/* Subroutine */ int igraphdlaexc_(logical *wantq, integer *n, doublereal *t,
integer *ldt, doublereal *q, integer *ldq, integer *j1, integer *n1,
integer *n2, doublereal *work, integer *info)
{
/* System generated locals */
integer q_dim1, q_offset, t_dim1, t_offset, i__1;
doublereal d__1, d__2, d__3;
/* Local variables */
doublereal d__[16] /* was [4][4] */;
integer k;
doublereal u[3], x[4] /* was [2][2] */;
integer j2, j3, j4;
doublereal u1[3], u2[3];
integer nd;
doublereal cs, t11, t22, t33, sn, wi1, wi2, wr1, wr2, eps, tau, tau1,
tau2;
integer ierr;
doublereal temp;
extern /* Subroutine */ int igraphdrot_(integer *, doublereal *, integer *,
doublereal *, integer *, doublereal *, doublereal *);
doublereal scale, dnorm, xnorm;
extern /* Subroutine */ int igraphdlanv2_(doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *, doublereal *), igraphdlasy2_(
logical *, logical *, integer *, integer *, integer *, doublereal
*, integer *, doublereal *, integer *, doublereal *, integer *,
doublereal *, doublereal *, integer *, doublereal *, integer *);
extern doublereal igraphdlamch_(char *), igraphdlange_(char *, integer *,
integer *, doublereal *, integer *, doublereal *);
extern /* Subroutine */ int igraphdlarfg_(integer *, doublereal *, doublereal *,
integer *, doublereal *), igraphdlacpy_(char *, integer *, integer *,
doublereal *, integer *, doublereal *, integer *),
igraphdlartg_(doublereal *, doublereal *, doublereal *, doublereal *,
doublereal *), igraphdlarfx_(char *, integer *, integer *, doublereal *,
doublereal *, doublereal *, integer *, doublereal *);
doublereal thresh, smlnum;
/* -- LAPACK auxiliary routine (version 3.2.2) --
-- LAPACK is a software package provided by Univ. of Tennessee, --
-- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
June 2010
Purpose
=======
DLAEXC swaps adjacent diagonal blocks T11 and T22 of order 1 or 2 in
an upper quasi-triangular matrix T by an orthogonal similarity
transformation.
T must be in Schur canonical form, that is, block upper triangular
with 1-by-1 and 2-by-2 diagonal blocks; each 2-by-2 diagonal block
has its diagonal elemnts equal and its off-diagonal elements of
opposite sign.
Arguments
=========
WANTQ (input) LOGICAL
= .TRUE. : accumulate the transformation in the matrix Q;
= .FALSE.: do not accumulate the transformation.
N (input) INTEGER
The order of the matrix T. N >= 0.
T (input/output) DOUBLE PRECISION array, dimension (LDT,N)
On entry, the upper quasi-triangular matrix T, in Schur
canonical form.
On exit, the updated matrix T, again in Schur canonical form.
LDT (input) INTEGER
The leading dimension of the array T. LDT >= max(1,N).
Q (input/output) DOUBLE PRECISION array, dimension (LDQ,N)
On entry, if WANTQ is .TRUE., the orthogonal matrix Q.
On exit, if WANTQ is .TRUE., the updated matrix Q.
If WANTQ is .FALSE., Q is not referenced.
LDQ (input) INTEGER
The leading dimension of the array Q.
LDQ >= 1; and if WANTQ is .TRUE., LDQ >= N.
J1 (input) INTEGER
The index of the first row of the first block T11.
N1 (input) INTEGER
The order of the first block T11. N1 = 0, 1 or 2.
N2 (input) INTEGER
The order of the second block T22. N2 = 0, 1 or 2.
WORK (workspace) DOUBLE PRECISION array, dimension (N)
INFO (output) INTEGER
= 0: successful exit
= 1: the transformed matrix T would be too far from Schur
form; the blocks are not swapped and T and Q are
unchanged.
=====================================================================
Parameter adjustments */
t_dim1 = *ldt;
t_offset = 1 + t_dim1;
t -= t_offset;
q_dim1 = *ldq;
q_offset = 1 + q_dim1;
q -= q_offset;
--work;
/* Function Body */
*info = 0;
/* Quick return if possible */
if (*n == 0 || *n1 == 0 || *n2 == 0) {
return 0;
}
if (*j1 + *n1 > *n) {
return 0;
}
j2 = *j1 + 1;
j3 = *j1 + 2;
j4 = *j1 + 3;
if (*n1 == 1 && *n2 == 1) {
/* Swap two 1-by-1 blocks. */
t11 = t[*j1 + *j1 * t_dim1];
t22 = t[j2 + j2 * t_dim1];
/* Determine the transformation to perform the interchange. */
d__1 = t22 - t11;
igraphdlartg_(&t[*j1 + j2 * t_dim1], &d__1, &cs, &sn, &temp);
/* Apply transformation to the matrix T. */
if (j3 <= *n) {
i__1 = *n - *j1 - 1;
igraphdrot_(&i__1, &t[*j1 + j3 * t_dim1], ldt, &t[j2 + j3 * t_dim1],
ldt, &cs, &sn);
}
i__1 = *j1 - 1;
igraphdrot_(&i__1, &t[*j1 * t_dim1 + 1], &c__1, &t[j2 * t_dim1 + 1], &c__1,
&cs, &sn);
t[*j1 + *j1 * t_dim1] = t22;
t[j2 + j2 * t_dim1] = t11;
if (*wantq) {
/* Accumulate transformation in the matrix Q. */
igraphdrot_(n, &q[*j1 * q_dim1 + 1], &c__1, &q[j2 * q_dim1 + 1], &c__1,
&cs, &sn);
}
} else {
/* Swapping involves at least one 2-by-2 block.
Copy the diagonal block of order N1+N2 to the local array D
and compute its norm. */
nd = *n1 + *n2;
igraphdlacpy_("Full", &nd, &nd, &t[*j1 + *j1 * t_dim1], ldt, d__, &c__4);
dnorm = igraphdlange_("Max", &nd, &nd, d__, &c__4, &work[1]);
/* Compute machine-dependent threshold for test for accepting
swap. */
eps = igraphdlamch_("P");
smlnum = igraphdlamch_("S") / eps;
/* Computing MAX */
d__1 = eps * 10. * dnorm;
thresh = max(d__1,smlnum);
/* Solve T11*X - X*T22 = scale*T12 for X. */
igraphdlasy2_(&c_false, &c_false, &c_n1, n1, n2, d__, &c__4, &d__[*n1 + 1 +
(*n1 + 1 << 2) - 5], &c__4, &d__[(*n1 + 1 << 2) - 4], &c__4, &
scale, x, &c__2, &xnorm, &ierr);
/* Swap the adjacent diagonal blocks. */
k = *n1 + *n1 + *n2 - 3;
switch (k) {
case 1: goto L10;
case 2: goto L20;
case 3: goto L30;
}
L10:
/* N1 = 1, N2 = 2: generate elementary reflector H so that:
( scale, X11, X12 ) H = ( 0, 0, * ) */
u[0] = scale;
u[1] = x[0];
u[2] = x[2];
igraphdlarfg_(&c__3, &u[2], u, &c__1, &tau);
u[2] = 1.;
t11 = t[*j1 + *j1 * t_dim1];
/* Perform swap provisionally on diagonal block in D. */
igraphdlarfx_("L", &c__3, &c__3, u, &tau, d__, &c__4, &work[1]);
igraphdlarfx_("R", &c__3, &c__3, u, &tau, d__, &c__4, &work[1]);
/* Test whether to reject swap.
Computing MAX */
d__2 = abs(d__[2]), d__3 = abs(d__[6]), d__2 = max(d__2,d__3), d__3 =
(d__1 = d__[10] - t11, abs(d__1));
if (max(d__2,d__3) > thresh) {
goto L50;
}
/* Accept swap: apply transformation to the entire matrix T. */
i__1 = *n - *j1 + 1;
igraphdlarfx_("L", &c__3, &i__1, u, &tau, &t[*j1 + *j1 * t_dim1], ldt, &
work[1]);
igraphdlarfx_("R", &j2, &c__3, u, &tau, &t[*j1 * t_dim1 + 1], ldt, &work[1]);
t[j3 + *j1 * t_dim1] = 0.;
t[j3 + j2 * t_dim1] = 0.;
t[j3 + j3 * t_dim1] = t11;
if (*wantq) {
/* Accumulate transformation in the matrix Q. */
igraphdlarfx_("R", n, &c__3, u, &tau, &q[*j1 * q_dim1 + 1], ldq, &work[
1]);
}
goto L40;
L20:
/* N1 = 2, N2 = 1: generate elementary reflector H so that:
H ( -X11 ) = ( * )
( -X21 ) = ( 0 )
( scale ) = ( 0 ) */
u[0] = -x[0];
u[1] = -x[1];
u[2] = scale;
igraphdlarfg_(&c__3, u, &u[1], &c__1, &tau);
u[0] = 1.;
t33 = t[j3 + j3 * t_dim1];
/* Perform swap provisionally on diagonal block in D. */
igraphdlarfx_("L", &c__3, &c__3, u, &tau, d__, &c__4, &work[1]);
igraphdlarfx_("R", &c__3, &c__3, u, &tau, d__, &c__4, &work[1]);
/* Test whether to reject swap.
