/* Subroutine */
int sgegv_(char *jobvl, char *jobvr, integer *n, real *a, integer *lda, real *b, integer *ldb, real *alphar, real *alphai, real *beta, real *vl, integer *ldvl, real *vr, integer *ldvr, real *work, integer *lwork, 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;
real r__1, r__2, r__3, r__4;
/* Local variables */
integer jc, nb, in, jr, nb1, nb2, nb3, ihi, ilo;
real eps;
logical ilv;
real absb, anrm, bnrm;
integer itau;
real temp;
logical ilvl, ilvr;
integer lopt;
real anrm1, anrm2, bnrm1, bnrm2, absai, scale, absar, sbeta;
extern logical lsame_(char *, char *);
integer ileft, iinfo, icols, iwork, irows;
real salfai;
extern /* Subroutine */
int sggbak_(char *, char *, integer *, integer *, integer *, real *, real *, integer *, real *, integer *, integer * ), sggbal_(char *, integer *, real *, integer *, real *, integer *, integer *, integer *, real *, real *, real *, integer *);
real salfar;
extern real slamch_(char *), slange_(char *, integer *, integer *, real *, integer *, real *);
real safmin;
extern /* Subroutine */
int sgghrd_(char *, char *, integer *, integer *, integer *, real *, integer *, real *, integer *, real *, integer * , real *, integer *, integer *);
real safmax;
char chtemp[1];
logical ldumma[1];
extern /* Subroutine */
int slascl_(char *, integer *, integer *, real *, real *, integer *, integer *, real *, integer *, integer *), xerbla_(char *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *, integer *, integer *);
integer ijobvl, iright;
logical ilimit;
extern /* Subroutine */
int sgeqrf_(integer *, integer *, real *, integer *, real *, real *, integer *, integer *);
integer ijobvr;
extern /* Subroutine */
int slacpy_(char *, integer *, integer *, real *, integer *, real *, integer *), slaset_(char *, integer *, integer *, real *, real *, real *, integer *), stgevc_( char *, char *, logical *, integer *, real *, integer *, real *, integer *, real *, integer *, real *, integer *, integer *, integer *, real *, integer *);
real onepls;
integer lwkmin;
extern /* Subroutine */
int shgeqz_(char *, char *, char *, integer *, integer *, integer *, real *, integer *, real *, integer *, real * , real *, real *, real *, integer *, real *, integer *, real *, integer *, integer *), sorgqr_(integer *, integer *, integer *, real *, integer *, real *, real *, integer * , integer *);
integer lwkopt;
logical lquery;
extern /* Subroutine */
int sormqr_(char *, char *, integer *, integer *, integer *, real *, integer *, real *, real *, integer *, real *, integer *, integer *);
/* -- LAPACK driver 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 .. */
/* .. */
/* .. Local Arrays .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Decode the input arguments */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
--alphar;
--alphai;
--beta;
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 */
if (lsame_(jobvl, "N"))
{
ijobvl = 1;
ilvl = FALSE_;
}
else if (lsame_(jobvl, "V"))
{
ijobvl = 2;
ilvl = TRUE_;
}
else
{
ijobvl = -1;
ilvl = FALSE_;
}
if (lsame_(jobvr, "N"))
{
ijobvr = 1;
ilvr = FALSE_;
}
else if (lsame_(jobvr, "V"))
{
ijobvr = 2;
ilvr = TRUE_;
}
else
{
ijobvr = -1;
ilvr = FALSE_;
}
ilv = ilvl || ilvr;
/* Test the input arguments */
/* Computing MAX */
i__1 = *n << 3;
lwkmin = max(i__1,1);
lwkopt = lwkmin;
work[1] = (real) lwkopt;
lquery = *lwork == -1;
*info = 0;
if (ijobvl <= 0)
{
*info = -1;
}
else if (ijobvr <= 0)
{
*info = -2;
}
else if (*n < 0)
{
*info = -3;
}
else if (*lda < max(1,*n))
{
*info = -5;
}
else if (*ldb < max(1,*n))
{
*info = -7;
}
else if (*ldvl < 1 || ilvl && *ldvl < *n)
{
*info = -12;
}
else if (*ldvr < 1 || ilvr && *ldvr < *n)
{
*info = -14;
}
else if (*lwork < lwkmin && ! lquery)
{
*info = -16;
}
if (*info == 0)
{
nb1 = ilaenv_(&c__1, "SGEQRF", " ", n, n, &c_n1, &c_n1);
nb2 = ilaenv_(&c__1, "SORMQR", " ", n, n, n, &c_n1);
nb3 = ilaenv_(&c__1, "SORGQR", " ", n, n, n, &c_n1);
/* Computing MAX */
i__1 = max(nb1,nb2);
nb = max(i__1,nb3);
/* Computing MAX */
i__1 = *n * 6;
i__2 = *n * (nb + 1); // , expr subst
lopt = (*n << 1) + max(i__1,i__2);
work[1] = (real) lopt;
}
if (*info != 0)
{
i__1 = -(*info);
xerbla_("SGEGV ", &i__1);
return 0;
}
else if (lquery)
{
return 0;
}
/* Quick return if possible */
if (*n == 0)
{
return 0;
}
/* Get machine constants */
eps = slamch_("E") * slamch_("B");
safmin = slamch_("S");
safmin += safmin;
safmax = 1.f / safmin;
onepls = eps * 4 + 1.f;
/* Scale A */
anrm = slange_("M", n, n, &a[a_offset], lda, &work[1]);
anrm1 = anrm;
anrm2 = 1.f;
if (anrm < 1.f)
{
if (safmax * anrm < 1.f)
{
anrm1 = safmin;
anrm2 = safmax * anrm;
}
}
if (anrm > 0.f)
{
slascl_("G", &c_n1, &c_n1, &anrm, &c_b27, n, n, &a[a_offset], lda, & iinfo);
if (iinfo != 0)
{
*info = *n + 10;
return 0;
}
}
/* Scale B */
bnrm = slange_("M", n, n, &b[b_offset], ldb, &work[1]);
bnrm1 = bnrm;
bnrm2 = 1.f;
if (bnrm < 1.f)
{
if (safmax * bnrm < 1.f)
{
bnrm1 = safmin;
bnrm2 = safmax * bnrm;
}
}
if (bnrm > 0.f)
{
slascl_("G", &c_n1, &c_n1, &bnrm, &c_b27, n, n, &b[b_offset], ldb, & iinfo);
if (iinfo != 0)
{
*info = *n + 10;
return 0;
}
}
/* Permute the matrix to make it more nearly triangular */
/* Workspace layout: (8*N words -- "work" requires 6*N words) */
/* left_permutation, right_permutation, work... */
ileft = 1;
iright = *n + 1;
iwork = iright + *n;
sggbal_("P", n, &a[a_offset], lda, &b[b_offset], ldb, &ilo, &ihi, &work[ ileft], &work[iright], &work[iwork], &iinfo);
if (iinfo != 0)
{
*info = *n + 1;
goto L120;
}
/* Reduce B to triangular form, and initialize VL and/or VR */
/* Workspace layout: ("work..." must have at least N words) */
/* left_permutation, right_permutation, tau, work... */
irows = ihi + 1 - ilo;
if (ilv)
{
icols = *n + 1 - ilo;
}
else
{
icols = irows;
}
itau = iwork;
iwork = itau + irows;
i__1 = *lwork + 1 - iwork;
sgeqrf_(&irows, &icols, &b[ilo + ilo * b_dim1], ldb, &work[itau], &work[ iwork], &i__1, &iinfo);
if (iinfo >= 0)
{
/* Computing MAX */
i__1 = lwkopt;
i__2 = (integer) work[iwork] + iwork - 1; // , expr subst
lwkopt = max(i__1,i__2);
}
if (iinfo != 0)
{
*info = *n + 2;
goto L120;
}
i__1 = *lwork + 1 - iwork;
sormqr_("L", "T", &irows, &icols, &irows, &b[ilo + ilo * b_dim1], ldb, & work[itau], &a[ilo + ilo * a_dim1], lda, &work[iwork], &i__1, & iinfo);
if (iinfo >= 0)
{
/* Computing MAX */
i__1 = lwkopt;
i__2 = (integer) work[iwork] + iwork - 1; // , expr subst
lwkopt = max(i__1,i__2);
}
if (iinfo != 0)
{
*info = *n + 3;
goto L120;
}
if (ilvl)
{
slaset_("Full", n, n, &c_b38, &c_b27, &vl[vl_offset], ldvl) ;
i__1 = irows - 1;
i__2 = irows - 1;
slacpy_("L", &i__1, &i__2, &b[ilo + 1 + ilo * b_dim1], ldb, &vl[ilo + 1 + ilo * vl_dim1], ldvl);
i__1 = *lwork + 1 - iwork;
sorgqr_(&irows, &irows, &irows, &vl[ilo + ilo * vl_dim1], ldvl, &work[ itau], &work[iwork], &i__1, &iinfo);
if (iinfo >= 0)
{
/* Computing MAX */
i__1 = lwkopt;
i__2 = (integer) work[iwork] + iwork - 1; // , expr subst
lwkopt = max(i__1,i__2);
}
if (iinfo != 0)
{
*info = *n + 4;
goto L120;
}
}
if (ilvr)
{
slaset_("Full", n, n, &c_b38, &c_b27, &vr[vr_offset], ldvr) ;
}
/* Reduce to generalized Hessenberg form */
if (ilv)
{
/* Eigenvectors requested -- work on whole matrix. */
sgghrd_(jobvl, jobvr, n, &ilo, &ihi, &a[a_offset], lda, &b[b_offset], ldb, &vl[vl_offset], ldvl, &vr[vr_offset], ldvr, &iinfo);
}
else
{
sgghrd_("N", "N", &irows, &c__1, &irows, &a[ilo + ilo * a_dim1], lda, &b[ilo + ilo * b_dim1], ldb, &vl[vl_offset], ldvl, &vr[ vr_offset], ldvr, &iinfo);
}
if (iinfo != 0)
{
*info = *n + 5;
goto L120;
}
/* Perform QZ algorithm */
/* Workspace layout: ("work..." must have at least 1 word) */
/* left_permutation, right_permutation, work... */
iwork = itau;
if (ilv)
{
*(unsigned char *)chtemp = 'S';
}
else
{
*(unsigned char *)chtemp = 'E';
}
i__1 = *lwork + 1 - iwork;
shgeqz_(chtemp, jobvl, jobvr, n, &ilo, &ihi, &a[a_offset], lda, &b[ b_offset], ldb, &alphar[1], &alphai[1], &beta[1], &vl[vl_offset], ldvl, &vr[vr_offset], ldvr, &work[iwork], &i__1, &iinfo);
if (iinfo >= 0)
{
/* Computing MAX */
i__1 = lwkopt;
i__2 = (integer) work[iwork] + iwork - 1; // , expr subst
lwkopt = max(i__1,i__2);
}
if (iinfo != 0)
{
if (iinfo > 0 && iinfo <= *n)
{
*info = iinfo;
}
else if (iinfo > *n && iinfo <= *n << 1)
{
*info = iinfo - *n;
}
else
{
*info = *n + 6;
}
goto L120;
}
if (ilv)
{
/* Compute Eigenvectors (STGEVC requires 6*N words of workspace) */
if (ilvl)
{
if (ilvr)
{
*(unsigned char *)chtemp = 'B';
}
else
{
*(unsigned char *)chtemp = 'L';
}
}
else
{
*(unsigned char *)chtemp = 'R';
}
stgevc_(chtemp, "B", ldumma, n, &a[a_offset], lda, &b[b_offset], ldb, &vl[vl_offset], ldvl, &vr[vr_offset], ldvr, n, &in, &work[ iwork], &iinfo);
if (iinfo != 0)
{
*info = *n + 7;
goto L120;
}
/* Undo balancing on VL and VR, rescale */
if (ilvl)
{
sggbak_("P", "L", n, &ilo, &ihi, &work[ileft], &work[iright], n, & vl[vl_offset], ldvl, &iinfo);
if (iinfo != 0)
{
*info = *n + 8;
goto L120;
}
i__1 = *n;
for (jc = 1;
jc <= i__1;
++jc)
{
if (alphai[jc] < 0.f)
{
goto L50;
}
temp = 0.f;
if (alphai[jc] == 0.f)
{
i__2 = *n;
for (jr = 1;
jr <= i__2;
++jr)
{
/* Computing MAX */
r__2 = temp;
r__3 = (r__1 = vl[jr + jc * vl_dim1], f2c_abs(r__1)); // , expr subst
temp = max(r__2,r__3);
/* L10: */
}
}
else
{
i__2 = *n;
for (jr = 1;
jr <= i__2;
++jr)
{
/* Computing MAX */
r__3 = temp;
r__4 = (r__1 = vl[jr + jc * vl_dim1], f2c_abs(r__1)) + (r__2 = vl[jr + (jc + 1) * vl_dim1], f2c_abs(r__2)); // , expr subst
temp = max(r__3,r__4);
/* L20: */
}
}
if (temp < safmin)
{
goto L50;
}
temp = 1.f / temp;
if (alphai[jc] == 0.f)
{
i__2 = *n;
for (jr = 1;
jr <= i__2;
++jr)
{
vl[jr + jc * vl_dim1] *= temp;
/* L30: */
}
}
else
{
i__2 = *n;
for (jr = 1;
jr <= i__2;
++jr)
{
vl[jr + jc * vl_dim1] *= temp;
vl[jr + (jc + 1) * vl_dim1] *= temp;
/* L40: */
}
}
L50:
;
}
}
if (ilvr)
{
sggbak_("P", "R", n, &ilo, &ihi, &work[ileft], &work[iright], n, & vr[vr_offset], ldvr, &iinfo);
if (iinfo != 0)
{
*info = *n + 9;
goto L120;
}
i__1 = *n;
for (jc = 1;
jc <= i__1;
++jc)
{
if (alphai[jc] < 0.f)
{
goto L100;
}
temp = 0.f;
if (alphai[jc] == 0.f)
{
i__2 = *n;
for (jr = 1;
jr <= i__2;
++jr)
{
/* Computing MAX */
r__2 = temp;
r__3 = (r__1 = vr[jr + jc * vr_dim1], f2c_abs(r__1)); // , expr subst
temp = max(r__2,r__3);
/* L60: */
}
}
else
{
i__2 = *n;
for (jr = 1;
jr <= i__2;
++jr)
{
/* Computing MAX */
r__3 = temp;
r__4 = (r__1 = vr[jr + jc * vr_dim1], f2c_abs(r__1)) + (r__2 = vr[jr + (jc + 1) * vr_dim1], f2c_abs(r__2)); // , expr subst
temp = max(r__3,r__4);
/* L70: */
}
}
if (temp < safmin)
{
goto L100;
}
temp = 1.f / temp;
if (alphai[jc] == 0.f)
{
i__2 = *n;
for (jr = 1;
jr <= i__2;
++jr)
{
vr[jr + jc * vr_dim1] *= temp;
/* L80: */
}
}
else
{
i__2 = *n;
for (jr = 1;
jr <= i__2;
++jr)
{
vr[jr + jc * vr_dim1] *= temp;
vr[jr + (jc + 1) * vr_dim1] *= temp;
/* L90: */
}
}
L100:
;
}
}
/* End of eigenvector calculation */
}
/* Undo scaling in alpha, beta */
/* Note: this does not give the alpha and beta for the unscaled */
/* problem. */
/* Un-scaling is limited to avoid underflow in alpha and beta */
/* if they are significant. */
i__1 = *n;
for (jc = 1;
jc <= i__1;
++jc)
{
absar = (r__1 = alphar[jc], f2c_abs(r__1));
absai = (r__1 = alphai[jc], f2c_abs(r__1));
absb = (r__1 = beta[jc], f2c_abs(r__1));
salfar = anrm * alphar[jc];
salfai = anrm * alphai[jc];
sbeta = bnrm * beta[jc];
ilimit = FALSE_;
scale = 1.f;
/* Check for significant underflow in ALPHAI */
/* Computing MAX */
r__1 = safmin, r__2 = eps * absar;
r__1 = max(r__1,r__2);
r__2 = eps * absb; // ; expr subst
if (f2c_abs(salfai) < safmin && absai >= max(r__1,r__2))
{
ilimit = TRUE_;
/* Computing MAX */
r__1 = onepls * safmin;
r__2 = anrm2 * absai; // , expr subst
scale = onepls * safmin / anrm1 / max(r__1,r__2);
}
else if (salfai == 0.f)
{
/* If insignificant underflow in ALPHAI, then make the */
/* conjugate eigenvalue real. */
if (alphai[jc] < 0.f && jc > 1)
{
alphai[jc - 1] = 0.f;
}
else if (alphai[jc] > 0.f && jc < *n)
{
alphai[jc + 1] = 0.f;
}
}
/* Check for significant underflow in ALPHAR */
/* Computing MAX */
r__1 = safmin, r__2 = eps * absai;
r__1 = max(r__1,r__2);
r__2 = eps * absb; // ; expr subst
if (f2c_abs(salfar) < safmin && absar >= max(r__1,r__2))
{
ilimit = TRUE_;
/* Computing MAX */
/* Computing MAX */
r__3 = onepls * safmin;
r__4 = anrm2 * absar; // , expr subst
r__1 = scale;
r__2 = onepls * safmin / anrm1 / max(r__3,r__4); // , expr subst
scale = max(r__1,r__2);
}
/* Check for significant underflow in BETA */
/* Computing MAX */
r__1 = safmin, r__2 = eps * absar;
r__1 = max(r__1,r__2);
r__2 = eps * absai; // ; expr subst
if (f2c_abs(sbeta) < safmin && absb >= max(r__1,r__2))
{
ilimit = TRUE_;
/* Computing MAX */
/* Computing MAX */
r__3 = onepls * safmin;
r__4 = bnrm2 * absb; // , expr subst
r__1 = scale;
r__2 = onepls * safmin / bnrm1 / max(r__3,r__4); // , expr subst
scale = max(r__1,r__2);
}
/* Check for possible overflow when limiting scaling */
if (ilimit)
{
/* Computing MAX */
r__1 = f2c_abs(salfar), r__2 = f2c_abs(salfai);
r__1 = max(r__1,r__2);
r__2 = f2c_abs(sbeta); // ; expr subst
temp = scale * safmin * max(r__1,r__2);
if (temp > 1.f)
{
scale /= temp;
}
if (scale < 1.f)
{
ilimit = FALSE_;
}
}
/* Recompute un-scaled ALPHAR, ALPHAI, BETA if necessary. */
if (ilimit)
{
salfar = scale * alphar[jc] * anrm;
salfai = scale * alphai[jc] * anrm;
sbeta = scale * beta[jc] * bnrm;
}
alphar[jc] = salfar;
alphai[jc] = salfai;
beta[jc] = sbeta;
/* L110: */
}
L120:
work[1] = (real) lwkopt;
return 0;
/* End of SGEGV */
}
/* Subroutine */ int slals0_(integer *icompq, integer *nl, integer *nr,
integer *sqre, integer *nrhs, real *b, integer *ldb, real *bx,
integer *ldbx, integer *perm, integer *givptr, integer *givcol,
integer *ldgcol, real *givnum, integer *ldgnum, real *poles, real *
difl, real *difr, real *z__, integer *k, real *c__, real *s, real *
work, integer *info)
{
/* System generated locals */
integer givcol_dim1, givcol_offset, b_dim1, b_offset, bx_dim1, bx_offset,
difr_dim1, difr_offset, givnum_dim1, givnum_offset, poles_dim1,
poles_offset, i__1, i__2;
real r__1;
/* Local variables */
integer i__, j, m, n;
real dj;
integer nlp1;
real temp;
real diflj, difrj, dsigj;
real dsigjp;
/* -- LAPACK routine (version 3.2) -- */
/* November 2006 */
/* Purpose */
/* ======= */
/* SLALS0 applies back the multiplying factors of either the left or the */
/* right singular vector matrix of a diagonal matrix appended by a row */
/* to the right hand side matrix B in solving the least squares problem */
/* using the divide-and-conquer SVD approach. */
/* For the left singular vector matrix, three types of orthogonal */
/* matrices are involved: */
/* (1L) Givens rotations: the number of such rotations is GIVPTR; the */
/* pairs of columns/rows they were applied to are stored in GIVCOL; */
/* and the C- and S-values of these rotations are stored in GIVNUM. */
/* (2L) Permutation. The (NL+1)-st row of B is to be moved to the first */
/* row, and for J=2:N, PERM(J)-th row of B is to be moved to the */
/* J-th row. */
/* (3L) The left singular vector matrix of the remaining matrix. */
/* For the right singular vector matrix, four types of orthogonal */
/* matrices are involved: */
/* (1R) The right singular vector matrix of the remaining matrix. */
/* (2R) If SQRE = 1, one extra Givens rotation to generate the right */
/* null space. */
/* (3R) The inverse transformation of (2L). */
/* (4R) The inverse transformation of (1L). */
/* Arguments */
/* ========= */
/* ICOMPQ (input) INTEGER */
/* Specifies whether singular vectors are to be computed in */
/* factored form: */
/* = 0: Left singular vector matrix. */
/* = 1: Right singular vector matrix. */
/* 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 row dimension N = NL + NR + 1, */
/* and column dimension M = N + SQRE. */
/* NRHS (input) INTEGER */
/* The number of columns of B and BX. NRHS must be at least 1. */
/* B (input/output) REAL array, dimension ( LDB, NRHS ) */
/* On input, B contains the right hand sides of the least */
/* squares problem in rows 1 through M. On output, B contains */
/* the solution X in rows 1 through N. */
/* LDB (input) INTEGER */
/* The leading dimension of B. LDB must be at least */
/* max(1,MAX( M, N ) ). */
/* BX (workspace) REAL array, dimension ( LDBX, NRHS ) */
/* LDBX (input) INTEGER */
/* The leading dimension of BX. */
/* PERM (input) INTEGER array, dimension ( N ) */
/* The permutations (from deflation and sorting) applied */
/* to the two blocks. */
/* GIVPTR (input) INTEGER */
/* The number of Givens rotations which took place in this */
/* subproblem. */
/* GIVCOL (input) INTEGER array, dimension ( LDGCOL, 2 ) */
/* Each pair of numbers indicates a pair of rows/columns */
/* involved in a Givens rotation. */
/* LDGCOL (input) INTEGER */
/* The leading dimension of GIVCOL, must be at least N. */
/* GIVNUM (input) REAL array, dimension ( LDGNUM, 2 ) */
/* Each number indicates the C or S value used in the */
/* corresponding Givens rotation. */
/* LDGNUM (input) INTEGER */
/* The leading dimension of arrays DIFR, POLES and */
/* GIVNUM, must be at least K. */
/* POLES (input) REAL array, dimension ( LDGNUM, 2 ) */
/* On entry, POLES(1:K, 1) contains the new singular */
/* values obtained from solving the secular equation, and */
/* POLES(1:K, 2) is an array containing the poles in the secular */
/* equation. */
/* DIFL (input) REAL array, dimension ( K ). */
/* On entry, DIFL(I) is the distance between I-th updated */
/* (undeflated) singular value and the I-th (undeflated) old */
/* singular value. */
/* DIFR (input) REAL array, dimension ( LDGNUM, 2 ). */
/* On entry, DIFR(I, 1) contains the distances between I-th */
/* updated (undeflated) singular value and the I+1-th */
/* (undeflated) old singular value. And DIFR(I, 2) is the */
/* normalizing factor for the I-th right singular vector. */
/* Z (input) REAL array, dimension ( K ) */
/* Contain the components of the deflation-adjusted updating row */
/* vector. */
/* K (input) INTEGER */
/* Contains the dimension of the non-deflated matrix, */
/* This is the order of the related secular equation. 1 <= K <=N. */
/* C (input) REAL */
/* C contains garbage if SQRE =0 and the C-value of a Givens */
/* rotation related to the right null space if SQRE = 1. */
/* S (input) REAL */
/* S contains garbage if SQRE =0 and the S-value of a Givens */
/* rotation related to the right null space if SQRE = 1. */
/* WORK (workspace) REAL array, dimension ( K ) */
/* 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 Ren-Cang Li, Computer Science Division, University of */
/* California at Berkeley, USA */
/* Osni Marques, LBNL/NERSC, USA */
/* ===================================================================== */
/* Test the input parameters. */
/* Parameter adjustments */
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
bx_dim1 = *ldbx;
bx_offset = 1 + bx_dim1;
bx -= bx_offset;
--perm;
givcol_dim1 = *ldgcol;
givcol_offset = 1 + givcol_dim1;
givcol -= givcol_offset;
difr_dim1 = *ldgnum;
difr_offset = 1 + difr_dim1;
difr -= difr_offset;
poles_dim1 = *ldgnum;
poles_offset = 1 + poles_dim1;
poles -= poles_offset;
givnum_dim1 = *ldgnum;
givnum_offset = 1 + givnum_dim1;
givnum -= givnum_offset;
--difl;
--z__;
--work;
/* Function Body */
*info = 0;
if (*icompq < 0 || *icompq > 1) {
*info = -1;
} else if (*nl < 1) {
*info = -2;
} else if (*nr < 1) {
*info = -3;
} else if (*sqre < 0 || *sqre > 1) {
*info = -4;
}
n = *nl + *nr + 1;
if (*nrhs < 1) {
*info = -5;
} else if (*ldb < n) {
*info = -7;
} else if (*ldbx < n) {
*info = -9;
} else if (*givptr < 0) {
*info = -11;
} else if (*ldgcol < n) {
*info = -13;
} else if (*ldgnum < n) {
*info = -15;
} else if (*k < 1) {
*info = -20;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SLALS0", &i__1);
return 0;
}
m = n + *sqre;
nlp1 = *nl + 1;
if (*icompq == 0) {
/* Apply back orthogonal transformations from the left. */
/* Step (1L): apply back the Givens rotations performed. */
i__1 = *givptr;
for (i__ = 1; i__ <= i__1; ++i__) {
srot_(nrhs, &b[givcol[i__ + (givcol_dim1 << 1)] + b_dim1], ldb, &
b[givcol[i__ + givcol_dim1] + b_dim1], ldb, &givnum[i__ +
(givnum_dim1 << 1)], &givnum[i__ + givnum_dim1]);
