + j * a_dim1;
i__3 = j + j * a_dim1;
d__1 = a[i__3].r;
z__1.r = d__1, z__1.i = 0.;
a[i__2].r = z__1.r, a[i__2].i = z__1.i;
/* L80: */
}
}
}
}
/* 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 L50;
}
L90:
return 0;
/* End of ZHETF2 */
} /* zhetf2_ */
/* Subroutine */ int dlaed2_(integer *k, integer *n, doublereal *d,
doublereal *q, integer *ldq, integer *indxq, doublereal *rho, integer
*cutpnt, doublereal *z, doublereal *dlamda, doublereal *q2, integer *
ldq2, integer *indxc, doublereal *w, integer *indxp, integer *indx,
integer *coltyp, integer *info)
{
/* -- LAPACK routine (version 2.0) --
Univ. of Tennessee, Oak Ridge National Lab, Argonne National Lab,
Courant Institute, NAG Ltd., and Rice University
September 30, 1994
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.
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), (CUTPNT,CUTPNT)
and (CUTPNT+1, CUTPNT+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 CUTPNT 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.
CUTPNT (input) INTEGER
The location of the last eigenvalue in the leading sub-matrix.
min(1,N) <= CUTPNT <= N.
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.
Q2 (output) DOUBLE PRECISION array, dimension (LDQ2, N)
A copy of the first K eigenvectors which will be used by
DLAED3 in a matrix multiply (DGEMM) to solve for the new
eigenvectors. Q2 is arranged into three blocks. The
first block contains non-zero elements only at and above
CUTPNT, the second contains non-zero elements only below
CUTPNT, and the third is dense.
LDQ2 (input) INTEGER
The leading dimension of the array Q2. LDQ2 >= max(1,N).
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 CUTPNT, the second contains
non-zero elements only below CUTPNT, and the third is dense.
W (output) DOUBLE PRECISION array, dimension (N)
The first k values of the final deflation-altered z-vector
which will be passed to DLAED3.
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.
INDX (workspace) INTEGER array, dimension (N)
The permutation used to sort the contents of D into ascending
order.
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 : non-zero in the lower half only;
3 : dense;
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.
=====================================================================
Test the input parameters.
Parameter adjustments */
/* Table of constant values */
static doublereal c_b3 = -1.;
static integer c__1 = 1;
/* System generated locals */
integer q_dim1, q_offset, q2_dim1, q2_offset, i__1;
doublereal d__1, d__2, d__3, d__4;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
static integer jlam, imax, jmax;
extern /* Subroutine */ int drot_(integer *, doublereal *, integer *,
doublereal *, integer *, doublereal *, doublereal *);
static integer ctot[4];
static doublereal c;
static integer i, j;
static doublereal s, t;
extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
integer *), dcopy_(integer *, doublereal *, integer *, doublereal
*, integer *);
static integer k2, n1, n2;
extern doublereal dlapy2_(doublereal *, doublereal *);
static integer ct;
extern doublereal dlamch_(char *);
static integer jp;
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 *);
static integer n1p1;
static doublereal eps, tau, tol;
static integer psm[4];
--d;
q_dim1 = *ldq;
q_offset = q_dim1 + 1;
q -= q_offset;
--indxq;
--z;
--dlamda;
q2_dim1 = *ldq2;
q2_offset = q2_dim1 + 1;
q2 -= q2_offset;
--indxc;
--w;
--indxp;
--indx;
--coltyp;
/* Function Body */
*info = 0;
if (*n < 0) {
*info = -2;
} else if (*ldq < max(1,*n)) {
*info = -5;
} else if (min(1,*n) > *cutpnt || *n < *cutpnt) {
*info = -8;
} else if (*ldq2 < max(1,*n)) {
*info = -12;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("DLAED2", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
n1 = *cutpnt;
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.);
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
indx[j] = j;
/* L10: */
}
dscal_(n, &t, &z[1], &c__1);
/* RHO = ABS( norm(z)**2 * RHO ) */
*rho = (d__1 = *rho * 2., abs(d__1));
i__1 = *cutpnt;
for (i = 1; i <= i__1; ++i) {
coltyp[i] = 1;
/* L20: */
}
i__1 = *n;
for (i = *cutpnt + 1; i <= i__1; ++i) {
coltyp[i] = 2;
/* L30: */
}
/* Sort the eigenvalues into increasing order */
i__1 = *n;
for (i = *cutpnt + 1; i <= i__1; ++i) {
indxq[i] += *cutpnt;
/* L40: */
}
/* re-integrate the deflated parts from the last pass */
i__1 = *n;
for (i = 1; i <= i__1; ++i) {
dlamda[i] = d[indxq[i]];
w[i] = z[indxq[i]];
indxc[i] = coltyp[indxq[i]];
/* L50: */
}
dlamrg_(&n1, &n2, &dlamda[1], &c__1, &c__1, &indx[1]);
i__1 = *n;
for (i = 1; i <= i__1; ++i) {
d[i] = dlamda[indx[i]];
z[i] = w[indx[i]];
coltyp[i] = indxc[indx[i]];
/* L60: */
}
/* 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;
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
dcopy_(n, &q[indxq[indx[j]] * q_dim1 + 1], &c__1, &q2[j * q2_dim1
+ 1], &c__1);
/* L70: */
}
dlacpy_("A", n, n, &q2[q2_offset], ldq2, &q[q_offset], ldq);
goto L180;
}
/* 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. */
*k = 0;
k2 = *n + 1;
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
if (*rho * (d__1 = z[j], abs(d__1)) <= tol) {
/* Deflate due to small z component. */
--k2;
indxp[k2] = j;
coltyp[j] = 4;
if (j == *n) {
goto L120;
}
} else {
jlam = j;
goto L90;
}
/* L80: */
}
L90:
++j;
if (j > *n) {
goto L110;
}
if (*rho * (d__1 = z[j], abs(d__1)) <= tol) {
/* Deflate due to small z component. */
--k2;
indxp[k2] = j;
coltyp[j] = 4;
} else {
/* Check if eigenvalues are close enough to allow deflation. */
s = z[jlam];
c = z[j];
/* Find sqrt(a**2+b**2) without overflow or
destructive underflow. */
tau = dlapy2_(&c, &s);
t = d[j] - d[jlam];
c /= tau;
s = -s / tau;
if ((d__1 = t * c * s, abs(d__1)) <= tol) {
/* Deflation is possible. */
z[j] = tau;
z[jlam] = 0.;
if (coltyp[j] != coltyp[jlam]) {
coltyp[j] = 3;
}
coltyp[jlam] = 4;
drot_(n, &q[indxq[indx[jlam]] * q_dim1 + 1], &c__1, &q[indxq[indx[
j]] * q_dim1 + 1], &c__1, &c, &s);
/* Computing 2nd power */
d__1 = c;
/* Computing 2nd power */
d__2 = s;
t = d[jlam] * (d__1 * d__1) + d[j] * (d__2 * d__2);
/* Computing 2nd power */
d__1 = s;
/* Computing 2nd power */
d__2 = c;
d[j] = d[jlam] * (d__1 * d__1) + d[j] * (d__2 * d__2);
d[jlam] = t;
--k2;
i = 1;
L100:
if (k2 + i <= *n) {
if (d[jlam] < d[indxp[k2 + i]]) {
indxp[k2 + i - 1] = indxp[k2 + i];
indxp[k2 + i] = jlam;
++i;
goto L100;
} else {
indxp[k2 + i - 1] = jlam;
}
} else {
indxp[k2 + i - 1] = jlam;
}
jlam = j;
} else {
++(*k);
w[*k] = z[jlam];
dlamda[*k] = d[jlam];
indxp[*k] = jlam;
jlam = j;
}
}
goto L90;
L110:
/* Record the last eigenvalue. */
++(*k);
w[*k] = z[jlam];
dlamda[*k] = d[jlam];
indxp[*k] = jlam;
L120:
/* 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;
/* L130: */
}
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
ct = coltyp[j];
++ctot[ct - 1];
/* L140: */
}
/* PSM(*) = Position in SubVISMatrix (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];
/* 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) {
jp = indxp[j];
ct = coltyp[jp];
indxc[psm[ct - 1]] = j;
++psm[ct - 1];
/* L150: */
}
/* 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 = *n;
for (j = 1; j <= i__1; ++j) {
jp = indxp[j];
dlamda[j] = d[jp];
dcopy_(n, &q[indxq[indx[indxp[indxc[j]]]] * q_dim1 + 1], &c__1, &q2[j
* q2_dim1 + 1], &c__1);
/* L160: */
}
/* The deflated eigenvalues and their corresponding vectors go back
into the last N - K slots of D and Q respectively. */
i__1 = *n - *k;
dcopy_(&i__1, &dlamda[*k + 1], &c__1, &d[*k + 1], &c__1);
i__1 = *n - *k;
dlacpy_("A", n, &i__1, &q2[(*k + 1) * q2_dim1 + 1], ldq2, &q[(*k + 1) *
q_dim1 + 1], ldq);
/* Copy CTOT into COLTYP for referencing in DLAED3. */
for (j = 1; j <= 4; ++j) {
coltyp[j] = ctot[j - 1];
/* L170: */
}
L180:
return 0;
/* End of DLAED2 */
} /* dlaed2_ */
/* Subroutine */ int ztgsna_(char *job, char *howmny, logical *select,
integer *n, doublecomplex *a, integer *lda, doublecomplex *b, integer
*ldb, doublecomplex *vl, integer *ldvl, doublecomplex *vr, integer *
ldvr, doublereal *s, doublereal *dif, integer *mm, integer *m,
doublecomplex *work, integer *lwork, integer *iwork, integer *info)
{
/* -- LAPACK routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
June 30, 1999
Purpose
=======
ZTGSNA estimates reciprocal condition numbers for specified
eigenvalues and/or eigenvectors of a matrix pair (A, B).
(A, B) must be in generalized Schur canonical form, that is, A and
B are both upper triangular.
Arguments
=========
JOB (input) CHARACTER*1
Specifies whether condition numbers are required for
eigenvalues (S) or eigenvectors (DIF):
= 'E': for eigenvalues only (S);
= 'V': for eigenvectors only (DIF);
= 'B': for both eigenvalues and eigenvectors (S and DIF).
HOWMNY (input) CHARACTER*1
= 'A': compute condition numbers for all eigenpairs;
= 'S': compute condition numbers for selected eigenpairs
specified by the array SELECT.
SELECT (input) LOGICAL array, dimension (N)
If HOWMNY = 'S', SELECT specifies the eigenpairs for which
condition numbers are required. To select condition numbers
for the corresponding j-th eigenvalue and/or eigenvector,
SELECT(j) must be set to .TRUE..
If HOWMNY = 'A', SELECT is not referenced.
N (input) INTEGER
The order of the square matrix pair (A, B). N >= 0.
A (input) COMPLEX*16 array, dimension (LDA,N)
The upper triangular matrix A in the pair (A,B).
LDA (input) INTEGER
The leading dimension of the array A. LDA >= max(1,N).
B (input) COMPLEX*16 array, dimension (LDB,N)
The upper triangular matrix B in the pair (A, B).
LDB (input) INTEGER
The leading dimension of the array B. LDB >= max(1,N).
VL (input) COMPLEX*16 array, dimension (LDVL,M)
IF JOB = 'E' or 'B', VL must contain left eigenvectors of
(A, B), corresponding to the eigenpairs specified by HOWMNY
and SELECT. The eigenvectors must be stored in consecutive
columns of VL, as returned by ZTGEVC.
If JOB = 'V', VL is not referenced.
LDVL (input) INTEGER
The leading dimension of the array VL. LDVL >= 1; and
If JOB = 'E' or 'B', LDVL >= N.
VR (input) COMPLEX*16 array, dimension (LDVR,M)
IF JOB = 'E' or 'B', VR must contain right eigenvectors of
(A, B), corresponding to the eigenpairs specified by HOWMNY
and SELECT. The eigenvectors must be stored in consecutive
columns of VR, as returned by ZTGEVC.
If JOB = 'V', VR is not referenced.
LDVR (input) INTEGER
The leading dimension of the array VR. LDVR >= 1;
If JOB = 'E' or 'B', LDVR >= N.
S (output) DOUBLE PRECISION array, dimension (MM)
If JOB = 'E' or 'B', the reciprocal condition numbers of the
selected eigenvalues, stored in consecutive elements of the
array.
If JOB = 'V', S is not referenced.
DIF (output) DOUBLE PRECISION array, dimension (MM)
If JOB = 'V' or 'B', the estimated reciprocal condition
numbers of the selected eigenvectors, stored in consecutive
elements of the array.
If the eigenvalues cannot be reordered to compute DIF(j),
DIF(j) is set to 0; this can only occur when the true value
would be very small anyway.
For each eigenvalue/vector specified by SELECT, DIF stores
a Frobenius norm-based estimate of Difl.
If JOB = 'E', DIF is not referenced.
MM (input) INTEGER
The number of elements in the arrays S and DIF. MM >= M.
M (output) INTEGER
The number of elements of the arrays S and DIF used to store
the specified condition numbers; for each selected eigenvalue
one element is used. If HOWMNY = 'A', M is set to N.
WORK (workspace/output) COMPLEX*16 array, dimension (LWORK)
If JOB = 'E', WORK is not referenced. Otherwise,
on exit, if INFO = 0, WORK(1) returns the optimal LWORK.
LWORK (input) INTEGER
The dimension of the array WORK. LWORK >= 1.
If JOB = 'V' or 'B', LWORK >= 2*N*N.
IWORK (workspace) INTEGER array, dimension (N+2)
If JOB = 'E', IWORK is not referenced.
INFO (output) INTEGER
= 0: Successful exit
< 0: If INFO = -i, the i-th argument had an illegal value
Further Details
===============
The reciprocal of the condition number of the i-th generalized
eigenvalue w = (a, b) is defined as
S(I) = (|v'Au|**2 + |v'Bu|**2)**(1/2) / (norm(u)*norm(v))
where u and v are the right and left eigenvectors of (A, B)
corresponding to w; |z| denotes the absolute value of the complex
number, and norm(u) denotes the 2-norm of the vector u. The pair
(a, b) corresponds to an eigenvalue w = a/b (= v'Au/v'Bu) of the
matrix pair (A, B). If both a and b equal zero, then (A,B) is
singular and S(I) = -1 is returned.
An approximate error bound on the chordal distance between the i-th
computed generalized eigenvalue w and the corresponding exact
eigenvalue lambda is
chord(w, lambda) <= EPS * norm(A, B) / S(I),
where EPS is the machine precision.
The reciprocal of the condition number of the right eigenvector u
and left eigenvector v corresponding to the generalized eigenvalue w
is defined as follows. Suppose
(A, B) = ( a * ) ( b * ) 1
( 0 A22 ),( 0 B22 ) n-1
1 n-1 1 n-1
Then the reciprocal condition number DIF(I) is
Difl[(a, b), (A22, B22)] = sigma-min( Zl )
where sigma-min(Zl) denotes the smallest singular value of
Zl = [ kron(a, In-1) -kron(1, A22) ]
[ kron(b, In-1) -kron(1, B22) ].
Here In-1 is the identity matrix of size n-1 and X' is the conjugate
transpose of X. kron(X, Y) is the Kronecker product between the
matrices X and Y.
We approximate the smallest singular value of Zl with an upper
bound. This is done by ZLATDF.
An approximate error bound for a computed eigenvector VL(i) or
VR(i) is given by
EPS * norm(A, B) / DIF(i).
See ref. [2-3] for more details and further references.
Based on contributions by
Bo Kagstrom and Peter Poromaa, Department of Computing Science,
Umea University, S-901 87 Umea, Sweden.
References
==========
[1] B. Kagstrom; A Direct Method for Reordering Eigenvalues in the
Generalized Real Schur Form of a Regular Matrix Pair (A, B), in
M.S. Moonen et al (eds), Linear Algebra for Large Scale and
Real-Time Applications, Kluwer Academic Publ. 1993, pp 195-218.
[2] B. Kagstrom and P. Poromaa; Computing Eigenspaces with Specified
Eigenvalues of a Regular Matrix Pair (A, B) and Condition
Estimation: Theory, Algorithms and Software, Report
UMINF - 94.04, Department of Computing Science, Umea University,
S-901 87 Umea, Sweden, 1994. Also as LAPACK Working Note 87.
To appear in Numerical Algorithms, 1996.
[3] B. Kagstrom and P. Poromaa, LAPACK-Style Algorithms and Software
for Solving the Generalized Sylvester Equation and Estimating the
Separation between Regular Matrix Pairs, Report UMINF - 93.23,
Department of Computing Science, Umea University, S-901 87 Umea,
Sweden, December 1993, Revised April 1994, Also as LAPACK Working
Note 75.