Computing MAX */
d__2 = abs(d__[1]), d__3 = abs(d__[2]), d__2 = max(d__2,d__3), d__3 =
(d__1 = d__[0] - t33, abs(d__1));
if (max(d__2,d__3) > thresh) {
goto L50;
}
/* Accept swap: apply transformation to the entire matrix T. */
igraphdlarfx_("R", &j3, &c__3, u, &tau, &t[*j1 * t_dim1 + 1], ldt, &work[1]);
i__1 = *n - *j1;
igraphdlarfx_("L", &c__3, &i__1, u, &tau, &t[*j1 + j2 * t_dim1], ldt, &work[
1]);
t[*j1 + *j1 * t_dim1] = t33;
t[j2 + *j1 * t_dim1] = 0.;
t[j3 + *j1 * t_dim1] = 0.;
if (*wantq) {
/* Accumulate transformation in the matrix Q. */
igraphdlarfx_("R", n, &c__3, u, &tau, &q[*j1 * q_dim1 + 1], ldq, &work[
1]);
}
goto L40;
L30:
/* N1 = 2, N2 = 2: generate elementary reflectors H(1) and H(2) so
that:
H(2) H(1) ( -X11 -X12 ) = ( * * )
( -X21 -X22 ) ( 0 * )
( scale 0 ) ( 0 0 )
( 0 scale ) ( 0 0 ) */
u1[0] = -x[0];
u1[1] = -x[1];
u1[2] = scale;
igraphdlarfg_(&c__3, u1, &u1[1], &c__1, &tau1);
u1[0] = 1.;
temp = -tau1 * (x[2] + u1[1] * x[3]);
u2[0] = -temp * u1[1] - x[3];
u2[1] = -temp * u1[2];
u2[2] = scale;
igraphdlarfg_(&c__3, u2, &u2[1], &c__1, &tau2);
u2[0] = 1.;
/* Perform swap provisionally on diagonal block in D. */
igraphdlarfx_("L", &c__3, &c__4, u1, &tau1, d__, &c__4, &work[1])
;
igraphdlarfx_("R", &c__4, &c__3, u1, &tau1, d__, &c__4, &work[1])
;
igraphdlarfx_("L", &c__3, &c__4, u2, &tau2, &d__[1], &c__4, &work[1]);
igraphdlarfx_("R", &c__4, &c__3, u2, &tau2, &d__[4], &c__4, &work[1]);
/* Test whether to reject swap.
Computing MAX */
d__1 = abs(d__[2]), d__2 = abs(d__[6]), d__1 = max(d__1,d__2), d__2 =
abs(d__[3]), d__1 = max(d__1,d__2), d__2 = abs(d__[7]);
if (max(d__1,d__2) > thresh) {
goto L50;
}
/* Accept swap: apply transformation to the entire matrix T. */
i__1 = *n - *j1 + 1;
igraphdlarfx_("L", &c__3, &i__1, u1, &tau1, &t[*j1 + *j1 * t_dim1], ldt, &
work[1]);
igraphdlarfx_("R", &j4, &c__3, u1, &tau1, &t[*j1 * t_dim1 + 1], ldt, &work[
1]);
i__1 = *n - *j1 + 1;
igraphdlarfx_("L", &c__3, &i__1, u2, &tau2, &t[j2 + *j1 * t_dim1], ldt, &
work[1]);
igraphdlarfx_("R", &j4, &c__3, u2, &tau2, &t[j2 * t_dim1 + 1], ldt, &work[1]
);
t[j3 + *j1 * t_dim1] = 0.;
t[j3 + j2 * t_dim1] = 0.;
t[j4 + *j1 * t_dim1] = 0.;
t[j4 + j2 * t_dim1] = 0.;
if (*wantq) {
/* Accumulate transformation in the matrix Q. */
igraphdlarfx_("R", n, &c__3, u1, &tau1, &q[*j1 * q_dim1 + 1], ldq, &
work[1]);
igraphdlarfx_("R", n, &c__3, u2, &tau2, &q[j2 * q_dim1 + 1], ldq, &work[
1]);
}
L40:
if (*n2 == 2) {
/* Standardize new 2-by-2 block T11 */
igraphdlanv2_(&t[*j1 + *j1 * t_dim1], &t[*j1 + j2 * t_dim1], &t[j2 + *
j1 * t_dim1], &t[j2 + j2 * t_dim1], &wr1, &wi1, &wr2, &
wi2, &cs, &sn);
i__1 = *n - *j1 - 1;
igraphdrot_(&i__1, &t[*j1 + (*j1 + 2) * t_dim1], ldt, &t[j2 + (*j1 + 2)
* t_dim1], ldt, &cs, &sn);
i__1 = *j1 - 1;
igraphdrot_(&i__1, &t[*j1 * t_dim1 + 1], &c__1, &t[j2 * t_dim1 + 1], &
c__1, &cs, &sn);
if (*wantq) {
igraphdrot_(n, &q[*j1 * q_dim1 + 1], &c__1, &q[j2 * q_dim1 + 1], &
c__1, &cs, &sn);
}
}
if (*n1 == 2) {
/* Standardize new 2-by-2 block T22 */
j3 = *j1 + *n2;
j4 = j3 + 1;
igraphdlanv2_(&t[j3 + j3 * t_dim1], &t[j3 + j4 * t_dim1], &t[j4 + j3 *
t_dim1], &t[j4 + j4 * t_dim1], &wr1, &wi1, &wr2, &wi2, &
cs, &sn);
if (j3 + 2 <= *n) {
i__1 = *n - j3 - 1;
igraphdrot_(&i__1, &t[j3 + (j3 + 2) * t_dim1], ldt, &t[j4 + (j3 + 2)
* t_dim1], ldt, &cs, &sn);
}
i__1 = j3 - 1;
igraphdrot_(&i__1, &t[j3 * t_dim1 + 1], &c__1, &t[j4 * t_dim1 + 1], &
c__1, &cs, &sn);
if (*wantq) {
igraphdrot_(n, &q[j3 * q_dim1 + 1], &c__1, &q[j4 * q_dim1 + 1], &
c__1, &cs, &sn);
}
}
}
return 0;
/* Exit with INFO = 1 if swap was rejected. */
L50:
*info = 1;
return 0;
/* End of DLAEXC */
} /* igraphdlaexc_ */
Subroutine */ int igraphdlaqr0_(logical *wantt, logical *wantz, integer *n,
integer *ilo, integer *ihi, doublereal *h__, integer *ldh, doublereal
*wr, doublereal *wi, integer *iloz, integer *ihiz, doublereal *z__,
integer *ldz, doublereal *work, integer *lwork, integer *info)
{
/* System generated locals */
integer h_dim1, h_offset, z_dim1, z_offset, i__1, i__2, i__3, i__4, i__5;
doublereal d__1, d__2, d__3, d__4;
/* Local variables */
integer i__, k;
doublereal aa, bb, cc, dd;
integer ld;
doublereal cs;
integer nh, it, ks, kt;
doublereal sn;
integer ku, kv, ls, ns;
doublereal ss;
integer nw, inf, kdu, nho, nve, kwh, nsr, nwr, kwv, ndec, ndfl, kbot,
nmin;
doublereal swap;
integer ktop;
doublereal zdum[1] /* was [1][1] */;
integer kacc22, itmax, nsmax, nwmax, kwtop;
extern /* Subroutine */ int igraphdlanv2_(doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *, doublereal *), igraphdlaqr3_(
logical *, logical *, integer *, integer *, integer *, integer *,
doublereal *, integer *, integer *, integer *, doublereal *,
integer *, integer *, integer *, doublereal *, doublereal *,
doublereal *, integer *, integer *, doublereal *, integer *,
integer *, doublereal *, integer *, doublereal *, integer *),
igraphdlaqr4_(logical *, logical *, integer *, integer *, integer *,
doublereal *, integer *, doublereal *, doublereal *, integer *,
integer *, doublereal *, integer *, doublereal *, integer *,
integer *), igraphdlaqr5_(logical *, logical *, integer *, integer *,
integer *, integer *, integer *, doublereal *, doublereal *,
doublereal *, integer *, integer *, integer *, doublereal *,
integer *, doublereal *, integer *, doublereal *, integer *,
integer *, doublereal *, integer *, integer *, doublereal *,
integer *);
integer nibble;
extern /* Subroutine */ int igraphdlahqr_(logical *, logical *, integer *,
integer *, integer *, doublereal *, integer *, doublereal *,
doublereal *, integer *, integer *, doublereal *, integer *,
integer *), igraphdlacpy_(char *, integer *, integer *, doublereal *,
integer *, doublereal *, integer *);
extern integer igraphilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *, ftnlen, ftnlen);
char jbcmpz[2];
integer nwupbd;
logical sorted;
integer lwkopt;
/* -- 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
================================================================
==== Matrices of order NTINY or smaller must be processed by
. DLAHQR because of insufficient subdiagonal scratch space.