}
/* Step (2L): permute rows of B. */
scopy_(nrhs, &b[nlp1 + b_dim1], ldb, &bx[bx_dim1 + 1], ldbx);
i__1 = n;
for (i__ = 2; i__ <= i__1; ++i__) {
scopy_(nrhs, &b[perm[i__] + b_dim1], ldb, &bx[i__ + bx_dim1],
ldbx);
}
/* Step (3L): apply the inverse of the left singular vector */
/* matrix to BX. */
if (*k == 1) {
scopy_(nrhs, &bx[bx_offset], ldbx, &b[b_offset], ldb);
if (z__[1] < 0.f) {
sscal_(nrhs, &c_b5, &b[b_offset], ldb);
}
} else {
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
diflj = difl[j];
dj = poles[j + poles_dim1];
dsigj = -poles[j + (poles_dim1 << 1)];
if (j < *k) {
difrj = -difr[j + difr_dim1];
dsigjp = -poles[j + 1 + (poles_dim1 << 1)];
}
if (z__[j] == 0.f || poles[j + (poles_dim1 << 1)] == 0.f) {
work[j] = 0.f;
} else {
work[j] = -poles[j + (poles_dim1 << 1)] * z__[j] / diflj /
(poles[j + (poles_dim1 << 1)] + dj);
}
i__2 = j - 1;
for (i__ = 1; i__ <= i__2; ++i__) {
if (z__[i__] == 0.f || poles[i__ + (poles_dim1 << 1)] ==
0.f) {
work[i__] = 0.f;
} else {
work[i__] = poles[i__ + (poles_dim1 << 1)] * z__[i__]
/ (slamc3_(&poles[i__ + (poles_dim1 << 1)], &
dsigj) - diflj) / (poles[i__ + (poles_dim1 <<
1)] + dj);
}
}
i__2 = *k;
for (i__ = j + 1; i__ <= i__2; ++i__) {
if (z__[i__] == 0.f || poles[i__ + (poles_dim1 << 1)] ==
0.f) {
work[i__] = 0.f;
} else {
work[i__] = poles[i__ + (poles_dim1 << 1)] * z__[i__]
/ (slamc3_(&poles[i__ + (poles_dim1 << 1)], &
dsigjp) + difrj) / (poles[i__ + (poles_dim1 <<
1)] + dj);
}
}
work[1] = -1.f;
temp = snrm2_(k, &work[1], &c__1);
sgemv_("T", k, nrhs, &c_b11, &bx[bx_offset], ldbx, &work[1], &
c__1, &c_b13, &b[j + b_dim1], ldb);
slascl_("G", &c__0, &c__0, &temp, &c_b11, &c__1, nrhs, &b[j +
b_dim1], ldb, info);
}
}
/* Move the deflated rows of BX to B also. */
if (*k < max(m,n)) {
i__1 = n - *k;
slacpy_("A", &i__1, nrhs, &bx[*k + 1 + bx_dim1], ldbx, &b[*k + 1
+ b_dim1], ldb);
}
} else {
/* Apply back the right orthogonal transformations. */
/* Step (1R): apply back the new right singular vector matrix */
/* to B. */
if (*k == 1) {
scopy_(nrhs, &b[b_offset], ldb, &bx[bx_offset], ldbx);
} else {
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
dsigj = poles[j + (poles_dim1 << 1)];
if (z__[j] == 0.f) {
work[j] = 0.f;
} else {
work[j] = -z__[j] / difl[j] / (dsigj + poles[j +
poles_dim1]) / difr[j + (difr_dim1 << 1)];
}
i__2 = j - 1;
for (i__ = 1; i__ <= i__2; ++i__) {
if (z__[j] == 0.f) {
work[i__] = 0.f;
} else {
r__1 = -poles[i__ + 1 + (poles_dim1 << 1)];
work[i__] = z__[j] / (slamc3_(&dsigj, &r__1) - difr[
i__ + difr_dim1]) / (dsigj + poles[i__ +
poles_dim1]) / difr[i__ + (difr_dim1 << 1)];
}
}
i__2 = *k;
for (i__ = j + 1; i__ <= i__2; ++i__) {
if (z__[j] == 0.f) {
work[i__] = 0.f;
} else {
r__1 = -poles[i__ + (poles_dim1 << 1)];
work[i__] = z__[j] / (slamc3_(&dsigj, &r__1) - difl[
i__]) / (dsigj + poles[i__ + poles_dim1]) /
difr[i__ + (difr_dim1 << 1)];
}
}
sgemv_("T", k, nrhs, &c_b11, &b[b_offset], ldb, &work[1], &
c__1, &c_b13, &bx[j + bx_dim1], ldbx);
}
}
/* Step (2R): if SQRE = 1, apply back the rotation that is */
/* related to the right null space of the subproblem. */
if (*sqre == 1) {
scopy_(nrhs, &b[m + b_dim1], ldb, &bx[m + bx_dim1], ldbx);
srot_(nrhs, &bx[bx_dim1 + 1], ldbx, &bx[m + bx_dim1], ldbx, c__,
s);
}
if (*k < max(m,n)) {
i__1 = n - *k;
slacpy_("A", &i__1, nrhs, &b[*k + 1 + b_dim1], ldb, &bx[*k + 1 +
bx_dim1], ldbx);
}
/* Step (3R): permute rows of B. */
scopy_(nrhs, &bx[bx_dim1 + 1], ldbx, &b[nlp1 + b_dim1], ldb);
if (*sqre == 1) {
scopy_(nrhs, &bx[m + bx_dim1], ldbx, &b[m + b_dim1], ldb);
}
i__1 = n;
for (i__ = 2; i__ <= i__1; ++i__) {
scopy_(nrhs, &bx[i__ + bx_dim1], ldbx, &b[perm[i__] + b_dim1],
ldb);
}
/* Step (4R): apply back the Givens rotations performed. */
for (i__ = *givptr; i__ >= 1; --i__) {
r__1 = -givnum[i__ + givnum_dim1];
srot_(nrhs, &b[givcol[i__ + (givcol_dim1 << 1)] + b_dim1], ldb, &
b[givcol[i__ + givcol_dim1] + b_dim1], ldb, &givnum[i__ +
(givnum_dim1 << 1)], &r__1);
}
}
return 0;
/* End of SLALS0 */
} /* slals0_ */
/* Subroutine */ int cgeevx_(char *balanc, char *jobvl, char *jobvr, char *
sense, integer *n, complex *a, integer *lda, complex *w, complex *vl,
integer *ldvl, complex *vr, integer *ldvr, integer *ilo, integer *ihi,
real *scale, real *abnrm, real *rconde, real *rcondv, complex *work,
integer *lwork, real *rwork, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, vl_dim1, vl_offset, vr_dim1, vr_offset, i__1,
i__2, i__3;
real r__1, r__2;
complex q__1, q__2;
/* Local variables */
integer i__, k;
char job[1];
real scl, dum[1], eps;
complex tmp;
char side[1];
real anrm;
integer ierr, itau, iwrk, nout;
integer icond;
logical scalea;
real cscale;
logical select[1];
real bignum;
integer minwrk, maxwrk;
logical wantvl, wntsnb;
integer hswork;
logical wntsne;
real smlnum;
logical lquery, wantvr, wntsnn, wntsnv;
/* -- LAPACK driver routine (version 3.2) -- */
/* November 2006 */
/* Purpose */
/* ======= */
/* CGEEVX computes for an N-by-N complex nonsymmetric matrix A, the */
/* eigenvalues and, optionally, the left and/or right eigenvectors. */
/* Optionally also, it computes a balancing transformation to improve */
/* the conditioning of the eigenvalues and eigenvectors (ILO, IHI, */
/* SCALE, and ABNRM), reciprocal condition numbers for the eigenvalues */
/* (RCONDE), and reciprocal condition numbers for the right */
/* eigenvectors (RCONDV). */
/* 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. */
/* Balancing a matrix means permuting the rows and columns to make it */
/* more nearly upper triangular, and applying a diagonal similarity */
/* transformation D * A * D**(-1), where D is a diagonal matrix, to */
/* make its rows and columns closer in norm and the condition numbers */
/* of its eigenvalues and eigenvectors smaller. The computed */
/* reciprocal condition numbers correspond to the balanced matrix. */
/* Permuting rows and columns will not change the condition numbers */
/* (in exact arithmetic) but diagonal scaling will. For further */
/* explanation of balancing, see section 4.10.2 of the LAPACK */
/* Users' Guide. */
/* Arguments */
/* ========= */
/* BALANC (input) CHARACTER*1 */
/* Indicates how the input matrix should be diagonally scaled */
/* and/or permuted to improve the conditioning of its */
/* eigenvalues. */
/* = 'N': Do not diagonally scale or permute; */
/* = 'P': Perform permutations to make the matrix more nearly */
/* upper triangular. Do not diagonally scale; */
/* = 'S': Diagonally scale the matrix, ie. replace A by */
/* D*A*D**(-1), where D is a diagonal matrix chosen */
/* to make the rows and columns of A more equal in */
/* norm. Do not permute; */
/* = 'B': Both diagonally scale and permute A. */
/* Computed reciprocal condition numbers will be for the matrix */
/* after balancing and/or permuting. Permuting does not change */
/* condition numbers (in exact arithmetic), but balancing does. */
/* JOBVL (input) CHARACTER*1 */
/* = 'N': left eigenvectors of A are not computed; */
/* = 'V': left eigenvectors of A are computed. */
/* If SENSE = 'E' or 'B', JOBVL must = 'V'. */
/* JOBVR (input) CHARACTER*1 */
/* = 'N': right eigenvectors of A are not computed; */
/* = 'V': right eigenvectors of A are computed. */
/* If SENSE = 'E' or 'B', JOBVR must = 'V'. */
/* SENSE (input) CHARACTER*1 */
/* Determines which reciprocal condition numbers are computed. */
/* = 'N': None are computed; */
/* = 'E': Computed for eigenvalues only; */
/* = 'V': Computed for right eigenvectors only; */
/* = 'B': Computed for eigenvalues and right eigenvectors. */
/* If SENSE = 'E' or 'B', both left and right eigenvectors */
/* must also be computed (JOBVL = 'V' and JOBVR = 'V'). */
/* N (input) INTEGER */
/* The order of the matrix A. N >= 0. */
/* A (input/output) COMPLEX array, dimension (LDA,N) */
/* On entry, the N-by-N matrix A. */
/* On exit, A has been overwritten. If JOBVL = 'V' or */
/* JOBVR = 'V', A contains the Schur form of the balanced */
/* version of the matrix A. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1,N). */
/* W (output) COMPLEX array, dimension (N) */
/* W contains the computed eigenvalues. */
/* VL (output) COMPLEX 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. */
/* u(j) = VL(:,j), the j-th column of VL. */
/* LDVL (input) INTEGER */
/* The leading dimension of the array VL. LDVL >= 1; if */
/* JOBVL = 'V', LDVL >= N. */
/* VR (output) COMPLEX 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. */
/* v(j) = VR(:,j), the j-th column of VR. */
/* LDVR (input) INTEGER */
/* The leading dimension of the array VR. LDVR >= 1; if */
/* JOBVR = 'V', LDVR >= N. */
/* ILO (output) INTEGER */
/* IHI (output) INTEGER */
/* ILO and IHI are integer values determined when A was */
/* balanced. The balanced A(i,j) = 0 if I > J and */
/* SCALE (output) REAL array, dimension (N) */
/* Details of the permutations and scaling factors applied */
/* when balancing A. If P(j) is the index of the row and column */
/* interchanged with row and column j, and D(j) is the scaling */
/* factor applied to row and column j, then */
/* The order in which the interchanges are made is N to IHI+1, */
/* then 1 to ILO-1. */
/* ABNRM (output) REAL */
/* The one-norm of the balanced matrix (the maximum */
/* of the sum of absolute values of elements of any column). */
/* RCONDE (output) REAL array, dimension (N) */
/* RCONDE(j) is the reciprocal condition number of the j-th */
/* eigenvalue. */
/* RCONDV (output) REAL array, dimension (N) */
/* RCONDV(j) is the reciprocal condition number of the j-th */
/* right eigenvector. */
/* WORK (workspace/output) COMPLEX 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. If SENSE = 'N' or 'E', */
/* LWORK >= max(1,2*N), and if SENSE = 'V' or 'B', */
/* LWORK >= N*N+2*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. */
/* RWORK (workspace) REAL array, dimension (2*N) */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value. */
/* > 0: if INFO = i, the QR algorithm failed to compute all the */
/* eigenvalues, and no eigenvectors or condition numbers */
/* have been computed; elements 1:ILO-1 and i+1:N of W */
/* contain eigenvalues which have converged. */
/* ===================================================================== */
/* Test the input arguments */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
--w;
vl_dim1 = *ldvl;
vl_offset = 1 + vl_dim1;
vl -= vl_offset;
vr_dim1 = *ldvr;
vr_offset = 1 + vr_dim1;
vr -= vr_offset;
--scale;
--rconde;
--rcondv;
--work;
--rwork;
/* Function Body */
*info = 0;
lquery = *lwork == -1;
wantvl = lsame_(jobvl, "V");
wantvr = lsame_(jobvr, "V");
wntsnn = lsame_(sense, "N");
wntsne = lsame_(sense, "E");
wntsnv = lsame_(sense, "V");
wntsnb = lsame_(sense, "B");
if (! (lsame_(balanc, "N") || lsame_(balanc, "S") || lsame_(balanc, "P")
|| lsame_(balanc, "B"))) {
*info = -1;
} else if (! wantvl && ! lsame_(jobvl, "N")) {
*info = -2;
} else if (! wantvr && ! lsame_(jobvr, "N")) {
*info = -3;
} else if (! (wntsnn || wntsne || wntsnb || wntsnv) || (wntsne || wntsnb)
&& ! (wantvl && wantvr)) {
*info = -4;
} else if (*n < 0) {
*info = -5;
} else if (*lda < max(1,*n)) {
*info = -7;
} else if (*ldvl < 1 || wantvl && *ldvl < *n) {
*info = -10;
} else if (*ldvr < 1 || wantvr && *ldvr < *n) {
*info = -12;
}
/* 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. */
/* CWorkspace refers to complex workspace, and RWorkspace to real */
/* workspace. NB refers to the optimal block size for the */
/* immediately following subroutine, as returned by ILAENV. */
/* HSWORK refers to the workspace preferred by CHSEQR, 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 + *n * ilaenv_(&c__1, "CGEHRD", " ", n, &c__1, n, &
c__0);
if (wantvl) {
chseqr_("S", "V", n, &c__1, n, &a[a_offset], lda, &w[1], &vl[
vl_offset], ldvl, &work[1], &c_n1, info);
} else if (wantvr) {
chseqr_("S", "V", n, &c__1, n, &a[a_offset], lda, &w[1], &vr[
vr_offset], ldvr, &work[1], &c_n1, info);
} else {
if (wntsnn) {
chseqr_("E", "N", n, &c__1, n, &a[a_offset], lda, &w[1], &
vr[vr_offset], ldvr, &work[1], &c_n1, info);
} else {
chseqr_("S", "N", n, &c__1, n, &a[a_offset], lda, &w[1], &
vr[vr_offset], ldvr, &work[1], &c_n1, info);
}
}
hswork = work[1].r;
if (! wantvl && ! wantvr) {
minwrk = *n << 1;
if (! (wntsnn || wntsne)) {
/* Computing MAX */
i__1 = minwrk, i__2 = *n * *n + (*n << 1);
minwrk = max(i__1,i__2);
}
maxwrk = max(maxwrk,hswork);
if (! (wntsnn || wntsne)) {
/* Computing MAX */
i__1 = maxwrk, i__2 = *n * *n + (*n << 1);
maxwrk = max(i__1,i__2);
}
} else {
minwrk = *n << 1;
if (! (wntsnn || wntsne)) {
/* Computing MAX */
i__1 = minwrk, i__2 = *n * *n + (*n << 1);
minwrk = max(i__1,i__2);
}
maxwrk = max(maxwrk,hswork);
/* Computing MAX */
i__1 = maxwrk, i__2 = *n + (*n - 1) * ilaenv_(&c__1, "CUNGHR",
" ", n, &c__1, n, &c_n1);
maxwrk = max(i__1,i__2);
if (! (wntsnn || wntsne)) {
/* Computing MAX */
i__1 = maxwrk, i__2 = *n * *n + (*n << 1);
maxwrk = max(i__1,i__2);
}
/* Computing MAX */
i__1 = maxwrk, i__2 = *n << 1;
maxwrk = max(i__1,i__2);
}
maxwrk = max(maxwrk,minwrk);
}
work[1].r = (real) maxwrk, work[1].i = 0.f;
if (*lwork < minwrk && ! lquery) {
*info = -20;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CGEEVX", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
/* Get machine constants */
eps = slamch_("P");
smlnum = slamch_("S");
bignum = 1.f / smlnum;
slabad_(&smlnum, &bignum);
smlnum = sqrt(smlnum) / eps;
bignum = 1.f / smlnum;
/* Scale A if max element outside range [SMLNUM,BIGNUM] */
icond = 0;
anrm = clange_("M", n, n, &a[a_offset], lda, dum);
scalea = FALSE_;
if (anrm > 0.f && anrm < smlnum) {
scalea = TRUE_;
cscale = smlnum;
} else if (anrm > bignum) {
scalea = TRUE_;
cscale = bignum;
}
if (scalea) {
clascl_("G", &c__0, &c__0, &anrm, &cscale, n, n, &a[a_offset], lda, &
ierr);
}
/* Balance the matrix and compute ABNRM */
cgebal_(balanc, n, &a[a_offset], lda, ilo, ihi, &scale[1], &ierr);
*abnrm = clange_("1", n, n, &a[a_offset], lda, dum);
if (scalea) {
dum[0] = *abnrm;
slascl_("G", &c__0, &c__0, &cscale, &anrm, &c__1, &c__1, dum, &c__1, &
ierr);
*abnrm = dum[0];
}
/* Reduce to upper Hessenberg form */
/* (CWorkspace: need 2*N, prefer N+N*NB) */
/* (RWorkspace: none) */
itau = 1;
iwrk = itau + *n;
i__1 = *lwork - iwrk + 1;
cgehrd_(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';
clacpy_("L", n, n, &a[a_offset], lda, &vl[vl_offset], ldvl)
;
/* Generate unitary matrix in VL */
/* (CWorkspace: need 2*N-1, prefer N+(N-1)*NB) */
/* (RWorkspace: none) */
i__1 = *lwork - iwrk + 1;
cunghr_(n, ilo, ihi, &vl[vl_offset], ldvl, &work[itau], &work[iwrk], &
i__1, &ierr);
/* Perform QR iteration, accumulating Schur vectors in VL */
/* (CWorkspace: need 1, prefer HSWORK (see comments) ) */
/* (RWorkspace: none) */
iwrk = itau;
i__1 = *lwork - iwrk + 1;
chseqr_("S", "V", n, ilo, ihi, &a[a_offset], lda, &w[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';
clacpy_("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';
clacpy_("L", n, n, &a[a_offset], lda, &vr[vr_offset], ldvr)
;
/* Generate unitary matrix in VR */
/* (CWorkspace: need 2*N-1, prefer N+(N-1)*NB) */
/* (RWorkspace: none) */
i__1 = *lwork - iwrk + 1;
cunghr_(n, ilo, ihi, &vr[vr_offset], ldvr, &work[itau], &work[iwrk], &
i__1, &ierr);
/* Perform QR iteration, accumulating Schur vectors in VR */
/* (CWorkspace: need 1, prefer HSWORK (see comments) ) */
/* (RWorkspace: none) */
iwrk = itau;
i__1 = *lwork - iwrk + 1;
chseqr_("S", "V", n, ilo, ihi, &a[a_offset], lda, &w[1], &vr[
vr_offset], ldvr, &work[iwrk], &i__1, info);
} else {
/* Compute eigenvalues only */
/* If condition numbers desired, compute Schur form */
if (wntsnn) {
*(unsigned char *)job = 'E';
} else {
*(unsigned char *)job = 'S';
}
/* (CWorkspace: need 1, prefer HSWORK (see comments) ) */
/* (RWorkspace: none) */
iwrk = itau;
i__1 = *lwork - iwrk + 1;
chseqr_(job, "N", n, ilo, ihi, &a[a_offset], lda, &w[1], &vr[
vr_offset], ldvr, &work[iwrk], &i__1, info);
}
/* If INFO > 0 from CHSEQR, then quit */
if (*info > 0) {
goto L50;
}
if (wantvl || wantvr) {
/* Compute left and/or right eigenvectors */
/* (CWorkspace: need 2*N) */
/* (RWorkspace: need N) */
ctrevc_(side, "B", select, n, &a[a_offset], lda, &vl[vl_offset], ldvl,
&vr[vr_offset], ldvr, n, &nout, &work[iwrk], &rwork[1], &
ierr);
}
/* Compute condition numbers if desired */
/* (CWorkspace: need N*N+2*N unless SENSE = 'E') */
/* (RWorkspace: need 2*N unless SENSE = 'E') */
if (! wntsnn) {
ctrsna_(sense, "A", select, n, &a[a_offset], lda, &vl[vl_offset],
ldvl, &vr[vr_offset], ldvr, &rconde[1], &rcondv[1], n, &nout,
&work[iwrk], n, &rwork[1], &icond);
}
if (wantvl) {
/* Undo balancing of left eigenvectors */
cgebak_(balanc, "L", n, ilo, ihi, &scale[1], n, &vl[vl_offset], ldvl,
&ierr);
/* Normalize left eigenvectors and make largest component real */
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
scl = 1.f / scnrm2_(n, &vl[i__ * vl_dim1 + 1], &c__1);
csscal_(n, &scl, &vl[i__ * vl_dim1 + 1], &c__1);
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
i__3 = k + i__ * vl_dim1;
/* Computing 2nd power */
r__1 = vl[i__3].r;
/* Computing 2nd power */
r__2 = r_imag(&vl[k + i__ * vl_dim1]);
rwork[k] = r__1 * r__1 + r__2 * r__2;
}
k = isamax_(n, &rwork[1], &c__1);
r_cnjg(&q__2, &vl[k + i__ * vl_dim1]);
r__1 = sqrt(rwork[k]);
q__1.r = q__2.r / r__1, q__1.i = q__2.i / r__1;
tmp.r = q__1.r, tmp.i = q__1.i;
cscal_(n, &tmp, &vl[i__ * vl_dim1 + 1], &c__1);
i__2 = k + i__ * vl_dim1;
i__3 = k + i__ * vl_dim1;
r__1 = vl[i__3].r;
q__1.r = r__1, q__1.i = 0.f;
vl[i__2].r = q__1.r, vl[i__2].i = q__1.i;
}
}
if (wantvr) {
/* Undo balancing of right eigenvectors */
cgebak_(balanc, "R", n, ilo, ihi, &scale[1], n, &vr[vr_offset], ldvr,
&ierr);
/* Normalize right eigenvectors and make largest component real */
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
scl = 1.f / scnrm2_(n, &vr[i__ * vr_dim1 + 1], &c__1);
csscal_(n, &scl, &vr[i__ * vr_dim1 + 1], &c__1);
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
i__3 = k + i__ * vr_dim1;
/* Computing 2nd power */
r__1 = vr[i__3].r;
/* Computing 2nd power */
r__2 = r_imag(&vr[k + i__ * vr_dim1]);
rwork[k] = r__1 * r__1 + r__2 * r__2;
}
k = isamax_(n, &rwork[1], &c__1);
r_cnjg(&q__2, &vr[k + i__ * vr_dim1]);
r__1 = sqrt(rwork[k]);
q__1.r = q__2.r / r__1, q__1.i = q__2.i / r__1;
tmp.r = q__1.r, tmp.i = q__1.i;
cscal_(n, &tmp, &vr[i__ * vr_dim1 + 1], &c__1);
i__2 = k + i__ * vr_dim1;
i__3 = k + i__ * vr_dim1;
r__1 = vr[i__3].r;
q__1.r = r__1, q__1.i = 0.f;
vr[i__2].r = q__1.r, vr[i__2].i = q__1.i;
}
}
/* Undo scaling if necessary */
L50:
if (scalea) {
i__1 = *n - *info;
/* Computing MAX */
i__3 = *n - *info;
i__2 = max(i__3,1);
clascl_("G", &c__0, &c__0, &cscale, &anrm, &i__1, &c__1, &w[*info + 1]
, &i__2, &ierr);
if (*info == 0) {
if ((wntsnv || wntsnb) && icond == 0) {
slascl_("G", &c__0, &c__0, &cscale, &anrm, n, &c__1, &rcondv[
1], n, &ierr);
}
} else {
i__1 = *ilo - 1;
clascl_("G", &c__0, &c__0, &cscale, &anrm, &i__1, &c__1, &w[1], n,
&ierr);
}
}
work[1].r = (real) maxwrk, work[1].i = 0.f;
return 0;
/* End of CGEEVX */
} /* cgeevx_ */
/* Subroutine */ int sgges_(char *jobvsl, char *jobvsr, char *sort, L_fp
selctg, integer *n, real *a, integer *lda, real *b, integer *ldb,
integer *sdim, real *alphar, real *alphai, real *beta, real *vsl,
integer *ldvsl, real *vsr, integer *ldvsr, real *work, integer *lwork,
logical *bwork, integer *info)
{
/* -- LAPACK driver 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
=======
SGGES computes for a pair of N-by-N real nonsymmetric matrices (A,B),
the generalized eigenvalues, the generalized real Schur form (S,T),
optionally, the left and/or right matrices of Schur vectors (VSL and
VSR). This gives the generalized Schur factorization
(A,B) = ( (VSL)*S*(VSR)**T, (VSL)*T*(VSR)**T )
Optionally, it also orders the eigenvalues so that a selected cluster
of eigenvalues appears in the leading diagonal blocks of the upper
quasi-triangular matrix S and the upper triangular matrix T.The
leading columns of VSL and VSR then form an orthonormal basis for the
corresponding left and right eigenspaces (deflating subspaces).