To appear in ACM Trans. on Math. Software, Vol 22, No 1, 1996.
=====================================================================
Decode and test the input parameters
Parameter adjustments */
/* Table of constant values */
static integer c__1 = 1;
static doublecomplex c_b19 = {1.,0.};
static doublecomplex c_b20 = {0.,0.};
static logical c_false = FALSE_;
static integer c__3 = 3;
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, vl_dim1, vl_offset, vr_dim1,
vr_offset, i__1, i__2;
doublereal d__1, d__2;
doublecomplex z__1;
/* Builtin functions */
double z_abs(doublecomplex *);
/* Local variables */
static doublereal cond;
static integer ierr, ifst;
static doublereal lnrm;
static doublecomplex yhax, yhbx;
static integer ilst;
static doublereal rnrm;
static integer i__, k;
static doublereal scale;
extern logical lsame_(char *, char *);
extern /* Double Complex */ VOID zdotc_(doublecomplex *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *);
static integer lwmin;
extern /* Subroutine */ int zgemv_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublecomplex *, doublecomplex *, integer *);
static logical wants;
static integer llwrk, n1, n2;
static doublecomplex dummy[1];
extern doublereal dlapy2_(doublereal *, doublereal *);
extern /* Subroutine */ int dlabad_(doublereal *, doublereal *);
static doublecomplex dummy1[1];
extern doublereal dznrm2_(integer *, doublecomplex *, integer *), dlamch_(
char *);
static integer ks;
extern /* Subroutine */ int xerbla_(char *, integer *);
static doublereal bignum;
static logical wantbh, wantdf, somcon;
extern /* Subroutine */ int zlacpy_(char *, integer *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *),
ztgexc_(logical *, logical *, integer *, doublecomplex *, integer
*, doublecomplex *, integer *, doublecomplex *, integer *,
doublecomplex *, integer *, integer *, integer *, integer *);
static doublereal smlnum;
static logical lquery;
extern /* Subroutine */ int ztgsyl_(char *, integer *, integer *, integer
*, doublecomplex *, integer *, doublecomplex *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *,
doublereal *, doublereal *, doublecomplex *, integer *, integer *,
integer *);
static doublereal eps;
#define a_subscr(a_1,a_2) (a_2)*a_dim1 + a_1
#define a_ref(a_1,a_2) a[a_subscr(a_1,a_2)]
#define b_subscr(a_1,a_2) (a_2)*b_dim1 + a_1
#define b_ref(a_1,a_2) b[b_subscr(a_1,a_2)]
#define vl_subscr(a_1,a_2) (a_2)*vl_dim1 + a_1
#define vl_ref(a_1,a_2) vl[vl_subscr(a_1,a_2)]
#define vr_subscr(a_1,a_2) (a_2)*vr_dim1 + a_1
#define vr_ref(a_1,a_2) vr[vr_subscr(a_1,a_2)]
--select;
a_dim1 = *lda;
a_offset = 1 + a_dim1 * 1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1 * 1;
b -= b_offset;
vl_dim1 = *ldvl;
vl_offset = 1 + vl_dim1 * 1;
vl -= vl_offset;
vr_dim1 = *ldvr;
vr_offset = 1 + vr_dim1 * 1;
vr -= vr_offset;
--s;
--dif;
--work;
--iwork;
/* Function Body */
wantbh = lsame_(job, "B");
wants = lsame_(job, "E") || wantbh;
wantdf = lsame_(job, "V") || wantbh;
somcon = lsame_(howmny, "S");
*info = 0;
lquery = *lwork == -1;
if (lsame_(job, "V") || lsame_(job, "B")) {
/* Computing MAX */
i__1 = 1, i__2 = (*n << 1) * *n;
lwmin = max(i__1,i__2);
} else {
lwmin = 1;
}
if (! wants && ! wantdf) {
*info = -1;
} else if (! lsame_(howmny, "A") && ! somcon) {
*info = -2;
} else if (*n < 0) {
*info = -4;
} else if (*lda < max(1,*n)) {
*info = -6;
} else if (*ldb < max(1,*n)) {
*info = -8;
} else if (wants && *ldvl < *n) {
*info = -10;
} else if (wants && *ldvr < *n) {
*info = -12;
} else {
/* Set M to the number of eigenpairs for which condition numbers
are required, and test MM. */
if (somcon) {
*m = 0;
i__1 = *n;
for (k = 1; k <= i__1; ++k) {
if (select[k]) {
++(*m);
}
/* L10: */
}
} else {
*m = *n;
}
if (*mm < *m) {
*info = -15;
} else if (*lwork < lwmin && ! lquery) {
*info = -18;
}
}
if (*info == 0) {
work[1].r = (doublereal) lwmin, work[1].i = 0.;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("ZTGSNA", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
/* Get machine constants */
eps = dlamch_("P");
smlnum = dlamch_("S") / eps;
bignum = 1. / smlnum;
dlabad_(&smlnum, &bignum);
llwrk = *lwork - (*n << 1) * *n;
ks = 0;
i__1 = *n;
for (k = 1; k <= i__1; ++k) {
/* Determine whether condition numbers are required for the k-th
eigenpair. */
if (somcon) {
if (! select[k]) {
goto L20;
}
}
++ks;
if (wants) {
/* Compute the reciprocal condition number of the k-th
eigenvalue. */
rnrm = dznrm2_(n, &vr_ref(1, ks), &c__1);
lnrm = dznrm2_(n, &vl_ref(1, ks), &c__1);
zgemv_("N", n, n, &c_b19, &a[a_offset], lda, &vr_ref(1, ks), &
c__1, &c_b20, &work[1], &c__1);
zdotc_(&z__1, n, &work[1], &c__1, &vl_ref(1, ks), &c__1);
yhax.r = z__1.r, yhax.i = z__1.i;
zgemv_("N", n, n, &c_b19, &b[b_offset], ldb, &vr_ref(1, ks), &
c__1, &c_b20, &work[1], &c__1);
zdotc_(&z__1, n, &work[1], &c__1, &vl_ref(1, ks), &c__1);
yhbx.r = z__1.r, yhbx.i = z__1.i;
d__1 = z_abs(&yhax);
d__2 = z_abs(&yhbx);
cond = dlapy2_(&d__1, &d__2);
if (cond == 0.) {
s[ks] = -1.;
} else {
s[ks] = cond / (rnrm * lnrm);
}
}
if (wantdf) {
if (*n == 1) {
d__1 = z_abs(&a_ref(1, 1));
d__2 = z_abs(&b_ref(1, 1));
dif[ks] = dlapy2_(&d__1, &d__2);
goto L20;
}
/* Estimate the reciprocal condition number of the k-th
eigenvectors.
Copy the matrix (A, B) to the array WORK and move the
(k,k)th pair to the (1,1) position. */
zlacpy_("Full", n, n, &a[a_offset], lda, &work[1], n);
zlacpy_("Full", n, n, &b[b_offset], ldb, &work[*n * *n + 1], n);
ifst = k;
ilst = 1;
ztgexc_(&c_false, &c_false, n, &work[1], n, &work[*n * *n + 1], n,
dummy, &c__1, dummy1, &c__1, &ifst, &ilst, &ierr);
if (ierr > 0) {
/* Ill-conditioned problem - swap rejected. */
dif[ks] = 0.;
} else {
/* Reordering successful, solve generalized Sylvester
equation for R and L,
A22 * R - L * A11 = A12
B22 * R - L * B11 = B12,
and compute estimate of Difl[(A11,B11), (A22, B22)]. */
n1 = 1;
n2 = *n - n1;
i__ = *n * *n + 1;
ztgsyl_("N", &c__3, &n2, &n1, &work[*n * n1 + n1 + 1], n, &
work[1], n, &work[n1 + 1], n, &work[*n * n1 + n1 +
i__], n, &work[i__], n, &work[n1 + i__], n, &scale, &
dif[ks], &work[(*n * *n << 1) + 1], &llwrk, &iwork[1],
&ierr);
}
}
L20:
;
}
work[1].r = (doublereal) lwmin, work[1].i = 0.;
return 0;
/* End of ZTGSNA */
} /* ztgsna_ */
/* Subroutine */ int dlarfg_(integer *n, doublereal *alpha, doublereal *x,
integer *incx, doublereal *tau)
{
/* System generated locals */
integer i__1;
doublereal d__1;
/* Builtin functions */
double d_sign(doublereal *, doublereal *);
/* Local variables */
integer j, knt;
doublereal beta;
extern doublereal dnrm2_(integer *, doublereal *, integer *);
extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
integer *);
doublereal xnorm;
extern doublereal dlapy2_(doublereal *, doublereal *), dlamch_(char *);
doublereal safmin, rsafmn;
/* -- LAPACK auxiliary routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* DLARFG generates a real elementary reflector H of order n, such */
/* that */
/* H * ( alpha ) = ( beta ), H' * H = I. */
/* ( x ) ( 0 ) */
/* where alpha and beta are scalars, and x is an (n-1)-element real */
/* vector. H is represented in the form */
/* H = I - tau * ( 1 ) * ( 1 v' ) , */
/* ( v ) */
/* where tau is a real scalar and v is a real (n-1)-element */
/* vector. */
/* If the elements of x are all zero, then tau = 0 and H is taken to be */
/* the unit matrix. */
/* Otherwise 1 <= tau <= 2. */
/* Arguments */
/* ========= */
/* N (input) INTEGER */
/* The order of the elementary reflector. */
/* ALPHA (input/output) DOUBLE PRECISION */
/* On entry, the value alpha. */
/* On exit, it is overwritten with the value beta. */
/* X (input/output) DOUBLE PRECISION array, dimension */
/* (1+(N-2)*abs(INCX)) */
/* On entry, the vector x. */
/* On exit, it is overwritten with the vector v. */
/* INCX (input) INTEGER */
/* The increment between elements of X. INCX > 0. */
/* TAU (output) DOUBLE PRECISION */
/* The value tau. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Executable Statements .. */
/* Parameter adjustments */
--x;
/* Function Body */
if (*n <= 1) {
*tau = 0.;
return 0;
}
i__1 = *n - 1;
xnorm = dnrm2_(&i__1, &x[1], incx);
if (xnorm == 0.) {
/* H = I */
*tau = 0.;
} else {
/* general case */
d__1 = dlapy2_(alpha, &xnorm);
beta = -d_sign(&d__1, alpha);
safmin = dlamch_("S") / dlamch_("E");
if (abs(beta) < safmin) {
/* XNORM, BETA may be inaccurate; scale X and recompute them */
rsafmn = 1. / safmin;
knt = 0;
L10:
++knt;
i__1 = *n - 1;
dscal_(&i__1, &rsafmn, &x[1], incx);
beta *= rsafmn;
*alpha *= rsafmn;
if (abs(beta) < safmin) {
goto L10;
}
/* New BETA is at most 1, at least SAFMIN */
i__1 = *n - 1;
xnorm = dnrm2_(&i__1, &x[1], incx);
d__1 = dlapy2_(alpha, &xnorm);
beta = -d_sign(&d__1, alpha);
*tau = (beta - *alpha) / beta;
i__1 = *n - 1;
d__1 = 1. / (*alpha - beta);
dscal_(&i__1, &d__1, &x[1], incx);
/* If ALPHA is subnormal, it may lose relative accuracy */
*alpha = beta;
i__1 = knt;
for (j = 1; j <= i__1; ++j) {
*alpha *= safmin;
/* L20: */
}
} else {
*tau = (beta - *alpha) / beta;
i__1 = *n - 1;
d__1 = 1. / (*alpha - beta);
dscal_(&i__1, &d__1, &x[1], incx);
*alpha = beta;
}
}
return 0;
/* End of DLARFG */
} /* dlarfg_ */
/* Subroutine */ int pdnaup2_(integer *comm, integer *ido, char *bmat,
integer *n, char *which, integer *nev, integer *np, doublereal *tol,
doublereal *resid, integer *mode, integer *iupd, integer *ishift,
integer *mxiter, doublereal *v, integer *ldv, doublereal *h__,
integer *ldh, doublereal *ritzr, doublereal *ritzi, doublereal *
bounds, doublereal *q, integer *ldq, doublereal *workl, integer *
ipntr, doublereal *workd, integer *info, ftnlen bmat_len, ftnlen
which_len)
{
/* System generated locals */
integer h_dim1, h_offset, q_dim1, q_offset, v_dim1, v_offset, i__1, i__2;
doublereal d__1, d__2;
/* Builtin functions */
double pow_dd(doublereal *, doublereal *);
integer s_cmp(char *, char *, ftnlen, ftnlen);
/* Subroutine */ int s_copy(char *, char *, ftnlen, ftnlen);
double sqrt(doublereal);
/* Local variables */
static integer j;
static real t0, t1, t2, t3;
static doublereal rnorm_buf__;
static integer kp[4], np0, nev0;
extern doublereal ddot_(integer *, doublereal *, integer *, doublereal *,
integer *);
static doublereal eps23;
static integer ierr, iter;
static doublereal temp;
static logical getv0, cnorm;
extern /* Subroutine */ int dcopy_(integer *, doublereal *, integer *,
doublereal *, integer *);
static integer nconv;
static logical initv;
static doublereal rnorm;
extern /* Subroutine */ int dvout_(integer *, integer *, doublereal *,
integer *, char *, ftnlen), mpi_allreduce__(doublereal *,
doublereal *, integer *, integer *, integer *, integer *, integer
*);
extern doublereal dlapy2_(doublereal *, doublereal *);
static integer nevbef;
static char wprime[2];
static logical update, ushift;
static integer kplusp, msglvl, nptemp, numcnv;
extern /* Subroutine */ int dnconv_(integer *, doublereal *, doublereal *,
doublereal *, doublereal *, integer *), pdvout_(integer *,
integer *, integer *, doublereal *, integer *, char *, ftnlen),
pivout_(integer *, integer *, integer *, integer *, integer *,
char *, ftnlen), second_(real *), dsortc_(char *, logical *,
integer *, doublereal *, doublereal *, doublereal *, ftnlen),
pdmout_(integer *, integer *, integer *, integer *, doublereal *,
integer *, integer *, char *, ftnlen), pdgetv0_(integer *,
integer *, char *, integer *, logical *, integer *, integer *,
doublereal *, integer *, doublereal *, doublereal *, integer *,
doublereal *, doublereal *, integer *, ftnlen);
extern doublereal pdnorm2_(integer *, integer *, doublereal *, integer *),
pdlamch_(integer *, char *, ftnlen);
extern /* Subroutine */ int pdneigh_(integer *, doublereal *, integer *,
doublereal *, integer *, doublereal *, doublereal *, doublereal *,
doublereal *, integer *, doublereal *, integer *), pdnaitr_(
integer *, integer *, char *, integer *, integer *, integer *,
integer *, doublereal *, doublereal *, doublereal *, integer *,
doublereal *, integer *, integer *, doublereal *, doublereal *,
integer *, ftnlen), pdngets_(integer *, integer *, char *,
integer *, integer *, doublereal *, doublereal *, doublereal *,
doublereal *, doublereal *, ftnlen), pdnapps_(integer *, integer *
, integer *, integer *, doublereal *, doublereal *, doublereal *,
integer *, doublereal *, integer *, doublereal *, doublereal *,
integer *, doublereal *, doublereal *);
/* %---------------% */
/* | MPI Variables | */
/* %---------------% */
/* /+ */
/* * */
/* * (C) 1993 by Argonne National Laboratory and Mississipi State University. */
/* * All rights reserved. See COPYRIGHT in top-level directory. */
/* +/ */
/* /+ user include file for MPI programs, with no dependencies +/ */
/* /+ return codes +/ */
/* We handle datatypes by putting the variables that hold them into */
/* common. This way, a Fortran program can directly use the various */