. (This is a hard limit.) ====
==== Exceptional deflation windows: try to cure rare
. slow convergence by varying the size of the
. deflation window after KEXNW iterations. ====
==== Exceptional shifts: try to cure rare slow convergence
. with ad-hoc exceptional shifts every KEXSH iterations.
. ====
==== The constants WILK1 and WILK2 are used to form the
. exceptional shifts. ====
Parameter adjustments */
h_dim1 = *ldh;
h_offset = 1 + h_dim1;
h__ -= h_offset;
--wr;
--wi;
z_dim1 = *ldz;
z_offset = 1 + z_dim1;
z__ -= z_offset;
--work;
/* Function Body */
*info = 0;
/* ==== Quick return for N = 0: nothing to do. ==== */
if (*n == 0) {
work[1] = 1.;
return 0;
}
if (*n <= 11) {
/* ==== Tiny matrices must use DLAHQR. ==== */
lwkopt = 1;
if (*lwork != -1) {
igraphdlahqr_(wantt, wantz, n, ilo, ihi, &h__[h_offset], ldh, &wr[1], &
wi[1], iloz, ihiz, &z__[z_offset], ldz, info);
}
} else {
/* ==== Use small bulge multi-shift QR with aggressive early
. deflation on larger-than-tiny matrices. ====
==== Hope for the best. ==== */
*info = 0;
/* ==== Set up job flags for ILAENV. ==== */
if (*wantt) {
*(unsigned char *)jbcmpz = 'S';
} else {
*(unsigned char *)jbcmpz = 'E';
}
if (*wantz) {
*(unsigned char *)&jbcmpz[1] = 'V';
} else {
*(unsigned char *)&jbcmpz[1] = 'N';
}
/* ==== NWR = recommended deflation window size. At this
. point, N .GT. NTINY = 11, so there is enough
. subdiagonal workspace for NWR.GE.2 as required.
. (In fact, there is enough subdiagonal space for
. NWR.GE.3.) ==== */
nwr = igraphilaenv_(&c__13, "DLAQR0", jbcmpz, n, ilo, ihi, lwork, (ftnlen)6,
(ftnlen)2);
nwr = max(2,nwr);
/* Computing MIN */
i__1 = *ihi - *ilo + 1, i__2 = (*n - 1) / 3, i__1 = min(i__1,i__2);
nwr = min(i__1,nwr);
/* ==== NSR = recommended number of simultaneous shifts.
. At this point N .GT. NTINY = 11, so there is at
. enough subdiagonal workspace for NSR to be even
. and greater than or equal to two as required. ==== */
nsr = igraphilaenv_(&c__15, "DLAQR0", jbcmpz, n, ilo, ihi, lwork, (ftnlen)6,
(ftnlen)2);
/* Computing MIN */
i__1 = nsr, i__2 = (*n + 6) / 9, i__1 = min(i__1,i__2), i__2 = *ihi -
*ilo;
nsr = min(i__1,i__2);
/* Computing MAX */
i__1 = 2, i__2 = nsr - nsr % 2;
nsr = max(i__1,i__2);
/* ==== Estimate optimal workspace ====
==== Workspace query call to DLAQR3 ==== */
i__1 = nwr + 1;
igraphdlaqr3_(wantt, wantz, n, ilo, ihi, &i__1, &h__[h_offset], ldh, iloz,
ihiz, &z__[z_offset], ldz, &ls, &ld, &wr[1], &wi[1], &h__[
h_offset], ldh, n, &h__[h_offset], ldh, n, &h__[h_offset],
ldh, &work[1], &c_n1);
/* ==== Optimal workspace = MAX(DLAQR5, DLAQR3) ====
Computing MAX */
i__1 = nsr * 3 / 2, i__2 = (integer) work[1];
lwkopt = max(i__1,i__2);
/* ==== Quick return in case of workspace query. ==== */
if (*lwork == -1) {
work[1] = (doublereal) lwkopt;
return 0;
}
/* ==== DLAHQR/DLAQR0 crossover point ==== */
nmin = igraphilaenv_(&c__12, "DLAQR0", jbcmpz, n, ilo, ihi, lwork, (ftnlen)
6, (ftnlen)2);
nmin = max(11,nmin);
/* ==== Nibble crossover point ==== */
nibble = igraphilaenv_(&c__14, "DLAQR0", jbcmpz, n, ilo, ihi, lwork, (
ftnlen)6, (ftnlen)2);
nibble = max(0,nibble);
/* ==== Accumulate reflections during ttswp? Use block
. 2-by-2 structure during matrix-matrix multiply? ==== */
kacc22 = igraphilaenv_(&c__16, "DLAQR0", jbcmpz, n, ilo, ihi, lwork, (
ftnlen)6, (ftnlen)2);
kacc22 = max(0,kacc22);
kacc22 = min(2,kacc22);
/* ==== NWMAX = the largest possible deflation window for
. which there is sufficient workspace. ====
Computing MIN */
i__1 = (*n - 1) / 3, i__2 = *lwork / 2;
nwmax = min(i__1,i__2);
nw = nwmax;
/* ==== NSMAX = the Largest number of simultaneous shifts
. for which there is sufficient workspace. ====
Computing MIN */
i__1 = (*n + 6) / 9, i__2 = (*lwork << 1) / 3;
nsmax = min(i__1,i__2);
nsmax -= nsmax % 2;
/* ==== NDFL: an iteration count restarted at deflation. ==== */
ndfl = 1;
/* ==== ITMAX = iteration limit ====
Computing MAX */
i__1 = 10, i__2 = *ihi - *ilo + 1;
itmax = max(i__1,i__2) * 30;
/* ==== Last row and column in the active block ==== */
kbot = *ihi;
/* ==== Main Loop ==== */
i__1 = itmax;
for (it = 1; it <= i__1; ++it) {
/* ==== Done when KBOT falls below ILO ==== */
if (kbot < *ilo) {
goto L90;
}
/* ==== Locate active block ==== */
i__2 = *ilo + 1;
for (k = kbot; k >= i__2; --k) {
if (h__[k + (k - 1) * h_dim1] == 0.) {
goto L20;
}
/* L10: */
}
k = *ilo;
L20:
ktop = k;
/* ==== Select deflation window size:
. Typical Case:
. If possible and advisable, nibble the entire
. active block. If not, use size MIN(NWR,NWMAX)
. or MIN(NWR+1,NWMAX) depending upon which has
. the smaller corresponding subdiagonal entry
. (a heuristic).
.
. Exceptional Case:
. If there have been no deflations in KEXNW or
. more iterations, then vary the deflation window
. size. At first, because, larger windows are,
. in general, more powerful than smaller ones,
. rapidly increase the window to the maximum possible.
. Then, gradually reduce the window size. ==== */
nh = kbot - ktop + 1;
nwupbd = min(nh,nwmax);
if (ndfl < 5) {
nw = min(nwupbd,nwr);
} else {
/* Computing MIN */
i__2 = nwupbd, i__3 = nw << 1;
nw = min(i__2,i__3);
}
if (nw < nwmax) {
if (nw >= nh - 1) {
nw = nh;
} else {
kwtop = kbot - nw + 1;
if ((d__1 = h__[kwtop + (kwtop - 1) * h_dim1], abs(d__1))
> (d__2 = h__[kwtop - 1 + (kwtop - 2) * h_dim1],
abs(d__2))) {
++nw;
}
}
}
if (ndfl < 5) {
ndec = -1;
} else if (ndec >= 0 || nw >= nwupbd) {
++ndec;
if (nw - ndec < 2) {
ndec = 0;
}
nw -= ndec;
}
/* ==== Aggressive early deflation:
. split workspace under the subdiagonal into
. - an nw-by-nw work array V in the lower
. left-hand-corner,
. - an NW-by-at-least-NW-but-more-is-better
. (NW-by-NHO) horizontal work array along
. the bottom edge,
. - an at-least-NW-but-more-is-better (NHV-by-NW)
. vertical work array along the left-hand-edge.
. ==== */
kv = *n - nw + 1;
kt = nw + 1;
nho = *n - nw - 1 - kt + 1;
kwv = nw + 2;
nve = *n - nw - kwv + 1;
/* ==== Aggressive early deflation ==== */
igraphdlaqr3_(wantt, wantz, n, &ktop, &kbot, &nw, &h__[h_offset], ldh,
iloz, ihiz, &z__[z_offset], ldz, &ls, &ld, &wr[1], &wi[1],
&h__[kv + h_dim1], ldh, &nho, &h__[kv + kt * h_dim1],
ldh, &nve, &h__[kwv + h_dim1], ldh, &work[1], lwork);
/* ==== Adjust KBOT accounting for new deflations. ==== */
kbot -= ld;
/* ==== KS points to the shifts. ==== */
ks = kbot - ls + 1;
/* ==== Skip an expensive QR sweep if there is a (partly
. heuristic) reason to expect that many eigenvalues
. will deflate without it. Here, the QR sweep is
. skipped if many eigenvalues have just been deflated
. or if the remaining active block is small. */
if (ld == 0 || ld * 100 <= nw * nibble && kbot - ktop + 1 > min(
nmin,nwmax)) {
/* ==== NS = nominal number of simultaneous shifts.