(If only the generalized eigenvalues are needed, use the driver
SGGEV instead, which is faster.)
A generalized eigenvalue for a pair of matrices (A,B) is a scalar w
or a ratio alpha/beta = w, such that A - w*B is singular. It is
usually represented as the pair (alpha,beta), as there is a
reasonable interpretation for beta=0 or both being zero.
A pair of matrices (S,T) is in generalized real Schur form if T is
upper triangular with non-negative diagonal and S is block upper
triangular with 1-by-1 and 2-by-2 blocks. 1-by-1 blocks correspond
to real generalized eigenvalues, while 2-by-2 blocks of S will be
"standardized" by making the corresponding elements of T have the
form:
[ a 0 ]
[ 0 b ]
and the pair of corresponding 2-by-2 blocks in S and T will have a
complex conjugate pair of generalized eigenvalues.
Arguments
=========
JOBVSL (input) CHARACTER*1
= 'N': do not compute the left Schur vectors;
= 'V': compute the left Schur vectors.
JOBVSR (input) CHARACTER*1
= 'N': do not compute the right Schur vectors;
= 'V': compute the right Schur vectors.
SORT (input) CHARACTER*1
Specifies whether or not to order the eigenvalues on the
diagonal of the generalized Schur form.
= 'N': Eigenvalues are not ordered;
= 'S': Eigenvalues are ordered (see SELCTG);
SELCTG (input) LOGICAL FUNCTION of three REAL arguments
SELCTG must be declared EXTERNAL in the calling subroutine.
If SORT = 'N', SELCTG is not referenced.
If SORT = 'S', SELCTG is used to select eigenvalues to sort
to the top left of the Schur form.
An eigenvalue (ALPHAR(j)+ALPHAI(j))/BETA(j) is selected if
SELCTG(ALPHAR(j),ALPHAI(j),BETA(j)) is true; i.e. if either
one of a complex conjugate pair of eigenvalues is selected,
then both complex eigenvalues are selected.
Note that in the ill-conditioned case, a selected complex
eigenvalue may no longer satisfy SELCTG(ALPHAR(j),ALPHAI(j),
BETA(j)) = .TRUE. after ordering. INFO is to be set to N+2
in this case.
N (input) INTEGER
The order of the matrices A, B, VSL, and VSR. N >= 0.
A (input/output) REAL array, dimension (LDA, N)
On entry, the first of the pair of matrices.
On exit, A has been overwritten by its generalized Schur
form S.
LDA (input) INTEGER
The leading dimension of A. LDA >= max(1,N).
B (input/output) REAL array, dimension (LDB, N)
On entry, the second of the pair of matrices.
On exit, B has been overwritten by its generalized Schur
form T.
LDB (input) INTEGER
The leading dimension of B. LDB >= max(1,N).
SDIM (output) INTEGER
If SORT = 'N', SDIM = 0.
If SORT = 'S', SDIM = number of eigenvalues (after sorting)
for which SELCTG is true. (Complex conjugate pairs for which
SELCTG is true for either eigenvalue count as 2.)
ALPHAR (output) REAL array, dimension (N)
ALPHAI (output) REAL array, dimension (N)
BETA (output) REAL array, dimension (N)
On exit, (ALPHAR(j) + ALPHAI(j)*i)/BETA(j), j=1,...,N, will
be the generalized eigenvalues. ALPHAR(j) + ALPHAI(j)*i,
and BETA(j),j=1,...,N are the diagonals of the complex Schur
form (S,T) that would result if the 2-by-2 diagonal blocks of
the real Schur form of (A,B) were further reduced to
triangular form using 2-by-2 complex unitary transformations.
If ALPHAI(j) is zero, then the j-th eigenvalue is real; if
positive, then the j-th and (j+1)-st eigenvalues are a
complex conjugate pair, with ALPHAI(j+1) negative.
Note: the quotients ALPHAR(j)/BETA(j) and ALPHAI(j)/BETA(j)
may easily over- or underflow, and BETA(j) may even be zero.
Thus, the user should avoid naively computing the ratio.
However, ALPHAR and ALPHAI will be always less than and
usually comparable with norm(A) in magnitude, and BETA always
less than and usually comparable with norm(B).
VSL (output) REAL array, dimension (LDVSL,N)
If JOBVSL = 'V', VSL will contain the left Schur vectors.
Not referenced if JOBVSL = 'N'.
LDVSL (input) INTEGER
The leading dimension of the matrix VSL. LDVSL >=1, and
if JOBVSL = 'V', LDVSL >= N.
VSR (output) REAL array, dimension (LDVSR,N)
If JOBVSR = 'V', VSR will contain the right Schur vectors.
Not referenced if JOBVSR = 'N'.
LDVSR (input) INTEGER
The leading dimension of the matrix VSR. LDVSR >= 1, and
if JOBVSR = 'V', LDVSR >= N.
WORK (workspace/output) REAL array, dimension (LWORK)
On exit, if INFO = 0, WORK(1) returns the optimal LWORK.
LWORK (input) INTEGER
The dimension of the array WORK. LWORK >= 8*N+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.
BWORK (workspace) LOGICAL array, dimension (N)
Not referenced if SORT = 'N'.
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value.
= 1,...,N:
The QZ iteration failed. (A,B) are not in Schur
form, but ALPHAR(j), ALPHAI(j), and BETA(j) should
be correct for j=INFO+1,...,N.
> N: =N+1: other than QZ iteration failed in SHGEQZ.
=N+2: after reordering, roundoff changed values of
some complex eigenvalues so that leading
eigenvalues in the Generalized Schur form no
longer satisfy SELCTG=.TRUE. This could also
be caused due to scaling.
=N+3: reordering failed in STGSEN.
=====================================================================
Decode the input arguments
Parameter adjustments */
/* Table of constant values */
static integer c__1 = 1;
static integer c__0 = 0;
static integer c_n1 = -1;
static real c_b33 = 0.f;
static real c_b34 = 1.f;
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, vsl_dim1, vsl_offset,
vsr_dim1, vsr_offset, i__1, i__2;
real r__1;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
static real anrm, bnrm;
static integer idum[1], ierr, itau, iwrk;
static real pvsl, pvsr;
static integer i__;
extern logical lsame_(char *, char *);
static integer ileft, icols;
static logical cursl, ilvsl, ilvsr;
static integer irows;
static logical lst2sl;
extern /* Subroutine */ int slabad_(real *, real *);
static integer ip;
extern /* Subroutine */ int sggbak_(char *, char *, integer *, integer *,
integer *, real *, real *, integer *, real *, integer *, integer *
), sggbal_(char *, integer *, real *, integer *,
real *, integer *, integer *, integer *, real *, real *, real *,
integer *);
static logical ilascl, ilbscl;
extern doublereal slamch_(char *), slange_(char *, integer *,
integer *, real *, integer *, real *);
static real safmin;
extern /* Subroutine */ int sgghrd_(char *, char *, integer *, integer *,
integer *, real *, integer *, real *, integer *, real *, integer *
, real *, integer *, integer *);
static real safmax;
extern /* Subroutine */ int xerbla_(char *, integer *);
static real bignum;
extern /* Subroutine */ int slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *, ftnlen, ftnlen);
static integer ijobvl, iright;
extern /* Subroutine */ int sgeqrf_(integer *, integer *, real *, integer
*, real *, real *, integer *, integer *);
static integer ijobvr;
extern /* Subroutine */ int slacpy_(char *, integer *, integer *, real *,
integer *, real *, integer *), slaset_(char *, integer *,
integer *, real *, real *, real *, integer *);
static real anrmto, bnrmto;
static logical lastsl;
extern /* Subroutine */ int shgeqz_(char *, char *, char *, integer *,
integer *, integer *, real *, integer *, real *, integer *, real *
, real *, real *, real *, integer *, real *, integer *, real *,
integer *, integer *), stgsen_(integer *,
logical *, logical *, logical *, integer *, real *, integer *,
real *, integer *, real *, real *, real *, real *, integer *,
real *, integer *, integer *, real *, real *, real *, real *,
integer *, integer *, integer *, integer *);
static integer minwrk, maxwrk;
static real smlnum;
extern /* Subroutine */ int sorgqr_(integer *, integer *, integer *, real
*, integer *, real *, real *, integer *, integer *);
static logical wantst, lquery;
extern /* Subroutine */ int sormqr_(char *, char *, integer *, integer *,
integer *, real *, integer *, real *, real *, integer *, real *,
integer *, integer *);
static real dif[2];
static integer ihi, ilo;
static real eps;
#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]
#define vsl_ref(a_1,a_2) vsl[(a_2)*vsl_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;
vsl_dim1 = *ldvsl;
vsl_offset = 1 + vsl_dim1 * 1;
vsl -= vsl_offset;
vsr_dim1 = *ldvsr;
vsr_offset = 1 + vsr_dim1 * 1;
vsr -= vsr_offset;
--work;
--bwork;
/* Function Body */
if (lsame_(jobvsl, "N")) {
ijobvl = 1;
ilvsl = FALSE_;
} else if (lsame_(jobvsl, "V")) {
ijobvl = 2;
ilvsl = TRUE_;
} else {
ijobvl = -1;
ilvsl = FALSE_;
}
if (lsame_(jobvsr, "N")) {
ijobvr = 1;
ilvsr = FALSE_;
} else if (lsame_(jobvsr, "V")) {
ijobvr = 2;
ilvsr = TRUE_;
} else {
ijobvr = -1;
ilvsr = FALSE_;
}
wantst = lsame_(sort, "S");
/* Test the input arguments */
*info = 0;
lquery = *lwork == -1;
if (ijobvl <= 0) {
*info = -1;
} else if (ijobvr <= 0) {
*info = -2;
} else if (! wantst && ! lsame_(sort, "N")) {
*info = -3;
} else if (*n < 0) {
*info = -5;
} else if (*lda < max(1,*n)) {
*info = -7;
} else if (*ldb < max(1,*n)) {
*info = -9;
} else if (*ldvsl < 1 || ilvsl && *ldvsl < *n) {
*info = -15;
} else if (*ldvsr < 1 || ilvsr && *ldvsr < *n) {
*info = -17;
}
/* 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.) */
minwrk = 1;
if (*info == 0 && (*lwork >= 1 || lquery)) {
minwrk = (*n + 1) * 7 + 16;
maxwrk = (*n + 1) * 7 + *n * ilaenv_(&c__1, "SGEQRF", " ", n, &c__1,
n, &c__0, (ftnlen)6, (ftnlen)1) + 16;
if (ilvsl) {
/* Computing MAX */
i__1 = maxwrk, i__2 = (*n + 1) * 7 + *n * ilaenv_(&c__1, "SORGQR",
" ", n, &c__1, n, &c_n1, (ftnlen)6, (ftnlen)1);
maxwrk = max(i__1,i__2);
}
work[1] = (real) maxwrk;
}
if (*lwork < minwrk && ! lquery) {
*info = -19;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SGGES ", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (*n == 0) {
*sdim = 0;
return 0;
}
/* Get machine constants */
eps = slamch_("P");
safmin = slamch_("S");
safmax = 1.f / safmin;
slabad_(&safmin, &safmax);
smlnum = sqrt(safmin) / eps;
bignum = 1.f / smlnum;
/* Scale A if max element outside range [SMLNUM,BIGNUM] */
anrm = slange_("M", n, n, &a[a_offset], lda, &work[1]);
ilascl = FALSE_;
if (anrm > 0.f && anrm < smlnum) {
anrmto = smlnum;
ilascl = TRUE_;
} else if (anrm > bignum) {
anrmto = bignum;
ilascl = TRUE_;
}
if (ilascl) {
slascl_("G", &c__0, &c__0, &anrm, &anrmto, n, n, &a[a_offset], lda, &
ierr);
}
/* Scale B if max element outside range [SMLNUM,BIGNUM] */
bnrm = slange_("M", n, n, &b[b_offset], ldb, &work[1]);
ilbscl = FALSE_;
if (bnrm > 0.f && bnrm < smlnum) {
bnrmto = smlnum;
ilbscl = TRUE_;
} else if (bnrm > bignum) {
bnrmto = bignum;
ilbscl = TRUE_;
}
if (ilbscl) {
slascl_("G", &c__0, &c__0, &bnrm, &bnrmto, n, n, &b[b_offset], ldb, &
ierr);
}
/* Permute the matrix to make it more nearly triangular
(Workspace: need 6*N + 2*N space for storing balancing factors) */
ileft = 1;
iright = *n + 1;
iwrk = iright + *n;
sggbal_("P", n, &a[a_offset], lda, &b[b_offset], ldb, &ilo, &ihi, &work[
ileft], &work[iright], &work[iwrk], &ierr);
/* Reduce B to triangular form (QR decomposition of B)
(Workspace: need N, prefer N*NB) */
irows = ihi + 1 - ilo;
icols = *n + 1 - ilo;
itau = iwrk;
iwrk = itau + irows;
i__1 = *lwork + 1 - iwrk;
sgeqrf_(&irows, &icols, &b_ref(ilo, ilo), ldb, &work[itau], &work[iwrk], &
i__1, &ierr);
/* Apply the orthogonal transformation to matrix A
(Workspace: need N, prefer N*NB) */
i__1 = *lwork + 1 - iwrk;
sormqr_("L", "T", &irows, &icols, &irows, &b_ref(ilo, ilo), ldb, &work[
itau], &a_ref(ilo, ilo), lda, &work[iwrk], &i__1, &ierr);
/* Initialize VSL
(Workspace: need N, prefer N*NB) */
if (ilvsl) {
slaset_("Full", n, n, &c_b33, &c_b34, &vsl[vsl_offset], ldvsl);
i__1 = irows - 1;
i__2 = irows - 1;
slacpy_("L", &i__1, &i__2, &b_ref(ilo + 1, ilo), ldb, &vsl_ref(ilo +
1, ilo), ldvsl);
i__1 = *lwork + 1 - iwrk;
sorgqr_(&irows, &irows, &irows, &vsl_ref(ilo, ilo), ldvsl, &work[itau]
, &work[iwrk], &i__1, &ierr);
}
/* Initialize VSR */
if (ilvsr) {
slaset_("Full", n, n, &c_b33, &c_b34, &vsr[vsr_offset], ldvsr);
}
/* Reduce to generalized Hessenberg form
(Workspace: none needed) */
sgghrd_(jobvsl, jobvsr, n, &ilo, &ihi, &a[a_offset], lda, &b[b_offset],
ldb, &vsl[vsl_offset], ldvsl, &vsr[vsr_offset], ldvsr, &ierr);
/* Perform QZ algorithm, computing Schur vectors if desired
(Workspace: need N) */
iwrk = itau;
i__1 = *lwork + 1 - iwrk;
shgeqz_("S", jobvsl, jobvsr, n, &ilo, &ihi, &a[a_offset], lda, &b[
b_offset], ldb, &alphar[1], &alphai[1], &beta[1], &vsl[vsl_offset]
, ldvsl, &vsr[vsr_offset], ldvsr, &work[iwrk], &i__1, &ierr);
if (ierr != 0) {
if (ierr > 0 && ierr <= *n) {
*info = ierr;
} else if (ierr > *n && ierr <= *n << 1) {
*info = ierr - *n;
} else {
*info = *n + 1;
}
goto L40;
}
/* Sort eigenvalues ALPHA/BETA if desired
(Workspace: need 4*N+16 ) */
*sdim = 0;
if (wantst) {
/* Undo scaling on eigenvalues before SELCTGing */
if (ilascl) {
slascl_("G", &c__0, &c__0, &anrmto, &anrm, n, &c__1, &alphar[1],
n, &ierr);
slascl_("G", &c__0, &c__0, &anrmto, &anrm, n, &c__1, &alphai[1],
n, &ierr);
}
if (ilbscl) {
slascl_("G", &c__0, &c__0, &bnrmto, &bnrm, n, &c__1, &beta[1], n,
&ierr);
}
/* Select eigenvalues */
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
bwork[i__] = (*selctg)(&alphar[i__], &alphai[i__], &beta[i__]);
/* L10: */
}
i__1 = *lwork - iwrk + 1;
stgsen_(&c__0, &ilvsl, &ilvsr, &bwork[1], n, &a[a_offset], lda, &b[
b_offset], ldb, &alphar[1], &alphai[1], &beta[1], &vsl[
vsl_offset], ldvsl, &vsr[vsr_offset], ldvsr, sdim, &pvsl, &
pvsr, dif, &work[iwrk], &i__1, idum, &c__1, &ierr);
if (ierr == 1) {
*info = *n + 3;
}
}
/* Apply back-permutation to VSL and VSR
(Workspace: none needed) */
if (ilvsl) {
sggbak_("P", "L", n, &ilo, &ihi, &work[ileft], &work[iright], n, &vsl[
vsl_offset], ldvsl, &ierr);
}
if (ilvsr) {
sggbak_("P", "R", n, &ilo, &ihi, &work[ileft], &work[iright], n, &vsr[
vsr_offset], ldvsr, &ierr);
}
/* Check if unscaling would cause over/underflow, if so, rescale
(ALPHAR(I),ALPHAI(I),BETA(I)) so BETA(I) is on the order of
B(I,I) and ALPHAR(I) and ALPHAI(I) are on the order of A(I,I) */
if (ilascl) {
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
if (alphai[i__] != 0.f) {
if (alphar[i__] / safmax > anrmto / anrm || safmin / alphar[
i__] > anrm / anrmto) {
work[1] = (r__1 = a_ref(i__, i__) / alphar[i__], dabs(
r__1));
beta[i__] *= work[1];
alphar[i__] *= work[1];
alphai[i__] *= work[1];
} else if (alphai[i__] / safmax > anrmto / anrm || safmin /
alphai[i__] > anrm / anrmto) {
work[1] = (r__1 = a_ref(i__, i__ + 1) / alphai[i__], dabs(
r__1));
beta[i__] *= work[1];
alphar[i__] *= work[1];
alphai[i__] *= work[1];
}
}
/* L50: */
}
}
if (ilbscl) {
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
if (alphai[i__] != 0.f) {
if (beta[i__] / safmax > bnrmto / bnrm || safmin / beta[i__]
> bnrm / bnrmto) {
work[1] = (r__1 = b_ref(i__, i__) / beta[i__], dabs(r__1))
;
beta[i__] *= work[1];
alphar[i__] *= work[1];
alphai[i__] *= work[1];
}
}
/* L60: */
}
}
/* Undo scaling */
if (ilascl) {
slascl_("H", &c__0, &c__0, &anrmto, &anrm, n, n, &a[a_offset], lda, &
ierr);
slascl_("G", &c__0, &c__0, &anrmto, &anrm, n, &c__1, &alphar[1], n, &
ierr);
slascl_("G", &c__0, &c__0, &anrmto, &anrm, n, &c__1, &alphai[1], n, &
ierr);
}
if (ilbscl) {
slascl_("U", &c__0, &c__0, &bnrmto, &bnrm, n, n, &b[b_offset], ldb, &
ierr);
slascl_("G", &c__0, &c__0, &bnrmto, &bnrm, n, &c__1, &beta[1], n, &
ierr);
}
if (wantst) {
/* Check if reordering is correct */
lastsl = TRUE_;
lst2sl = TRUE_;
*sdim = 0;
ip = 0;
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
cursl = (*selctg)(&alphar[i__], &alphai[i__], &beta[i__]);
if (alphai[i__] == 0.f) {
if (cursl) {
++(*sdim);
}
ip = 0;
if (cursl && ! lastsl) {
*info = *n + 2;
}
} else {
if (ip == 1) {
/* Last eigenvalue of conjugate pair */
cursl = cursl || lastsl;
lastsl = cursl;
if (cursl) {
*sdim += 2;
}
ip = -1;
if (cursl && ! lst2sl) {
*info = *n + 2;
}
} else {
/* First eigenvalue of conjugate pair */
ip = 1;
}
}
lst2sl = lastsl;
lastsl = cursl;
/* L30: */
}
}
L40:
work[1] = (real) maxwrk;
return 0;
/* End of SGGES */
} /* sgges_ */
/* Subroutine */ int sgegv_(char *jobvl, char *jobvr, integer *n, real *a,
integer *lda, real *b, integer *ldb, real *alphar, real *alphai, real
*beta, real *vl, integer *ldvl, real *vr, integer *ldvr, real *work,
integer *lwork, 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;
real r__1, r__2, r__3, r__4;
/* Local variables */
integer jc, nb, in, jr, nb1, nb2, nb3, ihi, ilo;
real eps;
logical ilv;
real absb, anrm, bnrm;
integer itau;
real temp;
logical ilvl, ilvr;
integer lopt;
real anrm1, anrm2, bnrm1, bnrm2, absai, scale, absar, sbeta;
integer ileft, iinfo, icols, iwork, irows;
real salfai;
real salfar;
real safmin;
real safmax;
char chtemp[1];
logical ldumma[1];
integer ijobvl, iright;
logical ilimit;
integer ijobvr;
real onepls;
integer lwkmin;
integer lwkopt;
logical lquery;
/* -- LAPACK driver routine (version 3.2) -- */
/* November 2006 */
/* Purpose */
/* ======= */
/* This routine is deprecated and has been replaced by routine SGGEV. */
/* SGEGV computes the eigenvalues and, optionally, the left and/or right */
/* eigenvectors of a real matrix pair (A,B). */
/* Given two square matrices A and B, */
/* the generalized nonsymmetric eigenvalue problem (GNEP) is to find the */
/* eigenvalues lambda and corresponding (non-zero) eigenvectors x such */
/* that */
/* A*x = lambda*B*x. */
/* An alternate form is to find the eigenvalues mu and corresponding */
/* eigenvectors y such that */
/* mu*A*y = B*y. */
/* These two forms are equivalent with mu = 1/lambda and x = y if */
/* neither lambda nor mu is zero. In order to deal with the case that */
/* lambda or mu is zero or small, two values alpha and beta are returned */
/* for each eigenvalue, such that lambda = alpha/beta and */
/* mu = beta/alpha. */
/* The vectors x and y in the above equations are right eigenvectors of */
/* the matrix pair (A,B). Vectors u and v satisfying */
/* u**H*A = lambda*u**H*B or mu*v**H*A = v**H*B */
/* are left eigenvectors of (A,B). */
/* Note: this routine performs "full balancing" on A and B -- see */
/* "Further Details", below. */
/* Arguments */
/* ========= */
/* JOBVL (input) CHARACTER*1 */
/* = 'N': do not compute the left generalized eigenvectors; */
/* = 'V': compute the left generalized eigenvectors (returned */