/* datatypes and can even give them to C programs. */
/* MPI_BOTTOM needs to be a known address; here we put it at the */
/* beginning of the common block. The point-to-point and collective */
/* routines know about MPI_BOTTOM, but MPI_TYPE_STRUCT as yet does not. */
/* The types MPI_INTEGER1,2,4 and MPI_REAL4,8 are OPTIONAL. */
/* Their values are zero if they are not available. Note that */
/* using these reduces the portability of code (though may enhance */
/* portability between Crays and other systems) */
/* All other MPI routines are subroutines */
/* The attribute copy/delete functions are symbols that can be passed */
/* to MPI routines */
/* %----------------------------------------------------% */
/* | Include files for debugging and timing information | */
/* %----------------------------------------------------% */
/* \SCCS Information: @(#) */
/* FILE: debug.h SID: 2.3 DATE OF SID: 11/16/95 RELEASE: 2 */
/* %---------------------------------% */
/* | See debug.doc for documentation | */
/* %---------------------------------% */
/* %------------------% */
/* | Scalar Arguments | */
/* %------------------% */
/* %--------------------------------% */
/* | See stat.doc for documentation | */
/* %--------------------------------% */
/* \SCCS Information: @(#) */
/* FILE: stat.h SID: 2.2 DATE OF SID: 11/16/95 RELEASE: 2 */
/* %-----------------% */
/* | Array Arguments | */
/* %-----------------% */
/* %------------% */
/* | Parameters | */
/* %------------% */
/* %---------------% */
/* | Local Scalars | */
/* %---------------% */
/* %-----------------------% */
/* | Local array arguments | */
/* %-----------------------% */
/* %----------------------% */
/* | External Subroutines | */
/* %----------------------% */
/* %--------------------% */
/* | External Functions | */
/* %--------------------% */
/* %---------------------% */
/* | Intrinsic Functions | */
/* %---------------------% */
/* %-----------------------% */
/* | Executable Statements | */
/* %-----------------------% */
/* Parameter adjustments */
--workd;
--resid;
--workl;
--bounds;
--ritzi;
--ritzr;
v_dim1 = *ldv;
v_offset = 1 + v_dim1;
v -= v_offset;
h_dim1 = *ldh;
h_offset = 1 + h_dim1;
h__ -= h_offset;
q_dim1 = *ldq;
q_offset = 1 + q_dim1;
q -= q_offset;
--ipntr;
/* Function Body */
if (*ido == 0) {
second_(&t0);
msglvl = debug_1.mnaup2;
/* %-------------------------------------% */
/* | Get the machine dependent constant. | */
/* %-------------------------------------% */
eps23 = pdlamch_(comm, "Epsilon-Machine", (ftnlen)15);
eps23 = pow_dd(&eps23, &c_b3);
nev0 = *nev;
np0 = *np;
/* %-------------------------------------% */
/* | kplusp is the bound on the largest | */
/* | Lanczos factorization built. | */
/* | nconv is the current number of | */
/* | "converged" eigenvlues. | */
/* | iter is the counter on the current | */
/* | iteration step. | */
/* %-------------------------------------% */
kplusp = *nev + *np;
nconv = 0;
iter = 0;
/* %---------------------------------------% */
/* | Set flags for computing the first NEV | */
/* | steps of the Arnoldi factorization. | */
/* %---------------------------------------% */
getv0 = TRUE_;
update = FALSE_;
ushift = FALSE_;
cnorm = FALSE_;
if (*info != 0) {
/* %--------------------------------------------% */
/* | User provides the initial residual vector. | */
/* %--------------------------------------------% */
initv = TRUE_;
*info = 0;
} else {
initv = FALSE_;
}
}
/* %---------------------------------------------% */
/* | Get a possibly random starting vector and | */
/* | force it into the range of the operator OP. | */
/* %---------------------------------------------% */
/* L10: */
if (getv0) {
pdgetv0_(comm, ido, bmat, &c__1, &initv, n, &c__1, &v[v_offset], ldv,
&resid[1], &rnorm, &ipntr[1], &workd[1], &workl[1], info, (
ftnlen)1);
if (*ido != 99) {
goto L9000;
}
if (rnorm == 0.) {
/* %-----------------------------------------% */
/* | The initial vector is zero. Error exit. | */
/* %-----------------------------------------% */
*info = -9;
goto L1100;
}
getv0 = FALSE_;
*ido = 0;
}
/* %-----------------------------------% */
/* | Back from reverse communication : | */
/* | continue with update step | */
/* %-----------------------------------% */
if (update) {
goto L20;
}
/* %-------------------------------------------% */
/* | Back from computing user specified shifts | */
/* %-------------------------------------------% */
if (ushift) {
goto L50;
}
/* %-------------------------------------% */
/* | Back from computing residual norm | */
/* | at the end of the current iteration | */
/* %-------------------------------------% */
if (cnorm) {
goto L100;
}
/* %----------------------------------------------------------% */
/* | Compute the first NEV steps of the Arnoldi factorization | */
/* %----------------------------------------------------------% */
pdnaitr_(comm, ido, bmat, n, &c__0, nev, mode, &resid[1], &rnorm, &v[
v_offset], ldv, &h__[h_offset], ldh, &ipntr[1], &workd[1], &workl[
1], info, (ftnlen)1);
/* %---------------------------------------------------% */
/* | ido .ne. 99 implies use of reverse communication | */
/* | to compute operations involving OP and possibly B | */
/* %---------------------------------------------------% */
if (*ido != 99) {
goto L9000;
}
if (*info > 0) {
*np = *info;
*mxiter = iter;
*info = -9999;
goto L1200;
}
/* %--------------------------------------------------------------% */
/* | | */
/* | M A I N ARNOLDI I T E R A T I O N L O O P | */
/* | Each iteration implicitly restarts the Arnoldi | */
/* | factorization in place. | */
/* | | */
/* %--------------------------------------------------------------% */
L1000:
++iter;
if (msglvl > 0) {
pivout_(comm, &debug_1.logfil, &c__1, &iter, &debug_1.ndigit, "_naup"
"2: **** Start of major iteration number ****", (ftnlen)49);
}
/* %-----------------------------------------------------------% */
/* | Compute NP additional steps of the Arnoldi factorization. | */
/* | Adjust NP since NEV might have been updated by last call | */
/* | to the shift application routine pdnapps . | */
/* %-----------------------------------------------------------% */
*np = kplusp - *nev;
if (msglvl > 1) {
pivout_(comm, &debug_1.logfil, &c__1, nev, &debug_1.ndigit, "_naup2:"
" The length of the current Arnoldi factorization", (ftnlen)55)
;
pivout_(comm, &debug_1.logfil, &c__1, np, &debug_1.ndigit, "_naup2: "
"Extend the Arnoldi factorization by", (ftnlen)43);
}
/* %-----------------------------------------------------------% */
/* | Compute NP additional steps of the Arnoldi factorization. | */
/* %-----------------------------------------------------------% */
*ido = 0;
L20:
update = TRUE_;
pdnaitr_(comm, ido, bmat, n, nev, np, mode, &resid[1], &rnorm, &v[
v_offset], ldv, &h__[h_offset], ldh, &ipntr[1], &workd[1], &workl[
1], info, (ftnlen)1);
/* %---------------------------------------------------% */
/* | ido .ne. 99 implies use of reverse communication | */
/* | to compute operations involving OP and possibly B | */
/* %---------------------------------------------------% */
if (*ido != 99) {
goto L9000;
}
if (*info > 0) {
*np = *info;
*mxiter = iter;
*info = -9999;
goto L1200;
}
update = FALSE_;
if (msglvl > 1) {
pdvout_(comm, &debug_1.logfil, &c__1, &rnorm, &debug_1.ndigit, "_nau"
"p2: Corresponding B-norm of the residual", (ftnlen)44);
}
/* %--------------------------------------------------------% */
/* | Compute the eigenvalues and corresponding error bounds | */
/* | of the current upper Hessenberg matrix. | */
/* %--------------------------------------------------------% */
pdneigh_(comm, &rnorm, &kplusp, &h__[h_offset], ldh, &ritzr[1], &ritzi[1],
&bounds[1], &q[q_offset], ldq, &workl[1], &ierr);
if (ierr != 0) {
*info = -8;
goto L1200;
}
/* %----------------------------------------------------% */
/* | Make a copy of eigenvalues and corresponding error | */
/* | bounds obtained from pdneigh . | */
/* %----------------------------------------------------% */
/* Computing 2nd power */
i__1 = kplusp;
dcopy_(&kplusp, &ritzr[1], &c__1, &workl[i__1 * i__1 + 1], &c__1);
/* Computing 2nd power */
i__1 = kplusp;
dcopy_(&kplusp, &ritzi[1], &c__1, &workl[i__1 * i__1 + kplusp + 1], &c__1)
;
/* Computing 2nd power */
i__1 = kplusp;
dcopy_(&kplusp, &bounds[1], &c__1, &workl[i__1 * i__1 + (kplusp << 1) + 1]
, &c__1);
/* %---------------------------------------------------% */
/* | Select the wanted Ritz values and their bounds | */
/* | to be used in the convergence test. | */
/* | The wanted part of the spectrum and corresponding | */
/* | error bounds are in the last NEV loc. of RITZR, | */
/* | RITZI and BOUNDS respectively. The variables NEV | */
/* | and NP may be updated if the NEV-th wanted Ritz | */
/* | value has a non zero imaginary part. In this case | */
/* | NEV is increased by one and NP decreased by one. | */
/* | NOTE: The last two arguments of pdngets are no | */
/* | longer used as of version 2.1. | */
/* %---------------------------------------------------% */
*nev = nev0;
*np = np0;
numcnv = *nev;
pdngets_(comm, ishift, which, nev, np, &ritzr[1], &ritzi[1], &bounds[1], &
workl[1], &workl[*np + 1], (ftnlen)2);
if (*nev == nev0 + 1) {
numcnv = nev0 + 1;
}
/* %-------------------% */
/* | Convergence test. | */
/* %-------------------% */
dcopy_(nev, &bounds[*np + 1], &c__1, &workl[(*np << 1) + 1], &c__1);
dnconv_(nev, &ritzr[*np + 1], &ritzi[*np + 1], &workl[(*np << 1) + 1],
tol, &nconv);
if (msglvl > 2) {
kp[0] = *nev;
kp[1] = *np;
kp[2] = numcnv;
kp[3] = nconv;
pivout_(comm, &debug_1.logfil, &c__4, kp, &debug_1.ndigit, "_naup2: "
"NEV, NP, NUMCNV, NCONV are", (ftnlen)34);
pdvout_(comm, &debug_1.logfil, &kplusp, &ritzr[1], &debug_1.ndigit,
"_naup2: Real part of the eigenvalues of H", (ftnlen)41);
pdvout_(comm, &debug_1.logfil, &kplusp, &ritzi[1], &debug_1.ndigit,
"_naup2: Imaginary part of the eigenvalues of H", (ftnlen)46);
pdvout_(comm, &debug_1.logfil, &kplusp, &bounds[1], &debug_1.ndigit,
"_naup2: Ritz estimates of the current NCV Ritz values", (
ftnlen)53);
}
/* %---------------------------------------------------------% */
/* | Count the number of unwanted Ritz values that have zero | */
/* | Ritz estimates. If any Ritz estimates are equal to zero | */
/* | then a leading block of H of order equal to at least | */
/* | the number of Ritz values with zero Ritz estimates has | */
/* | split off. None of these Ritz values may be removed by | */
/* | shifting. Decrease NP the number of shifts to apply. If | */
/* | no shifts may be applied, then prepare to exit | */
/* %---------------------------------------------------------% */
nptemp = *np;
i__1 = nptemp;
for (j = 1; j <= i__1; ++j) {
if (bounds[j] == 0.) {
--(*np);
++(*nev);
}
/* L30: */
}
if (nconv >= numcnv || iter > *mxiter || *np == 0) {
if (msglvl > 4) {
/* Computing 2nd power */
i__1 = kplusp;
dvout_(&debug_1.logfil, &kplusp, &workl[i__1 * i__1 + 1], &
debug_1.ndigit, "_naup2: Real part of the eig computed b"
"y _neigh:", (ftnlen)48);
/* Computing 2nd power */
i__1 = kplusp;
dvout_(&debug_1.logfil, &kplusp, &workl[i__1 * i__1 + kplusp + 1],
&debug_1.ndigit, "_naup2: Imag part of the eig computed"
" by _neigh:", (ftnlen)48);
/* Computing 2nd power */
i__1 = kplusp;
dvout_(&debug_1.logfil, &kplusp, &workl[i__1 * i__1 + (kplusp <<
1) + 1], &debug_1.ndigit, "_naup2: Ritz estimates comput"
"ed by _neigh:", (ftnlen)42);
}
/* %------------------------------------------------% */
/* | Prepare to exit. Put the converged Ritz values | */
/* | and corresponding bounds in RITZ(1:NCONV) and | */
/* | BOUNDS(1:NCONV) respectively. Then sort. Be | */
/* | careful when NCONV > NP | */
/* %------------------------------------------------% */
/* %------------------------------------------% */
/* | Use h( 3,1 ) as storage to communicate | */
/* | rnorm to _neupd if needed | */
/* %------------------------------------------% */
h__[h_dim1 + 3] = rnorm;
/* %----------------------------------------------% */
/* | To be consistent with dngets , we first do a | */
/* | pre-processing sort in order to keep complex | */
/* | conjugate pairs together. This is similar | */
/* | to the pre-processing sort used in dngets | */
/* | except that the sort is done in the opposite | */
/* | order. | */
/* %----------------------------------------------% */
if (s_cmp(which, "LM", (ftnlen)2, (ftnlen)2) == 0) {
s_copy(wprime, "SR", (ftnlen)2, (ftnlen)2);
}
if (s_cmp(which, "SM", (ftnlen)2, (ftnlen)2) == 0) {
s_copy(wprime, "LR", (ftnlen)2, (ftnlen)2);
}
if (s_cmp(which, "LR", (ftnlen)2, (ftnlen)2) == 0) {
s_copy(wprime, "SM", (ftnlen)2, (ftnlen)2);
}
if (s_cmp(which, "SR", (ftnlen)2, (ftnlen)2) == 0) {
s_copy(wprime, "LM", (ftnlen)2, (ftnlen)2);
}
if (s_cmp(which, "LI", (ftnlen)2, (ftnlen)2) == 0) {
s_copy(wprime, "SM", (ftnlen)2, (ftnlen)2);