. This may be lowered (slightly) if DLAQR3
. did not provide that many shifts. ====
Computing MIN
Computing MAX */
i__4 = 2, i__5 = kbot - ktop;
i__2 = min(nsmax,nsr), i__3 = max(i__4,i__5);
ns = min(i__2,i__3);
ns -= ns % 2;
/* ==== If there have been no deflations
. in a multiple of KEXSH iterations,
. then try exceptional shifts.
. Otherwise use shifts provided by
. DLAQR3 above or from the eigenvalues
. of a trailing principal submatrix. ==== */
if (ndfl % 6 == 0) {
ks = kbot - ns + 1;
/* Computing MAX */
i__3 = ks + 1, i__4 = ktop + 2;
i__2 = max(i__3,i__4);
for (i__ = kbot; i__ >= i__2; i__ += -2) {
ss = (d__1 = h__[i__ + (i__ - 1) * h_dim1], abs(d__1))
+ (d__2 = h__[i__ - 1 + (i__ - 2) * h_dim1],
abs(d__2));
aa = ss * .75 + h__[i__ + i__ * h_dim1];
bb = ss;
cc = ss * -.4375;
dd = aa;
igraphdlanv2_(&aa, &bb, &cc, &dd, &wr[i__ - 1], &wi[i__ - 1]
, &wr[i__], &wi[i__], &cs, &sn);
/* L30: */
}
if (ks == ktop) {
wr[ks + 1] = h__[ks + 1 + (ks + 1) * h_dim1];
wi[ks + 1] = 0.;
wr[ks] = wr[ks + 1];
wi[ks] = wi[ks + 1];
}
} else {
/* ==== Got NS/2 or fewer shifts? Use DLAQR4 or
. DLAHQR on a trailing principal submatrix to
. get more. (Since NS.LE.NSMAX.LE.(N+6)/9,
. there is enough space below the subdiagonal
. to fit an NS-by-NS scratch array.) ==== */
if (kbot - ks + 1 <= ns / 2) {
ks = kbot - ns + 1;
kt = *n - ns + 1;
igraphdlacpy_("A", &ns, &ns, &h__[ks + ks * h_dim1], ldh, &
h__[kt + h_dim1], ldh);
if (ns > nmin) {
igraphdlaqr4_(&c_false, &c_false, &ns, &c__1, &ns, &h__[
kt + h_dim1], ldh, &wr[ks], &wi[ks], &
c__1, &c__1, zdum, &c__1, &work[1], lwork,
&inf);
} else {
igraphdlahqr_(&c_false, &c_false, &ns, &c__1, &ns, &h__[
kt + h_dim1], ldh, &wr[ks], &wi[ks], &
c__1, &c__1, zdum, &c__1, &inf);
}
ks += inf;
/* ==== In case of a rare QR failure use
. eigenvalues of the trailing 2-by-2
. principal submatrix. ==== */
if (ks >= kbot) {
aa = h__[kbot - 1 + (kbot - 1) * h_dim1];
cc = h__[kbot + (kbot - 1) * h_dim1];
bb = h__[kbot - 1 + kbot * h_dim1];
dd = h__[kbot + kbot * h_dim1];
igraphdlanv2_(&aa, &bb, &cc, &dd, &wr[kbot - 1], &wi[
kbot - 1], &wr[kbot], &wi[kbot], &cs, &sn)
;
ks = kbot - 1;
}
}
if (kbot - ks + 1 > ns) {
/* ==== Sort the shifts (Helps a little)
. Bubble sort keeps complex conjugate
. pairs together. ==== */
sorted = FALSE_;
i__2 = ks + 1;
for (k = kbot; k >= i__2; --k) {
if (sorted) {
goto L60;
}
sorted = TRUE_;
i__3 = k - 1;
for (i__ = ks; i__ <= i__3; ++i__) {
if ((d__1 = wr[i__], abs(d__1)) + (d__2 = wi[
i__], abs(d__2)) < (d__3 = wr[i__ + 1]
, abs(d__3)) + (d__4 = wi[i__ + 1],
abs(d__4))) {
sorted = FALSE_;
swap = wr[i__];
wr[i__] = wr[i__ + 1];
wr[i__ + 1] = swap;
swap = wi[i__];
wi[i__] = wi[i__ + 1];
wi[i__ + 1] = swap;
}
/* L40: */
}
/* L50: */
}
L60:
;
}
/* ==== Shuffle shifts into pairs of real shifts
. and pairs of complex conjugate shifts
. assuming complex conjugate shifts are
. already adjacent to one another. (Yes,
. they are.) ==== */
i__2 = ks + 2;
for (i__ = kbot; i__ >= i__2; i__ += -2) {
if (wi[i__] != -wi[i__ - 1]) {
swap = wr[i__];
wr[i__] = wr[i__ - 1];
wr[i__ - 1] = wr[i__ - 2];
wr[i__ - 2] = swap;
swap = wi[i__];
wi[i__] = wi[i__ - 1];
wi[i__ - 1] = wi[i__ - 2];
wi[i__ - 2] = swap;
}
/* L70: */
}
}
/* ==== If there are only two shifts and both are
. real, then use only one. ==== */
if (kbot - ks + 1 == 2) {
if (wi[kbot] == 0.) {
if ((d__1 = wr[kbot] - h__[kbot + kbot * h_dim1], abs(
d__1)) < (d__2 = wr[kbot - 1] - h__[kbot +
kbot * h_dim1], abs(d__2))) {
wr[kbot - 1] = wr[kbot];
} else {
wr[kbot] = wr[kbot - 1];
}
}
}
/* ==== Use up to NS of the the smallest magnatiude
. shifts. If there aren't NS shifts available,
. then use them all, possibly dropping one to
. make the number of shifts even. ====
Computing MIN */
i__2 = ns, i__3 = kbot - ks + 1;
ns = min(i__2,i__3);
ns -= ns % 2;
ks = kbot - ns + 1;
/* ==== Small-bulge multi-shift QR sweep:
. split workspace under the subdiagonal into
. - a KDU-by-KDU work array U in the lower
. left-hand-corner,
. - a KDU-by-at-least-KDU-but-more-is-better
. (KDU-by-NHo) horizontal work array WH along
. the bottom edge,
. - and an at-least-KDU-but-more-is-better-by-KDU
. (NVE-by-KDU) vertical work WV arrow along
. the left-hand-edge. ==== */
kdu = ns * 3 - 3;
ku = *n - kdu + 1;
kwh = kdu + 1;
nho = *n - kdu - 3 - (kdu + 1) + 1;
kwv = kdu + 4;
nve = *n - kdu - kwv + 1;
/* ==== Small-bulge multi-shift QR sweep ==== */
igraphdlaqr5_(wantt, wantz, &kacc22, n, &ktop, &kbot, &ns, &wr[ks],
&wi[ks], &h__[h_offset], ldh, iloz, ihiz, &z__[
z_offset], ldz, &work[1], &c__3, &h__[ku + h_dim1],
ldh, &nve, &h__[kwv + h_dim1], ldh, &nho, &h__[ku +
kwh * h_dim1], ldh);
}
/* ==== Note progress (or the lack of it). ==== */
if (ld > 0) {
ndfl = 1;
} else {
++ndfl;
}
/* ==== End of main loop ====
L80: */
}
/* ==== Iteration limit exceeded. Set INFO to show where
. the problem occurred and exit. ==== */
*info = kbot;
L90:
;
}
/* ==== Return the optimal value of LWORK. ==== */
work[1] = (doublereal) lwkopt;
/* ==== End of DLAQR0 ==== */
return 0;
} /* igraphdlaqr0_ */
/* Subroutine */ int igraphdlaqr2_(logical *wantt, logical *wantz, integer *n,
integer *ktop, integer *kbot, integer *nw, doublereal *h__, integer *
ldh, integer *iloz, integer *ihiz, doublereal *z__, integer *ldz,
integer *ns, integer *nd, doublereal *sr, doublereal *si, doublereal *
v, integer *ldv, integer *nh, doublereal *t, integer *ldt, integer *
nv, doublereal *wv, integer *ldwv, doublereal *work, integer *lwork)
{
/* System generated locals */
integer h_dim1, h_offset, t_dim1, t_offset, v_dim1, v_offset, wv_dim1,
wv_offset, z_dim1, z_offset, i__1, i__2, i__3, i__4;
doublereal d__1, d__2, d__3, d__4, d__5, d__6;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
integer i__, j, k;
doublereal s, aa, bb, cc, dd, cs, sn;
integer jw;
doublereal evi, evk, foo;
integer kln;
doublereal tau, ulp;
integer lwk1, lwk2;
doublereal beta;
integer kend, kcol, info, ifst, ilst, ltop, krow;
extern /* Subroutine */ int igraphdlarf_(char *, integer *, integer *,
doublereal *, integer *, doublereal *, doublereal *, integer *,
doublereal *), igraphdgemm_(char *, char *, integer *, integer *
, integer *, doublereal *, doublereal *, integer *, doublereal *,
integer *, doublereal *, doublereal *, integer *);
logical bulge;
extern /* Subroutine */ int igraphdcopy_(integer *, doublereal *, integer *,
doublereal *, integer *);
integer infqr, kwtop;
extern /* Subroutine */ int igraphdlanv2_(doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *, doublereal *), igraphdlabad_(
doublereal *, doublereal *);
extern doublereal igraphdlamch_(char *);
extern /* Subroutine */ int igraphdgehrd_(integer *, integer *, integer *,
doublereal *, integer *, doublereal *, doublereal *, integer *,
integer *), igraphdlarfg_(integer *, doublereal *, doublereal *,
integer *, doublereal *), igraphdlahqr_(logical *, logical *, integer *,
integer *, integer *, doublereal *, integer *, doublereal *,
doublereal *, integer *, integer *, doublereal *, integer *,
integer *), igraphdlacpy_(char *, integer *, integer *, doublereal *,
integer *, doublereal *, integer *);
doublereal safmin;
extern /* Subroutine */ int igraphdlaset_(char *, integer *, integer *,
doublereal *, doublereal *, doublereal *, integer *);
doublereal safmax;
extern /* Subroutine */ int igraphdtrexc_(char *, integer *, doublereal *,
integer *, doublereal *, integer *, integer *, integer *,
doublereal *, integer *), igraphdormhr_(char *, char *, integer
*, integer *, integer *, integer *, doublereal *, integer *,
doublereal *, doublereal *, integer *, doublereal *, integer *,
integer *);
logical sorted;
doublereal smlnum;
integer lwkopt;
/* -- LAPACK auxiliary routine (version 3.2.2) --
Univ. of Tennessee, Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..