/* in VL). */
/* JOBVR (input) CHARACTER*1 */
/* = 'N': do not compute the right generalized eigenvectors; */
/* = 'V': compute the right generalized eigenvectors (returned */
/* in VR). */
/* N (input) INTEGER */
/* The order of the matrices A, B, VL, and VR. N >= 0. */
/* A (input/output) REAL array, dimension (LDA, N) */
/* On entry, the matrix A. */
/* If JOBVL = 'V' or JOBVR = 'V', then on exit A */
/* contains the real Schur form of A from the generalized Schur */
/* factorization of the pair (A,B) after balancing. */
/* If no eigenvectors were computed, then only the diagonal */
/* blocks from the Schur form will be correct. See SGGHRD and */
/* SHGEQZ for details. */
/* LDA (input) INTEGER */
/* The leading dimension of A. LDA >= max(1,N). */
/* B (input/output) REAL array, dimension (LDB, N) */
/* On entry, the matrix B. */
/* If JOBVL = 'V' or JOBVR = 'V', then on exit B contains the */
/* upper triangular matrix obtained from B in the generalized */
/* Schur factorization of the pair (A,B) after balancing. */
/* If no eigenvectors were computed, then only those elements of */
/* B corresponding to the diagonal blocks from the Schur form of */
/* A will be correct. See SGGHRD and SHGEQZ for details. */
/* LDB (input) INTEGER */
/* The leading dimension of B. LDB >= max(1,N). */
/* ALPHAR (output) REAL array, dimension (N) */
/* The real parts of each scalar alpha defining an eigenvalue of */
/* GNEP. */
/* ALPHAI (output) REAL array, dimension (N) */
/* The imaginary parts of each scalar alpha defining an */
/* eigenvalue of GNEP. If ALPHAI(j) is zero, then the j-th */
/* eigenvalue is real; if positive, then the j-th and */
/* (j+1)-st eigenvalues are a complex conjugate pair, with */
/* ALPHAI(j+1) = -ALPHAI(j). */
/* BETA (output) REAL array, dimension (N) */
/* The scalars beta that define the eigenvalues of GNEP. */
/* Together, the quantities alpha = (ALPHAR(j),ALPHAI(j)) and */
/* beta = BETA(j) represent the j-th eigenvalue of the matrix */
/* pair (A,B), in one of the forms lambda = alpha/beta or */
/* mu = beta/alpha. Since either lambda or mu may overflow, */
/* they should not, in general, be computed. */
/* VL (output) REAL array, dimension (LDVL,N) */
/* If JOBVL = 'V', the left eigenvectors u(j) are stored */
/* in the columns of VL, in the same order as their eigenvalues. */
/* If the j-th eigenvalue is real, then u(j) = VL(:,j). */
/* 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). */
/* Each eigenvector is scaled so that its largest component has */
/* abs(real part) + abs(imag. part) = 1, except for eigenvectors */
/* corresponding to an eigenvalue with alpha = beta = 0, which */
/* are set to zero. */
/* Not referenced if JOBVL = 'N'. */
/* LDVL (input) INTEGER */
/* The leading dimension of the matrix VL. LDVL >= 1, and */
/* if JOBVL = 'V', LDVL >= N. */
/* VR (output) REAL array, dimension (LDVR,N) */
/* If JOBVR = 'V', the right eigenvectors x(j) are stored */
/* in the columns of VR, in the same order as their eigenvalues. */
/* If the j-th eigenvalue is real, then x(j) = VR(:,j). */
/* If the j-th and (j+1)-st eigenvalues form a complex conjugate */
/* pair, then */
/* x(j) = VR(:,j) + i*VR(:,j+1) */
/* and */
/* x(j+1) = VR(:,j) - i*VR(:,j+1). */
/* Each eigenvector is scaled so that its largest component has */
/* abs(real part) + abs(imag. part) = 1, except for eigenvalues */
/* corresponding to an eigenvalue with alpha = beta = 0, which */
/* are set to zero. */
/* Not referenced if JOBVR = 'N'. */
/* LDVR (input) INTEGER */
/* The leading dimension of the matrix VR. LDVR >= 1, and */
/* if JOBVR = 'V', LDVR >= N. */
/* WORK (workspace/output) REAL 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(1,8*N). */
/* For good performance, LWORK must generally be larger. */
/* To compute the optimal value of LWORK, call ILAENV to get */
/* blocksizes (for SGEQRF, SORMQR, and SORGQR.) Then compute: */
/* NB -- MAX of the blocksizes for SGEQRF, SORMQR, and SORGQR; */
/* The optimal LWORK is: */
/* 2*N + MAX( 6*N, N*(NB+1) ). */
/* 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. */
/* The QZ iteration failed. No eigenvectors have been */
/* calculated, but ALPHAR(j), ALPHAI(j), and BETA(j) */
/* > N: errors that usually indicate LAPACK problems: */
/* =N+1: error return from SGGBAL */
/* =N+2: error return from SGEQRF */
/* =N+3: error return from SORMQR */
/* =N+4: error return from SORGQR */
/* =N+5: error return from SGGHRD */
/* =N+6: error return from SHGEQZ (other than failed */
/* iteration) */
/* =N+7: error return from STGEVC */
/* =N+8: error return from SGGBAK (computing VL) */
/* =N+9: error return from SGGBAK (computing VR) */
/* =N+10: error return from SLASCL (various calls) */
/* Further Details */
/* =============== */
/* Balancing */
/* --------- */
/* This driver calls SGGBAL to both permute and scale rows and columns */
/* of A and B. The permutations PL and PR are chosen so that PL*A*PR */
/* and PL*B*R will be upper triangular except for the diagonal blocks */
/* A(i:j,i:j) and B(i:j,i:j), with i and j as close together as */
/* possible. The diagonal scaling matrices DL and DR are chosen so */
/* that the pair DL*PL*A*PR*DR, DL*PL*B*PR*DR have elements close to */
/* one (except for the elements that start out zero.) */
/* After the eigenvalues and eigenvectors of the balanced matrices */
/* have been computed, SGGBAK transforms the eigenvectors back to what */
/* they would have been (in perfect arithmetic) if they had not been */
/* balanced. */
/* Contents of A and B on Exit */
/* -------- -- - --- - -- ---- */
/* If any eigenvectors are computed (either JOBVL='V' or JOBVR='V' or */
/* both), then on exit the arrays A and B will contain the real Schur */
/* form[*] of the "balanced" versions of A and B. If no eigenvectors */
/* are computed, then only the diagonal blocks will be correct. */
/* [*] See SHGEQZ, SGEGS, or read the book "Matrix Computations", */
/* by Golub & van Loan, pub. by Johns Hopkins U. Press. */
/* ===================================================================== */
/* Decode the input arguments */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
--alphar;
--alphai;
--beta;
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 */
if (lsame_(jobvl, "N")) {
ijobvl = 1;
ilvl = FALSE_;
} else if (lsame_(jobvl, "V")) {
ijobvl = 2;
ilvl = TRUE_;
} else {
ijobvl = -1;
ilvl = FALSE_;
}
if (lsame_(jobvr, "N")) {
ijobvr = 1;
ilvr = FALSE_;
} else if (lsame_(jobvr, "V")) {
ijobvr = 2;
ilvr = TRUE_;
} else {
ijobvr = -1;
ilvr = FALSE_;
}
ilv = ilvl || ilvr;
/* Test the input arguments */
/* Computing MAX */
i__1 = *n << 3;
lwkmin = max(i__1,1);
lwkopt = lwkmin;
work[1] = (real) lwkopt;
lquery = *lwork == -1;
*info = 0;
if (ijobvl <= 0) {
*info = -1;
} else if (ijobvr <= 0) {
*info = -2;
} else if (*n < 0) {
*info = -3;
} else if (*lda < max(1,*n)) {
*info = -5;
} else if (*ldb < max(1,*n)) {
*info = -7;
} else if (*ldvl < 1 || ilvl && *ldvl < *n) {
*info = -12;
} else if (*ldvr < 1 || ilvr && *ldvr < *n) {
*info = -14;
} else if (*lwork < lwkmin && ! lquery) {
*info = -16;
}
if (*info == 0) {
nb1 = ilaenv_(&c__1, "SGEQRF", " ", n, n, &c_n1, &c_n1);
nb2 = ilaenv_(&c__1, "SORMQR", " ", n, n, n, &c_n1);
nb3 = ilaenv_(&c__1, "SORGQR", " ", n, n, n, &c_n1);
/* Computing MAX */
i__1 = max(nb1,nb2);
nb = max(i__1,nb3);
/* Computing MAX */
i__1 = *n * 6, i__2 = *n * (nb + 1);
lopt = (*n << 1) + max(i__1,i__2);
work[1] = (real) lopt;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SGEGV ", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
/* Get machine constants */
eps = slamch_("E") * slamch_("B");
safmin = slamch_("S");
safmin += safmin;
safmax = 1.f / safmin;
onepls = eps * 4 + 1.f;
/* Scale A */
anrm = slange_("M", n, n, &a[a_offset], lda, &work[1]);
anrm1 = anrm;
anrm2 = 1.f;
if (anrm < 1.f) {
if (safmax * anrm < 1.f) {
anrm1 = safmin;
anrm2 = safmax * anrm;
}
}
if (anrm > 0.f) {
slascl_("G", &c_n1, &c_n1, &anrm, &c_b27, n, n, &a[a_offset], lda, &
iinfo);
if (iinfo != 0) {
*info = *n + 10;
return 0;
}
}
/* Scale B */
bnrm = slange_("M", n, n, &b[b_offset], ldb, &work[1]);
bnrm1 = bnrm;
bnrm2 = 1.f;
if (bnrm < 1.f) {
if (safmax * bnrm < 1.f) {
bnrm1 = safmin;
bnrm2 = safmax * bnrm;
}
}
if (bnrm > 0.f) {
slascl_("G", &c_n1, &c_n1, &bnrm, &c_b27, n, n, &b[b_offset], ldb, &
iinfo);
if (iinfo != 0) {
*info = *n + 10;
return 0;
}
}
/* Permute the matrix to make it more nearly triangular */
/* Workspace layout: (8*N words -- "work" requires 6*N words) */
ileft = 1;
iright = *n + 1;
iwork = iright + *n;
sggbal_("P", n, &a[a_offset], lda, &b[b_offset], ldb, &ilo, &ihi, &work[
ileft], &work[iright], &work[iwork], &iinfo);
if (iinfo != 0) {
*info = *n + 1;
goto L120;
}
/* Reduce B to triangular form, and initialize VL and/or VR */
irows = ihi + 1 - ilo;
if (ilv) {
icols = *n + 1 - ilo;
} else {
icols = irows;
}
itau = iwork;
iwork = itau + irows;
i__1 = *lwork + 1 - iwork;
sgeqrf_(&irows, &icols, &b[ilo + ilo * b_dim1], ldb, &work[itau], &work[
iwork], &i__1, &iinfo);
if (iinfo >= 0) {
/* Computing MAX */
i__1 = lwkopt, i__2 = (integer) work[iwork] + iwork - 1;
lwkopt = max(i__1,i__2);
}
if (iinfo != 0) {
*info = *n + 2;
goto L120;
}
i__1 = *lwork + 1 - iwork;
sormqr_("L", "T", &irows, &icols, &irows, &b[ilo + ilo * b_dim1], ldb, &
work[itau], &a[ilo + ilo * a_dim1], lda, &work[iwork], &i__1, &
iinfo);
if (iinfo >= 0) {
/* Computing MAX */
i__1 = lwkopt, i__2 = (integer) work[iwork] + iwork - 1;
lwkopt = max(i__1,i__2);
}
if (iinfo != 0) {
*info = *n + 3;
goto L120;
}
if (ilvl) {
slaset_("Full", n, n, &c_b38, &c_b27, &vl[vl_offset], ldvl)
;
i__1 = irows - 1;
i__2 = irows - 1;
slacpy_("L", &i__1, &i__2, &b[ilo + 1 + ilo * b_dim1], ldb, &vl[ilo +
1 + ilo * vl_dim1], ldvl);
i__1 = *lwork + 1 - iwork;
sorgqr_(&irows, &irows, &irows, &vl[ilo + ilo * vl_dim1], ldvl, &work[
itau], &work[iwork], &i__1, &iinfo);
if (iinfo >= 0) {
/* Computing MAX */
i__1 = lwkopt, i__2 = (integer) work[iwork] + iwork - 1;
lwkopt = max(i__1,i__2);
}
if (iinfo != 0) {
*info = *n + 4;
goto L120;
}
}
if (ilvr) {
slaset_("Full", n, n, &c_b38, &c_b27, &vr[vr_offset], ldvr)
;
}
/* Reduce to generalized Hessenberg form */
if (ilv) {
/* Eigenvectors requested -- work on whole matrix. */
sgghrd_(jobvl, jobvr, n, &ilo, &ihi, &a[a_offset], lda, &b[b_offset],
ldb, &vl[vl_offset], ldvl, &vr[vr_offset], ldvr, &iinfo);
} else {
sgghrd_("N", "N", &irows, &c__1, &irows, &a[ilo + ilo * a_dim1], lda,
&b[ilo + ilo * b_dim1], ldb, &vl[vl_offset], ldvl, &vr[
vr_offset], ldvr, &iinfo);
}
if (iinfo != 0) {
*info = *n + 5;
goto L120;
}
/* Perform QZ algorithm */
iwork = itau;
if (ilv) {
*(unsigned char *)chtemp = 'S';
} else {
*(unsigned char *)chtemp = 'E';
}
i__1 = *lwork + 1 - iwork;
shgeqz_(chtemp, jobvl, jobvr, n, &ilo, &ihi, &a[a_offset], lda, &b[
b_offset], ldb, &alphar[1], &alphai[1], &beta[1], &vl[vl_offset],
ldvl, &vr[vr_offset], ldvr, &work[iwork], &i__1, &iinfo);
if (iinfo >= 0) {
/* Computing MAX */
i__1 = lwkopt, i__2 = (integer) work[iwork] + iwork - 1;
lwkopt = max(i__1,i__2);
}
if (iinfo != 0) {
if (iinfo > 0 && iinfo <= *n) {
*info = iinfo;
} else if (iinfo > *n && iinfo <= *n << 1) {
*info = iinfo - *n;
} else {
*info = *n + 6;
}
goto L120;
}
if (ilv) {
/* Compute Eigenvectors (STGEVC requires 6*N words of workspace) */
if (ilvl) {
if (ilvr) {
*(unsigned char *)chtemp = 'B';
} else {
*(unsigned char *)chtemp = 'L';
}
} else {
*(unsigned char *)chtemp = 'R';
}
stgevc_(chtemp, "B", ldumma, n, &a[a_offset], lda, &b[b_offset], ldb,
&vl[vl_offset], ldvl, &vr[vr_offset], ldvr, n, &in, &work[
iwork], &iinfo);
if (iinfo != 0) {
*info = *n + 7;
goto L120;
}
/* Undo balancing on VL and VR, rescale */
if (ilvl) {
sggbak_("P", "L", n, &ilo, &ihi, &work[ileft], &work[iright], n, &
vl[vl_offset], ldvl, &iinfo);
if (iinfo != 0) {
*info = *n + 8;
goto L120;
}
i__1 = *n;
for (jc = 1; jc <= i__1; ++jc) {
if (alphai[jc] < 0.f) {
goto L50;
}
temp = 0.f;
if (alphai[jc] == 0.f) {
i__2 = *n;
for (jr = 1; jr <= i__2; ++jr) {
/* Computing MAX */
r__2 = temp, r__3 = (r__1 = vl[jr + jc * vl_dim1],
dabs(r__1));
temp = dmax(r__2,r__3);
}
} else {
i__2 = *n;
for (jr = 1; jr <= i__2; ++jr) {
/* Computing MAX */
r__3 = temp, r__4 = (r__1 = vl[jr + jc * vl_dim1],
dabs(r__1)) + (r__2 = vl[jr + (jc + 1) *
vl_dim1], dabs(r__2));
temp = dmax(r__3,r__4);
}
}
if (temp < safmin) {
goto L50;
}
temp = 1.f / temp;
if (alphai[jc] == 0.f) {
i__2 = *n;
for (jr = 1; jr <= i__2; ++jr) {
vl[jr + jc * vl_dim1] *= temp;
}
} else {
i__2 = *n;
for (jr = 1; jr <= i__2; ++jr) {
vl[jr + jc * vl_dim1] *= temp;
vl[jr + (jc + 1) * vl_dim1] *= temp;
}
}
L50:
;
}
}
if (ilvr) {
sggbak_("P", "R", n, &ilo, &ihi, &work[ileft], &work[iright], n, &
vr[vr_offset], ldvr, &iinfo);
if (iinfo != 0) {
*info = *n + 9;
goto L120;
}
i__1 = *n;
for (jc = 1; jc <= i__1; ++jc) {
if (alphai[jc] < 0.f) {
goto L100;
}
temp = 0.f;
if (alphai[jc] == 0.f) {
i__2 = *n;
for (jr = 1; jr <= i__2; ++jr) {
/* Computing MAX */
r__2 = temp, r__3 = (r__1 = vr[jr + jc * vr_dim1],
dabs(r__1));
temp = dmax(r__2,r__3);
}
} else {
i__2 = *n;
for (jr = 1; jr <= i__2; ++jr) {
/* Computing MAX */
r__3 = temp, r__4 = (r__1 = vr[jr + jc * vr_dim1],
dabs(r__1)) + (r__2 = vr[jr + (jc + 1) *
vr_dim1], dabs(r__2));
temp = dmax(r__3,r__4);
}
}
if (temp < safmin) {
goto L100;
}
temp = 1.f / temp;
if (alphai[jc] == 0.f) {
i__2 = *n;
for (jr = 1; jr <= i__2; ++jr) {
vr[jr + jc * vr_dim1] *= temp;
}
} else {
i__2 = *n;
for (jr = 1; jr <= i__2; ++jr) {
vr[jr + jc * vr_dim1] *= temp;
vr[jr + (jc + 1) * vr_dim1] *= temp;
}
}
L100:
;
}
}
/* End of eigenvector calculation */
}
/* Undo scaling in alpha, beta */
/* Note: this does not give the alpha and beta for the unscaled */
/* problem. */
/* Un-scaling is limited to avoid underflow in alpha and beta */
/* if they are significant. */
i__1 = *n;
for (jc = 1; jc <= i__1; ++jc) {
absar = (r__1 = alphar[jc], dabs(r__1));
absai = (r__1 = alphai[jc], dabs(r__1));
absb = (r__1 = beta[jc], dabs(r__1));
salfar = anrm * alphar[jc];
salfai = anrm * alphai[jc];
sbeta = bnrm * beta[jc];
ilimit = FALSE_;
scale = 1.f;
/* Check for significant underflow in ALPHAI */
/* Computing MAX */
r__1 = safmin, r__2 = eps * absar, r__1 = max(r__1,r__2), r__2 = eps *
absb;
if (dabs(salfai) < safmin && absai >= dmax(r__1,r__2)) {
ilimit = TRUE_;
/* Computing MAX */
r__1 = onepls * safmin, r__2 = anrm2 * absai;
scale = onepls * safmin / anrm1 / dmax(r__1,r__2);
} else if (salfai == 0.f) {
/* If insignificant underflow in ALPHAI, then make the */
/* conjugate eigenvalue real. */
if (alphai[jc] < 0.f && jc > 1) {
alphai[jc - 1] = 0.f;
} else if (alphai[jc] > 0.f && jc < *n) {
alphai[jc + 1] = 0.f;
}
}
/* Check for significant underflow in ALPHAR */
/* Computing MAX */
r__1 = safmin, r__2 = eps * absai, r__1 = max(r__1,r__2), r__2 = eps *
absb;
if (dabs(salfar) < safmin && absar >= dmax(r__1,r__2)) {
ilimit = TRUE_;
/* Computing MAX */
/* Computing MAX */
r__3 = onepls * safmin, r__4 = anrm2 * absar;
r__1 = scale, r__2 = onepls * safmin / anrm1 / dmax(r__3,r__4);
scale = dmax(r__1,r__2);
}
/* Check for significant underflow in BETA */
/* Computing MAX */
r__1 = safmin, r__2 = eps * absar, r__1 = max(r__1,r__2), r__2 = eps *
absai;
if (dabs(sbeta) < safmin && absb >= dmax(r__1,r__2)) {
ilimit = TRUE_;
/* Computing MAX */
/* Computing MAX */
r__3 = onepls * safmin, r__4 = bnrm2 * absb;
r__1 = scale, r__2 = onepls * safmin / bnrm1 / dmax(r__3,r__4);
scale = dmax(r__1,r__2);
}
/* Check for possible overflow when limiting scaling */
if (ilimit) {
/* Computing MAX */
r__1 = dabs(salfar), r__2 = dabs(salfai), r__1 = max(r__1,r__2),
r__2 = dabs(sbeta);
temp = scale * safmin * dmax(r__1,r__2);
if (temp > 1.f) {
scale /= temp;
}
if (scale < 1.f) {
ilimit = FALSE_;
}
}
/* Recompute un-scaled ALPHAR, ALPHAI, BETA if necessary. */
if (ilimit) {
salfar = scale * alphar[jc] * anrm;
salfai = scale * alphai[jc] * anrm;
sbeta = scale * beta[jc] * bnrm;
}
alphar[jc] = salfar;
alphai[jc] = salfai;
beta[jc] = sbeta;
}
L120:
work[1] = (real) lwkopt;
return 0;
/* End of SGEGV */
} /* sgegv_ */
/* Subroutine */ int ssterf_(integer *n, real *d__, real *e, integer *info)
{
/* System generated locals */
integer i__1;
real r__1, r__2, r__3;
/* Builtin functions */
double sqrt(doublereal), r_sign(real *, real *);
/* Local variables */
real c__;
integer i__, l, m;
real p, r__, s;
integer l1;
real bb, rt1, rt2, eps, rte;
integer lsv;
real eps2, oldc;
integer lend, jtot;
extern /* Subroutine */ int slae2_(real *, real *, real *, real *, real *)
;
real gamma, alpha, sigma, anorm;
extern doublereal slapy2_(real *, real *);
integer iscale;
real oldgam;
extern doublereal slamch_(char *);
real safmin;
extern /* Subroutine */ int xerbla_(char *, integer *);
real safmax;
extern /* Subroutine */ int slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *);
integer lendsv;
real ssfmin;
integer nmaxit;
real ssfmax;
extern doublereal slanst_(char *, integer *, real *, real *);
extern /* Subroutine */ int slasrt_(char *, integer *, real *, integer *);
/* -- LAPACK routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SSTERF 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) REAL 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) REAL 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. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
--e;
--d__;
/* Function Body */
*info = 0;
/* Quick return if possible */
if (*n < 0) {
*info = -1;
i__1 = -(*info);
xerbla_("SSTERF", &i__1);
return 0;
}
if (*n <= 1) {
return 0;
}
/* Determine the unit roundoff for this environment. */
eps = slamch_("E");
/* Computing 2nd power */
r__1 = eps;
eps2 = r__1 * r__1;
safmin = slamch_("S");
safmax = 1.f / safmin;
ssfmax = sqrt(safmax) / 3.f;
ssfmin = sqrt(safmin) / eps2;
/* Compute the eigenvalues of the tridiagonal matrix. */
nmaxit = *n * 30;
sigma = 0.f;
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.f;
}
i__1 = *n - 1;
for (m = l1; m <= i__1; ++m) {
if ((r__3 = e[m], dabs(r__3)) <= sqrt((r__1 = d__[m], dabs(r__1))) *
sqrt((r__2 = d__[m + 1], dabs(r__2))) * eps) {
e[m] = 0.f;