}
if (s_cmp(which, "SI", (ftnlen)2, (ftnlen)2) == 0) {
s_copy(wprime, "LM", (ftnlen)2, (ftnlen)2);
}
dsortc_(wprime, &c_true, &kplusp, &ritzr[1], &ritzi[1], &bounds[1], (
ftnlen)2);
/* %----------------------------------------------% */
/* | Now sort Ritz values so that converged Ritz | */
/* | values appear within the first NEV locations | */
/* | of ritzr, ritzi and bounds, and the most | */
/* | desired one appears at the front. | */
/* %----------------------------------------------% */
if (s_cmp(which, "LM", (ftnlen)2, (ftnlen)2) == 0) {
s_copy(wprime, "SM", (ftnlen)2, (ftnlen)2);
}
if (s_cmp(which, "SM", (ftnlen)2, (ftnlen)2) == 0) {
s_copy(wprime, "LM", (ftnlen)2, (ftnlen)2);
}
if (s_cmp(which, "LR", (ftnlen)2, (ftnlen)2) == 0) {
s_copy(wprime, "SR", (ftnlen)2, (ftnlen)2);
}
if (s_cmp(which, "SR", (ftnlen)2, (ftnlen)2) == 0) {
s_copy(wprime, "LR", (ftnlen)2, (ftnlen)2);
}
if (s_cmp(which, "LI", (ftnlen)2, (ftnlen)2) == 0) {
s_copy(wprime, "SI", (ftnlen)2, (ftnlen)2);
}
if (s_cmp(which, "SI", (ftnlen)2, (ftnlen)2) == 0) {
s_copy(wprime, "LI", (ftnlen)2, (ftnlen)2);
}
dsortc_(wprime, &c_true, &kplusp, &ritzr[1], &ritzi[1], &bounds[1], (
ftnlen)2);
/* %--------------------------------------------------% */
/* | Scale the Ritz estimate of each Ritz value | */
/* | by 1 / max(eps23,magnitude of the Ritz value). | */
/* %--------------------------------------------------% */
i__1 = numcnv;
for (j = 1; j <= i__1; ++j) {
/* Computing MAX */
d__1 = eps23, d__2 = dlapy2_(&ritzr[j], &ritzi[j]);
temp = max(d__1,d__2);
bounds[j] /= temp;
/* L35: */
}
/* %----------------------------------------------------% */
/* | Sort the Ritz values according to the scaled Ritz | */
/* | esitmates. This will push all the converged ones | */
/* | towards the front of ritzr, ritzi, bounds | */
/* | (in the case when NCONV < NEV.) | */
/* %----------------------------------------------------% */
s_copy(wprime, "LR", (ftnlen)2, (ftnlen)2);
dsortc_(wprime, &c_true, &numcnv, &bounds[1], &ritzr[1], &ritzi[1], (
ftnlen)2);
/* %----------------------------------------------% */
/* | Scale the Ritz estimate back to its original | */
/* | value. | */
/* %----------------------------------------------% */
i__1 = numcnv;
for (j = 1; j <= i__1; ++j) {
/* Computing MAX */
d__1 = eps23, d__2 = dlapy2_(&ritzr[j], &ritzi[j]);
temp = max(d__1,d__2);
bounds[j] *= temp;
/* L40: */
}
/* %------------------------------------------------% */
/* | Sort the converged Ritz values again so that | */
/* | the "threshold" value appears at the front of | */
/* | ritzr, ritzi and bound. | */
/* %------------------------------------------------% */
dsortc_(which, &c_true, &nconv, &ritzr[1], &ritzi[1], &bounds[1], (
ftnlen)2);
if (msglvl > 1) {
dvout_(&debug_1.logfil, &kplusp, &ritzr[1], &debug_1.ndigit,
"_naup2: Sorted real part of the eigenvalues", (ftnlen)43)
;
dvout_(&debug_1.logfil, &kplusp, &ritzi[1], &debug_1.ndigit,
"_naup2: Sorted imaginary part of the eigenvalues", (
ftnlen)48);
dvout_(&debug_1.logfil, &kplusp, &bounds[1], &debug_1.ndigit,
"_naup2: Sorted ritz estimates.", (ftnlen)30);
}
/* %------------------------------------% */
/* | Max iterations have been exceeded. | */
/* %------------------------------------% */
if (iter > *mxiter && nconv < numcnv) {
*info = 1;
}
/* %---------------------% */
/* | No shifts to apply. | */
/* %---------------------% */
if (*np == 0 && nconv < numcnv) {
*info = 2;
}
*np = nconv;
goto L1100;
} else if (nconv < numcnv && *ishift == 1) {
/* %-------------------------------------------------% */
/* | Do not have all the requested eigenvalues yet. | */
/* | To prevent possible stagnation, adjust the size | */
/* | of NEV. | */
/* %-------------------------------------------------% */
nevbef = *nev;
/* Computing MIN */
i__1 = nconv, i__2 = *np / 2;
*nev += min(i__1,i__2);
if (*nev == 1 && kplusp >= 6) {
*nev = kplusp / 2;
} else if (*nev == 1 && kplusp > 3) {
*nev = 2;
}
*np = kplusp - *nev;
/* %---------------------------------------% */
/* | If the size of NEV was just increased | */
/* | resort the eigenvalues. | */
/* %---------------------------------------% */
if (nevbef < *nev) {
pdngets_(comm, ishift, which, nev, np, &ritzr[1], &ritzi[1], &
bounds[1], &workl[1], &workl[*np + 1], (ftnlen)2);
}
}
if (msglvl > 0) {
pivout_(comm, &debug_1.logfil, &c__1, &nconv, &debug_1.ndigit, "_nau"
"p2: no. of \"converged\" Ritz values at this iter.", (ftnlen)
52);
if (msglvl > 1) {
kp[0] = *nev;
kp[1] = *np;
pivout_(comm, &debug_1.logfil, &c__2, kp, &debug_1.ndigit, "_nau"
"p2: NEV and NP are", (ftnlen)22);
pdvout_(comm, &debug_1.logfil, nev, &ritzr[*np + 1], &
debug_1.ndigit, "_naup2: \"wanted\" Ritz values -- real "
"part", (ftnlen)41);
pdvout_(comm, &debug_1.logfil, nev, &ritzi[*np + 1], &
debug_1.ndigit, "_naup2: \"wanted\" Ritz values -- imag "
"part", (ftnlen)41);
pdvout_(comm, &debug_1.logfil, nev, &bounds[*np + 1], &
debug_1.ndigit, "_naup2: Ritz estimates of the \"wante"
"d\" values ", (ftnlen)46);
}
}
if (*ishift == 0) {
/* %-------------------------------------------------------% */
/* | User specified shifts: reverse comminucation to | */
/* | compute the shifts. They are returned in the first | */
/* | 2*NP locations of WORKL. | */
/* %-------------------------------------------------------% */
ushift = TRUE_;
*ido = 3;
goto L9000;
}
L50:
/* %------------------------------------% */
/* | Back from reverse communication; | */
/* | User specified shifts are returned | */
/* | in WORKL(1:2*NP) | */
/* %------------------------------------% */
ushift = FALSE_;
if (*ishift == 0) {
/* %----------------------------------% */
/* | Move the NP shifts from WORKL to | */
/* | RITZR, RITZI to free up WORKL | */
/* | for non-exact shift case. | */
/* %----------------------------------% */
dcopy_(np, &workl[1], &c__1, &ritzr[1], &c__1);
dcopy_(np, &workl[*np + 1], &c__1, &ritzi[1], &c__1);
}
if (msglvl > 2) {
pivout_(comm, &debug_1.logfil, &c__1, np, &debug_1.ndigit, "_naup2: "
"The number of shifts to apply ", (ftnlen)38);
pdvout_(comm, &debug_1.logfil, np, &ritzr[1], &debug_1.ndigit, "_nau"
"p2: Real part of the shifts", (ftnlen)31);
pdvout_(comm, &debug_1.logfil, np, &ritzi[1], &debug_1.ndigit, "_nau"
"p2: Imaginary part of the shifts", (ftnlen)36);
if (*ishift == 1) {
pdvout_(comm, &debug_1.logfil, np, &bounds[1], &debug_1.ndigit,
"_naup2: Ritz estimates of the shifts", (ftnlen)36);
}
}
/* %---------------------------------------------------------% */
/* | Apply the NP implicit shifts by QR bulge chasing. | */
/* | Each shift is applied to the whole upper Hessenberg | */
/* | matrix H. | */
/* | The first 2*N locations of WORKD are used as workspace. | */
/* %---------------------------------------------------------% */
pdnapps_(comm, n, nev, np, &ritzr[1], &ritzi[1], &v[v_offset], ldv, &h__[
h_offset], ldh, &resid[1], &q[q_offset], ldq, &workl[1], &workd[1]
);
/* %---------------------------------------------% */
/* | Compute the B-norm of the updated residual. | */
/* | Keep B*RESID in WORKD(1:N) to be used in | */
/* | the first step of the next call to pdnaitr . | */
/* %---------------------------------------------% */
cnorm = TRUE_;
second_(&t2);
if (*(unsigned char *)bmat == 'G') {
++timing_1.nbx;
dcopy_(n, &resid[1], &c__1, &workd[*n + 1], &c__1);
ipntr[1] = *n + 1;
ipntr[2] = 1;
*ido = 2;
/* %----------------------------------% */
/* | Exit in order to compute B*RESID | */
/* %----------------------------------% */
goto L9000;
} else if (*(unsigned char *)bmat == 'I') {
dcopy_(n, &resid[1], &c__1, &workd[1], &c__1);
}
L100:
/* %----------------------------------% */
/* | Back from reverse communication; | */
/* | WORKD(1:N) := B*RESID | */
/* %----------------------------------% */
if (*(unsigned char *)bmat == 'G') {
second_(&t3);
timing_1.tmvbx += t3 - t2;
}
if (*(unsigned char *)bmat == 'G') {
rnorm_buf__ = ddot_(n, &resid[1], &c__1, &workd[1], &c__1);
mpi_allreduce__(&rnorm_buf__, &rnorm, &c__1, &
mpipriv_1.mpi_double_precision__, &mpipriv_1.mpi_sum__, comm,
&ierr);
rnorm = sqrt((abs(rnorm)));
} else if (*(unsigned char *)bmat == 'I') {
rnorm = pdnorm2_(comm, n, &resid[1], &c__1);
}
cnorm = FALSE_;
if (msglvl > 2) {
pdvout_(comm, &debug_1.logfil, &c__1, &rnorm, &debug_1.ndigit, "_nau"
"p2: B-norm of residual for compressed factorization", (ftnlen)
55);
pdmout_(comm, &debug_1.logfil, nev, nev, &h__[h_offset], ldh, &
debug_1.ndigit, "_naup2: Compressed upper Hessenberg matrix H"
, (ftnlen)44);
}
goto L1000;
/* %---------------------------------------------------------------% */
/* | | */
/* | E N D O F M A I N I T E R A T I O N L O O P | */
/* | | */
/* %---------------------------------------------------------------% */
L1100:
*mxiter = iter;
*nev = numcnv;
L1200:
*ido = 99;
/* %------------% */
/* | Error Exit | */
/* %------------% */
second_(&t1);
timing_1.tnaup2 = t1 - t0;
L9000:
/* %----------------% */
/* | End of pdnaup2 | */
/* %----------------% */
return 0;
} /* pdnaup2_ */
/* Subroutine */
int dsteqr_(char *compz, integer *n, doublereal *d__, doublereal *e, doublereal *z__, integer *ldz, doublereal *work, integer *info)
{
/* System generated locals */
integer z_dim1, z_offset, i__1, i__2;
doublereal d__1, d__2;
/* Builtin functions */
double sqrt(doublereal), d_sign(doublereal *, doublereal *);
/* Local variables */
doublereal b, c__, f, g;
integer i__, j, k, l, m;
doublereal p, r__, s;
integer l1, ii, mm, lm1, mm1, nm1;
doublereal rt1, rt2, eps;
integer lsv;
doublereal tst, eps2;
integer lend, jtot;
extern /* Subroutine */
int dlae2_(doublereal *, doublereal *, doublereal *, doublereal *, doublereal *);
extern logical lsame_(char *, char *);
extern /* Subroutine */
int dlasr_(char *, char *, char *, integer *, integer *, doublereal *, doublereal *, doublereal *, integer *);
doublereal anorm;
extern /* Subroutine */
int dswap_(integer *, doublereal *, integer *, doublereal *, integer *), dlaev2_(doublereal *, doublereal *, doublereal *, doublereal *, doublereal *, doublereal *, doublereal *);
integer lendm1, lendp1;
extern doublereal dlapy2_(doublereal *, doublereal *), dlamch_(char *);
integer iscale;
extern /* Subroutine */
int dlascl_(char *, integer *, integer *, doublereal *, doublereal *, integer *, integer *, doublereal *, integer *, integer *), dlaset_(char *, integer *, integer *, doublereal *, doublereal *, doublereal *, integer *);
doublereal safmin;
extern /* Subroutine */
int dlartg_(doublereal *, doublereal *, doublereal *, doublereal *, doublereal *);
doublereal safmax;
extern /* Subroutine */
int xerbla_(char *, integer *);
extern doublereal dlanst_(char *, integer *, doublereal *, doublereal *);
extern /* Subroutine */
int dlasrt_(char *, integer *, doublereal *, integer *);
integer lendsv;
doublereal ssfmin;
integer nmaxit, icompz;
doublereal ssfmax;
/* -- LAPACK computational routine (version 3.4.0) -- */
/* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
/* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
/* November 2011 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
--d__;
--e;
z_dim1 = *ldz;
z_offset = 1 + z_dim1;
z__ -= z_offset;
--work;
/* Function Body */
*info = 0;
if (lsame_(compz, "N"))
{
icompz = 0;
}
else if (lsame_(compz, "V"))
{
icompz = 1;
}
else if (lsame_(compz, "I"))
{
icompz = 2;
}
else
{
icompz = -1;
}
if (icompz < 0)
{
*info = -1;
}
else if (*n < 0)
{
*info = -2;
}
else if (*ldz < 1 || icompz > 0 && *ldz < max(1,*n))
{
*info = -6;
}
if (*info != 0)
{
i__1 = -(*info);
xerbla_("DSTEQR", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0)
{
return 0;
}
if (*n == 1)
{
if (icompz == 2)
{
z__[z_dim1 + 1] = 1.;
}
return 0;
}
/* Determine the unit roundoff and over/underflow thresholds. */
eps = dlamch_("E");
/* Computing 2nd power */
d__1 = eps;
eps2 = d__1 * d__1;
safmin = dlamch_("S");
safmax = 1. / safmin;
ssfmax = sqrt(safmax) / 3.;
ssfmin = sqrt(safmin) / eps2;
/* Compute the eigenvalues and eigenvectors of the tridiagonal */
/* matrix. */
if (icompz == 2)
{
dlaset_("Full", n, n, &c_b9, &c_b10, &z__[z_offset], ldz);
}
nmaxit = *n * 30;
jtot = 0;
/* Determine where the matrix splits and choose QL or QR iteration */
/* for each block, according to whether top or bottom diagonal */
/* element is smaller. */
l1 = 1;
nm1 = *n - 1;
L10:
if (l1 > *n)
{
goto L160;
}
if (l1 > 1)
{
e[l1 - 1] = 0.;
}
if (l1 <= nm1)
{
i__1 = nm1;
for (m = l1;
m <= i__1;
++m)
{
tst = (d__1 = e[m], abs(d__1));
if (tst == 0.)
{
goto L30;
}
if (tst <= sqrt((d__1 = d__[m], abs(d__1))) * sqrt((d__2 = d__[m + 1], abs(d__2))) * eps)
{
e[m] = 0.;
goto L30;
}
/* L20: */
}
}
m = *n;
L30:
l = l1;
lsv = l;
lend = m;
lendsv = lend;
l1 = m + 1;
if (lend == l)
{
goto L10;
}
/* Scale submatrix in rows and columns L to LEND */
i__1 = lend - l + 1;
anorm = dlanst_("M", &i__1, &d__[l], &e[l]);
iscale = 0;
if (anorm == 0.)