-- June 2010 --
This subroutine is identical to DLAQR3 except that it avoids
recursion by calling DLAHQR instead of DLAQR4.
******************************************************************
Aggressive early deflation:
This subroutine accepts as input an upper Hessenberg matrix
H and performs an orthogonal similarity transformation
designed to detect and deflate fully converged eigenvalues from
a trailing principal submatrix. On output H has been over-
written by a new Hessenberg matrix that is a perturbation of
an orthogonal similarity transformation of H. It is to be
hoped that the final version of H has many zero subdiagonal
entries.
******************************************************************
WANTT (input) LOGICAL
If .TRUE., then the Hessenberg matrix H is fully updated
so that the quasi-triangular Schur factor may be
computed (in cooperation with the calling subroutine).
If .FALSE., then only enough of H is updated to preserve
the eigenvalues.
WANTZ (input) LOGICAL
If .TRUE., then the orthogonal matrix Z is updated so
so that the orthogonal Schur factor may be computed
(in cooperation with the calling subroutine).
If .FALSE., then Z is not referenced.
N (input) INTEGER
The order of the matrix H and (if WANTZ is .TRUE.) the
order of the orthogonal matrix Z.
KTOP (input) INTEGER
It is assumed that either KTOP = 1 or H(KTOP,KTOP-1)=0.
KBOT and KTOP together determine an isolated block
along the diagonal of the Hessenberg matrix.
KBOT (input) INTEGER
It is assumed without a check that either
KBOT = N or H(KBOT+1,KBOT)=0. KBOT and KTOP together
determine an isolated block along the diagonal of the
Hessenberg matrix.
NW (input) INTEGER
Deflation window size. 1 .LE. NW .LE. (KBOT-KTOP+1).
H (input/output) DOUBLE PRECISION array, dimension (LDH,N)
On input the initial N-by-N section of H stores the
Hessenberg matrix undergoing aggressive early deflation.
On output H has been transformed by an orthogonal
similarity transformation, perturbed, and the returned
to Hessenberg form that (it is to be hoped) has some
zero subdiagonal entries.
LDH (input) integer
Leading dimension of H just as declared in the calling
subroutine. N .LE. LDH
ILOZ (input) INTEGER
IHIZ (input) INTEGER
Specify the rows of Z to which transformations must be
applied if WANTZ is .TRUE.. 1 .LE. ILOZ .LE. IHIZ .LE. N.
Z (input/output) DOUBLE PRECISION array, dimension (LDZ,N)
IF WANTZ is .TRUE., then on output, the orthogonal
similarity transformation mentioned above has been
accumulated into Z(ILOZ:IHIZ,ILO:IHI) from the right.
If WANTZ is .FALSE., then Z is unreferenced.
LDZ (input) integer
The leading dimension of Z just as declared in the
calling subroutine. 1 .LE. LDZ.
NS (output) integer
The number of unconverged (ie approximate) eigenvalues
returned in SR and SI that may be used as shifts by the
calling subroutine.
ND (output) integer
The number of converged eigenvalues uncovered by this
subroutine.
SR (output) DOUBLE PRECISION array, dimension (KBOT)
SI (output) DOUBLE PRECISION array, dimension (KBOT)
On output, the real and imaginary parts of approximate
eigenvalues that may be used for shifts are stored in
SR(KBOT-ND-NS+1) through SR(KBOT-ND) and
SI(KBOT-ND-NS+1) through SI(KBOT-ND), respectively.
The real and imaginary parts of converged eigenvalues
are stored in SR(KBOT-ND+1) through SR(KBOT) and
SI(KBOT-ND+1) through SI(KBOT), respectively.
V (workspace) DOUBLE PRECISION array, dimension (LDV,NW)
An NW-by-NW work array.
LDV (input) integer scalar
The leading dimension of V just as declared in the
calling subroutine. NW .LE. LDV
NH (input) integer scalar
The number of columns of T. NH.GE.NW.
T (workspace) DOUBLE PRECISION array, dimension (LDT,NW)
LDT (input) integer
The leading dimension of T just as declared in the
calling subroutine. NW .LE. LDT
NV (input) integer
The number of rows of work array WV available for
workspace. NV.GE.NW.
WV (workspace) DOUBLE PRECISION array, dimension (LDWV,NW)
LDWV (input) integer
The leading dimension of W just as declared in the
calling subroutine. NW .LE. LDV
WORK (workspace) DOUBLE PRECISION array, dimension (LWORK)
On exit, WORK(1) is set to an estimate of the optimal value
of LWORK for the given values of N, NW, KTOP and KBOT.
LWORK (input) integer
The dimension of the work array WORK. LWORK = 2*NW
suffices, but greater efficiency may result from larger
values of LWORK.
If LWORK = -1, then a workspace query is assumed; DLAQR2
only estimates the optimal workspace size for the given
values of N, NW, KTOP and KBOT. The estimate is returned
in WORK(1). No error message related to LWORK is issued
by XERBLA. Neither H nor Z are accessed.
================================================================
Based on contributions by
Karen Braman and Ralph Byers, Department of Mathematics,
University of Kansas, USA
================================================================
==== Estimate optimal workspace. ====
Parameter adjustments */
h_dim1 = *ldh;
h_offset = 1 + h_dim1;
h__ -= h_offset;
z_dim1 = *ldz;
z_offset = 1 + z_dim1;
z__ -= z_offset;
--sr;
--si;
v_dim1 = *ldv;
v_offset = 1 + v_dim1;
v -= v_offset;
t_dim1 = *ldt;
t_offset = 1 + t_dim1;
t -= t_offset;
wv_dim1 = *ldwv;
wv_offset = 1 + wv_dim1;
wv -= wv_offset;
--work;
/* Function Body
Computing MIN */
i__1 = *nw, i__2 = *kbot - *ktop + 1;
jw = min(i__1,i__2);
if (jw <= 2) {
lwkopt = 1;
} else {
/* ==== Workspace query call to DGEHRD ==== */
i__1 = jw - 1;
igraphdgehrd_(&jw, &c__1, &i__1, &t[t_offset], ldt, &work[1], &work[1], &
c_n1, &info);
lwk1 = (integer) work[1];
/* ==== Workspace query call to DORMHR ==== */
i__1 = jw - 1;
igraphdormhr_("R", "N", &jw, &jw, &c__1, &i__1, &t[t_offset], ldt, &work[1],
&v[v_offset], ldv, &work[1], &c_n1, &info);
lwk2 = (integer) work[1];
/* ==== Optimal workspace ==== */
lwkopt = jw + max(lwk1,lwk2);
}
/* ==== Quick return in case of workspace query. ==== */
if (*lwork == -1) {
work[1] = (doublereal) lwkopt;
return 0;
}
/* ==== Nothing to do ...