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 = slanst_("I", &i__1, &d__[l], &e[l]);
iscale = 0;
if (anorm > ssfmax) {
iscale = 1;
i__1 = lend - l + 1;
slascl_("G", &c__0, &c__0, &anorm, &ssfmax, &i__1, &c__1, &d__[l], n,
info);
i__1 = lend - l;
slascl_("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;
slascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &d__[l], n,
info);
i__1 = lend - l;
slascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &e[l], n,
info);
}
i__1 = lend - 1;
for (i__ = l; i__ <= i__1; ++i__) {
/* Computing 2nd power */
r__1 = e[i__];
e[i__] = r__1 * r__1;
/* L40: */
}
/* Choose between QL and QR iteration */
if ((r__1 = d__[lend], dabs(r__1)) < (r__2 = d__[l], dabs(r__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) {
if ((r__2 = e[m], dabs(r__2)) <= eps2 * (r__1 = d__[m] * d__[
m + 1], dabs(r__1))) {
goto L70;
}
/* L60: */
}
}
m = lend;
L70:
if (m < lend) {
e[m] = 0.f;
}
p = d__[l];
if (m == l) {
goto L90;
}
/* If remaining matrix is 2 by 2, use SLAE2 to compute its */
/* eigenvalues. */
if (m == l + 1) {
rte = sqrt(e[l]);
slae2_(&d__[l], &rte, &d__[l + 1], &rt1, &rt2);
d__[l] = rt1;
d__[l + 1] = rt2;
e[l] = 0.f;
l += 2;
if (l <= lend) {
goto L50;
}
goto L150;
}
if (jtot == nmaxit) {
goto L150;
}
++jtot;
/* Form shift. */
rte = sqrt(e[l]);
sigma = (d__[l + 1] - p) / (rte * 2.f);
r__ = slapy2_(&sigma, &c_b32);
sigma = p - rte / (sigma + r_sign(&r__, &sigma));
c__ = 1.f;
s = 0.f;
gamma = d__[m] - sigma;
p = gamma * gamma;
/* Inner loop */
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.f) {
p = gamma * gamma / c__;
} else {
p = oldc * bb;
}
/* L80: */
}
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) {
if ((r__2 = e[m - 1], dabs(r__2)) <= eps2 * (r__1 = d__[m] * d__[
m - 1], dabs(r__1))) {
goto L120;
}
/* L110: */
}
m = lend;
L120:
if (m > lend) {
e[m - 1] = 0.f;
}
p = d__[l];
if (m == l) {
goto L140;
}
/* If remaining matrix is 2 by 2, use SLAE2 to compute its */
/* eigenvalues. */
if (m == l - 1) {
rte = sqrt(e[l - 1]);
slae2_(&d__[l], &rte, &d__[l - 1], &rt1, &rt2);
d__[l] = rt1;
d__[l - 1] = rt2;
e[l - 1] = 0.f;
l += -2;
if (l >= lend) {
goto L100;
}
goto L150;
}
if (jtot == nmaxit) {
goto L150;
}
++jtot;
/* Form shift. */
rte = sqrt(e[l - 1]);
sigma = (d__[l - 1] - p) / (rte * 2.f);
r__ = slapy2_(&sigma, &c_b32);
sigma = p - rte / (sigma + r_sign(&r__, &sigma));
c__ = 1.f;
s = 0.f;
gamma = d__[m] - sigma;
p = gamma * gamma;
/* Inner loop */
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.f) {
p = gamma * gamma / c__;
} else {
p = oldc * bb;
}
/* L130: */
}
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) {
i__1 = lendsv - lsv + 1;
slascl_("G", &c__0, &c__0, &ssfmax, &anorm, &i__1, &c__1, &d__[lsv],
n, info);
}
if (iscale == 2) {
i__1 = lendsv - lsv + 1;
slascl_("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.f) {
++(*info);
}
/* L160: */
}
goto L180;
/* Sort eigenvalues in increasing order. */
L170:
slasrt_("I", n, &d__[1], info);
L180:
return 0;
/* End of SSTERF */
} /* ssterf_ */
int ssbev_(char *jobz, char *uplo, int *n, int *kd,
float *ab, int *ldab, float *w, float *z__, int *ldz, float *work,
int *info)
{
/* System generated locals */
int ab_dim1, ab_offset, z_dim1, z_offset, i__1;
float r__1;
/* Builtin functions */
double sqrt(double);
/* Local variables */
float eps;
int inde;
float anrm;
int imax;
float rmin, rmax, sigma;
extern int lsame_(char *, char *);
int iinfo;
extern int sscal_(int *, float *, float *, int *);
int lower, wantz;
int iscale;
extern double slamch_(char *);
float safmin;
extern int xerbla_(char *, int *);
float bignum;
extern double slansb_(char *, char *, int *, int *, float *,
int *, float *);
extern int slascl_(char *, int *, int *, float *,
float *, int *, int *, float *, int *, int *);
int indwrk;
extern int ssbtrd_(char *, char *, int *, int *,
float *, int *, float *, float *, float *, int *, float *,
int *), ssterf_(int *, float *, float *,
int *);
float smlnum;
extern int ssteqr_(char *, int *, float *, float *,
float *, int *, float *, int *);
/* -- LAPACK driver routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SSBEV computes all the eigenvalues and, optionally, eigenvectors of */
/* a float symmetric band matrix A. */
/* Arguments */
/* ========= */
/* JOBZ (input) CHARACTER*1 */
/* = 'N': Compute eigenvalues only; */
/* = 'V': Compute eigenvalues and eigenvectors. */
/* UPLO (input) CHARACTER*1 */
/* = 'U': Upper triangle of A is stored; */
/* = 'L': Lower triangle of A is stored. */
/* N (input) INTEGER */
/* The order of the matrix A. N >= 0. */
/* KD (input) INTEGER */
/* The number of superdiagonals of the matrix A if UPLO = 'U', */
/* or the number of subdiagonals if UPLO = 'L'. KD >= 0. */
/* AB (input/output) REAL array, dimension (LDAB, N) */
/* On entry, the upper or lower triangle of the symmetric band */
/* matrix A, stored in the first KD+1 rows of the array. The */
/* j-th column of A is stored in the j-th column of the array AB */
/* as follows: */
/* if UPLO = 'U', AB(kd+1+i-j,j) = A(i,j) for MAX(1,j-kd)<=i<=j; */
/* if UPLO = 'L', AB(1+i-j,j) = A(i,j) for j<=i<=MIN(n,j+kd). */
/* On exit, AB is overwritten by values generated during the */
/* reduction to tridiagonal form. If UPLO = 'U', the first */
/* superdiagonal and the diagonal of the tridiagonal matrix T */
/* are returned in rows KD and KD+1 of AB, and if UPLO = 'L', */
/* the diagonal and first subdiagonal of T are returned in the */
/* first two rows of AB. */
/* LDAB (input) INTEGER */
/* The leading dimension of the array AB. LDAB >= KD + 1. */
/* W (output) REAL array, dimension (N) */
/* If INFO = 0, the eigenvalues in ascending order. */
/* Z (output) REAL array, dimension (LDZ, N) */
/* If JOBZ = 'V', then if INFO = 0, Z contains the orthonormal */
/* eigenvectors of the matrix A, with the i-th column of Z */
/* holding the eigenvector associated with W(i). */
/* If JOBZ = 'N', then Z is not referenced. */
/* LDZ (input) INTEGER */
/* The leading dimension of the array Z. LDZ >= 1, and if */
/* JOBZ = 'V', LDZ >= MAX(1,N). */
/* WORK (workspace) REAL array, dimension (MAX(1,3*N-2)) */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* > 0: if INFO = i, the algorithm failed to converge; i */
/* off-diagonal elements of an intermediate tridiagonal */
/* form did not converge to zero. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
ab_dim1 = *ldab;
ab_offset = 1 + ab_dim1;
ab -= ab_offset;
--w;
z_dim1 = *ldz;
z_offset = 1 + z_dim1;
z__ -= z_offset;
--work;
/* Function Body */
wantz = lsame_(jobz, "V");
lower = lsame_(uplo, "L");
*info = 0;
if (! (wantz || lsame_(jobz, "N"))) {
*info = -1;
} else if (! (lower || lsame_(uplo, "U"))) {
*info = -2;
} else if (*n < 0) {
*info = -3;
} else if (*kd < 0) {
*info = -4;
} else if (*ldab < *kd + 1) {
*info = -6;
} else if (*ldz < 1 || wantz && *ldz < *n) {
*info = -9;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SSBEV ", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
if (*n == 1) {
if (lower) {
w[1] = ab[ab_dim1 + 1];
} else {
w[1] = ab[*kd + 1 + ab_dim1];
}
if (wantz) {
z__[z_dim1 + 1] = 1.f;
}
return 0;
}
/* Get machine constants. */
safmin = slamch_("Safe minimum");
eps = slamch_("Precision");
smlnum = safmin / eps;
bignum = 1.f / smlnum;
rmin = sqrt(smlnum);
rmax = sqrt(bignum);
/* Scale matrix to allowable range, if necessary. */
anrm = slansb_("M", uplo, n, kd, &ab[ab_offset], ldab, &work[1]);
iscale = 0;
if (anrm > 0.f && anrm < rmin) {
iscale = 1;
sigma = rmin / anrm;
} else if (anrm > rmax) {
iscale = 1;
sigma = rmax / anrm;
}
if (iscale == 1) {
if (lower) {
slascl_("B", kd, kd, &c_b11, &sigma, n, n, &ab[ab_offset], ldab,
info);
} else {
slascl_("Q", kd, kd, &c_b11, &sigma, n, n, &ab[ab_offset], ldab,
info);
}
}
/* Call SSBTRD to reduce symmetric band matrix to tridiagonal form. */
inde = 1;
indwrk = inde + *n;
ssbtrd_(jobz, uplo, n, kd, &ab[ab_offset], ldab, &w[1], &work[inde], &z__[
z_offset], ldz, &work[indwrk], &iinfo);
/* For eigenvalues only, call SSTERF. For eigenvectors, call SSTEQR. */
if (! wantz) {
ssterf_(n, &w[1], &work[inde], info);
} else {
ssteqr_(jobz, n, &w[1], &work[inde], &z__[z_offset], ldz, &work[
indwrk], info);
}
/* If matrix was scaled, then rescale eigenvalues appropriately. */
if (iscale == 1) {
if (*info == 0) {
imax = *n;
} else {
imax = *info - 1;
}
r__1 = 1.f / sigma;
sscal_(&imax, &r__1, &w[1], &c__1);
}
return 0;
/* End of SSBEV */
} /* ssbev_ */
/* Subroutine */ int cqrt15_(integer *scale, integer *rksel, integer *m,
integer *n, integer *nrhs, complex *a, integer *lda, complex *b,
integer *ldb, real *s, integer *rank, real *norma, real *normb,
integer *iseed, complex *work, integer *lwork)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, i__1, i__2;
real r__1;
/* Local variables */
integer j, mn;
real eps;
integer info;
real temp;
extern /* Subroutine */ int cgemm_(char *, char *, integer *, integer *,
integer *, complex *, complex *, integer *, complex *, integer *,
complex *, complex *, integer *), clarf_(char *,
integer *, integer *, complex *, integer *, complex *, complex *,
integer *, complex *);
extern doublereal sasum_(integer *, real *, integer *);
real dummy[1];
extern doublereal scnrm2_(integer *, complex *, integer *);
extern /* Subroutine */ int slabad_(real *, real *);
extern doublereal clange_(char *, integer *, integer *, complex *,
integer *, real *);
extern /* Subroutine */ int clascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, complex *, integer *, integer *);
extern doublereal slamch_(char *);
extern /* Subroutine */ int csscal_(integer *, real *, complex *, integer
*), claset_(char *, integer *, integer *, complex *, complex *,
complex *, integer *), xerbla_(char *, integer *);
real bignum;
extern /* Subroutine */ int claror_(char *, char *, integer *, integer *,
complex *, integer *, integer *, complex *, integer *);
extern doublereal slarnd_(integer *, integer *);
extern /* Subroutine */ int slaord_(char *, integer *, real *, integer *), clarnv_(integer *, integer *, integer *, complex *),
slascl_(char *, integer *, integer *, real *, real *, integer *,
integer *, real *, integer *, integer *);
real smlnum;
/* -- LAPACK test routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* CQRT15 generates a matrix with full or deficient rank and of various */
/* norms. */
/* Arguments */
/* ========= */
/* SCALE (input) INTEGER */
/* SCALE = 1: normally scaled matrix */
/* SCALE = 2: matrix scaled up */
/* SCALE = 3: matrix scaled down */
/* RKSEL (input) INTEGER */
/* RKSEL = 1: full rank matrix */
/* RKSEL = 2: rank-deficient matrix */
/* M (input) INTEGER */
/* The number of rows of the matrix A. */
/* N (input) INTEGER */
/* The number of columns of A. */
/* NRHS (input) INTEGER */
/* The number of columns of B. */
/* A (output) COMPLEX array, dimension (LDA,N) */
/* The M-by-N matrix A. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. */
/* B (output) COMPLEX array, dimension (LDB, NRHS) */
/* A matrix that is in the range space of matrix A. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. */
/* S (output) REAL array, dimension MIN(M,N) */
/* Singular values of A. */
/* RANK (output) INTEGER */
/* number of nonzero singular values of A. */
/* NORMA (output) REAL */
/* one-norm norm of A. */
/* NORMB (output) REAL */
/* one-norm norm of B. */
/* ISEED (input/output) integer array, dimension (4) */
/* seed for random number generator. */
/* WORK (workspace) COMPLEX array, dimension (LWORK) */
/* LWORK (input) INTEGER */
/* length of work space required. */
/* LWORK >= MAX(M+MIN(M,N),NRHS*MIN(M,N),2*N+M) */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. Local Arrays .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
--s;
--iseed;
--work;
/* Function Body */
mn = min(*m,*n);
/* Computing MAX */
i__1 = *m + mn, i__2 = mn * *nrhs, i__1 = max(i__1,i__2), i__2 = (*n << 1)
+ *m;
if (*lwork < max(i__1,i__2)) {
xerbla_("CQRT15", &c__16);
return 0;
}
smlnum = slamch_("Safe minimum");
bignum = 1.f / smlnum;
slabad_(&smlnum, &bignum);
eps = slamch_("Epsilon");
smlnum = smlnum / eps / eps;
bignum = 1.f / smlnum;
/* Determine rank and (unscaled) singular values */
if (*rksel == 1) {
*rank = mn;
} else if (*rksel == 2) {
*rank = mn * 3 / 4;
i__1 = mn;
for (j = *rank + 1; j <= i__1; ++j) {
s[j] = 0.f;
/* L10: */
}
} else {
xerbla_("CQRT15", &c__2);
}
if (*rank > 0) {
/* Nontrivial case */
s[1] = 1.f;
i__1 = *rank;
for (j = 2; j <= i__1; ++j) {
L20:
temp = slarnd_(&c__1, &iseed[1]);
if (temp > .1f) {
s[j] = dabs(temp);
} else {
goto L20;
}
/* L30: */
}
slaord_("Decreasing", rank, &s[1], &c__1);
/* Generate 'rank' columns of a random orthogonal matrix in A */
clarnv_(&c__2, &iseed[1], m, &work[1]);
r__1 = 1.f / scnrm2_(m, &work[1], &c__1);
csscal_(m, &r__1, &work[1], &c__1);
claset_("Full", m, rank, &c_b1, &c_b2, &a[a_offset], lda);
clarf_("Left", m, rank, &work[1], &c__1, &c_b22, &a[a_offset], lda, &
work[*m + 1]);
/* workspace used: m+mn */
/* Generate consistent rhs in the range space of A */
i__1 = *rank * *nrhs;
clarnv_(&c__2, &iseed[1], &i__1, &work[1]);
cgemm_("No transpose", "No transpose", m, nrhs, rank, &c_b2, &a[
a_offset], lda, &work[1], rank, &c_b1, &b[b_offset], ldb);
/* work space used: <= mn *nrhs */
/* generate (unscaled) matrix A */
i__1 = *rank;
for (j = 1; j <= i__1; ++j) {
csscal_(m, &s[j], &a[j * a_dim1 + 1], &c__1);
/* L40: */
}
if (*rank < *n) {
i__1 = *n - *rank;
claset_("Full", m, &i__1, &c_b1, &c_b1, &a[(*rank + 1) * a_dim1 +
1], lda);
}
claror_("Right", "No initialization", m, n, &a[a_offset], lda, &iseed[
1], &work[1], &info);
} else {
/* work space used 2*n+m */
/* Generate null matrix and rhs */
i__1 = mn;
for (j = 1; j <= i__1; ++j) {
s[j] = 0.f;
/* L50: */
}
claset_("Full", m, n, &c_b1, &c_b1, &a[a_offset], lda);
claset_("Full", m, nrhs, &c_b1, &c_b1, &b[b_offset], ldb);
}
/* Scale the matrix */
if (*scale != 1) {
*norma = clange_("Max", m, n, &a[a_offset], lda, dummy);
if (*norma != 0.f) {
if (*scale == 2) {
/* matrix scaled up */
clascl_("General", &c__0, &c__0, norma, &bignum, m, n, &a[
a_offset], lda, &info);
slascl_("General", &c__0, &c__0, norma, &bignum, &mn, &c__1, &
s[1], &mn, &info);
clascl_("General", &c__0, &c__0, norma, &bignum, m, nrhs, &b[
b_offset], ldb, &info);
} else if (*scale == 3) {
/* matrix scaled down */
clascl_("General", &c__0, &c__0, norma, &smlnum, m, n, &a[
a_offset], lda, &info);
slascl_("General", &c__0, &c__0, norma, &smlnum, &mn, &c__1, &
s[1], &mn, &info);
clascl_("General", &c__0, &c__0, norma, &smlnum, m, nrhs, &b[
b_offset], ldb, &info);
} else {
xerbla_("CQRT15", &c__1);
return 0;
}
}
}
*norma = sasum_(&mn, &s[1], &c__1);
*normb = clange_("One-norm", m, nrhs, &b[b_offset], ldb, dummy)
;
return 0;
/* End of CQRT15 */
} /* cqrt15_ */
/* Subroutine */ int slasd8_(integer *icompq, integer *k, real *d__, real *
z__, real *vf, real *vl, real *difl, real *difr, integer *lddifr,
real *dsigma, real *work, integer *info)
{
/* System generated locals */
integer difr_dim1, difr_offset, i__1, i__2;
real r__1, r__2;
/* Builtin functions */
double sqrt(doublereal), r_sign(real *, real *);
/* Local variables */
integer i__, j;
real dj, rho;
integer iwk1, iwk2, iwk3;
real temp;
extern doublereal sdot_(integer *, real *, integer *, real *, integer *);
integer iwk2i, iwk3i;
extern doublereal snrm2_(integer *, real *, integer *);
real diflj, difrj, dsigj;
extern /* Subroutine */ int scopy_(integer *, real *, integer *, real *,
integer *);
extern doublereal slamc3_(real *, real *);
extern /* Subroutine */ int slasd4_(integer *, integer *, real *, real *,
real *, real *, real *, real *, integer *), xerbla_(char *,
integer *);
real dsigjp;
extern /* Subroutine */ int slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *), slaset_(char *, integer *, integer *, real *, real *,
real *, integer *);
/* -- LAPACK auxiliary routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* October 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SLASD8 finds the square roots of the roots of the secular equation, */
/* as defined by the values in DSIGMA and Z. It makes the appropriate */
/* calls to SLASD4, and stores, for each element in D, the distance */
/* to its two nearest poles (elements in DSIGMA). It also updates */
/* the arrays VF and VL, the first and last components of all the */
/* right singular vectors of the original bidiagonal matrix. */
/* SLASD8 is called from SLASD6. */
/* Arguments */
/* ========= */
/* ICOMPQ (input) INTEGER */
/* Specifies whether singular vectors are to be computed in */
/* factored form in the calling routine: */
/* = 0: Compute singular values only. */
/* = 1: Compute singular vectors in factored form as well. */
/* K (input) INTEGER */
/* The number of terms in the rational function to be solved */
/* by SLASD4. K >= 1. */
/* D (output) REAL array, dimension ( K ) */
/* On output, D contains the updated singular values. */
/* Z (input/output) REAL array, dimension ( K ) */
/* On entry, the first K elements of this array contain the */
/* components of the deflation-adjusted updating row vector. */
/* On exit, Z is updated. */
/* VF (input/output) REAL array, dimension ( K ) */
/* On entry, VF contains information passed through DBEDE8. */
/* On exit, VF contains the first K components of the first */
/* components of all right singular vectors of the bidiagonal */
/* matrix. */
/* VL (input/output) REAL array, dimension ( K ) */
/* On entry, VL contains information passed through DBEDE8. */
/* On exit, VL contains the first K components of the last */
/* components of all right singular vectors of the bidiagonal */
/* matrix. */
/* DIFL (output) REAL array, dimension ( K ) */
/* On exit, DIFL(I) = D(I) - DSIGMA(I). */
/* DIFR (output) REAL array, */
/* dimension ( LDDIFR, 2 ) if ICOMPQ = 1 and */
/* dimension ( K ) if ICOMPQ = 0. */
/* On exit, DIFR(I,1) = D(I) - DSIGMA(I+1), DIFR(K,1) is not */
/* defined and will not be referenced. */
/* If ICOMPQ = 1, DIFR(1:K,2) is an array containing the */
/* normalizing factors for the right singular vector matrix. */
/* LDDIFR (input) INTEGER */