{
goto L10;
}
if (anorm > ssfmax)
{
iscale = 1;
i__1 = lend - l + 1;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmax, &i__1, &c__1, &d__[l], n, info);
i__1 = lend - l;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmax, &i__1, &c__1, &e[l], n, info);
}
else if (anorm < ssfmin)
{
iscale = 2;
i__1 = lend - l + 1;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &d__[l], n, info);
i__1 = lend - l;
dlascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &e[l], n, info);
}
/* Choose between QL and QR iteration */
if ((d__1 = d__[lend], abs(d__1)) < (d__2 = d__[l], abs(d__2)))
{
lend = lsv;
l = lendsv;
}
if (lend > l)
{
/* QL Iteration */
/* Look for small subdiagonal element. */
L40:
if (l != lend)
{
lendm1 = lend - 1;
i__1 = lendm1;
for (m = l;
m <= i__1;
++m)
{
/* Computing 2nd power */
d__2 = (d__1 = e[m], abs(d__1));
tst = d__2 * d__2;
if (tst <= eps2 * (d__1 = d__[m], abs(d__1)) * (d__2 = d__[m + 1], abs(d__2)) + safmin)
{
goto L60;
}
/* L50: */
}
}
m = lend;
L60:
if (m < lend)
{
e[m] = 0.;
}
p = d__[l];
if (m == l)
{
goto L80;
}
/* If remaining matrix is 2-by-2, use DLAE2 or SLAEV2 */
/* to compute its eigensystem. */
if (m == l + 1)
{
if (icompz > 0)
{
dlaev2_(&d__[l], &e[l], &d__[l + 1], &rt1, &rt2, &c__, &s);
work[l] = c__;
work[*n - 1 + l] = s;
dlasr_("R", "V", "B", n, &c__2, &work[l], &work[*n - 1 + l], & z__[l * z_dim1 + 1], ldz);
}
else
{
dlae2_(&d__[l], &e[l], &d__[l + 1], &rt1, &rt2);
}
d__[l] = rt1;
d__[l + 1] = rt2;
e[l] = 0.;
l += 2;
if (l <= lend)
{
goto L40;
}
goto L140;
}
if (jtot == nmaxit)
{
goto L140;
}
++jtot;
/* Form shift. */
g = (d__[l + 1] - p) / (e[l] * 2.);
r__ = dlapy2_(&g, &c_b10);
g = d__[m] - p + e[l] / (g + d_sign(&r__, &g));
s = 1.;
c__ = 1.;
p = 0.;
/* Inner loop */
mm1 = m - 1;
i__1 = l;
for (i__ = mm1;
i__ >= i__1;
--i__)
{
f = s * e[i__];
b = c__ * e[i__];
dlartg_(&g, &f, &c__, &s, &r__);
if (i__ != m - 1)
{
e[i__ + 1] = r__;
}
g = d__[i__ + 1] - p;
r__ = (d__[i__] - g) * s + c__ * 2. * b;
p = s * r__;
d__[i__ + 1] = g + p;
g = c__ * r__ - b;
/* If eigenvectors are desired, then save rotations. */
if (icompz > 0)
{
work[i__] = c__;
work[*n - 1 + i__] = -s;
}
/* L70: */
}
/* If eigenvectors are desired, then apply saved rotations. */
if (icompz > 0)
{
mm = m - l + 1;
dlasr_("R", "V", "B", n, &mm, &work[l], &work[*n - 1 + l], &z__[l * z_dim1 + 1], ldz);
}
d__[l] -= p;
e[l] = g;
goto L40;
/* Eigenvalue found. */
L80:
d__[l] = p;
++l;
if (l <= lend)
{
goto L40;
}
goto L140;
}
else
{
/* QR Iteration */
/* Look for small superdiagonal element. */
L90:
if (l != lend)
{
lendp1 = lend + 1;
i__1 = lendp1;
for (m = l;
m >= i__1;
--m)
{
/* Computing 2nd power */
d__2 = (d__1 = e[m - 1], abs(d__1));
tst = d__2 * d__2;
if (tst <= eps2 * (d__1 = d__[m], abs(d__1)) * (d__2 = d__[m - 1], abs(d__2)) + safmin)
{
goto L110;
}
/* L100: */
}
}
m = lend;
L110:
if (m > lend)
{
e[m - 1] = 0.;
}
p = d__[l];
if (m == l)
{
goto L130;
}
/* If remaining matrix is 2-by-2, use DLAE2 or SLAEV2 */
/* to compute its eigensystem. */
if (m == l - 1)
{
if (icompz > 0)
{
dlaev2_(&d__[l - 1], &e[l - 1], &d__[l], &rt1, &rt2, &c__, &s) ;
work[m] = c__;
work[*n - 1 + m] = s;
dlasr_("R", "V", "F", n, &c__2, &work[m], &work[*n - 1 + m], & z__[(l - 1) * z_dim1 + 1], ldz);
}
else
{
dlae2_(&d__[l - 1], &e[l - 1], &d__[l], &rt1, &rt2);
}
d__[l - 1] = rt1;
d__[l] = rt2;
e[l - 1] = 0.;
l += -2;
if (l >= lend)
{
goto L90;
}
goto L140;
}
if (jtot == nmaxit)
{
goto L140;
}
++jtot;
/* Form shift. */
g = (d__[l - 1] - p) / (e[l - 1] * 2.);
r__ = dlapy2_(&g, &c_b10);
g = d__[m] - p + e[l - 1] / (g + d_sign(&r__, &g));
s = 1.;
c__ = 1.;
p = 0.;
/* Inner loop */
lm1 = l - 1;
i__1 = lm1;
for (i__ = m;
i__ <= i__1;
++i__)
{
f = s * e[i__];
b = c__ * e[i__];
dlartg_(&g, &f, &c__, &s, &r__);
if (i__ != m)
{
e[i__ - 1] = r__;
}
g = d__[i__] - p;
r__ = (d__[i__ + 1] - g) * s + c__ * 2. * b;
p = s * r__;
d__[i__] = g + p;
g = c__ * r__ - b;
/* If eigenvectors are desired, then save rotations. */
if (icompz > 0)
{
work[i__] = c__;
work[*n - 1 + i__] = s;
}
/* L120: */
}
/* If eigenvectors are desired, then apply saved rotations. */
if (icompz > 0)
{
mm = l - m + 1;
dlasr_("R", "V", "F", n, &mm, &work[m], &work[*n - 1 + m], &z__[m * z_dim1 + 1], ldz);
}
d__[l] -= p;
e[lm1] = g;
goto L90;
/* Eigenvalue found. */
L130:
d__[l] = p;
--l;
if (l >= lend)
{
goto L90;
}
goto L140;
}
/* Undo scaling if necessary */
L140:
if (iscale == 1)
{
i__1 = lendsv - lsv + 1;
dlascl_("G", &c__0, &c__0, &ssfmax, &anorm, &i__1, &c__1, &d__[lsv], n, info);
i__1 = lendsv - lsv;
dlascl_("G", &c__0, &c__0, &ssfmax, &anorm, &i__1, &c__1, &e[lsv], n, info);
}
else if (iscale == 2)
{
i__1 = lendsv - lsv + 1;
dlascl_("G", &c__0, &c__0, &ssfmin, &anorm, &i__1, &c__1, &d__[lsv], n, info);
i__1 = lendsv - lsv;
dlascl_("G", &c__0, &c__0, &ssfmin, &anorm, &i__1, &c__1, &e[lsv], n, info);
}
/* Check for no convergence to an eigenvalue after a total */
/* of N*MAXIT iterations. */
if (jtot < nmaxit)
{
goto L10;
}
i__1 = *n - 1;
for (i__ = 1;
i__ <= i__1;
++i__)
{
if (e[i__] != 0.)
{
++(*info);
}
/* L150: */
}
goto L190;
/* Order eigenvalues and eigenvectors. */
L160:
if (icompz == 0)
{
/* Use Quick Sort */
dlasrt_("I", n, &d__[1], info);
}
else
{
/* Use Selection Sort to minimize swaps of eigenvectors */
i__1 = *n;
for (ii = 2;
ii <= i__1;
++ii)
{
i__ = ii - 1;
k = i__;
p = d__[i__];
i__2 = *n;
for (j = ii;
j <= i__2;
++j)
{
if (d__[j] < p)
{
k = j;
p = d__[j];
}
/* L170: */
}
if (k != i__)
{
d__[k] = d__[i__];
d__[i__] = p;
dswap_(n, &z__[i__ * z_dim1 + 1], &c__1, &z__[k * z_dim1 + 1], &c__1);
}
/* L180: */
}
}
L190:
return 0;
/* End of DSTEQR */
}
/* Subroutine */ int dlanv2_(doublereal *a, doublereal *b, doublereal *c__,
doublereal *d__, doublereal *rt1r, doublereal *rt1i, doublereal *rt2r,
doublereal *rt2i, doublereal *cs, doublereal *sn)
{
/* System generated locals */
doublereal d__1, d__2;
/* Builtin functions */
double d_sign(doublereal *, doublereal *), sqrt(doublereal);
/* Local variables */
doublereal p, z__, aa, bb, cc, dd, cs1, sn1, sab, sac, eps, tau, temp,
scale, bcmax, bcmis, sigma;
extern doublereal dlapy2_(doublereal *, doublereal *), dlamch_(char *);
/* -- LAPACK driver routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* DLANV2 computes the Schur factorization of a real 2-by-2 nonsymmetric */
/* matrix in standard form: */
/* [ A B ] = [ CS -SN ] [ AA BB ] [ CS SN ] */
/* [ C D ] [ SN CS ] [ CC DD ] [-SN CS ] */
/* where either */
/* 1) CC = 0 so that AA and DD are real eigenvalues of the matrix, or */
/* 2) AA = DD and BB*CC < 0, so that AA + or - sqrt(BB*CC) are complex */
/* conjugate eigenvalues. */
/* Arguments */
/* ========= */
/* A (input/output) DOUBLE PRECISION */
/* B (input/output) DOUBLE PRECISION */
/* C (input/output) DOUBLE PRECISION */
/* D (input/output) DOUBLE PRECISION */
/* On entry, the elements of the input matrix. */
/* On exit, they are overwritten by the elements of the */
/* standardised Schur form. */
/* RT1R (output) DOUBLE PRECISION */
/* RT1I (output) DOUBLE PRECISION */
/* RT2R (output) DOUBLE PRECISION */
/* RT2I (output) DOUBLE PRECISION */
/* The real and imaginary parts of the eigenvalues. If the */
/* eigenvalues are a complex conjugate pair, RT1I > 0. */
/* CS (output) DOUBLE PRECISION */
/* SN (output) DOUBLE PRECISION */
/* Parameters of the rotation matrix. */
/* Further Details */
/* =============== */
/* Modified by V. Sima, Research Institute for Informatics, Bucharest, */
/* Romania, to reduce the risk of cancellation errors, */
/* when computing real eigenvalues, and to ensure, if possible, that */
/* abs(RT1R) >= abs(RT2R). */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
eps = dlamch_("P");
if (*c__ == 0.) {
*cs = 1.;
*sn = 0.;
goto L10;
} else if (*b == 0.) {
/* Swap rows and columns */
*cs = 0.;
*sn = 1.;
temp = *d__;
*d__ = *a;
*a = temp;
*b = -(*c__);
*c__ = 0.;
goto L10;
} else if (*a - *d__ == 0. && d_sign(&c_b4, b) != d_sign(&c_b4, c__)) {
*cs = 1.;
*sn = 0.;
goto L10;
} else {
temp = *a - *d__;
p = temp * .5;
/* Computing MAX */
d__1 = abs(*b), d__2 = abs(*c__);
bcmax = max(d__1,d__2);
/* Computing MIN */
d__1 = abs(*b), d__2 = abs(*c__);
bcmis = min(d__1,d__2) * d_sign(&c_b4, b) * d_sign(&c_b4, c__);
/* Computing MAX */
d__1 = abs(p);
scale = max(d__1,bcmax);
z__ = p / scale * p + bcmax / scale * bcmis;
/* If Z is of the order of the machine accuracy, postpone the */
/* decision on the nature of eigenvalues */
if (z__ >= eps * 4.) {
/* Real eigenvalues. Compute A and D. */
d__1 = sqrt(scale) * sqrt(z__);
z__ = p + d_sign(&d__1, &p);
*a = *d__ + z__;
*d__ -= bcmax / z__ * bcmis;
/* Compute B and the rotation matrix */
tau = dlapy2_(c__, &z__);
*cs = z__ / tau;
*sn = *c__ / tau;
*b -= *c__;
*c__ = 0.;
} else {
/* Complex eigenvalues, or real (almost) equal eigenvalues. */
/* Make diagonal elements equal. */
sigma = *b + *c__;
tau = dlapy2_(&sigma, &temp);
*cs = sqrt((abs(sigma) / tau + 1.) * .5);
*sn = -(p / (tau * *cs)) * d_sign(&c_b4, &sigma);
/* Compute [ AA BB ] = [ A B ] [ CS -SN ] */
/* [ CC DD ] [ C D ] [ SN CS ] */
aa = *a * *cs + *b * *sn;
bb = -(*a) * *sn + *b * *cs;
cc = *c__ * *cs + *d__ * *sn;
dd = -(*c__) * *sn + *d__ * *cs;
/* Compute [ A B ] = [ CS SN ] [ AA BB ] */
/* [ C D ] [-SN CS ] [ CC DD ] */
*a = aa * *cs + cc * *sn;
*b = bb * *cs + dd * *sn;
*c__ = -aa * *sn + cc * *cs;
*d__ = -bb * *sn + dd * *cs;
temp = (*a + *d__) * .5;
*a = temp;
*d__ = temp;
if (*c__ != 0.) {
if (*b != 0.) {
if (d_sign(&c_b4, b) == d_sign(&c_b4, c__)) {
/* Real eigenvalues: reduce to upper triangular form */
sab = sqrt((abs(*b)));
sac = sqrt((abs(*c__)));
d__1 = sab * sac;
p = d_sign(&d__1, c__);
tau = 1. / sqrt((d__1 = *b + *c__, abs(d__1)));
*a = temp + p;
*d__ = temp - p;
*b -= *c__;
*c__ = 0.;
cs1 = sab * tau;
sn1 = sac * tau;
temp = *cs * cs1 - *sn * sn1;
*sn = *cs * sn1 + *sn * cs1;
*cs = temp;
}
} else {
*b = -(*c__);
*c__ = 0.;
temp = *cs;
*cs = -(*sn);
*sn = temp;
}
}
}
}
L10:
/* Store eigenvalues in (RT1R,RT1I) and (RT2R,RT2I). */
*rt1r = *a;
*rt2r = *d__;
if (*c__ == 0.) {
*rt1i = 0.;
*rt2i = 0.;
} else {
*rt1i = sqrt((abs(*b))) * sqrt((abs(*c__)));
*rt2i = -(*rt1i);
}
return 0;
/* End of DLANV2 */
} /* dlanv2_ */
/* Subroutine */ int zlaed8_(integer *k, integer *n, integer *qsiz,
doublecomplex *q, integer *ldq, doublereal *d__, doublereal *rho,
integer *cutpnt, doublereal *z__, doublereal *dlamda, doublecomplex *
q2, integer *ldq2, doublereal *w, integer *indxp, integer *indx,
integer *indxq, integer *perm, integer *givptr, integer *givcol,
doublereal *givnum, integer *info)
{
/* -- LAPACK routine (version 3.0) --
Univ. of Tennessee, Oak Ridge National Lab, Argonne National Lab,
Courant Institute, NAG Ltd., and Rice University
September 30, 1994
Purpose
=======
ZLAED8 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 element in the
Z vector. For each such occurrence the order of the related secular
equation problem is reduced by one.
Arguments
=========
K (output) INTEGER
Contains the number of non-deflated eigenvalues.
This is the order of the related secular equation.
N (input) INTEGER
The dimension of the symmetric tridiagonal matrix. N >= 0.
QSIZ (input) INTEGER
The dimension of the unitary matrix used to reduce
the dense or band matrix to tridiagonal form.
QSIZ >= N if ICOMPQ = 1.
Q (input/output) COMPLEX*16 array, dimension (LDQ,N)
On entry, Q contains the eigenvectors of the partially solved
system which has been previously updated in matrix
multiplies with other partially solved eigensystems.
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 ).
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.
RHO (input/output) DOUBLE PRECISION
Contains the off diagonal element associated with the rank-1
cut which originally split the two submatrices which are now
being recombined. RHO is modified during the computation to
the value required by DLAED3.
CUTPNT (input) INTEGER
Contains the location of the last eigenvalue in the leading
sub-matrix. MIN(1,N) <= CUTPNT <= N.
Z (input) DOUBLE PRECISION array, dimension (N)
On input this vector contains the updating vector (the last
row of the first sub-eigenvector matrix and the first row of
the second sub-eigenvector matrix). The contents of Z are
destroyed during the updating process.
DLAMDA (output) DOUBLE PRECISION array, dimension (N)
Contains a copy of the first K eigenvalues which will be used
by DLAED3 to form the secular equation.
Q2 (output) COMPLEX*16 array, dimension (LDQ2,N)
If ICOMPQ = 0, Q2 is not referenced. Otherwise,
Contains a copy of the first K eigenvectors which will be used
by DLAED7 in a matrix multiply (DGEMM) to update the new
eigenvectors.
LDQ2 (input) INTEGER
The leading dimension of the array Q2. LDQ2 >= max( 1, N ).
W (output) DOUBLE PRECISION array, dimension (N)
This will hold the first k values of the final
deflation-altered z-vector and will be passed to DLAED3.
INDXP (workspace) INTEGER array, dimension (N)
This will contain the permutation used to place deflated
values of D at the end of the array. On output INDXP(1:K)
points to the nondeflated D-values and INDXP(K+1:N)
points to the deflated eigenvalues.
INDX (workspace) INTEGER array, dimension (N)
This will contain the permutation used to sort the contents of
D into ascending order.
INDXQ (input) INTEGER array, dimension (N)
This contains 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 CUTPNT
added to their values in order to be accurate.
PERM (output) INTEGER array, dimension (N)
Contains the permutations (from deflation and sorting) to be
applied to each eigenblock.
GIVPTR (output) INTEGER
Contains the number of Givens rotations which took place in
this subproblem.
GIVCOL (output) INTEGER array, dimension (2, N)
Each pair of numbers indicates a pair of columns to take place
in a Givens rotation.
GIVNUM (output) DOUBLE PRECISION array, dimension (2, N)
Each number indicates the S value to be used in the
corresponding Givens rotation.
INFO (output) INTEGER
= 0: successful exit.
< 0: if INFO = -i, the i-th argument had an illegal value.
=====================================================================
Test the input parameters.