... for an empty active block ... ==== */
*ns = 0;
*nd = 0;
work[1] = 1.;
if (*ktop > *kbot) {
return 0;
}
/* ... nor for an empty deflation window. ==== */
if (*nw < 1) {
return 0;
}
/* ==== Machine constants ==== */
safmin = igraphdlamch_("SAFE MINIMUM");
safmax = 1. / safmin;
igraphdlabad_(&safmin, &safmax);
ulp = igraphdlamch_("PRECISION");
smlnum = safmin * ((doublereal) (*n) / ulp);
/* ==== Setup deflation window ====
Computing MIN */
i__1 = *nw, i__2 = *kbot - *ktop + 1;
jw = min(i__1,i__2);
kwtop = *kbot - jw + 1;
if (kwtop == *ktop) {
s = 0.;
} else {
s = h__[kwtop + (kwtop - 1) * h_dim1];
}
if (*kbot == kwtop) {
/* ==== 1-by-1 deflation window: not much to do ==== */
sr[kwtop] = h__[kwtop + kwtop * h_dim1];
si[kwtop] = 0.;
*ns = 1;
*nd = 0;
/* Computing MAX */
d__2 = smlnum, d__3 = ulp * (d__1 = h__[kwtop + kwtop * h_dim1], abs(
d__1));
if (abs(s) <= max(d__2,d__3)) {
*ns = 0;
*nd = 1;
if (kwtop > *ktop) {
h__[kwtop + (kwtop - 1) * h_dim1] = 0.;
}
}
work[1] = 1.;
return 0;
}
/* ==== Convert to spike-triangular form. (In case of a
. rare QR failure, this routine continues to do
. aggressive early deflation using that part of
. the deflation window that converged using INFQR
. here and there to keep track.) ==== */
igraphdlacpy_("U", &jw, &jw, &h__[kwtop + kwtop * h_dim1], ldh, &t[t_offset],
ldt);
i__1 = jw - 1;
i__2 = *ldh + 1;
i__3 = *ldt + 1;
igraphdcopy_(&i__1, &h__[kwtop + 1 + kwtop * h_dim1], &i__2, &t[t_dim1 + 2], &
i__3);
igraphdlaset_("A", &jw, &jw, &c_b12, &c_b13, &v[v_offset], ldv);
igraphdlahqr_(&c_true, &c_true, &jw, &c__1, &jw, &t[t_offset], ldt, &sr[kwtop],
&si[kwtop], &c__1, &jw, &v[v_offset], ldv, &infqr);
/* ==== DTREXC needs a clean margin near the diagonal ==== */
i__1 = jw - 3;
for (j = 1; j <= i__1; ++j) {
t[j + 2 + j * t_dim1] = 0.;
t[j + 3 + j * t_dim1] = 0.;
/* L10: */
}
if (jw > 2) {
t[jw + (jw - 2) * t_dim1] = 0.;
}
/* ==== Deflation detection loop ==== */
*ns = jw;
ilst = infqr + 1;
L20:
if (ilst <= *ns) {
if (*ns == 1) {
bulge = FALSE_;
} else {
bulge = t[*ns + (*ns - 1) * t_dim1] != 0.;
}
/* ==== Small spike tip test for deflation ==== */
if (! bulge) {
/* ==== Real eigenvalue ==== */
foo = (d__1 = t[*ns + *ns * t_dim1], abs(d__1));
if (foo == 0.) {
foo = abs(s);
}
/* Computing MAX */
d__2 = smlnum, d__3 = ulp * foo;
if ((d__1 = s * v[*ns * v_dim1 + 1], abs(d__1)) <= max(d__2,d__3))
{
/* ==== Deflatable ==== */
--(*ns);
} else {
/* ==== Undeflatable. Move it up out of the way.
. (DTREXC can not fail in this case.) ==== */
ifst = *ns;
igraphdtrexc_("V", &jw, &t[t_offset], ldt, &v[v_offset], ldv, &ifst,
&ilst, &work[1], &info);
++ilst;
}
} else {
/* ==== Complex conjugate pair ==== */
foo = (d__3 = t[*ns + *ns * t_dim1], abs(d__3)) + sqrt((d__1 = t[*
ns + (*ns - 1) * t_dim1], abs(d__1))) * sqrt((d__2 = t[*
ns - 1 + *ns * t_dim1], abs(d__2)));
if (foo == 0.) {
foo = abs(s);
}
/* Computing MAX */
d__3 = (d__1 = s * v[*ns * v_dim1 + 1], abs(d__1)), d__4 = (d__2 =
s * v[(*ns - 1) * v_dim1 + 1], abs(d__2));
/* Computing MAX */
d__5 = smlnum, d__6 = ulp * foo;
if (max(d__3,d__4) <= max(d__5,d__6)) {
/* ==== Deflatable ==== */
*ns += -2;
} else {
/* ==== Undeflatable. Move them up out of the way.
. Fortunately, DTREXC does the right thing with
. ILST in case of a rare exchange failure. ==== */
ifst = *ns;
igraphdtrexc_("V", &jw, &t[t_offset], ldt, &v[v_offset], ldv, &ifst,
&ilst, &work[1], &info);
ilst += 2;
}
}
/* ==== End deflation detection loop ==== */
goto L20;
}
/* ==== Return to Hessenberg form ==== */
if (*ns == 0) {
s = 0.;
}
if (*ns < jw) {
/* ==== sorting diagonal blocks of T improves accuracy for
. graded matrices. Bubble sort deals well with
. exchange failures. ==== */
sorted = FALSE_;
i__ = *ns + 1;
L30:
if (sorted) {
goto L50;
}
sorted = TRUE_;
kend = i__ - 1;
i__ = infqr + 1;
if (i__ == *ns) {
k = i__ + 1;
} else if (t[i__ + 1 + i__ * t_dim1] == 0.) {
k = i__ + 1;
} else {
k = i__ + 2;
}
L40:
if (k <= kend) {
if (k == i__ + 1) {
evi = (d__1 = t[i__ + i__ * t_dim1], abs(d__1));
} else {
evi = (d__3 = t[i__ + i__ * t_dim1], abs(d__3)) + sqrt((d__1 =
t[i__ + 1 + i__ * t_dim1], abs(d__1))) * sqrt((d__2 =
t[i__ + (i__ + 1) * t_dim1], abs(d__2)));
}
if (k == kend) {
evk = (d__1 = t[k + k * t_dim1], abs(d__1));
} else if (t[k + 1 + k * t_dim1] == 0.) {
evk = (d__1 = t[k + k * t_dim1], abs(d__1));
} else {
evk = (d__3 = t[k + k * t_dim1], abs(d__3)) + sqrt((d__1 = t[
k + 1 + k * t_dim1], abs(d__1))) * sqrt((d__2 = t[k +
(k + 1) * t_dim1], abs(d__2)));
}
if (evi >= evk) {
i__ = k;
} else {
sorted = FALSE_;
ifst = i__;
ilst = k;
igraphdtrexc_("V", &jw, &t[t_offset], ldt, &v[v_offset], ldv, &ifst,
&ilst, &work[1], &info);
if (info == 0) {
i__ = ilst;
} else {
i__ = k;
}
}
if (i__ == kend) {
k = i__ + 1;
} else if (t[i__ + 1 + i__ * t_dim1] == 0.) {
k = i__ + 1;
} else {
k = i__ + 2;
}
goto L40;
}
goto L30;
L50:
;
}
/* ==== Restore shift/eigenvalue array from T ==== */
i__ = jw;
L60:
if (i__ >= infqr + 1) {
if (i__ == infqr + 1) {
sr[kwtop + i__ - 1] = t[i__ + i__ * t_dim1];
si[kwtop + i__ - 1] = 0.;
--i__;
} else if (t[i__ + (i__ - 1) * t_dim1] == 0.) {
sr[kwtop + i__ - 1] = t[i__ + i__ * t_dim1];
si[kwtop + i__ - 1] = 0.;
--i__;
} else {
aa = t[i__ - 1 + (i__ - 1) * t_dim1];
cc = t[i__ + (i__ - 1) * t_dim1];
bb = t[i__ - 1 + i__ * t_dim1];
dd = t[i__ + i__ * t_dim1];
igraphdlanv2_(&aa, &bb, &cc, &dd, &sr[kwtop + i__ - 2], &si[kwtop + i__
- 2], &sr[kwtop + i__ - 1], &si[kwtop + i__ - 1], &cs, &
sn);
i__ += -2;
}
goto L60;
}
if (*ns < jw || s == 0.) {
if (*ns > 1 && s != 0.) {
/* ==== Reflect spike back into lower triangle ==== */
igraphdcopy_(ns, &v[v_offset], ldv, &work[1], &c__1);
beta = work[1];
igraphdlarfg_(ns, &beta, &work[2], &c__1, &tau);
work[1] = 1.;
i__1 = jw - 2;
i__2 = jw - 2;
igraphdlaset_("L", &i__1, &i__2, &c_b12, &c_b12, &t[t_dim1 + 3], ldt);
igraphdlarf_("L", ns, &jw, &work[1], &c__1, &tau, &t[t_offset], ldt, &
work[jw + 1]);
igraphdlarf_("R", ns, ns, &work[1], &c__1, &tau, &t[t_offset], ldt, &
work[jw + 1]);
igraphdlarf_("R", &jw, ns, &work[1], &c__1, &tau, &v[v_offset], ldv, &
work[jw + 1]);
i__1 = *lwork - jw;
igraphdgehrd_(&jw, &c__1, ns, &t[t_offset], ldt, &work[1], &work[jw + 1]
, &i__1, &info);
}
/* ==== Copy updated reduced window into place ==== */
if (kwtop > 1) {
h__[kwtop + (kwtop - 1) * h_dim1] = s * v[v_dim1 + 1];
}
igraphdlacpy_("U", &jw, &jw, &t[t_offset], ldt, &h__[kwtop + kwtop * h_dim1]
, ldh);
i__1 = jw - 1;
i__2 = *ldt + 1;
i__3 = *ldh + 1;
igraphdcopy_(&i__1, &t[t_dim1 + 2], &i__2, &h__[kwtop + 1 + kwtop * h_dim1],
&i__3);
/* ==== Accumulate orthogonal matrix in order update
. H and Z, if requested. ==== */
if (*ns > 1 && s != 0.) {
i__1 = *lwork - jw;
igraphdormhr_("R", "N", &jw, ns, &c__1, ns, &t[t_offset], ldt, &work[1],
&v[v_offset], ldv, &work[jw + 1], &i__1, &info);
}