/* The leading dimension of DIFR, must be at least K. */
/* DSIGMA (input/output) REAL array, dimension ( K ) */
/* On entry, the first K elements of this array contain the old */
/* roots of the deflated updating problem. These are the poles */
/* of the secular equation. */
/* On exit, the elements of DSIGMA may be very slightly altered */
/* in value. */
/* WORK (workspace) REAL array, dimension at least 3 * K */
/* INFO (output) INTEGER */
/* = 0: successful exit. */
/* < 0: if INFO = -i, the i-th argument had an illegal value. */
/* > 0: if INFO = 1, an singular value did not converge */
/* Further Details */
/* =============== */
/* Based on contributions by */
/* Ming Gu and Huan Ren, Computer Science Division, University of */
/* California at Berkeley, USA */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
--d__;
--z__;
--vf;
--vl;
--difl;
difr_dim1 = *lddifr;
difr_offset = 1 + difr_dim1;
difr -= difr_offset;
--dsigma;
--work;
/* Function Body */
*info = 0;
if (*icompq < 0 || *icompq > 1) {
*info = -1;
} else if (*k < 1) {
*info = -2;
} else if (*lddifr < *k) {
*info = -9;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SLASD8", &i__1);
return 0;
}
/* Quick return if possible */
if (*k == 1) {
d__[1] = dabs(z__[1]);
difl[1] = d__[1];
if (*icompq == 1) {
difl[2] = 1.f;
difr[(difr_dim1 << 1) + 1] = 1.f;
}
return 0;
}
/* Modify values DSIGMA(i) to make sure all DSIGMA(i)-DSIGMA(j) can */
/* be computed with high relative accuracy (barring over/underflow). */
/* This is a problem on machines without a guard digit in */
/* add/subtract (Cray XMP, Cray YMP, Cray C 90 and Cray 2). */
/* The following code replaces DSIGMA(I) by 2*DSIGMA(I)-DSIGMA(I), */
/* which on any of these machines zeros out the bottommost */
/* bit of DSIGMA(I) if it is 1; this makes the subsequent */
/* subtractions DSIGMA(I)-DSIGMA(J) unproblematic when cancellation */
/* occurs. On binary machines with a guard digit (almost all */
/* machines) it does not change DSIGMA(I) at all. On hexadecimal */
/* and decimal machines with a guard digit, it slightly */
/* changes the bottommost bits of DSIGMA(I). It does not account */
/* for hexadecimal or decimal machines without guard digits */
/* (we know of none). We use a subroutine call to compute */
/* 2*DLAMBDA(I) to prevent optimizing compilers from eliminating */
/* this code. */
i__1 = *k;
for (i__ = 1; i__ <= i__1; ++i__) {
dsigma[i__] = slamc3_(&dsigma[i__], &dsigma[i__]) - dsigma[i__];
/* L10: */
}
/* Book keeping. */
iwk1 = 1;
iwk2 = iwk1 + *k;
iwk3 = iwk2 + *k;
iwk2i = iwk2 - 1;
iwk3i = iwk3 - 1;
/* Normalize Z. */
rho = snrm2_(k, &z__[1], &c__1);
slascl_("G", &c__0, &c__0, &rho, &c_b8, k, &c__1, &z__[1], k, info);
rho *= rho;
/* Initialize WORK(IWK3). */
slaset_("A", k, &c__1, &c_b8, &c_b8, &work[iwk3], k);
/* Compute the updated singular values, the arrays DIFL, DIFR, */
/* and the updated Z. */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
slasd4_(k, &j, &dsigma[1], &z__[1], &work[iwk1], &rho, &d__[j], &work[
iwk2], info);
/* If the root finder fails, the computation is terminated. */
if (*info != 0) {
return 0;
}
work[iwk3i + j] = work[iwk3i + j] * work[j] * work[iwk2i + j];
difl[j] = -work[j];
difr[j + difr_dim1] = -work[j + 1];
i__2 = j - 1;
for (i__ = 1; i__ <= i__2; ++i__) {
work[iwk3i + i__] = work[iwk3i + i__] * work[i__] * work[iwk2i +
i__] / (dsigma[i__] - dsigma[j]) / (dsigma[i__] + dsigma[
j]);
/* L20: */
}
i__2 = *k;
for (i__ = j + 1; i__ <= i__2; ++i__) {
work[iwk3i + i__] = work[iwk3i + i__] * work[i__] * work[iwk2i +
i__] / (dsigma[i__] - dsigma[j]) / (dsigma[i__] + dsigma[
j]);
/* L30: */
}
/* L40: */
}
/* Compute updated Z. */
i__1 = *k;
for (i__ = 1; i__ <= i__1; ++i__) {
r__2 = sqrt((r__1 = work[iwk3i + i__], dabs(r__1)));
z__[i__] = r_sign(&r__2, &z__[i__]);
/* L50: */
}
/* Update VF and VL. */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
diflj = difl[j];
dj = d__[j];
dsigj = -dsigma[j];
if (j < *k) {
difrj = -difr[j + difr_dim1];
dsigjp = -dsigma[j + 1];
}
work[j] = -z__[j] / diflj / (dsigma[j] + dj);
i__2 = j - 1;
for (i__ = 1; i__ <= i__2; ++i__) {
work[i__] = z__[i__] / (slamc3_(&dsigma[i__], &dsigj) - diflj) / (
dsigma[i__] + dj);
/* L60: */
}
i__2 = *k;
for (i__ = j + 1; i__ <= i__2; ++i__) {
work[i__] = z__[i__] / (slamc3_(&dsigma[i__], &dsigjp) + difrj) /
(dsigma[i__] + dj);
/* L70: */
}
temp = snrm2_(k, &work[1], &c__1);
work[iwk2i + j] = sdot_(k, &work[1], &c__1, &vf[1], &c__1) / temp;
work[iwk3i + j] = sdot_(k, &work[1], &c__1, &vl[1], &c__1) / temp;
if (*icompq == 1) {
difr[j + (difr_dim1 << 1)] = temp;
}
/* L80: */
}
scopy_(k, &work[iwk2], &c__1, &vf[1], &c__1);
scopy_(k, &work[iwk3], &c__1, &vl[1], &c__1);
return 0;
/* End of SLASD8 */
} /* slasd8_ */
/* Subroutine */ int cgeesx_(char *jobvs, char *sort, L_fp select, char *
sense, integer *n, complex *a, integer *lda, integer *sdim, complex *
w, complex *vs, integer *ldvs, real *rconde, real *rcondv, complex *
work, integer *lwork, real *rwork, logical *bwork, 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
March 31, 1993
Purpose
=======
CGEESX computes for an N-by-N complex nonsymmetric matrix A, the
eigenvalues, the Schur form T, and, optionally, the matrix of Schur
vectors Z. This gives the Schur factorization A = Z*T*(Z**H).
Optionally, it also orders the eigenvalues on the diagonal of the
Schur form so that selected eigenvalues are at the top left;
computes a reciprocal condition number for the average of the
selected eigenvalues (RCONDE); and computes a reciprocal condition
number for the right invariant subspace corresponding to the
selected eigenvalues (RCONDV). The leading columns of Z form an
orthonormal basis for this invariant subspace.
For further explanation of the reciprocal condition numbers RCONDE
and RCONDV, see Section 4.10 of the LAPACK Users' Guide (where
these quantities are called s and sep respectively).
A complex matrix is in Schur form if it is upper triangular.
Arguments
=========
JOBVS (input) CHARACTER*1
= 'N': Schur vectors are not computed;
= 'V': Schur vectors are computed.
SORT (input) CHARACTER*1
Specifies whether or not to order the eigenvalues on the
diagonal of the Schur form.
= 'N': Eigenvalues are not ordered;
= 'S': Eigenvalues are ordered (see SELECT).
SELECT (input) LOGICAL FUNCTION of one COMPLEX argument
SELECT must be declared EXTERNAL in the calling subroutine.
If SORT = 'S', SELECT is used to select eigenvalues to order
to the top left of the Schur form.
If SORT = 'N', SELECT is not referenced.
An eigenvalue W(j) is selected if SELECT(W(j)) is true.
SENSE (input) CHARACTER*1
Determines which reciprocal condition numbers are computed.
= 'N': None are computed;
= 'E': Computed for average of selected eigenvalues only;
= 'V': Computed for selected right invariant subspace only;
= 'B': Computed for both.
If SENSE = 'E', 'V' or 'B', SORT must equal 'S'.
N (input) INTEGER
The order of the matrix A. N >= 0.
A (input/output) COMPLEX array, dimension (LDA, N)
On entry, the N-by-N matrix A.
On exit, A is overwritten by its Schur form T.
LDA (input) INTEGER
The leading dimension of the array A. LDA >= max(1,N).
SDIM (output) INTEGER
If SORT = 'N', SDIM = 0.
If SORT = 'S', SDIM = number of eigenvalues for which
SELECT is true.
W (output) COMPLEX array, dimension (N)
W contains the computed eigenvalues, in the same order
that they appear on the diagonal of the output Schur form T.
VS (output) COMPLEX array, dimension (LDVS,N)
If JOBVS = 'V', VS contains the unitary matrix Z of Schur
vectors.
If JOBVS = 'N', VS is not referenced.
LDVS (input) INTEGER
The leading dimension of the array VS. LDVS >= 1, and if
JOBVS = 'V', LDVS >= N.
RCONDE (output) REAL
If SENSE = 'E' or 'B', RCONDE contains the reciprocal
condition number for the average of the selected eigenvalues.
Not referenced if SENSE = 'N' or 'V'.
RCONDV (output) REAL
If SENSE = 'V' or 'B', RCONDV contains the reciprocal
condition number for the selected right invariant subspace.
Not referenced if SENSE = 'N' or 'E'.
WORK (workspace/output) COMPLEX 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(1,2*N).
Also, if SENSE = 'E' or 'V' or 'B', LWORK >= 2*SDIM*(N-SDIM),
where SDIM is the number of selected eigenvalues computed by
this routine. Note that 2*SDIM*(N-SDIM) <= N*N/2.
For good performance, LWORK must generally be larger.
RWORK (workspace) REAL array, dimension (N)
BWORK (workspace) LOGICAL array, dimension (N)
Not referenced if SORT = 'N'.
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value.
> 0: if INFO = i, and i is
<= N: the QR algorithm failed to compute all the
eigenvalues; elements 1:ILO-1 and i+1:N of W
contain those eigenvalues which have converged; if
JOBVS = 'V', VS contains the transformation which
reduces A to its partially converged Schur form.
= N+1: the eigenvalues could not be reordered because some
eigenvalues were too close to separate (the problem
is very ill-conditioned);
= N+2: after reordering, roundoff changed values of some
complex eigenvalues so that leading eigenvalues in
the Schur form no longer satisfy SELECT=.TRUE. This
could also be caused by underflow due to scaling.
=====================================================================
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, vs_dim1, vs_offset, i__1, i__2, i__3, i__4;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
static integer ibal, maxb;
static real anrm;
static integer ierr, itau, iwrk, i, k, icond, ieval;
extern logical lsame_(char *, char *);
extern /* Subroutine */ int ccopy_(integer *, complex *, integer *,
complex *, integer *), cgebak_(char *, char *, integer *, integer
*, integer *, real *, integer *, complex *, integer *, integer *), cgebal_(char *, integer *, complex *, integer *,
integer *, integer *, real *, integer *), slabad_(real *,
real *);
static logical scalea;
extern doublereal clange_(char *, integer *, integer *, complex *,
integer *, real *);
static real cscale;
extern /* Subroutine */ int cgehrd_(integer *, integer *, integer *,
complex *, integer *, complex *, complex *, integer *, integer *),
clascl_(char *, integer *, integer *, real *, real *, integer *,
integer *, complex *, integer *, integer *);
extern doublereal slamch_(char *);
extern /* Subroutine */ int clacpy_(char *, integer *, integer *, complex
*, integer *, complex *, integer *), xerbla_(char *,
integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *, ftnlen, ftnlen);
static real bignum;
extern /* Subroutine */ int slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *), chseqr_(char *, char *, integer *, integer *, integer *,
complex *, integer *, complex *, complex *, integer *, complex *,
integer *, integer *), cunghr_(integer *, integer
*, integer *, complex *, integer *, complex *, complex *, integer
*, integer *);
static logical wantsb;
extern /* Subroutine */ int ctrsen_(char *, char *, logical *, integer *,
complex *, integer *, complex *, integer *, complex *, integer *,
real *, real *, complex *, integer *, integer *);
static logical wantse;
static integer minwrk, maxwrk;
static logical wantsn;
static real smlnum;
static integer hswork;
static logical wantst, wantsv, wantvs;
static integer ihi, ilo;
static real dum[1], eps;
#define W(I) w[(I)-1]
#define WORK(I) work[(I)-1]
#define RWORK(I) rwork[(I)-1]
#define BWORK(I) bwork[(I)-1]
#define A(I,J) a[(I)-1 + ((J)-1)* ( *lda)]
#define VS(I,J) vs[(I)-1 + ((J)-1)* ( *ldvs)]
*info = 0;
wantvs = lsame_(jobvs, "V");
wantst = lsame_(sort, "S");
wantsn = lsame_(sense, "N");
wantse = lsame_(sense, "E");
wantsv = lsame_(sense, "V");
wantsb = lsame_(sense, "B");
if (! wantvs && ! lsame_(jobvs, "N")) {
*info = -1;
} else if (! wantst && ! lsame_(sort, "N")) {
*info = -2;
} else if (! (wantsn || wantse || wantsv || wantsb) || ! wantst && !
wantsn) {
*info = -4;
} else if (*n < 0) {
*info = -5;
} else if (*lda < max(1,*n)) {
*info = -7;
} else if (*ldvs < 1 || wantvs && *ldvs < *n) {
*info = -11;
}
/* Compute workspace
(Note: Comments in the code beginning "Workspace:" describe the
minimal amount of real workspace needed at that point in the
code, as well as the preferred amount for good performance.
CWorkspace refers to complex workspace, and RWorkspace to real
workspace. NB refers to the optimal block size for the
immediately following subroutine, as returned by ILAENV.
HSWORK refers to the workspace preferred by CHSEQR, as
calculated below. HSWORK is computed assuming ILO=1 and IHI=N,
the worst case.
If SENSE = 'E', 'V' or 'B', then the amount of workspace needed
depends on SDIM, which is computed by the routine CTRSEN later
in the code.) */
minwrk = 1;
if (*info == 0 && *lwork >= 1) {
maxwrk = *n + *n * ilaenv_(&c__1, "CGEHRD", " ", n, &c__1, n, &c__0,
6L, 1L);
/* Computing MAX */
i__1 = 1, i__2 = *n << 1;
minwrk = max(i__1,i__2);
if (! wantvs) {
/* Computing MAX */
i__1 = ilaenv_(&c__8, "CHSEQR", "SN", 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, "CHSEQR", "SN", 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 = max(maxwrk,hswork);
maxwrk = max(i__1,1);
} else {
/* Computing MAX */
i__1 = maxwrk, i__2 = *n + (*n - 1) * ilaenv_(&c__1, "CUNGHR",
" ", n, &c__1, n, &c_n1, 6L, 1L);
maxwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = ilaenv_(&c__8, "CHSEQR", "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, "CHSEQR", "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 = max(maxwrk,hswork);
maxwrk = max(i__1,1);
}
WORK(1).r = (real) maxwrk, WORK(1).i = 0.f;
}
if (*lwork < minwrk) {
*info = -15;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CGEESX", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0) {
*sdim = 0;
return 0;
}
/* Get machine constants */
eps = slamch_("P");
smlnum = slamch_("S");
bignum = 1.f / smlnum;
slabad_(&smlnum, &bignum);
smlnum = sqrt(smlnum) / eps;
bignum = 1.f / smlnum;
/* Scale A if max element outside range [SMLNUM,BIGNUM] */
anrm = clange_("M", n, n, &A(1,1), lda, dum);
scalea = FALSE_;
if (anrm > 0.f && anrm < smlnum) {
scalea = TRUE_;
cscale = smlnum;
} else if (anrm > bignum) {
scalea = TRUE_;
cscale = bignum;
}
if (scalea) {
clascl_("G", &c__0, &c__0, &anrm, &cscale, n, n, &A(1,1), lda, &
ierr);
}
/* Permute the matrix to make it more nearly triangular
(CWorkspace: none)
(RWorkspace: need N) */
ibal = 1;
cgebal_("P", n, &A(1,1), lda, &ilo, &ihi, &RWORK(ibal), &ierr);
/* Reduce to upper Hessenberg form
(CWorkspace: need 2*N, prefer N+N*NB)
(RWorkspace: none) */
itau = 1;
iwrk = *n + itau;
i__1 = *lwork - iwrk + 1;
cgehrd_(n, &ilo, &ihi, &A(1,1), lda, &WORK(itau), &WORK(iwrk), &i__1,
&ierr);
if (wantvs) {
/* Copy Householder vectors to VS */
clacpy_("L", n, n, &A(1,1), lda, &VS(1,1), ldvs);
/* Generate unitary matrix in VS
(CWorkspace: need 2*N-1, prefer N+(N-1)*NB)
(RWorkspace: none) */
i__1 = *lwork - iwrk + 1;
cunghr_(n, &ilo, &ihi, &VS(1,1), ldvs, &WORK(itau), &WORK(iwrk),
&i__1, &ierr);
}
*sdim = 0;
/* Perform QR iteration, accumulating Schur vectors in VS if desired
(CWorkspace: need 1, prefer HSWORK (see comments) )
(RWorkspace: none) */
iwrk = itau;
i__1 = *lwork - iwrk + 1;
chseqr_("S", jobvs, n, &ilo, &ihi, &A(1,1), lda, &W(1), &VS(1,1), ldvs, &WORK(iwrk), &i__1, &ieval);
if (ieval > 0) {
*info = ieval;
}
/* Sort eigenvalues if desired */
if (wantst && *info == 0) {
if (scalea) {
clascl_("G", &c__0, &c__0, &cscale, &anrm, n, &c__1, &W(1), n, &
ierr);
}
i__1 = *n;
for (i = 1; i <= *n; ++i) {
BWORK(i) = (*select)(&W(i));
/* L10: */
}
/* Reorder eigenvalues, transform Schur vectors, and compute
reciprocal condition numbers
(CWorkspace: if SENSE is not 'N', need 2*SDIM*(N-SDIM)
otherwise, need none )
(RWorkspace: none) */
i__1 = *lwork - iwrk + 1;
ctrsen_(sense, jobvs, &BWORK(1), n, &A(1,1), lda, &VS(1,1),
ldvs, &W(1), sdim, rconde, rcondv, &WORK(iwrk), &i__1, &
icond);
if (! wantsn) {
/* Computing MAX */
i__1 = maxwrk, i__2 = (*sdim << 1) * (*n - *sdim);
maxwrk = max(i__1,i__2);
}
if (icond == -14) {
/* Not enough complex workspace */
*info = -15;
}
}
if (wantvs) {
/* Undo balancing
(CWorkspace: none)
(RWorkspace: need N) */
cgebak_("P", "R", n, &ilo, &ihi, &RWORK(ibal), n, &VS(1,1),
ldvs, &ierr);
}
if (scalea) {
/* Undo scaling for the Schur form of A */
clascl_("U", &c__0, &c__0, &cscale, &anrm, n, n, &A(1,1), lda, &
ierr);
i__1 = *lda + 1;
ccopy_(n, &A(1,1), &i__1, &W(1), &c__1);
if ((wantsv || wantsb) && *info == 0) {
dum[0] = *rcondv;
slascl_("G", &c__0, &c__0, &cscale, &anrm, &c__1, &c__1, dum, &
c__1, &ierr);
*rcondv = dum[0];
}
}
WORK(1).r = (real) maxwrk, WORK(1).i = 0.f;
return 0;
/* End of CGEESX */
} /* cgeesx_ */
doublereal sqrt12_(integer *m, integer *n, real *a, integer *lda, real *s,
real *work, integer *lwork)
{
/* System generated locals */
integer a_dim1, a_offset, i__1, i__2;
real ret_val;
/* Local variables */
static integer iscl, info;
static real anrm;
extern doublereal snrm2_(integer *, real *, integer *);
static integer i__, j;
extern doublereal sasum_(integer *, real *, integer *);
static real dummy[1];
extern /* Subroutine */ int saxpy_(integer *, real *, real *, integer *,
real *, integer *), sgebd2_(integer *, integer *, real *, integer
*, real *, real *, real *, real *, real *, integer *), slabad_(
real *, real *);
static integer mn;
extern doublereal slamch_(char *), slange_(char *, integer *,
integer *, real *, integer *, real *);
extern /* Subroutine */ int xerbla_(char *, integer *);
static real bignum;
extern /* Subroutine */ int slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *), slaset_(char *, integer *, integer *, real *, real *,
real *, integer *), sbdsqr_(char *, integer *, integer *,
integer *, integer *, real *, real *, real *, integer *, real *,
integer *, real *, integer *, real *, integer *);
static real smlnum, nrmsvl;
#define a_ref(a_1,a_2) a[(a_2)*a_dim1 + a_1]
/* -- LAPACK test routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
September 30, 1994
Purpose
=======
SQRT12 computes the singular values `svlues' of the upper trapezoid
of A(1:M,1:N) and returns the ratio
|| s - svlues||/(||svlues||*eps*max(M,N))
Arguments
=========
M (input) INTEGER
The number of rows of the matrix A.