Parameter adjustments */
/* Table of constant values */
static doublereal c_b3 = -1.;
static integer c__1 = 1;
/* System generated locals */
integer q_dim1, q_offset, q2_dim1, q2_offset, i__1;
doublereal d__1;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
static integer jlam, imax, jmax;
static doublereal c__;
static integer i__, j;
static doublereal s, t;
extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
integer *), dcopy_(integer *, doublereal *, integer *, doublereal
*, integer *);
static integer k2, n1, n2;
extern /* Subroutine */ int zdrot_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *, doublereal *, doublereal *), zcopy_(
integer *, doublecomplex *, integer *, doublecomplex *, integer *)
;
extern doublereal dlapy2_(doublereal *, doublereal *), dlamch_(char *);
static integer jp;
extern integer idamax_(integer *, doublereal *, integer *);
extern /* Subroutine */ int dlamrg_(integer *, integer *, doublereal *,
integer *, integer *, integer *), xerbla_(char *, integer *), zlacpy_(char *, integer *, integer *, doublecomplex *,
integer *, doublecomplex *, integer *);
static integer n1p1;
static doublereal eps, tau, tol;
#define q_subscr(a_1,a_2) (a_2)*q_dim1 + a_1
#define q_ref(a_1,a_2) q[q_subscr(a_1,a_2)]
#define q2_subscr(a_1,a_2) (a_2)*q2_dim1 + a_1
#define q2_ref(a_1,a_2) q2[q2_subscr(a_1,a_2)]
#define givcol_ref(a_1,a_2) givcol[(a_2)*2 + a_1]
#define givnum_ref(a_1,a_2) givnum[(a_2)*2 + a_1]
q_dim1 = *ldq;
q_offset = 1 + q_dim1 * 1;
q -= q_offset;
--d__;
--z__;
--dlamda;
q2_dim1 = *ldq2;
q2_offset = 1 + q2_dim1 * 1;
q2 -= q2_offset;
--w;
--indxp;
--indx;
--indxq;
--perm;
givcol -= 3;
givnum -= 3;
/* Function Body */
*info = 0;
if (*n < 0) {
*info = -2;
} else if (*qsiz < *n) {
*info = -3;
} else if (*ldq < max(1,*n)) {
*info = -5;
} else if (*cutpnt < min(1,*n) || *cutpnt > *n) {
*info = -8;
} else if (*ldq2 < max(1,*n)) {
*info = -12;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("ZLAED8", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
n1 = *cutpnt;
n2 = *n - n1;
n1p1 = n1 + 1;
if (*rho < 0.) {
dscal_(&n2, &c_b3, &z__[n1p1], &c__1);
}
/* Normalize z so that norm(z) = 1 */
t = 1. / sqrt(2.);
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
indx[j] = j;
/* L10: */
}
dscal_(n, &t, &z__[1], &c__1);
*rho = (d__1 = *rho * 2., abs(d__1));
/* Sort the eigenvalues into increasing order */
i__1 = *n;
for (i__ = *cutpnt + 1; i__ <= i__1; ++i__) {
indxq[i__] += *cutpnt;
/* L20: */
}
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
dlamda[i__] = d__[indxq[i__]];
w[i__] = z__[indxq[i__]];
/* L30: */
}
i__ = 1;
j = *cutpnt + 1;
dlamrg_(&n1, &n2, &dlamda[1], &c__1, &c__1, &indx[1]);
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
d__[i__] = dlamda[indx[i__]];
z__[i__] = w[indx[i__]];
/* L40: */
}
/* Calculate the allowable deflation tolerance */
imax = idamax_(n, &z__[1], &c__1);
jmax = idamax_(n, &d__[1], &c__1);
eps = dlamch_("Epsilon");
tol = eps * 8. * (d__1 = d__[jmax], abs(d__1));
/* 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;
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
perm[j] = indxq[indx[j]];
zcopy_(qsiz, &q_ref(1, perm[j]), &c__1, &q2_ref(1, j), &c__1);
/* L50: */
}
zlacpy_("A", qsiz, n, &q2_ref(1, 1), ldq2, &q_ref(1, 1), ldq);
return 0;
}
/* 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. */
*k = 0;
*givptr = 0;
k2 = *n + 1;
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
if (*rho * (d__1 = z__[j], abs(d__1)) <= tol) {
/* Deflate due to small z component. */
--k2;
indxp[k2] = j;
if (j == *n) {
goto L100;
}
} else {
jlam = j;
goto L70;
}
/* L60: */
}
L70:
++j;
if (j > *n) {
goto L90;
}
if (*rho * (d__1 = z__[j], abs(d__1)) <= tol) {
/* Deflate due to small z component. */
--k2;
indxp[k2] = j;
} else {
/* Check if eigenvalues are close enough to allow deflation. */
s = z__[jlam];
c__ = z__[j];
/* Find sqrt(a**2+b**2) without overflow or
destructive underflow. */
tau = dlapy2_(&c__, &s);
t = d__[j] - d__[jlam];
c__ /= tau;
s = -s / tau;
if ((d__1 = t * c__ * s, abs(d__1)) <= tol) {
/* Deflation is possible. */
z__[j] = tau;
z__[jlam] = 0.;
/* Record the appropriate Givens rotation */
++(*givptr);
givcol_ref(1, *givptr) = indxq[indx[jlam]];
givcol_ref(2, *givptr) = indxq[indx[j]];
givnum_ref(1, *givptr) = c__;
givnum_ref(2, *givptr) = s;
zdrot_(qsiz, &q_ref(1, indxq[indx[jlam]]), &c__1, &q_ref(1, indxq[
indx[j]]), &c__1, &c__, &s);
t = d__[jlam] * c__ * c__ + d__[j] * s * s;
d__[j] = d__[jlam] * s * s + d__[j] * c__ * c__;
d__[jlam] = t;
--k2;
i__ = 1;
L80:
if (k2 + i__ <= *n) {
if (d__[jlam] < d__[indxp[k2 + i__]]) {
indxp[k2 + i__ - 1] = indxp[k2 + i__];
indxp[k2 + i__] = jlam;
++i__;
goto L80;
} else {
indxp[k2 + i__ - 1] = jlam;
}
} else {
indxp[k2 + i__ - 1] = jlam;
}
jlam = j;
} else {
++(*k);
w[*k] = z__[jlam];
dlamda[*k] = d__[jlam];
indxp[*k] = jlam;
jlam = j;
}
}
goto L70;
L90:
/* Record the last eigenvalue. */
++(*k);
w[*k] = z__[jlam];
dlamda[*k] = d__[jlam];
indxp[*k] = jlam;
L100:
/* 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 = *n;
for (j = 1; j <= i__1; ++j) {
jp = indxp[j];
dlamda[j] = d__[jp];
perm[j] = indxq[indx[jp]];
zcopy_(qsiz, &q_ref(1, perm[j]), &c__1, &q2_ref(1, j), &c__1);
/* L110: */
}
/* The deflated eigenvalues and their corresponding vectors go back
into the last N - K slots of D and Q respectively. */
if (*k < *n) {
i__1 = *n - *k;
dcopy_(&i__1, &dlamda[*k + 1], &c__1, &d__[*k + 1], &c__1);
i__1 = *n - *k;
zlacpy_("A", qsiz, &i__1, &q2_ref(1, *k + 1), ldq2, &q_ref(1, *k + 1),
ldq);
}
return 0;
/* End of ZLAED8 */
} /* zlaed8_ */
/* Subroutine */ int znaitr_(integer *ido, char *bmat, integer *n, integer *k,
integer *np, integer *nb, doublecomplex *resid, doublereal *rnorm,
doublecomplex *v, integer *ldv, doublecomplex *h__, integer *ldh,
integer *ipntr, doublecomplex *workd, integer *info, ftnlen bmat_len)
{
/* Initialized data */
static logical first = TRUE_;
/* System generated locals */
integer h_dim1, h_offset, v_dim1, v_offset, i__1, i__2, i__3;
doublereal d__1, d__2, d__3, d__4;
doublecomplex z__1;
/* Builtin functions */
double d_imag(doublecomplex *), sqrt(doublereal);
/* Local variables */
static integer i__, j;
static real t0, t1, t2, t3, t4, t5;
static integer jj, ipj, irj, ivj;
static doublereal ulp, tst1;
static integer ierr, iter;
static doublereal unfl, ovfl;
static integer itry;
static doublereal temp1;
static logical orth1, orth2, step3, step4;
static doublereal betaj;
static integer infol;
static doublecomplex cnorm;
extern /* Double Complex */ VOID zdotc_(doublecomplex *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *);
static doublereal rtemp[2];
extern /* Subroutine */ int zgemv_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublecomplex *, doublecomplex *, integer *, ftnlen);
static doublereal wnorm;
extern /* Subroutine */ int dvout_(integer *, integer *, doublereal *,
integer *, char *, ftnlen), zcopy_(integer *, doublecomplex *,
integer *, doublecomplex *, integer *), ivout_(integer *, integer
*, integer *, integer *, char *, ftnlen), zaxpy_(integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *), zmout_(integer *, integer *, integer *, doublecomplex
*, integer *, integer *, char *, ftnlen), zvout_(integer *,
integer *, doublecomplex *, integer *, char *, ftnlen);
extern doublereal dlapy2_(doublereal *, doublereal *);
extern /* Subroutine */ int dlabad_(doublereal *, doublereal *);
extern doublereal dznrm2_(integer *, doublecomplex *, integer *);
static doublereal rnorm1;
extern /* Subroutine */ int zgetv0_(integer *, char *, integer *, logical
*, integer *, integer *, doublecomplex *, integer *,
doublecomplex *, doublereal *, integer *, doublecomplex *,
integer *, ftnlen);
extern doublereal dlamch_(char *, ftnlen);
extern /* Subroutine */ int second_(real *), zdscal_(integer *,
doublereal *, doublecomplex *, integer *);
static logical rstart;
static integer msglvl;
static doublereal smlnum;
extern doublereal zlanhs_(char *, integer *, doublecomplex *, integer *,
doublecomplex *, ftnlen);
extern /* Subroutine */ int zlascl_(char *, integer *, integer *,
doublereal *, doublereal *, integer *, integer *, doublecomplex *,
integer *, integer *, ftnlen);
/* %----------------------------------------------------% */
/* | Include files for debugging and timing information | */
/* %----------------------------------------------------% */
/* \SCCS Information: @(#) */
/* FILE: debug.h SID: 2.3 DATE OF SID: 11/16/95 RELEASE: 2 */
/* %---------------------------------% */
/* | See debug.doc for documentation | */
/* %---------------------------------% */
/* %------------------% */
/* | Scalar Arguments | */
/* %------------------% */
/* %--------------------------------% */
/* | See stat.doc for documentation | */
/* %--------------------------------% */
/* \SCCS Information: @(#) */
/* FILE: stat.h SID: 2.2 DATE OF SID: 11/16/95 RELEASE: 2 */
/* %-----------------% */
/* | Array Arguments | */
/* %-----------------% */
/* %------------% */
/* | Parameters | */
/* %------------% */
/* %--------------% */
/* | Local Arrays | */
/* %--------------% */
/* %---------------% */
/* | Local Scalars | */
/* %---------------% */
/* %----------------------% */
/* | External Subroutines | */
/* %----------------------% */
/* %--------------------% */
/* | External Functions | */
/* %--------------------% */
/* %---------------------% */
/* | Intrinsic Functions | */
/* %---------------------% */
/* %-----------------% */
/* | Data statements | */
/* %-----------------% */
/* Parameter adjustments */
--workd;
--resid;
v_dim1 = *ldv;
v_offset = 1 + v_dim1;
v -= v_offset;
h_dim1 = *ldh;
h_offset = 1 + h_dim1;
h__ -= h_offset;
--ipntr;
/* Function Body */
/* %-----------------------% */
/* | Executable Statements | */
/* %-----------------------% */
if (first) {
/* %-----------------------------------------% */
/* | Set machine-dependent constants for the | */
/* | the splitting and deflation criterion. | */
/* | If norm(H) <= sqrt(OVFL), | */
/* | overflow should not occur. | */
/* | REFERENCE: LAPACK subroutine zlahqr | */
/* %-----------------------------------------% */
unfl = dlamch_("safe minimum", (ftnlen)12);
z__1.r = 1. / unfl, z__1.i = 0. / unfl;
ovfl = z__1.r;
dlabad_(&unfl, &ovfl);
ulp = dlamch_("precision", (ftnlen)9);
smlnum = unfl * (*n / ulp);
first = FALSE_;
}
if (*ido == 0) {
/* %-------------------------------% */
/* | Initialize timing statistics | */
/* | & message level for debugging | */
/* %-------------------------------% */
second_(&t0);
msglvl = debug_1.mcaitr;
/* %------------------------------% */
/* | Initial call to this routine | */
/* %------------------------------% */
*info = 0;
step3 = FALSE_;
step4 = FALSE_;
rstart = FALSE_;
orth1 = FALSE_;
orth2 = FALSE_;
j = *k + 1;
ipj = 1;
irj = ipj + *n;
ivj = irj + *n;
}
/* %-------------------------------------------------% */
/* | When in reverse communication mode one of: | */
/* | STEP3, STEP4, ORTH1, ORTH2, RSTART | */
/* | will be .true. when .... | */
/* | STEP3: return from computing OP*v_{j}. | */
/* | STEP4: return from computing B-norm of OP*v_{j} | */
/* | ORTH1: return from computing B-norm of r_{j+1} | */
/* | ORTH2: return from computing B-norm of | */
/* | correction to the residual vector. | */
/* | RSTART: return from OP computations needed by | */
/* | zgetv0. | */
/* %-------------------------------------------------% */
if (step3) {
goto L50;
}
if (step4) {
goto L60;
}
if (orth1) {
goto L70;
}
if (orth2) {
goto L90;
}
if (rstart) {
goto L30;
}
/* %-----------------------------% */
/* | Else this is the first step | */
/* %-----------------------------% */
/* %--------------------------------------------------------------% */
/* | | */
/* | A R N O L D I I T E R A T I O N L O O P | */
/* | | */
/* | Note: B*r_{j-1} is already in WORKD(1:N)=WORKD(IPJ:IPJ+N-1) | */
/* %--------------------------------------------------------------% */
L1000:
if (msglvl > 1) {
ivout_(&debug_1.logfil, &c__1, &j, &debug_1.ndigit, "_naitr: generat"
"ing Arnoldi vector number", (ftnlen)40);
dvout_(&debug_1.logfil, &c__1, rnorm, &debug_1.ndigit, "_naitr: B-no"
"rm of the current residual is", (ftnlen)41);
}
/* %---------------------------------------------------% */
/* | STEP 1: Check if the B norm of j-th residual | */
/* | vector is zero. Equivalent to determine whether | */
/* | an exact j-step Arnoldi factorization is present. | */
/* %---------------------------------------------------% */
betaj = *rnorm;
if (*rnorm > 0.) {
goto L40;
}
/* %---------------------------------------------------% */
/* | Invariant subspace found, generate a new starting | */
/* | vector which is orthogonal to the current Arnoldi | */
/* | basis and continue the iteration. | */
/* %---------------------------------------------------% */
if (msglvl > 0) {
ivout_(&debug_1.logfil, &c__1, &j, &debug_1.ndigit, "_naitr: ****** "
"RESTART AT STEP ******", (ftnlen)37);
}
/* %---------------------------------------------% */
/* | ITRY is the loop variable that controls the | */
/* | maximum amount of times that a restart is | */
/* | attempted. NRSTRT is used by stat.h | */
/* %---------------------------------------------% */
betaj = 0.;
++timing_1.nrstrt;
itry = 1;
L20:
rstart = TRUE_;
*ido = 0;
L30:
/* %--------------------------------------% */
/* | If in reverse communication mode and | */
/* | RSTART = .true. flow returns here. | */
/* %--------------------------------------% */
zgetv0_(ido, bmat, &itry, &c_false, n, &j, &v[v_offset], ldv, &resid[1],
rnorm, &ipntr[1], &workd[1], &ierr, (ftnlen)1);
if (*ido != 99) {
goto L9000;
}
if (ierr < 0) {
++itry;
if (itry <= 3) {
goto L20;
}
/* %------------------------------------------------% */
/* | Give up after several restart attempts. | */
/* | Set INFO to the size of the invariant subspace | */
/* | which spans OP and exit. | */
/* %------------------------------------------------% */
*info = j - 1;
second_(&t1);
timing_1.tcaitr += t1 - t0;
*ido = 99;
goto L9000;
}
L40:
/* %---------------------------------------------------------% */
/* | STEP 2: v_{j} = r_{j-1}/rnorm and p_{j} = p_{j}/rnorm | */
/* | Note that p_{j} = B*r_{j-1}. In order to avoid overflow | */
/* | when reciprocating a small RNORM, test against lower | */
/* | machine bound. | */
/* %---------------------------------------------------------% */
zcopy_(n, &resid[1], &c__1, &v[j * v_dim1 + 1], &c__1);
if (*rnorm >= unfl) {
temp1 = 1. / *rnorm;