/* ==== Update vertical slab in H ==== */
if (*wantt) {
ltop = 1;
} else {
ltop = *ktop;
}
i__1 = kwtop - 1;
i__2 = *nv;
for (krow = ltop; i__2 < 0 ? krow >= i__1 : krow <= i__1; krow +=
i__2) {
/* Computing MIN */
i__3 = *nv, i__4 = kwtop - krow;
kln = min(i__3,i__4);
igraphdgemm_("N", "N", &kln, &jw, &jw, &c_b13, &h__[krow + kwtop *
h_dim1], ldh, &v[v_offset], ldv, &c_b12, &wv[wv_offset],
ldwv);
igraphdlacpy_("A", &kln, &jw, &wv[wv_offset], ldwv, &h__[krow + kwtop *
h_dim1], ldh);
/* L70: */
}
/* ==== Update horizontal slab in H ==== */
if (*wantt) {
i__2 = *n;
i__1 = *nh;
for (kcol = *kbot + 1; i__1 < 0 ? kcol >= i__2 : kcol <= i__2;
kcol += i__1) {
/* Computing MIN */
i__3 = *nh, i__4 = *n - kcol + 1;
kln = min(i__3,i__4);
igraphdgemm_("C", "N", &jw, &kln, &jw, &c_b13, &v[v_offset], ldv, &
h__[kwtop + kcol * h_dim1], ldh, &c_b12, &t[t_offset],
ldt);
igraphdlacpy_("A", &jw, &kln, &t[t_offset], ldt, &h__[kwtop + kcol *
h_dim1], ldh);
/* L80: */
}
}
/* ==== Update vertical slab in Z ==== */
if (*wantz) {
i__1 = *ihiz;
i__2 = *nv;
for (krow = *iloz; i__2 < 0 ? krow >= i__1 : krow <= i__1; krow +=
i__2) {
/* Computing MIN */
i__3 = *nv, i__4 = *ihiz - krow + 1;
kln = min(i__3,i__4);
igraphdgemm_("N", "N", &kln, &jw, &jw, &c_b13, &z__[krow + kwtop *
z_dim1], ldz, &v[v_offset], ldv, &c_b12, &wv[
wv_offset], ldwv);
igraphdlacpy_("A", &kln, &jw, &wv[wv_offset], ldwv, &z__[krow +
kwtop * z_dim1], ldz);
/* L90: */
}
}
}
/* ==== Return the number of deflations ... ==== */
*nd = jw - *ns;
/* ==== ... and the number of shifts. (Subtracting
. INFQR from the spike length takes care
. of the case of a rare QR failure while
. calculating eigenvalues of the deflation
. window.) ==== */
*ns -= infqr;
/* ==== Return optimal workspace. ==== */
work[1] = (doublereal) lwkopt;
/* ==== End of DLAQR2 ==== */
return 0;
} /* igraphdlaqr2_ */
/* Subroutine */ int igraphdlaqrb_(logical *wantt, integer *n, integer *ilo,
integer *ihi, doublereal *h__, integer *ldh, doublereal *wr,
doublereal *wi, doublereal *z__, integer *info)
{
/* System generated locals */
integer h_dim1, h_offset, i__1, i__2, i__3, i__4;
doublereal d__1, d__2;
/* Local variables */
static integer i__, j, k, l, m;
static doublereal s, v[3];
static integer i1, i2;
static doublereal t1, t2, t3, v1, v2, v3, h00, h10, h11, h12, h21, h22,
h33, h44;
static integer nh;
static doublereal cs;
static integer nr;
static doublereal sn, h33s, h44s;
static integer itn, its;
static doublereal ulp, sum, tst1, h43h34, unfl, ovfl;
extern /* Subroutine */ int igraphdrot_(integer *, doublereal *, integer *,
doublereal *, integer *, doublereal *, doublereal *);
static doublereal work[1];
extern /* Subroutine */ int igraphdcopy_(integer *, doublereal *, integer *,
doublereal *, integer *), igraphdlanv2_(doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *, doublereal *), igraphdlabad_(
doublereal *, doublereal *);
extern doublereal igraphdlamch_(char *);
extern /* Subroutine */ int igraphdlarfg_(integer *, doublereal *, doublereal *,
integer *, doublereal *);
extern doublereal igraphdlanhs_(char *, integer *, doublereal *, integer *,
doublereal *);
static doublereal smlnum;
/* %------------------% */
/* | Scalar Arguments | */
/* %------------------% */
/* %-----------------% */
/* | Array Arguments | */
/* %-----------------% */
/* %------------% */
/* | Parameters | */
/* %------------% */
/* %------------------------% */
/* | Local Scalars & Arrays | */
/* %------------------------% */
/* %--------------------% */
/* | External Functions | */
/* %--------------------% */
/* %----------------------% */
/* | External Subroutines | */
/* %----------------------% */
/* %-----------------------% */
/* | Executable Statements | */
/* %-----------------------% */
/* Parameter adjustments */
h_dim1 = *ldh;
h_offset = 1 + h_dim1;
h__ -= h_offset;
--wr;
--wi;
--z__;
/* Function Body */
*info = 0;
/* %--------------------------% */
/* | Quick return if possible | */
/* %--------------------------% */
if (*n == 0) {
return 0;
}
if (*ilo == *ihi) {
wr[*ilo] = h__[*ilo + *ilo * h_dim1];
wi[*ilo] = 0.;
return 0;
}
/* %---------------------------------------------% */
/* | Initialize the vector of last components of | */
/* | the Schur vectors for accumulation. | */
/* %---------------------------------------------% */
i__1 = *n - 1;
for (j = 1; j <= i__1; ++j) {
z__[j] = 0.;
/* L5: */
}
z__[*n] = 1.;
nh = *ihi - *ilo + 1;
/* %-------------------------------------------------------------% */
/* | Set machine-dependent constants for the stopping criterion. | */
/* | If norm(H) <= sqrt(OVFL), overflow should not occur. | */
/* %-------------------------------------------------------------% */
unfl = igraphdlamch_("safe minimum");
ovfl = 1. / unfl;
igraphdlabad_(&unfl, &ovfl);
ulp = igraphdlamch_("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 computed, I1 and I2 are set inside the main loop. | */
/* | Zero out H(J+2,J) = ZERO for J=1:N if WANTT = .TRUE. | */
/* | else H(J+2,J) for J=ILO:IHI-ILO-1 if WANTT = .FALSE. | */
/* %---------------------------------------------------------------% */
if (*wantt) {
i1 = 1;
i2 = *n;
i__1 = i2 - 2;
for (i__ = 1; i__ <= i__1; ++i__) {
h__[i1 + i__ + 1 + i__ * h_dim1] = 0.;
/* L8: */
}
} else {
i__1 = *ihi - *ilo - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
h__[*ilo + i__ + 1 + (*ilo + i__ - 1) * h_dim1] = 0.;
/* L9: */
}
}
/* %---------------------------------------------------% */
/* | ITN is the total number of 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 1 or 2. 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;
L10:
l = *ilo;
if (i__ < *ilo) {
goto L150;
}
/* %--------------------------------------------------------------% */
/* | Perform QR iterations on rows and columns ILO to I until a | */
/* | submatrix of order 1 or 2 splits off at the bottom because a | */
/* | subdiagonal element has become negligible. | */
/* %--------------------------------------------------------------% */
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) {
tst1 = (d__1 = h__[k - 1 + (k - 1) * h_dim1], abs(d__1)) + (d__2 =
h__[k + k * h_dim1], abs(d__2));
if (tst1 == 0.) {
i__3 = i__ - l + 1;
tst1 = igraphdlanhs_("1", &i__3, &h__[l + l * h_dim1], ldh, work);
}
/* Computing MAX */
d__2 = ulp * tst1;
if ((d__1 = h__[k + (k - 1) * h_dim1], abs(d__1)) <= max(d__2,
smlnum)) {
goto L30;
}
/* L20: */
}
L30:
l = k;
if (l > *ilo) {
/* %------------------------% */
/* | H(L,L-1) is negligible | */
/* %------------------------% */
h__[l + (l - 1) * h_dim1] = 0.;
}
/* %-------------------------------------------------------------% */
/* | Exit from loop if a submatrix of order 1 or 2 has split off | */
/* %-------------------------------------------------------------% */
if (l >= i__ - 1) {
goto L140;
}