N (input) INTEGER
The number of columns of the matrix A.
A (input) REAL array, dimension (LDA,N)
The M-by-N matrix A. Only the upper trapezoid is referenced.
LDA (input) INTEGER
The leading dimension of the array A.
S (input) REAL array, dimension (min(M,N))
The singular values of the matrix A.
WORK (workspace) REAL array, dimension (LWORK)
LWORK (input) INTEGER
The length of the array WORK. LWORK >= M*N + 4*min(M,N) +
max(M,N).
=====================================================================
Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1 * 1;
a -= a_offset;
--s;
--work;
/* Function Body */
ret_val = 0.f;
/* Test that enough workspace is supplied */
if (*lwork < *m * *n + (min(*m,*n) << 2) + max(*m,*n)) {
xerbla_("SQRT12", &c__7);
return ret_val;
}
/* Quick return if possible */
mn = min(*m,*n);
if ((real) mn <= 0.f) {
return ret_val;
}
nrmsvl = snrm2_(&mn, &s[1], &c__1);
/* Copy upper triangle of A into work */
slaset_("Full", m, n, &c_b6, &c_b6, &work[1], m);
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
i__2 = min(j,*m);
for (i__ = 1; i__ <= i__2; ++i__) {
work[(j - 1) * *m + i__] = a_ref(i__, j);
/* L10: */
}
/* L20: */
}
/* Get machine parameters */
smlnum = slamch_("S") / slamch_("P");
bignum = 1.f / smlnum;
slabad_(&smlnum, &bignum);
/* Scale work if max entry outside range [SMLNUM,BIGNUM] */
anrm = slange_("M", m, n, &work[1], m, dummy);
iscl = 0;
if (anrm > 0.f && anrm < smlnum) {
/* Scale matrix norm up to SMLNUM */
slascl_("G", &c__0, &c__0, &anrm, &smlnum, m, n, &work[1], m, &info);
iscl = 1;
} else if (anrm > bignum) {
/* Scale matrix norm down to BIGNUM */
slascl_("G", &c__0, &c__0, &anrm, &bignum, m, n, &work[1], m, &info);
iscl = 1;
}
if (anrm != 0.f) {
/* Compute SVD of work */
sgebd2_(m, n, &work[1], m, &work[*m * *n + 1], &work[*m * *n + mn + 1]
, &work[*m * *n + (mn << 1) + 1], &work[*m * *n + mn * 3 + 1],
&work[*m * *n + (mn << 2) + 1], &info);
sbdsqr_("Upper", &mn, &c__0, &c__0, &c__0, &work[*m * *n + 1], &work[*
m * *n + mn + 1], dummy, &mn, dummy, &c__1, dummy, &mn, &work[
*m * *n + (mn << 1) + 1], &info);
if (iscl == 1) {
if (anrm > bignum) {
slascl_("G", &c__0, &c__0, &bignum, &anrm, &mn, &c__1, &work[*
m * *n + 1], &mn, &info);
}
if (anrm < smlnum) {
slascl_("G", &c__0, &c__0, &smlnum, &anrm, &mn, &c__1, &work[*
m * *n + 1], &mn, &info);
}
}
} else {
i__1 = mn;
for (i__ = 1; i__ <= i__1; ++i__) {
work[*m * *n + i__] = 0.f;
/* L30: */
}
}
/* Compare s and singular values of work */
saxpy_(&mn, &c_b33, &s[1], &c__1, &work[*m * *n + 1], &c__1);
ret_val = sasum_(&mn, &work[*m * *n + 1], &c__1) / (slamch_("Epsilon") * (real) max(*m,*n));
if (nrmsvl != 0.f) {
ret_val /= nrmsvl;
}
return ret_val;
/* End of SQRT12 */
} /* sqrt12_ */
/* Subroutine */ int ssterf_(integer *n, real *d, real *e, 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
=======
SSTERF 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) REAL 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) REAL 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
Function Body */
/* Table of constant values */
static integer c__0 = 0;
static integer c__1 = 1;
static real c_b32 = 1.f;
/* System generated locals */
integer i__1;
real r__1, r__2;
/* Builtin functions */
double sqrt(doublereal), r_sign(real *, real *);
/* Local variables */
static real oldc;
static integer lend, jtot;
extern /* Subroutine */ int slae2_(real *, real *, real *, real *, real *)
;
static real c;
static integer i, l, m;
static real p, gamma, r, s, alpha, sigma, anorm;
static integer l1, lendm1, lendp1;
static real bb;
extern doublereal slapy2_(real *, real *);
static integer iscale;
static real oldgam;
extern doublereal slamch_(char *);
static real safmin;
extern /* Subroutine */ int xerbla_(char *, integer *);
static real safmax;
extern /* Subroutine */ int slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *);
static integer lendsv;
static real ssfmin;
static integer nmaxit;
static real ssfmax;
extern doublereal slanst_(char *, integer *, real *, real *);
extern /* Subroutine */ int slasrt_(char *, integer *, real *, integer *);
static integer lm1, mm1, nm1;
static real rt1, rt2, eps, rte;
static integer lsv;
static real tst, eps2;
#define E(I) e[(I)-1]
#define D(I) d[(I)-1]
*info = 0;
/* Quick return if possible */
if (*n < 0) {
*info = -1;
i__1 = -(*info);
xerbla_("SSTERF", &i__1);
return 0;
}
if (*n <= 1) {
return 0;
}
/* Determine the unit roundoff for this environment. */
eps = slamch_("E");
/* Computing 2nd power */
r__1 = eps;
eps2 = r__1 * r__1;
safmin = slamch_("S");
safmax = 1.f / safmin;
ssfmax = sqrt(safmax) / 3.f;
ssfmin = sqrt(safmin) / eps2;
/* Compute the eigenvalues of the tridiagonal matrix. */
nmaxit = *n * 30;
sigma = 0.f;
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 L170;
}
if (l1 > 1) {
E(l1 - 1) = 0.f;
}
if (l1 <= nm1) {
i__1 = nm1;
for (m = l1; m <= nm1; ++m) {
tst = (r__1 = E(m), dabs(r__1));
if (tst == 0.f) {
goto L30;
}
if (tst <= sqrt((r__1 = D(m), dabs(r__1))) * sqrt((r__2 = D(m + 1)
, dabs(r__2))) * eps) {
E(m) = 0.f;
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 = slanst_("I", &i__1, &D(l), &E(l));
iscale = 0;
if (anorm > ssfmax) {
iscale = 1;
i__1 = lend - l + 1;
slascl_("G", &c__0, &c__0, &anorm, &ssfmax, &i__1, &c__1, &D(l), n,
info);
i__1 = lend - l;
slascl_("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;
slascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &D(l), n,
info);
i__1 = lend - l;
slascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &E(l), n,
info);
}
i__1 = lend - 1;
for (i = l; i <= lend-1; ++i) {
/* Computing 2nd power */
r__1 = E(i);
E(i) = r__1 * r__1;
/* L40: */
}
/* Choose between QL and QR iteration */
if ((r__1 = D(lend), dabs(r__1)) < (r__2 = D(l), dabs(r__2))) {
lend = lsv;
l = lendsv;
}
if (lend >= l) {
/* QL Iteration
Look for small subdiagonal element. */
L50:
if (l != lend) {
lendm1 = lend - 1;
i__1 = lendm1;
for (m = l; m <= lendm1; ++m) {
tst = (r__1 = E(m), dabs(r__1));
if (tst <= eps2 * (r__1 = D(m) * D(m + 1), dabs(r__1))) {
goto L70;
}
/* L60: */
}
}
m = lend;
L70:
if (m < lend) {
E(m) = 0.f;
}
p = D(l);
if (m == l) {
goto L90;
}
/* If remaining matrix is 2 by 2, use SLAE2 to compute its
eigenvalues. */
if (m == l + 1) {
rte = sqrt(E(l));
slae2_(&D(l), &rte, &D(l + 1), &rt1, &rt2);
D(l) = rt1;
D(l + 1) = rt2;
E(l) = 0.f;
l += 2;
if (l <= lend) {
goto L50;
}
goto L150;
}
if (jtot == nmaxit) {
goto L150;
}
++jtot;
/* Form shift. */
rte = sqrt(E(l));
sigma = (D(l + 1) - p) / (rte * 2.f);
r = slapy2_(&sigma, &c_b32);
sigma = p - rte / (sigma + r_sign(&r, &sigma));
c = 1.f;
s = 0.f;
gamma = D(m) - sigma;
p = gamma * gamma;
/* Inner loop */
mm1 = m - 1;
i__1 = l;
for (i = mm1; i >= l; --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.f) {
p = gamma * gamma / c;
} else {
p = oldc * bb;
}
/* L80: */
}
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:
if (l != lend) {
lendp1 = lend + 1;
i__1 = lendp1;
for (m = l; m >= lendp1; --m) {
tst = (r__1 = E(m - 1), dabs(r__1));
if (tst <= eps2 * (r__1 = D(m) * D(m - 1), dabs(r__1))) {
goto L120;
}
/* L110: */
}
}
m = lend;
L120:
if (m > lend) {
E(m - 1) = 0.f;
}
p = D(l);
if (m == l) {
goto L140;
}
/* If remaining matrix is 2 by 2, use SLAE2 to compute its
eigenvalues. */
if (m == l - 1) {
rte = sqrt(E(l - 1));
slae2_(&D(l), &rte, &D(l - 1), &rt1, &rt2);
D(l) = rt1;
D(l - 1) = rt2;
E(l - 1) = 0.f;
l += -2;
if (l >= lend) {
goto L100;
}
goto L150;
}
if (jtot == nmaxit) {
goto L150;
}
++jtot;
/* Form shift. */
rte = sqrt(E(l - 1));
sigma = (D(l - 1) - p) / (rte * 2.f);
r = slapy2_(&sigma, &c_b32);
sigma = p - rte / (sigma + r_sign(&r, &sigma));
c = 1.f;
s = 0.f;
gamma = D(m) - sigma;
p = gamma * gamma;
/* Inner loop */
lm1 = l - 1;
i__1 = lm1;
for (i = m; i <= lm1; ++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.f) {
p = gamma * gamma / c;
} else {
p = oldc * bb;
}
/* L130: */
}
E(lm1) = 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) {
i__1 = lendsv - lsv + 1;
slascl_("G", &c__0, &c__0, &ssfmax, &anorm, &i__1, &c__1, &D(lsv), n,
info);
}
if (iscale == 2) {
i__1 = lendsv - lsv + 1;
slascl_("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) {
i__1 = *n - 1;
for (i = 1; i <= *n-1; ++i) {
if (E(i) != 0.f) {
++(*info);
}
/* L160: */
}
return 0;
}
goto L10;
/* Sort eigenvalues in increasing order. */
L170:
slasrt_("I", n, &D(1), info);
return 0;
/* End of SSTERF */
} /* ssterf_ */
/* Subroutine */ int sggesx_(char *jobvsl, char *jobvsr, char *sort, L_fp
selctg, char *sense, integer *n, real *a, integer *lda, real *b,
integer *ldb, integer *sdim, real *alphar, real *alphai, real *beta,
real *vsl, integer *ldvsl, real *vsr, integer *ldvsr, real *rconde,
real *rcondv, real *work, integer *lwork, integer *iwork, integer *
liwork, logical *bwork, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, vsl_dim1, vsl_offset,
vsr_dim1, vsr_offset, i__1, i__2;
real r__1;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
integer i__, ip;
real pl, pr, dif[2];
integer ihi, ilo;
real eps;
integer ijob;
real anrm, bnrm;
integer ierr, itau, iwrk, lwrk;
extern logical lsame_(char *, char *);
integer ileft, icols;
logical cursl, ilvsl, ilvsr;
integer irows;
logical lst2sl;
extern /* Subroutine */ int slabad_(real *, real *), sggbak_(char *, char
*, integer *, integer *, integer *, real *, real *, integer *,
real *, integer *, integer *), sggbal_(char *,
integer *, real *, integer *, real *, integer *, integer *,
integer *, real *, real *, real *, integer *);
logical ilascl, ilbscl;
extern doublereal slamch_(char *), slange_(char *, integer *,
integer *, real *, integer *, real *);
real safmin;
extern /* Subroutine */ int sgghrd_(char *, char *, integer *, integer *,
integer *, real *, integer *, real *, integer *, real *, integer *
, real *, integer *, integer *);
real safmax;
extern /* Subroutine */ int xerbla_(char *, integer *);
real bignum;
extern /* Subroutine */ int slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *);
integer ijobvl, iright;
extern /* Subroutine */ int sgeqrf_(integer *, integer *, real *, integer
*, real *, real *, integer *, integer *);
integer ijobvr;
extern /* Subroutine */ int slacpy_(char *, integer *, integer *, real *,
integer *, real *, integer *);
logical wantsb, wantse, lastsl;
integer liwmin;
real anrmto, bnrmto;
integer minwrk, maxwrk;
logical wantsn;
real smlnum;
extern /* Subroutine */ int shgeqz_(char *, char *, char *, integer *,
integer *, integer *, real *, integer *, real *, integer *, real *
, real *, real *, real *, integer *, real *, integer *, real *,
integer *, integer *), slaset_(char *,
integer *, integer *, real *, real *, real *, integer *),
sorgqr_(integer *, integer *, integer *, real *, integer *, real *
, real *, integer *, integer *), stgsen_(integer *, logical *,
logical *, logical *, integer *, real *, integer *, real *,
integer *, real *, real *, real *, real *, integer *, real *,
integer *, integer *, real *, real *, real *, real *, integer *,
integer *, integer *, integer *);
logical wantst, lquery, wantsv;
extern /* Subroutine */ int sormqr_(char *, char *, integer *, integer *,
integer *, real *, integer *, real *, real *, integer *, real *,
integer *, integer *);
/* -- LAPACK driver routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* .. Function Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SGGESX computes for a pair of N-by-N real nonsymmetric matrices */
/* (A,B), the generalized eigenvalues, the real Schur form (S,T), and, */
/* optionally, the left and/or right matrices of Schur vectors (VSL and */
/* VSR). This gives the generalized Schur factorization */
/* (A,B) = ( (VSL) S (VSR)**T, (VSL) T (VSR)**T ) */
/* Optionally, it also orders the eigenvalues so that a selected cluster */
/* of eigenvalues appears in the leading diagonal blocks of the upper */
/* quasi-triangular matrix S and the upper triangular matrix T; computes */
/* a reciprocal condition number for the average of the selected */
/* eigenvalues (RCONDE); and computes a reciprocal condition number for */
/* the right and left deflating subspaces corresponding to the selected */
/* eigenvalues (RCONDV). The leading columns of VSL and VSR then form */
/* an orthonormal basis for the corresponding left and right eigenspaces */
/* (deflating subspaces). */
/* A generalized eigenvalue for a pair of matrices (A,B) is a scalar w */
/* or a ratio alpha/beta = w, such that A - w*B is singular. It is */
/* usually represented as the pair (alpha,beta), as there is a */
/* reasonable interpretation for beta=0 or for both being zero. */
/* A pair of matrices (S,T) is in generalized real Schur form if T is */
/* upper triangular with non-negative diagonal and S is block upper */
/* triangular with 1-by-1 and 2-by-2 blocks. 1-by-1 blocks correspond */
/* to real generalized eigenvalues, while 2-by-2 blocks of S will be */
/* "standardized" by making the corresponding elements of T have the */
/* form: */
/* [ a 0 ] */
/* [ 0 b ] */
/* and the pair of corresponding 2-by-2 blocks in S and T will have a */
/* complex conjugate pair of generalized eigenvalues. */
/* Arguments */
/* ========= */
/* JOBVSL (input) CHARACTER*1 */
/* = 'N': do not compute the left Schur vectors; */
/* = 'V': compute the left Schur vectors. */
/* JOBVSR (input) CHARACTER*1 */
/* = 'N': do not compute the right Schur vectors; */
/* = 'V': compute the right Schur vectors. */
/* SORT (input) CHARACTER*1 */
/* Specifies whether or not to order the eigenvalues on the */
/* diagonal of the generalized Schur form. */
/* = 'N': Eigenvalues are not ordered; */
/* = 'S': Eigenvalues are ordered (see SELCTG). */
/* SELCTG (external procedure) LOGICAL FUNCTION of three REAL arguments */
/* SELCTG must be declared EXTERNAL in the calling subroutine. */
/* If SORT = 'N', SELCTG is not referenced. */
/* If SORT = 'S', SELCTG is used to select eigenvalues to sort */
/* to the top left of the Schur form. */
/* An eigenvalue (ALPHAR(j)+ALPHAI(j))/BETA(j) is selected if */
/* SELCTG(ALPHAR(j),ALPHAI(j),BETA(j)) is true; i.e. if either */
/* one of a complex conjugate pair of eigenvalues is selected, */
/* then both complex eigenvalues are selected. */
/* Note that a selected complex eigenvalue may no longer satisfy */
/* SELCTG(ALPHAR(j),ALPHAI(j),BETA(j)) = .TRUE. after ordering, */
/* since ordering may change the value of complex eigenvalues */
/* (especially if the eigenvalue is ill-conditioned), in this */
/* case INFO is set to N+3. */
/* SENSE (input) CHARACTER*1 */
/* Determines which reciprocal condition numbers are computed. */
/* = 'N' : None are computed; */
/* = 'E' : Computed for average of selected eigenvalues only; */
/* = 'V' : Computed for selected deflating subspaces only; */
/* = 'B' : Computed for both. */
/* If SENSE = 'E', 'V', or 'B', SORT must equal 'S'. */
/* N (input) INTEGER */
/* The order of the matrices A, B, VSL, and VSR. N >= 0. */
/* A (input/output) REAL array, dimension (LDA, N) */
/* On entry, the first of the pair of matrices. */
/* On exit, A has been overwritten by its generalized Schur */
/* form S. */
/* LDA (input) INTEGER */
/* The leading dimension of A. LDA >= max(1,N). */
/* B (input/output) REAL array, dimension (LDB, N) */
/* On entry, the second of the pair of matrices. */
/* On exit, B has been overwritten by its generalized Schur */
/* form T. */
/* LDB (input) INTEGER */
/* The leading dimension of B. LDB >= max(1,N). */
/* SDIM (output) INTEGER */
/* If SORT = 'N', SDIM = 0. */
/* If SORT = 'S', SDIM = number of eigenvalues (after sorting) */
/* for which SELCTG is true. (Complex conjugate pairs for which */
/* SELCTG is true for either eigenvalue count as 2.) */
/* ALPHAR (output) REAL array, dimension (N) */
/* ALPHAI (output) REAL array, dimension (N) */
/* BETA (output) REAL array, dimension (N) */
/* On exit, (ALPHAR(j) + ALPHAI(j)*i)/BETA(j), j=1,...,N, will */
/* be the generalized eigenvalues. ALPHAR(j) + ALPHAI(j)*i */
/* and BETA(j),j=1,...,N are the diagonals of the complex Schur */
/* form (S,T) that would result if the 2-by-2 diagonal blocks of */
/* the real Schur form of (A,B) were further reduced to */
/* triangular form using 2-by-2 complex unitary transformations. */
/* If ALPHAI(j) is zero, then the j-th eigenvalue is real; if */
/* positive, then the j-th and (j+1)-st eigenvalues are a */
/* complex conjugate pair, with ALPHAI(j+1) negative. */
/* Note: the quotients ALPHAR(j)/BETA(j) and ALPHAI(j)/BETA(j) */
/* may easily over- or underflow, and BETA(j) may even be zero. */
/* Thus, the user should avoid naively computing the ratio. */
/* However, ALPHAR and ALPHAI will be always less than and */
/* usually comparable with norm(A) in magnitude, and BETA always */
/* less than and usually comparable with norm(B). */
/* VSL (output) REAL array, dimension (LDVSL,N) */
/* If JOBVSL = 'V', VSL will contain the left Schur vectors. */
/* Not referenced if JOBVSL = 'N'. */
/* LDVSL (input) INTEGER */
/* The leading dimension of the matrix VSL. LDVSL >=1, and */
/* if JOBVSL = 'V', LDVSL >= N. */
/* VSR (output) REAL array, dimension (LDVSR,N) */
/* If JOBVSR = 'V', VSR will contain the right Schur vectors. */
/* Not referenced if JOBVSR = 'N'. */
/* LDVSR (input) INTEGER */
/* The leading dimension of the matrix VSR. LDVSR >= 1, and */
/* if JOBVSR = 'V', LDVSR >= N. */
/* RCONDE (output) REAL array, dimension ( 2 ) */
/* If SENSE = 'E' or 'B', RCONDE(1) and RCONDE(2) contain the */
/* reciprocal condition numbers for the average of the selected */
/* eigenvalues. */
/* Not referenced if SENSE = 'N' or 'V'. */
/* RCONDV (output) REAL array, dimension ( 2 ) */
/* If SENSE = 'V' or 'B', RCONDV(1) and RCONDV(2) contain the */
/* reciprocal condition numbers for the selected deflating */
/* subspaces. */
/* Not referenced if SENSE = 'N' or 'E'. */
/* WORK (workspace/output) REAL 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. */
/* If N = 0, LWORK >= 1, else if SENSE = 'E', 'V', or 'B', */
/* LWORK >= max( 8*N, 6*N+16, 2*SDIM*(N-SDIM) ), else */
/* LWORK >= max( 8*N, 6*N+16 ). */
/* Note that 2*SDIM*(N-SDIM) <= N*N/2. */
/* Note also that an error is only returned if */
/* LWORK < max( 8*N, 6*N+16), but if SENSE = 'E' or 'V' or 'B' */
/* this may not be large enough. */
/* If LWORK = -1, then a workspace query is assumed; the routine */
/* only calculates the bound on the optimal size of the WORK */
/* array and the minimum size of the IWORK array, returns these */
/* values as the first entries of the WORK and IWORK arrays, and */
/* no error message related to LWORK or LIWORK is issued by */
/* XERBLA. */
/* IWORK (workspace) INTEGER array, dimension (MAX(1,LIWORK)) */
/* On exit, if INFO = 0, IWORK(1) returns the minimum LIWORK. */
/* LIWORK (input) INTEGER */
/* The dimension of the array IWORK. */
/* If SENSE = 'N' or N = 0, LIWORK >= 1, otherwise */
/* LIWORK >= N+6. */
/* If LIWORK = -1, then a workspace query is assumed; the */
/* routine only calculates the bound on the optimal size of the */
/* WORK array and the minimum size of the IWORK array, returns */
/* these values as the first entries of the WORK and IWORK */
/* arrays, and no error message related to LWORK or LIWORK is */
/* issued by XERBLA. */
/* BWORK (workspace) LOGICAL array, dimension (N) */
/* Not referenced if SORT = 'N'. */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value. */
/* = 1,...,N: */
/* The QZ iteration failed. (A,B) are not in Schur */
/* form, but ALPHAR(j), ALPHAI(j), and BETA(j) should */
/* be correct for j=INFO+1,...,N. */
/* > N: =N+1: other than QZ iteration failed in SHGEQZ */
/* =N+2: after reordering, roundoff changed values of */
/* some complex eigenvalues so that leading */
/* eigenvalues in the Generalized Schur form no */
/* longer satisfy SELCTG=.TRUE. This could also */
/* be caused due to scaling. */
/* =N+3: reordering failed in STGSEN. */
/* Further details */
/* =============== */
/* An approximate (asymptotic) bound on the average absolute error of */
/* the selected eigenvalues is */
/* EPS * norm((A, B)) / RCONDE( 1 ). */
/* An approximate (asymptotic) bound on the maximum angular error in */
/* the computed deflating subspaces is */
/* EPS * norm((A, B)) / RCONDV( 2 ). */
/* See LAPACK User's Guide, section 4.11 for more information. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. Local Arrays .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Decode the input arguments */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
--alphar;
--alphai;
--beta;
vsl_dim1 = *ldvsl;
vsl_offset = 1 + vsl_dim1;
vsl -= vsl_offset;
vsr_dim1 = *ldvsr;
vsr_offset = 1 + vsr_dim1;
vsr -= vsr_offset;
--rconde;
--rcondv;
--work;
--iwork;
--bwork;
/* Function Body */
if (lsame_(jobvsl, "N")) {
ijobvl = 1;
ilvsl = FALSE_;
} else if (lsame_(jobvsl, "V")) {
ijobvl = 2;
ilvsl = TRUE_;
} else {
ijobvl = -1;
ilvsl = FALSE_;
}
if (lsame_(jobvsr, "N")) {
ijobvr = 1;
ilvsr = FALSE_;
} else if (lsame_(jobvsr, "V")) {
ijobvr = 2;
ilvsr = TRUE_;
} else {
ijobvr = -1;
ilvsr = FALSE_;
}
wantst = lsame_(sort, "S");
wantsn = lsame_(sense, "N");
wantse = lsame_(sense, "E");
wantsv = lsame_(sense, "V");
wantsb = lsame_(sense, "B");
lquery = *lwork == -1 || *liwork == -1;
if (wantsn) {
ijob = 0;
} else if (wantse) {
ijob = 1;
} else if (wantsv) {
ijob = 2;
} else if (wantsb) {
ijob = 4;
}
/* Test the input arguments */
*info = 0;
if (ijobvl <= 0) {
*info = -1;
} else if (ijobvr <= 0) {
*info = -2;
} else if (! wantst && ! lsame_(sort, "N")) {
*info = -3;
} else if (! (wantsn || wantse || wantsv || wantsb) || ! wantst && !