zdscal_(n, &temp1, &v[j * v_dim1 + 1], &c__1);
zdscal_(n, &temp1, &workd[ipj], &c__1);
} else {
/* %-----------------------------------------% */
/* | To scale both v_{j} and p_{j} carefully | */
/* | use LAPACK routine zlascl | */
/* %-----------------------------------------% */
zlascl_("General", &i__, &i__, rnorm, &c_b27, n, &c__1, &v[j * v_dim1
+ 1], n, &infol, (ftnlen)7);
zlascl_("General", &i__, &i__, rnorm, &c_b27, n, &c__1, &workd[ipj],
n, &infol, (ftnlen)7);
}
/* %------------------------------------------------------% */
/* | STEP 3: r_{j} = OP*v_{j}; Note that p_{j} = B*v_{j} | */
/* | Note that this is not quite yet r_{j}. See STEP 4 | */
/* %------------------------------------------------------% */
step3 = TRUE_;
++timing_1.nopx;
second_(&t2);
zcopy_(n, &v[j * v_dim1 + 1], &c__1, &workd[ivj], &c__1);
ipntr[1] = ivj;
ipntr[2] = irj;
ipntr[3] = ipj;
*ido = 1;
/* %-----------------------------------% */
/* | Exit in order to compute OP*v_{j} | */
/* %-----------------------------------% */
goto L9000;
L50:
/* %----------------------------------% */
/* | Back from reverse communication; | */
/* | WORKD(IRJ:IRJ+N-1) := OP*v_{j} | */
/* | if step3 = .true. | */
/* %----------------------------------% */
second_(&t3);
timing_1.tmvopx += t3 - t2;
step3 = FALSE_;
/* %------------------------------------------% */
/* | Put another copy of OP*v_{j} into RESID. | */
/* %------------------------------------------% */
zcopy_(n, &workd[irj], &c__1, &resid[1], &c__1);
/* %---------------------------------------% */
/* | STEP 4: Finish extending the Arnoldi | */
/* | factorization to length j. | */
/* %---------------------------------------% */
second_(&t2);
if (*(unsigned char *)bmat == 'G') {
++timing_1.nbx;
step4 = TRUE_;
ipntr[1] = irj;
ipntr[2] = ipj;
*ido = 2;
/* %-------------------------------------% */
/* | Exit in order to compute B*OP*v_{j} | */
/* %-------------------------------------% */
goto L9000;
} else if (*(unsigned char *)bmat == 'I') {
zcopy_(n, &resid[1], &c__1, &workd[ipj], &c__1);
}
L60:
/* %----------------------------------% */
/* | Back from reverse communication; | */
/* | WORKD(IPJ:IPJ+N-1) := B*OP*v_{j} | */
/* | if step4 = .true. | */
/* %----------------------------------% */
if (*(unsigned char *)bmat == 'G') {
second_(&t3);
timing_1.tmvbx += t3 - t2;
}
step4 = FALSE_;
/* %-------------------------------------% */
/* | The following is needed for STEP 5. | */
/* | Compute the B-norm of OP*v_{j}. | */
/* %-------------------------------------% */
if (*(unsigned char *)bmat == 'G') {
zdotc_(&z__1, n, &resid[1], &c__1, &workd[ipj], &c__1);
cnorm.r = z__1.r, cnorm.i = z__1.i;
d__1 = cnorm.r;
d__2 = d_imag(&cnorm);
wnorm = sqrt(dlapy2_(&d__1, &d__2));
} else if (*(unsigned char *)bmat == 'I') {
wnorm = dznrm2_(n, &resid[1], &c__1);
}
/* %-----------------------------------------% */
/* | Compute the j-th residual corresponding | */
/* | to the j step factorization. | */
/* | Use Classical Gram Schmidt and compute: | */
/* | w_{j} <- V_{j}^T * B * OP * v_{j} | */
/* | r_{j} <- OP*v_{j} - V_{j} * w_{j} | */
/* %-----------------------------------------% */
/* %------------------------------------------% */
/* | Compute the j Fourier coefficients w_{j} | */
/* | WORKD(IPJ:IPJ+N-1) contains B*OP*v_{j}. | */
/* %------------------------------------------% */
zgemv_("C", n, &j, &c_b1, &v[v_offset], ldv, &workd[ipj], &c__1, &c_b2, &
h__[j * h_dim1 + 1], &c__1, (ftnlen)1);
/* %--------------------------------------% */
/* | Orthogonalize r_{j} against V_{j}. | */
/* | RESID contains OP*v_{j}. See STEP 3. | */
/* %--------------------------------------% */
z__1.r = -1., z__1.i = -0.;
zgemv_("N", n, &j, &z__1, &v[v_offset], ldv, &h__[j * h_dim1 + 1], &c__1,
&c_b1, &resid[1], &c__1, (ftnlen)1);
if (j > 1) {
i__1 = j + (j - 1) * h_dim1;
z__1.r = betaj, z__1.i = 0.;
h__[i__1].r = z__1.r, h__[i__1].i = z__1.i;
}
second_(&t4);
orth1 = TRUE_;
second_(&t2);
if (*(unsigned char *)bmat == 'G') {
++timing_1.nbx;
zcopy_(n, &resid[1], &c__1, &workd[irj], &c__1);
ipntr[1] = irj;
ipntr[2] = ipj;
*ido = 2;
/* %----------------------------------% */
/* | Exit in order to compute B*r_{j} | */
/* %----------------------------------% */
goto L9000;
} else if (*(unsigned char *)bmat == 'I') {
zcopy_(n, &resid[1], &c__1, &workd[ipj], &c__1);
}
L70:
/* %---------------------------------------------------% */
/* | Back from reverse communication if ORTH1 = .true. | */
/* | WORKD(IPJ:IPJ+N-1) := B*r_{j}. | */
/* %---------------------------------------------------% */
if (*(unsigned char *)bmat == 'G') {
second_(&t3);
timing_1.tmvbx += t3 - t2;
}
orth1 = FALSE_;
/* %------------------------------% */
/* | Compute the B-norm of r_{j}. | */
/* %------------------------------% */
if (*(unsigned char *)bmat == 'G') {
zdotc_(&z__1, n, &resid[1], &c__1, &workd[ipj], &c__1);
cnorm.r = z__1.r, cnorm.i = z__1.i;
d__1 = cnorm.r;
d__2 = d_imag(&cnorm);
*rnorm = sqrt(dlapy2_(&d__1, &d__2));
} else if (*(unsigned char *)bmat == 'I') {
*rnorm = dznrm2_(n, &resid[1], &c__1);
}
/* %-----------------------------------------------------------% */
/* | STEP 5: Re-orthogonalization / Iterative refinement phase | */
/* | Maximum NITER_ITREF tries. | */
/* | | */
/* | s = V_{j}^T * B * r_{j} | */
/* | r_{j} = r_{j} - V_{j}*s | */
/* | alphaj = alphaj + s_{j} | */
/* | | */
/* | The stopping criteria used for iterative refinement is | */
/* | discussed in Parlett's book SEP, page 107 and in Gragg & | */
/* | Reichel ACM TOMS paper; Algorithm 686, Dec. 1990. | */
/* | Determine if we need to correct the residual. The goal is | */
/* | to enforce ||v(:,1:j)^T * r_{j}|| .le. eps * || r_{j} || | */
/* | The following test determines whether the sine of the | */
/* | angle between OP*x and the computed residual is less | */
/* | than or equal to 0.717. | */
/* %-----------------------------------------------------------% */
if (*rnorm > wnorm * .717f) {
goto L100;
}
iter = 0;
++timing_1.nrorth;
/* %---------------------------------------------------% */
/* | Enter the Iterative refinement phase. If further | */
/* | refinement is necessary, loop back here. The loop | */
/* | variable is ITER. Perform a step of Classical | */
/* | Gram-Schmidt using all the Arnoldi vectors V_{j} | */
/* %---------------------------------------------------% */
L80:
if (msglvl > 2) {
rtemp[0] = wnorm;
rtemp[1] = *rnorm;
dvout_(&debug_1.logfil, &c__2, rtemp, &debug_1.ndigit, "_naitr: re-o"
"rthogonalization; wnorm and rnorm are", (ftnlen)49);
zvout_(&debug_1.logfil, &j, &h__[j * h_dim1 + 1], &debug_1.ndigit,
"_naitr: j-th column of H", (ftnlen)24);
}
/* %----------------------------------------------------% */
/* | Compute V_{j}^T * B * r_{j}. | */
/* | WORKD(IRJ:IRJ+J-1) = v(:,1:J)'*WORKD(IPJ:IPJ+N-1). | */
/* %----------------------------------------------------% */
zgemv_("C", n, &j, &c_b1, &v[v_offset], ldv, &workd[ipj], &c__1, &c_b2, &
workd[irj], &c__1, (ftnlen)1);
/* %---------------------------------------------% */
/* | Compute the correction to the residual: | */
/* | r_{j} = r_{j} - V_{j} * WORKD(IRJ:IRJ+J-1). | */
/* | The correction to H is v(:,1:J)*H(1:J,1:J) | */
/* | + v(:,1:J)*WORKD(IRJ:IRJ+J-1)*e'_j. | */
/* %---------------------------------------------% */
z__1.r = -1., z__1.i = -0.;
zgemv_("N", n, &j, &z__1, &v[v_offset], ldv, &workd[irj], &c__1, &c_b1, &
resid[1], &c__1, (ftnlen)1);
zaxpy_(&j, &c_b1, &workd[irj], &c__1, &h__[j * h_dim1 + 1], &c__1);
orth2 = TRUE_;
second_(&t2);
if (*(unsigned char *)bmat == 'G') {
++timing_1.nbx;
zcopy_(n, &resid[1], &c__1, &workd[irj], &c__1);
ipntr[1] = irj;
ipntr[2] = ipj;
*ido = 2;
/* %-----------------------------------% */
/* | Exit in order to compute B*r_{j}. | */
/* | r_{j} is the corrected residual. | */
/* %-----------------------------------% */
goto L9000;
} else if (*(unsigned char *)bmat == 'I') {
zcopy_(n, &resid[1], &c__1, &workd[ipj], &c__1);
}
L90:
/* %---------------------------------------------------% */
/* | Back from reverse communication if ORTH2 = .true. | */
/* %---------------------------------------------------% */
if (*(unsigned char *)bmat == 'G') {
second_(&t3);
timing_1.tmvbx += t3 - t2;
}
/* %-----------------------------------------------------% */
/* | Compute the B-norm of the corrected residual r_{j}. | */
/* %-----------------------------------------------------% */
if (*(unsigned char *)bmat == 'G') {
zdotc_(&z__1, n, &resid[1], &c__1, &workd[ipj], &c__1);
cnorm.r = z__1.r, cnorm.i = z__1.i;
d__1 = cnorm.r;
d__2 = d_imag(&cnorm);
rnorm1 = sqrt(dlapy2_(&d__1, &d__2));
} else if (*(unsigned char *)bmat == 'I') {
rnorm1 = dznrm2_(n, &resid[1], &c__1);
}
if (msglvl > 0 && iter > 0) {
ivout_(&debug_1.logfil, &c__1, &j, &debug_1.ndigit, "_naitr: Iterati"
"ve refinement for Arnoldi residual", (ftnlen)49);
if (msglvl > 2) {
rtemp[0] = *rnorm;
rtemp[1] = rnorm1;
dvout_(&debug_1.logfil, &c__2, rtemp, &debug_1.ndigit, "_naitr: "
"iterative refinement ; rnorm and rnorm1 are", (ftnlen)51);
}
}
/* %-----------------------------------------% */
/* | Determine if we need to perform another | */
/* | step of re-orthogonalization. | */
/* %-----------------------------------------% */
if (rnorm1 > *rnorm * .717f) {
/* %---------------------------------------% */
/* | No need for further refinement. | */
/* | The cosine of the angle between the | */
/* | corrected residual vector and the old | */
/* | residual vector is greater than 0.717 | */
/* | In other words the corrected residual | */
/* | and the old residual vector share an | */
/* | angle of less than arcCOS(0.717) | */
/* %---------------------------------------% */
*rnorm = rnorm1;
} else {
/* %-------------------------------------------% */
/* | Another step of iterative refinement step | */
/* | is required. NITREF is used by stat.h | */
/* %-------------------------------------------% */
++timing_1.nitref;
*rnorm = rnorm1;
++iter;
if (iter <= 1) {
goto L80;
}
/* %-------------------------------------------------% */
/* | Otherwise RESID is numerically in the span of V | */
/* %-------------------------------------------------% */
i__1 = *n;
for (jj = 1; jj <= i__1; ++jj) {
i__2 = jj;
resid[i__2].r = 0., resid[i__2].i = 0.;
/* L95: */
}
*rnorm = 0.;
}
/* %----------------------------------------------% */
/* | Branch here directly if iterative refinement | */
/* | wasn't necessary or after at most NITER_REF | */
/* | steps of iterative refinement. | */
/* %----------------------------------------------% */
L100:
rstart = FALSE_;
orth2 = FALSE_;
second_(&t5);
timing_1.titref += t5 - t4;
/* %------------------------------------% */
/* | STEP 6: Update j = j+1; Continue | */
/* %------------------------------------% */
++j;
if (j > *k + *np) {
second_(&t1);
timing_1.tcaitr += t1 - t0;
*ido = 99;
i__1 = *k + *np - 1;
for (i__ = max(1,*k); i__ <= i__1; ++i__) {
/* %--------------------------------------------% */
/* | Check for splitting and deflation. | */
/* | Use a standard test as in the QR algorithm | */
/* | REFERENCE: LAPACK subroutine zlahqr | */
/* %--------------------------------------------% */
i__2 = i__ + i__ * h_dim1;
d__1 = h__[i__2].r;
d__2 = d_imag(&h__[i__ + i__ * h_dim1]);
i__3 = i__ + 1 + (i__ + 1) * h_dim1;
d__3 = h__[i__3].r;
d__4 = d_imag(&h__[i__ + 1 + (i__ + 1) * h_dim1]);
tst1 = dlapy2_(&d__1, &d__2) + dlapy2_(&d__3, &d__4);
if (tst1 == 0.) {
i__2 = *k + *np;
tst1 = zlanhs_("1", &i__2, &h__[h_offset], ldh, &workd[*n + 1]
, (ftnlen)1);
}
i__2 = i__ + 1 + i__ * h_dim1;
d__1 = h__[i__2].r;
d__2 = d_imag(&h__[i__ + 1 + i__ * h_dim1]);
/* Computing MAX */
d__3 = ulp * tst1;
if (dlapy2_(&d__1, &d__2) <= max(d__3,smlnum)) {
i__3 = i__ + 1 + i__ * h_dim1;
h__[i__3].r = 0., h__[i__3].i = 0.;
}
/* L110: */
}
if (msglvl > 2) {
i__1 = *k + *np;
i__2 = *k + *np;
zmout_(&debug_1.logfil, &i__1, &i__2, &h__[h_offset], ldh, &
debug_1.ndigit, "_naitr: Final upper Hessenberg matrix H"
" of order K+NP", (ftnlen)53);
}
goto L9000;
}
/* %--------------------------------------------------------% */
/* | Loop back to extend the factorization by another step. | */
/* %--------------------------------------------------------% */
goto L1000;
/* %---------------------------------------------------------------% */
/* | | */
/* | E N D O F M A I N I T E R A T I O N L O O P | */
/* | | */
/* %---------------------------------------------------------------% */
L9000:
return 0;
/* %---------------% */
/* | End of znaitr | */
/* %---------------% */
} /* znaitr_ */
/* Subroutine */
int dlaed8_(integer *icompq, integer *k, integer *n, integer *qsiz, doublereal *d__, doublereal *q, integer *ldq, integer *indxq, doublereal *rho, integer *cutpnt, doublereal *z__, doublereal *dlamda, doublereal *q2, integer *ldq2, doublereal *w, integer *perm, integer *givptr, integer *givcol, doublereal *givnum, integer *indxp, integer *indx, integer *info)
{
/* System generated locals */
integer q_dim1, q_offset, q2_dim1, q2_offset, i__1;
doublereal d__1;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
doublereal c__;
integer i__, j;
doublereal s, t;
integer k2, n1, n2, jp, n1p1;
doublereal eps, tau, tol;
integer jlam, imax, jmax;
extern /* Subroutine */
int drot_(integer *, doublereal *, integer *, doublereal *, integer *, doublereal *, doublereal *), 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 computational 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 .. */
/* Test the input parameters. */
/* Parameter adjustments */
--d__;
q_dim1 = *ldq;
q_offset = 1 + q_dim1;
q -= q_offset;
--indxq;
--z__;
--dlamda;
q2_dim1 = *ldq2;
q2_offset = 1 + q2_dim1;
q2 -= q2_offset;
--w;
--perm;
givcol -= 3;
givnum -= 3;
--indxp;
--indx;
/* Function Body */
*info = 0;
if (*icompq < 0 || *icompq > 1)
{
*info = -1;
}
else if (*n < 0)
{
*info = -3;
}
else if (*icompq == 1 && *qsiz < *n)
{
*info = -4;
}
else if (*ldq < max(1,*n))
{
*info = -7;
}
else if (*cutpnt < min(1,*n) || *cutpnt > *n)
{
*info = -10;
}
else if (*ldq2 < max(1,*n))
{
*info = -14;
}
if (*info != 0)
{
i__1 = -(*info);
xerbla_("DLAED8", &i__1);
return 0;
}
/* Need to initialize GIVPTR to O here in case of quick exit */
/* to prevent an unspecified code behavior (usually sigfault) */
/* when IWORK array on entry to *stedc is not zeroed */
/* (or at least some IWORK entries which used in *laed7 for GIVPTR). */
*givptr = 0;
/* Quick return if possible */
if (*n == 0)
{
return 0;
}
n1 = *cutpnt;
n2 = *n - n1;
n1p1 = n1 + 1;
if (*rho < 0.)