/* %---------------------------------------------------------% */
/* | 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 == 10 || its == 20) {
/* %-------------------% */
/* | Exceptional shift | */
/* %-------------------% */
s = (d__1 = h__[i__ + (i__ - 1) * h_dim1], abs(d__1)) + (d__2 =
h__[i__ - 1 + (i__ - 2) * h_dim1], abs(d__2));
h44 = s * .75;
h33 = h44;
h43h34 = s * -.4375 * s;
} else {
/* %-----------------------------------------% */
/* | Prepare to use Wilkinson's double shift | */
/* %-----------------------------------------% */
h44 = h__[i__ + i__ * h_dim1];
h33 = h__[i__ - 1 + (i__ - 1) * h_dim1];
h43h34 = h__[i__ + (i__ - 1) * h_dim1] * h__[i__ - 1 + i__ *
h_dim1];
}
/* %-----------------------------------------------------% */
/* | Look for two consecutive small subdiagonal elements | */
/* %-----------------------------------------------------% */
i__2 = l;
for (m = i__ - 2; m >= i__2; --m) {
/* %---------------------------------------------------------% */
/* | Determine the effect of starting the double-shift QR | */
/* | iteration at row M, and see if this would make H(M,M-1) | */
/* | negligible. | */
/* %---------------------------------------------------------% */
h11 = h__[m + m * h_dim1];
h22 = h__[m + 1 + (m + 1) * h_dim1];
h21 = h__[m + 1 + m * h_dim1];
h12 = h__[m + (m + 1) * h_dim1];
h44s = h44 - h11;
h33s = h33 - h11;
v1 = (h33s * h44s - h43h34) / h21 + h12;
v2 = h22 - h11 - h33s - h44s;
v3 = h__[m + 2 + (m + 1) * h_dim1];
s = abs(v1) + abs(v2) + abs(v3);
v1 /= s;
v2 /= s;
v3 /= s;
v[0] = v1;
v[1] = v2;
v[2] = v3;
if (m == l) {
goto L50;
}
h00 = h__[m - 1 + (m - 1) * h_dim1];
h10 = h__[m + (m - 1) * h_dim1];
tst1 = abs(v1) * (abs(h00) + abs(h11) + abs(h22));
if (abs(h10) * (abs(v2) + abs(v3)) <= ulp * tst1) {
goto L50;
}
/* L40: */
}
L50:
/* %----------------------% */
/* | Double-shift QR step | */
/* %----------------------% */
i__2 = i__ - 1;
for (k = m; 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 = 3, i__4 = i__ - k + 1;
nr = min(i__3,i__4);
if (k > m) {
igraphdcopy_(&nr, &h__[k + (k - 1) * h_dim1], &c__1, v, &c__1);
}
igraphdlarfg_(&nr, v, &v[1], &c__1, &t1);
if (k > m) {
h__[k + (k - 1) * h_dim1] = v[0];
h__[k + 1 + (k - 1) * h_dim1] = 0.;
if (k < i__ - 1) {
h__[k + 2 + (k - 1) * h_dim1] = 0.;
}
} else if (m > l) {
h__[k + (k - 1) * h_dim1] = -h__[k + (k - 1) * h_dim1];
}
v2 = v[1];
t2 = t1 * v2;
if (nr == 3) {
v3 = v[2];
t3 = t1 * v3;
/* %------------------------------------------------% */
/* | Apply G from the left to transform the rows of | */
/* | the matrix in columns K to I2. | */
/* %------------------------------------------------% */
i__3 = i2;
for (j = k; j <= i__3; ++j) {
sum = h__[k + j * h_dim1] + v2 * h__[k + 1 + j * h_dim1]
+ v3 * h__[k + 2 + j * h_dim1];
h__[k + j * h_dim1] -= sum * t1;
h__[k + 1 + j * h_dim1] -= sum * t2;
h__[k + 2 + j * h_dim1] -= sum * t3;
/* L60: */
}
/* %----------------------------------------------------% */
/* | Apply G from the right to transform the columns of | */
/* | the matrix in rows I1 to min(K+3,I). | */
/* %----------------------------------------------------% */
/* Computing MIN */
i__4 = k + 3;
i__3 = min(i__4,i__);
for (j = i1; j <= i__3; ++j) {
sum = h__[j + k * h_dim1] + v2 * h__[j + (k + 1) * h_dim1]
+ v3 * h__[j + (k + 2) * h_dim1];
h__[j + k * h_dim1] -= sum * t1;
h__[j + (k + 1) * h_dim1] -= sum * t2;
h__[j + (k + 2) * h_dim1] -= sum * t3;
/* L70: */
}
/* %----------------------------------% */
/* | Accumulate transformations for Z | */
/* %----------------------------------% */
sum = z__[k] + v2 * z__[k + 1] + v3 * z__[k + 2];
z__[k] -= sum * t1;
z__[k + 1] -= sum * t2;
z__[k + 2] -= sum * t3;
} else if (nr == 2) {
/* %------------------------------------------------% */
/* | Apply G from the left to transform the rows of | */
/* | the matrix in columns K to I2. | */
/* %------------------------------------------------% */
i__3 = i2;
for (j = k; j <= i__3; ++j) {
sum = h__[k + j * h_dim1] + v2 * h__[k + 1 + j * h_dim1];
h__[k + j * h_dim1] -= sum * t1;
h__[k + 1 + j * h_dim1] -= sum * t2;
/* L90: */
}
/* %----------------------------------------------------% */
/* | Apply G from the right to transform the columns of | */
/* | the matrix in rows I1 to min(K+3,I). | */
/* %----------------------------------------------------% */
i__3 = i__;
for (j = i1; j <= i__3; ++j) {
sum = h__[j + k * h_dim1] + v2 * h__[j + (k + 1) * h_dim1]
;
h__[j + k * h_dim1] -= sum * t1;
h__[j + (k + 1) * h_dim1] -= sum * t2;
/* L100: */
}
/* %----------------------------------% */
/* | Accumulate transformations for Z | */
/* %----------------------------------% */
sum = z__[k] + v2 * z__[k + 1];
z__[k] -= sum * t1;
z__[k + 1] -= sum * t2;
}
/* L120: */
}
/* L130: */
}
/* %-------------------------------------------------------% */
/* | Failure to converge in remaining number of iterations | */
/* %-------------------------------------------------------% */
*info = i__;
return 0;
L140:
if (l == i__) {
/* %------------------------------------------------------% */
/* | H(I,I-1) is negligible: one eigenvalue has converged | */
/* %------------------------------------------------------% */
wr[i__] = h__[i__ + i__ * h_dim1];
wi[i__] = 0.;
} else if (l == i__ - 1) {
/* %--------------------------------------------------------% */
/* | H(I-1,I-2) is negligible; | */
/* | a pair of eigenvalues have converged. | */
/* | | */
/* | Transform the 2-by-2 submatrix to standard Schur form, | */
/* | and compute and store the eigenvalues. | */
/* %--------------------------------------------------------% */
igraphdlanv2_(&h__[i__ - 1 + (i__ - 1) * h_dim1], &h__[i__ - 1 + i__ *
h_dim1], &h__[i__ + (i__ - 1) * h_dim1], &h__[i__ + i__ *
h_dim1], &wr[i__ - 1], &wi[i__ - 1], &wr[i__], &wi[i__], &cs,
&sn);
if (*wantt) {
/* %-----------------------------------------------------% */
/* | Apply the transformation to the rest of H and to Z, | */
/* | as required. | */
/* %-----------------------------------------------------% */
if (i2 > i__) {
i__1 = i2 - i__;
igraphdrot_(&i__1, &h__[i__ - 1 + (i__ + 1) * h_dim1], ldh, &h__[
i__ + (i__ + 1) * h_dim1], ldh, &cs, &sn);
}
i__1 = i__ - i1 - 1;
igraphdrot_(&i__1, &h__[i1 + (i__ - 1) * h_dim1], &c__1, &h__[i1 + i__ *
h_dim1], &c__1, &cs, &sn);
sum = cs * z__[i__ - 1] + sn * z__[i__];
z__[i__] = cs * z__[i__] - sn * z__[i__ - 1];
z__[i__ - 1] = sum;
}
}
/* %---------------------------------------------------------% */
/* | Decrement number of remaining iterations, and return to | */
/* | start of the main loop with new value of I. | */
/* %---------------------------------------------------------% */
itn -= its;
i__ = l - 1;
goto L10;
L150:
return 0;
/* %---------------% */
/* | End of igraphdlaqrb | */
/* %---------------% */
} /* igraphdlaqrb_ */