{
dscal_(&n2, &c_b3, &z__[n1p1], &c__1);
}
/* Normalize z so that norm(z) = 1 */
t = 1. / sqrt(2.);
i__1 = *n;
for (j = 1;
j <= i__1;
++j)
{
indx[j] = j;
/* L10: */
}
dscal_(n, &t, &z__[1], &c__1);
*rho = (d__1 = *rho * 2., abs(d__1));
/* Sort the eigenvalues into increasing order */
i__1 = *n;
for (i__ = *cutpnt + 1;
i__ <= i__1;
++i__)
{
indxq[i__] += *cutpnt;
/* L20: */
}
i__1 = *n;
for (i__ = 1;
i__ <= i__1;
++i__)
{
dlamda[i__] = d__[indxq[i__]];
w[i__] = z__[indxq[i__]];
/* L30: */
}
i__ = 1;
j = *cutpnt + 1;
dlamrg_(&n1, &n2, &dlamda[1], &c__1, &c__1, &indx[1]);
i__1 = *n;
for (i__ = 1;
i__ <= i__1;
++i__)
{
d__[i__] = dlamda[indx[i__]];
z__[i__] = w[indx[i__]];
/* L40: */
}
/* Calculate the allowable deflation tolerence */
imax = idamax_(n, &z__[1], &c__1);
jmax = idamax_(n, &d__[1], &c__1);
eps = dlamch_("Epsilon");
tol = eps * 8. * (d__1 = d__[jmax], abs(d__1));
/* 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;
if (*icompq == 0)
{
i__1 = *n;
for (j = 1;
j <= i__1;
++j)
{
perm[j] = indxq[indx[j]];
/* L50: */
}
}
else
{
i__1 = *n;
for (j = 1;
j <= i__1;
++j)
{
perm[j] = indxq[indx[j]];
dcopy_(qsiz, &q[perm[j] * q_dim1 + 1], &c__1, &q2[j * q2_dim1 + 1], &c__1);
/* L60: */
}
dlacpy_("A", qsiz, n, &q2[q2_dim1 + 1], ldq2, &q[q_dim1 + 1], ldq);
}
return 0;
}
/* 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. */
*k = 0;
k2 = *n + 1;
i__1 = *n;
for (j = 1;
j <= i__1;
++j)
{
if (*rho * (d__1 = z__[j], abs(d__1)) <= tol)
{
/* Deflate due to small z component. */
--k2;
indxp[k2] = j;
if (j == *n)
{
goto L110;
}
}
else
{
jlam = j;
goto L80;
}
/* L70: */
}
L80:
++j;
if (j > *n)
{
goto L100;
}
if (*rho * (d__1 = z__[j], abs(d__1)) <= tol)
{
/* Deflate due to small z component. */
--k2;
indxp[k2] = j;
}
else
{
/* Check if eigenvalues are close enough to allow deflation. */
s = z__[jlam];
c__ = z__[j];
/* Find sqrt(a**2+b**2) without overflow or */
/* destructive underflow. */
tau = dlapy2_(&c__, &s);
t = d__[j] - d__[jlam];
c__ /= tau;
s = -s / tau;
if ((d__1 = t * c__ * s, abs(d__1)) <= tol)
{
/* Deflation is possible. */
z__[j] = tau;
z__[jlam] = 0.;
/* Record the appropriate Givens rotation */
++(*givptr);
givcol[(*givptr << 1) + 1] = indxq[indx[jlam]];
givcol[(*givptr << 1) + 2] = indxq[indx[j]];
givnum[(*givptr << 1) + 1] = c__;
givnum[(*givptr << 1) + 2] = s;
if (*icompq == 1)
{
drot_(qsiz, &q[indxq[indx[jlam]] * q_dim1 + 1], &c__1, &q[ indxq[indx[j]] * q_dim1 + 1], &c__1, &c__, &s);
}
t = d__[jlam] * c__ * c__ + d__[j] * s * s;
d__[j] = d__[jlam] * s * s + d__[j] * c__ * c__;
d__[jlam] = t;
--k2;
i__ = 1;
L90:
if (k2 + i__ <= *n)
{
if (d__[jlam] < d__[indxp[k2 + i__]])
{
indxp[k2 + i__ - 1] = indxp[k2 + i__];
indxp[k2 + i__] = jlam;
++i__;
goto L90;
}
else
{
indxp[k2 + i__ - 1] = jlam;
}
}
else
{
indxp[k2 + i__ - 1] = jlam;
}
jlam = j;
}
else
{
++(*k);
w[*k] = z__[jlam];
dlamda[*k] = d__[jlam];
indxp[*k] = jlam;
jlam = j;
}
}
goto L80;
L100: /* Record the last eigenvalue. */
++(*k);
w[*k] = z__[jlam];
dlamda[*k] = d__[jlam];
indxp[*k] = jlam;
L110: /* 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. */
if (*icompq == 0)
{
i__1 = *n;
for (j = 1;
j <= i__1;
++j)
{
jp = indxp[j];
dlamda[j] = d__[jp];
perm[j] = indxq[indx[jp]];
/* L120: */
}
}
else
{
i__1 = *n;
for (j = 1;
j <= i__1;
++j)
{
jp = indxp[j];
dlamda[j] = d__[jp];
perm[j] = indxq[indx[jp]];
dcopy_(qsiz, &q[perm[j] * q_dim1 + 1], &c__1, &q2[j * q2_dim1 + 1] , &c__1);
/* L130: */
}
}
/* The deflated eigenvalues and their corresponding vectors go back */
/* into the last N - K slots of D and Q respectively. */
if (*k < *n)
{
if (*icompq == 0)
{
i__1 = *n - *k;
dcopy_(&i__1, &dlamda[*k + 1], &c__1, &d__[*k + 1], &c__1);
}
else
{
i__1 = *n - *k;
dcopy_(&i__1, &dlamda[*k + 1], &c__1, &d__[*k + 1], &c__1);
i__1 = *n - *k;
dlacpy_("A", qsiz, &i__1, &q2[(*k + 1) * q2_dim1 + 1], ldq2, &q[(* k + 1) * q_dim1 + 1], ldq);
}
}
return 0;
/* End of DLAED8 */
}
/* 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 zlargv_(integer *n, doublecomplex *x, integer *incx,
doublecomplex *y, integer *incy, doublereal *c__, integer *incc)
{
/* -- 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
=======
ZLARGV generates a vector of complex plane rotations with real
cosines, determined by elements of the complex vectors x and y.
For i = 1,2,...,n
( c(i) s(i) ) ( x(i) ) = ( r(i) )
( -conjg(s(i)) c(i) ) ( y(i) ) = ( 0 )
where c(i)**2 + ABS(s(i))**2 = 1
The following conventions are used (these are the same as in ZLARTG,
but differ from the BLAS1 routine ZROTG):
If y(i)=0, then c(i)=1 and s(i)=0.
If x(i)=0, then c(i)=0 and s(i) is chosen so that r(i) is real.
Arguments
=========
N (input) INTEGER
The number of plane rotations to be generated.
X (input/output) COMPLEX*16 array, dimension (1+(N-1)*INCX)
On entry, the vector x.
On exit, x(i) is overwritten by r(i), for i = 1,...,n.
INCX (input) INTEGER
The increment between elements of X. INCX > 0.
Y (input/output) COMPLEX*16 array, dimension (1+(N-1)*INCY)
On entry, the vector y.
On exit, the sines of the plane rotations.
INCY (input) INTEGER
The increment between elements of Y. INCY > 0.
C (output) DOUBLE PRECISION array, dimension (1+(N-1)*INCC)
The cosines of the plane rotations.
INCC (input) INTEGER
The increment between elements of C. INCC > 0.
Further Details
======= =======
6-6-96 - Modified with a new algorithm by W. Kahan and J. Demmel
=====================================================================
Parameter adjustments */
/* Initialized data */
static logical first = TRUE_;
/* System generated locals */
integer i__1, i__2;
doublereal d__1, d__2, d__3, d__4, d__5, d__6, d__7, d__8, d__9, d__10;
doublecomplex z__1, z__2, z__3;
/* Builtin functions */
double log(doublereal), pow_di(doublereal *, integer *), d_imag(
doublecomplex *), sqrt(doublereal);
void d_cnjg(doublecomplex *, doublecomplex *);
/* Local variables */
static doublereal d__;
static doublecomplex f, g;
static integer i__, j;
static doublecomplex r__;
static doublereal scale;
static integer count;
static doublereal f2, g2, safmn2;
extern doublereal dlapy2_(doublereal *, doublereal *);
static doublereal safmx2;
static integer ic;
static doublereal di;
static doublecomplex ff;
static doublereal cs, dr;
extern doublereal dlamch_(char *);
static doublecomplex fs, gs;
static integer ix, iy;
static doublecomplex sn;
static doublereal safmin, f2s, g2s, eps;
--c__;
--y;
--x;
/* Function Body */
if (first) {
first = FALSE_;
safmin = dlamch_("S");
eps = dlamch_("E");
d__1 = dlamch_("B");
i__1 = (integer) (log(safmin / eps) / log(dlamch_("B")) /
2.);
safmn2 = pow_di(&d__1, &i__1);
safmx2 = 1. / safmn2;
}
ix = 1;
iy = 1;
ic = 1;
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = ix;
f.r = x[i__2].r, f.i = x[i__2].i;
i__2 = iy;
g.r = y[i__2].r, g.i = y[i__2].i;
/* Use identical algorithm as in ZLARTG
Computing MAX
Computing MAX */
d__7 = (d__1 = f.r, abs(d__1)), d__8 = (d__2 = d_imag(&f), abs(d__2));
/* Computing MAX */
d__9 = (d__3 = g.r, abs(d__3)), d__10 = (d__4 = d_imag(&g), abs(d__4))
;
d__5 = max(d__7,d__8), d__6 = max(d__9,d__10);
scale = max(d__5,d__6);
fs.r = f.r, fs.i = f.i;
gs.r = g.r, gs.i = g.i;
count = 0;
if (scale >= safmx2) {
L10:
++count;
z__1.r = safmn2 * fs.r, z__1.i = safmn2 * fs.i;
fs.r = z__1.r, fs.i = z__1.i;
z__1.r = safmn2 * gs.r, z__1.i = safmn2 * gs.i;
gs.r = z__1.r, gs.i = z__1.i;
scale *= safmn2;
if (scale >= safmx2) {
goto L10;
}
} else if (scale <= safmn2) {
if (g.r == 0. && g.i == 0.) {
cs = 1.;
sn.r = 0., sn.i = 0.;
r__.r = f.r, r__.i = f.i;
goto L50;
}
L20:
--count;
z__1.r = safmx2 * fs.r, z__1.i = safmx2 * fs.i;
fs.r = z__1.r, fs.i = z__1.i;
z__1.r = safmx2 * gs.r, z__1.i = safmx2 * gs.i;
gs.r = z__1.r, gs.i = z__1.i;
scale *= safmx2;
if (scale <= safmn2) {
goto L20;
}
}
/* Computing 2nd power */
d__1 = fs.r;
/* Computing 2nd power */
d__2 = d_imag(&fs);
f2 = d__1 * d__1 + d__2 * d__2;
/* Computing 2nd power */
d__1 = gs.r;
/* Computing 2nd power */
d__2 = d_imag(&gs);
g2 = d__1 * d__1 + d__2 * d__2;
if (f2 <= max(g2,1.) * safmin) {
/* This is a rare case: F is very small. */
if (f.r == 0. && f.i == 0.) {
cs = 0.;
d__2 = g.r;
d__3 = d_imag(&g);
d__1 = dlapy2_(&d__2, &d__3);
r__.r = d__1, r__.i = 0.;
/* Do complex/real division explicitly with two real
divisions */
d__1 = gs.r;
d__2 = d_imag(&gs);
d__ = dlapy2_(&d__1, &d__2);
d__1 = gs.r / d__;
d__2 = -d_imag(&gs) / d__;
z__1.r = d__1, z__1.i = d__2;
sn.r = z__1.r, sn.i = z__1.i;
goto L50;
}
d__1 = fs.r;
d__2 = d_imag(&fs);
f2s = dlapy2_(&d__1, &d__2);
/* G2 and G2S are accurate
G2 is at least SAFMIN, and G2S is at least SAFMN2 */
g2s = sqrt(g2);
/* Error in CS from underflow in F2S is at most
UNFL / SAFMN2 .lt. sqrt(UNFL*EPS) .lt. EPS
If MAX(G2,ONE)=G2, then F2 .lt. G2*SAFMIN,
and so CS .lt. sqrt(SAFMIN)
If MAX(G2,ONE)=ONE, then F2 .lt. SAFMIN
and so CS .lt. sqrt(SAFMIN)/SAFMN2 = sqrt(EPS)
Therefore, CS = F2S/G2S / sqrt( 1 + (F2S/G2S)**2 ) = F2S/G2S */
cs = f2s / g2s;
/* Make sure abs(FF) = 1
Do complex/real division explicitly with 2 real divisions
Computing MAX */
d__3 = (d__1 = f.r, abs(d__1)), d__4 = (d__2 = d_imag(&f), abs(
d__2));
if (max(d__3,d__4) > 1.) {
d__1 = f.r;
d__2 = d_imag(&f);
d__ = dlapy2_(&d__1, &d__2);
d__1 = f.r / d__;
d__2 = d_imag(&f) / d__;
z__1.r = d__1, z__1.i = d__2;
ff.r = z__1.r, ff.i = z__1.i;
} else {
dr = safmx2 * f.r;
di = safmx2 * d_imag(&f);
d__ = dlapy2_(&dr, &di);
d__1 = dr / d__;
d__2 = di / d__;
z__1.r = d__1, z__1.i = d__2;
ff.r = z__1.r, ff.i = z__1.i;
}
d__1 = gs.r / g2s;
d__2 = -d_imag(&gs) / g2s;
z__2.r = d__1, z__2.i = d__2;
z__1.r = ff.r * z__2.r - ff.i * z__2.i, z__1.i = ff.r * z__2.i +
ff.i * z__2.r;
sn.r = z__1.r, sn.i = z__1.i;
z__2.r = cs * f.r, z__2.i = cs * f.i;
z__3.r = sn.r * g.r - sn.i * g.i, z__3.i = sn.r * g.i + sn.i *
g.r;
z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i;
r__.r = z__1.r, r__.i = z__1.i;
} else {
/* This is the most common case.
Neither F2 nor F2/G2 are less than SAFMIN
F2S cannot overflow, and it is accurate */
f2s = sqrt(g2 / f2 + 1.);
/* Do the F2S(real)*FS(complex) multiply with two real
multiplies */
d__1 = f2s * fs.r;
d__2 = f2s * d_imag(&fs);
z__1.r = d__1, z__1.i = d__2;
r__.r = z__1.r, r__.i = z__1.i;
cs = 1. / f2s;
d__ = f2 + g2;
/* Do complex/real division explicitly with two real divisions */
d__1 = r__.r / d__;
d__2 = d_imag(&r__) / d__;
z__1.r = d__1, z__1.i = d__2;
sn.r = z__1.r, sn.i = z__1.i;
d_cnjg(&z__2, &gs);
z__1.r = sn.r * z__2.r - sn.i * z__2.i, z__1.i = sn.r * z__2.i +
sn.i * z__2.r;
sn.r = z__1.r, sn.i = z__1.i;
if (count != 0) {
if (count > 0) {
i__2 = count;
for (j = 1; j <= i__2; ++j) {
z__1.r = safmx2 * r__.r, z__1.i = safmx2 * r__.i;
r__.r = z__1.r, r__.i = z__1.i;
/* L30: */
}
} else {
i__2 = -count;
for (j = 1; j <= i__2; ++j) {
z__1.r = safmn2 * r__.r, z__1.i = safmn2 * r__.i;
r__.r = z__1.r, r__.i = z__1.i;
/* L40: */
}
}
}
}
L50:
c__[ic] = cs;
i__2 = iy;
y[i__2].r = sn.r, y[i__2].i = sn.i;
i__2 = ix;
x[i__2].r = r__.r, x[i__2].i = r__.i;
ic += *incc;
iy += *incy;
ix += *incx;
/* L60: */
}
return 0;
/* End of ZLARGV */
} /* zlargv_ */