wantsn) {
*info = -5;
} else if (*n < 0) {
*info = -6;
} else if (*lda < max(1,*n)) {
*info = -8;
} else if (*ldb < max(1,*n)) {
*info = -10;
} else if (*ldvsl < 1 || ilvsl && *ldvsl < *n) {
*info = -16;
} else if (*ldvsr < 1 || ilvsr && *ldvsr < *n) {
*info = -18;
}
/* 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.) */
if (*info == 0) {
if (*n > 0) {
/* Computing MAX */
i__1 = *n << 3, i__2 = *n * 6 + 16;
minwrk = max(i__1,i__2);
maxwrk = minwrk - *n + *n * ilaenv_(&c__1, "SGEQRF", " ", n, &
c__1, n, &c__0);
/* Computing MAX */
i__1 = maxwrk, i__2 = minwrk - *n + *n * ilaenv_(&c__1, "SORMQR",
" ", n, &c__1, n, &c_n1);
maxwrk = max(i__1,i__2);
if (ilvsl) {
/* Computing MAX */
i__1 = maxwrk, i__2 = minwrk - *n + *n * ilaenv_(&c__1, "SOR"
"GQR", " ", n, &c__1, n, &c_n1);
maxwrk = max(i__1,i__2);
}
lwrk = maxwrk;
if (ijob >= 1) {
/* Computing MAX */
i__1 = lwrk, i__2 = *n * *n / 2;
lwrk = max(i__1,i__2);
}
} else {
minwrk = 1;
maxwrk = 1;
lwrk = 1;
}
work[1] = (real) lwrk;
if (wantsn || *n == 0) {
liwmin = 1;
} else {
liwmin = *n + 6;
}
iwork[1] = liwmin;
if (*lwork < minwrk && ! lquery) {
*info = -22;
} else if (*liwork < liwmin && ! lquery) {
*info = -24;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SGGESX", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (*n == 0) {
*sdim = 0;
return 0;
}
/* Get machine constants */
eps = slamch_("P");
safmin = slamch_("S");
safmax = 1.f / safmin;
slabad_(&safmin, &safmax);
smlnum = sqrt(safmin) / eps;
bignum = 1.f / smlnum;
/* Scale A if max element outside range [SMLNUM,BIGNUM] */
anrm = slange_("M", n, n, &a[a_offset], lda, &work[1]);
ilascl = FALSE_;
if (anrm > 0.f && anrm < smlnum) {
anrmto = smlnum;
ilascl = TRUE_;
} else if (anrm > bignum) {
anrmto = bignum;
ilascl = TRUE_;
}
if (ilascl) {
slascl_("G", &c__0, &c__0, &anrm, &anrmto, n, n, &a[a_offset], lda, &
ierr);
}
/* Scale B if max element outside range [SMLNUM,BIGNUM] */
bnrm = slange_("M", n, n, &b[b_offset], ldb, &work[1]);
ilbscl = FALSE_;
if (bnrm > 0.f && bnrm < smlnum) {
bnrmto = smlnum;
ilbscl = TRUE_;
} else if (bnrm > bignum) {
bnrmto = bignum;
ilbscl = TRUE_;
}
if (ilbscl) {
slascl_("G", &c__0, &c__0, &bnrm, &bnrmto, n, n, &b[b_offset], ldb, &
ierr);
}
/* Permute the matrix to make it more nearly triangular */
/* (Workspace: need 6*N + 2*N for permutation parameters) */
ileft = 1;
iright = *n + 1;
iwrk = iright + *n;
sggbal_("P", n, &a[a_offset], lda, &b[b_offset], ldb, &ilo, &ihi, &work[
ileft], &work[iright], &work[iwrk], &ierr);
/* Reduce B to triangular form (QR decomposition of B) */
/* (Workspace: need N, prefer N*NB) */
irows = ihi + 1 - ilo;
icols = *n + 1 - ilo;
itau = iwrk;
iwrk = itau + irows;
i__1 = *lwork + 1 - iwrk;
sgeqrf_(&irows, &icols, &b[ilo + ilo * b_dim1], ldb, &work[itau], &work[
iwrk], &i__1, &ierr);
/* Apply the orthogonal transformation to matrix A */
/* (Workspace: need N, prefer N*NB) */
i__1 = *lwork + 1 - iwrk;
sormqr_("L", "T", &irows, &icols, &irows, &b[ilo + ilo * b_dim1], ldb, &
work[itau], &a[ilo + ilo * a_dim1], lda, &work[iwrk], &i__1, &
ierr);
/* Initialize VSL */
/* (Workspace: need N, prefer N*NB) */
if (ilvsl) {
slaset_("Full", n, n, &c_b42, &c_b43, &vsl[vsl_offset], ldvsl);
if (irows > 1) {
i__1 = irows - 1;
i__2 = irows - 1;
slacpy_("L", &i__1, &i__2, &b[ilo + 1 + ilo * b_dim1], ldb, &vsl[
ilo + 1 + ilo * vsl_dim1], ldvsl);
}
i__1 = *lwork + 1 - iwrk;
sorgqr_(&irows, &irows, &irows, &vsl[ilo + ilo * vsl_dim1], ldvsl, &
work[itau], &work[iwrk], &i__1, &ierr);
}
/* Initialize VSR */
if (ilvsr) {
slaset_("Full", n, n, &c_b42, &c_b43, &vsr[vsr_offset], ldvsr);
}
/* Reduce to generalized Hessenberg form */
/* (Workspace: none needed) */
sgghrd_(jobvsl, jobvsr, n, &ilo, &ihi, &a[a_offset], lda, &b[b_offset],
ldb, &vsl[vsl_offset], ldvsl, &vsr[vsr_offset], ldvsr, &ierr);
*sdim = 0;
/* Perform QZ algorithm, computing Schur vectors if desired */
/* (Workspace: need N) */
iwrk = itau;
i__1 = *lwork + 1 - iwrk;
shgeqz_("S", jobvsl, jobvsr, n, &ilo, &ihi, &a[a_offset], lda, &b[
b_offset], ldb, &alphar[1], &alphai[1], &beta[1], &vsl[vsl_offset]
, ldvsl, &vsr[vsr_offset], ldvsr, &work[iwrk], &i__1, &ierr);
if (ierr != 0) {
if (ierr > 0 && ierr <= *n) {
*info = ierr;
} else if (ierr > *n && ierr <= *n << 1) {
*info = ierr - *n;
} else {
*info = *n + 1;
}
goto L50;
}
/* Sort eigenvalues ALPHA/BETA and compute the reciprocal of */
/* condition number(s) */
/* (Workspace: If IJOB >= 1, need MAX( 8*(N+1), 2*SDIM*(N-SDIM) ) */
/* otherwise, need 8*(N+1) ) */
if (wantst) {
/* Undo scaling on eigenvalues before SELCTGing */
if (ilascl) {
slascl_("G", &c__0, &c__0, &anrmto, &anrm, n, &c__1, &alphar[1],
n, &ierr);
slascl_("G", &c__0, &c__0, &anrmto, &anrm, n, &c__1, &alphai[1],
n, &ierr);
}
if (ilbscl) {
slascl_("G", &c__0, &c__0, &bnrmto, &bnrm, n, &c__1, &beta[1], n,
&ierr);
}
/* Select eigenvalues */
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
bwork[i__] = (*selctg)(&alphar[i__], &alphai[i__], &beta[i__]);
/* L10: */
}
/* Reorder eigenvalues, transform Generalized Schur vectors, and */
/* compute reciprocal condition numbers */
i__1 = *lwork - iwrk + 1;
stgsen_(&ijob, &ilvsl, &ilvsr, &bwork[1], n, &a[a_offset], lda, &b[
b_offset], ldb, &alphar[1], &alphai[1], &beta[1], &vsl[
vsl_offset], ldvsl, &vsr[vsr_offset], ldvsr, sdim, &pl, &pr,
dif, &work[iwrk], &i__1, &iwork[1], liwork, &ierr);
if (ijob >= 1) {
/* Computing MAX */
i__1 = maxwrk, i__2 = (*sdim << 1) * (*n - *sdim);
maxwrk = max(i__1,i__2);
}
if (ierr == -22) {
/* not enough real workspace */
*info = -22;
} else {
if (ijob == 1 || ijob == 4) {
rconde[1] = pl;
rconde[2] = pr;
}
if (ijob == 2 || ijob == 4) {
rcondv[1] = dif[0];
rcondv[2] = dif[1];
}
if (ierr == 1) {
*info = *n + 3;
}
}
}
/* Apply permutation to VSL and VSR */
/* (Workspace: none needed) */
if (ilvsl) {
sggbak_("P", "L", n, &ilo, &ihi, &work[ileft], &work[iright], n, &vsl[
vsl_offset], ldvsl, &ierr);
}
if (ilvsr) {
sggbak_("P", "R", n, &ilo, &ihi, &work[ileft], &work[iright], n, &vsr[
vsr_offset], ldvsr, &ierr);
}
/* Check if unscaling would cause over/underflow, if so, rescale */
/* (ALPHAR(I),ALPHAI(I),BETA(I)) so BETA(I) is on the order of */
/* B(I,I) and ALPHAR(I) and ALPHAI(I) are on the order of A(I,I) */
if (ilascl) {
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
if (alphai[i__] != 0.f) {
if (alphar[i__] / safmax > anrmto / anrm || safmin / alphar[
i__] > anrm / anrmto) {
work[1] = (r__1 = a[i__ + i__ * a_dim1] / alphar[i__],
dabs(r__1));
beta[i__] *= work[1];
alphar[i__] *= work[1];
alphai[i__] *= work[1];
} else if (alphai[i__] / safmax > anrmto / anrm || safmin /
alphai[i__] > anrm / anrmto) {
work[1] = (r__1 = a[i__ + (i__ + 1) * a_dim1] / alphai[
i__], dabs(r__1));
beta[i__] *= work[1];
alphar[i__] *= work[1];
alphai[i__] *= work[1];
}
}
/* L20: */
}
}
if (ilbscl) {
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
if (alphai[i__] != 0.f) {
if (beta[i__] / safmax > bnrmto / bnrm || safmin / beta[i__]
> bnrm / bnrmto) {
work[1] = (r__1 = b[i__ + i__ * b_dim1] / beta[i__], dabs(
r__1));
beta[i__] *= work[1];
alphar[i__] *= work[1];
alphai[i__] *= work[1];
}
}
/* L25: */
}
}
/* Undo scaling */
if (ilascl) {
slascl_("H", &c__0, &c__0, &anrmto, &anrm, n, n, &a[a_offset], lda, &
ierr);
slascl_("G", &c__0, &c__0, &anrmto, &anrm, n, &c__1, &alphar[1], n, &
ierr);
slascl_("G", &c__0, &c__0, &anrmto, &anrm, n, &c__1, &alphai[1], n, &
ierr);
}
if (ilbscl) {
slascl_("U", &c__0, &c__0, &bnrmto, &bnrm, n, n, &b[b_offset], ldb, &
ierr);
slascl_("G", &c__0, &c__0, &bnrmto, &bnrm, n, &c__1, &beta[1], n, &
ierr);
}
if (wantst) {
/* Check if reordering is correct */
lastsl = TRUE_;
lst2sl = TRUE_;
*sdim = 0;
ip = 0;
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
cursl = (*selctg)(&alphar[i__], &alphai[i__], &beta[i__]);
if (alphai[i__] == 0.f) {
if (cursl) {
++(*sdim);
}
ip = 0;
if (cursl && ! lastsl) {
*info = *n + 2;
}
} else {
if (ip == 1) {
/* Last eigenvalue of conjugate pair */
cursl = cursl || lastsl;
lastsl = cursl;
if (cursl) {
*sdim += 2;
}
ip = -1;
if (cursl && ! lst2sl) {
*info = *n + 2;
}
} else {
/* First eigenvalue of conjugate pair */
ip = 1;
}
}
lst2sl = lastsl;
lastsl = cursl;
/* L40: */
}
}
L50:
work[1] = (real) maxwrk;
iwork[1] = liwmin;
return 0;
/* End of SGGESX */
} /* sggesx_ */
/* Subroutine */ int slasd6_(integer *icompq, integer *nl, integer *nr,
integer *sqre, real *d__, real *vf, real *vl, real *alpha, real *beta,
integer *idxq, integer *perm, integer *givptr, integer *givcol,
integer *ldgcol, real *givnum, integer *ldgnum, real *poles, real *
difl, real *difr, real *z__, integer *k, real *c__, real *s, real *
work, integer *iwork, integer *info)
{
/* System generated locals */
integer givcol_dim1, givcol_offset, givnum_dim1, givnum_offset,
poles_dim1, poles_offset, i__1;
real r__1, r__2;
/* Local variables */
integer i__, m, n, n1, n2, iw, idx, idxc, idxp, ivfw, ivlw;
extern /* Subroutine */ int scopy_(integer *, real *, integer *, real *,
integer *), slasd7_(integer *, integer *, integer *, integer *,
integer *, real *, real *, real *, real *, real *, real *, real *,
real *, real *, real *, integer *, integer *, integer *, integer
*, integer *, integer *, integer *, real *, integer *, real *,
real *, integer *), slasd8_(integer *, integer *, real *, real *,
real *, real *, real *, real *, integer *, real *, real *,
integer *);
integer isigma;
extern /* Subroutine */ int xerbla_(char *, integer *), slascl_(
char *, integer *, integer *, real *, real *, integer *, integer *
, real *, integer *, integer *), slamrg_(integer *,
integer *, real *, integer *, integer *, integer *);
real orgnrm;
/* -- LAPACK auxiliary routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SLASD6 computes the SVD of an updated upper bidiagonal matrix B */
/* obtained by merging two smaller ones by appending a row. This */
/* routine is used only for the problem which requires all singular */
/* values and optionally singular vector matrices in factored form. */
/* B is an N-by-M matrix with N = NL + NR + 1 and M = N + SQRE. */
/* A related subroutine, SLASD1, handles the case in which all singular */
/* values and singular vectors of the bidiagonal matrix are desired. */
/* SLASD6 computes the SVD as follows: */
/* ( D1(in) 0 0 0 ) */
/* B = U(in) * ( Z1' a Z2' b ) * VT(in) */
/* ( 0 0 D2(in) 0 ) */
/* = U(out) * ( D(out) 0) * VT(out) */
/* where Z' = (Z1' a Z2' b) = u' VT', and u is a vector of dimension M */
/* with ALPHA and BETA in the NL+1 and NL+2 th entries and zeros */
/* elsewhere; and the entry b is empty if SQRE = 0. */
/* The singular values of B can be computed using D1, D2, the first */
/* components of all the right singular vectors of the lower block, and */
/* the last components of all the right singular vectors of the upper */
/* block. These components are stored and updated in VF and VL, */
/* respectively, in SLASD6. Hence U and VT are not explicitly */
/* referenced. */
/* The singular values are stored in D. The algorithm consists of two */
/* stages: */
/* The first stage consists of deflating the size of the problem */
/* when there are multiple singular values or if there is a zero */
/* in the Z vector. For each such occurence the dimension of the */
/* secular equation problem is reduced by one. This stage is */
/* performed by the routine SLASD7. */
/* The second stage consists of calculating the updated */
/* singular values. This is done by finding the roots of the */
/* secular equation via the routine SLASD4 (as called by SLASD8). */
/* This routine also updates VF and VL and computes the distances */
/* between the updated singular values and the old singular */
/* values. */
/* SLASD6 is called from SLASDA. */
/* Arguments */
/* ========= */
/* ICOMPQ (input) INTEGER */
/* Specifies whether singular vectors are to be computed in */
/* factored form: */
/* = 0: Compute singular values only. */
/* = 1: Compute singular vectors in factored form as well. */
/* 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 row dimension N = NL + NR + 1, */
/* and column dimension M = N + SQRE. */
/* D (input/output) REAL array, dimension (NL+NR+1). */
/* On entry D(1:NL,1:NL) contains the singular values of the */
/* upper block, and D(NL+2:N) contains the singular values */
/* of the lower block. On exit D(1:N) contains the singular */
/* values of the modified matrix. */
/* VF (input/output) REAL array, dimension (M) */
/* On entry, VF(1:NL+1) contains the first components of all */
/* right singular vectors of the upper block; and VF(NL+2:M) */
/* contains the first components of all right singular vectors */
/* of the lower block. On exit, VF contains the first components */
/* of all right singular vectors of the bidiagonal matrix. */
/* VL (input/output) REAL array, dimension (M) */
/* On entry, VL(1:NL+1) contains the last components of all */
/* right singular vectors of the upper block; and VL(NL+2:M) */
/* contains the last components of all right singular vectors of */
/* the lower block. On exit, VL contains the last components of */
/* all right singular vectors of the bidiagonal matrix. */
/* ALPHA (input/output) REAL */
/* Contains the diagonal element associated with the added row. */
/* BETA (input/output) REAL */
/* Contains the off-diagonal element associated with the added */
/* row. */
/* IDXQ (output) INTEGER array, dimension (N) */
/* This contains the permutation which will reintegrate the */
/* subproblem just solved back into sorted order, i.e. */
/* D( IDXQ( I = 1, N ) ) will be in ascending order. */
/* PERM (output) INTEGER array, dimension ( N ) */
/* The permutations (from deflation and sorting) to be applied */
/* to each block. Not referenced if ICOMPQ = 0. */
/* GIVPTR (output) INTEGER */
/* The number of Givens rotations which took place in this */
/* subproblem. Not referenced if ICOMPQ = 0. */
/* GIVCOL (output) INTEGER array, dimension ( LDGCOL, 2 ) */
/* Each pair of numbers indicates a pair of columns to take place */
/* in a Givens rotation. Not referenced if ICOMPQ = 0. */
/* LDGCOL (input) INTEGER */
/* leading dimension of GIVCOL, must be at least N. */
/* GIVNUM (output) REAL array, dimension ( LDGNUM, 2 ) */
/* Each number indicates the C or S value to be used in the */
/* corresponding Givens rotation. Not referenced if ICOMPQ = 0. */
/* LDGNUM (input) INTEGER */
/* The leading dimension of GIVNUM and POLES, must be at least N. */
/* POLES (output) REAL array, dimension ( LDGNUM, 2 ) */
/* On exit, POLES(1,*) is an array containing the new singular */
/* values obtained from solving the secular equation, and */
/* POLES(2,*) is an array containing the poles in the secular */
/* equation. Not referenced if ICOMPQ = 0. */
/* DIFL (output) REAL array, dimension ( N ) */
/* On exit, DIFL(I) is the distance between I-th updated */
/* (undeflated) singular value and the I-th (undeflated) old */
/* singular value. */
/* DIFR (output) REAL array, */
/* dimension ( LDGNUM, 2 ) if ICOMPQ = 1 and */
/* dimension ( N ) if ICOMPQ = 0. */
/* On exit, DIFR(I, 1) is the distance between I-th updated */
/* (undeflated) singular value and the I+1-th (undeflated) old */
/* singular value. */
/* If ICOMPQ = 1, DIFR(1:K,2) is an array containing the */
/* normalizing factors for the right singular vector matrix. */
/* See SLASD8 for details on DIFL and DIFR. */
/* Z (output) REAL array, dimension ( M ) */
/* The first elements of this array contain the components */
/* of the deflation-adjusted updating row vector. */
/* K (output) INTEGER */
/* Contains the dimension of the non-deflated matrix, */
/* This is the order of the related secular equation. 1 <= K <=N. */
/* C (output) REAL */
/* C contains garbage if SQRE =0 and the C-value of a Givens */
/* rotation related to the right null space if SQRE = 1. */
/* S (output) REAL */
/* S contains garbage if SQRE =0 and the S-value of a Givens */
/* rotation related to the right null space if SQRE = 1. */
/* WORK (workspace) REAL array, dimension ( 4 * M ) */
/* IWORK (workspace) INTEGER array, dimension ( 3 * N ) */
/* INFO (output) INTEGER */
/* = 0: successful exit. */
/* < 0: if INFO = -i, the i-th argument had an illegal value. */
/* > 0: if INFO = 1, an singular value did not converge */
/* Further Details */
/* =============== */
/* Based on contributions by */
/* Ming Gu and Huan Ren, Computer Science Division, University of */
/* California at Berkeley, USA */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
--d__;
--vf;
--vl;
--idxq;
--perm;
givcol_dim1 = *ldgcol;
givcol_offset = 1 + givcol_dim1;
givcol -= givcol_offset;
poles_dim1 = *ldgnum;
poles_offset = 1 + poles_dim1;
poles -= poles_offset;
givnum_dim1 = *ldgnum;
givnum_offset = 1 + givnum_dim1;
givnum -= givnum_offset;
--difl;
--difr;
--z__;
--work;
--iwork;
/* Function Body */
*info = 0;
n = *nl + *nr + 1;
m = n + *sqre;
if (*icompq < 0 || *icompq > 1) {
*info = -1;
} else if (*nl < 1) {
*info = -2;
} else if (*nr < 1) {
*info = -3;
} else if (*sqre < 0 || *sqre > 1) {
*info = -4;
} else if (*ldgcol < n) {
*info = -14;
} else if (*ldgnum < n) {
*info = -16;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SLASD6", &i__1);
return 0;
}
/* The following values are for bookkeeping purposes only. They are */
/* integer pointers which indicate the portion of the workspace */
/* used by a particular array in SLASD7 and SLASD8. */
isigma = 1;
iw = isigma + n;
ivfw = iw + m;
ivlw = ivfw + m;
idx = 1;
idxc = idx + n;
idxp = idxc + n;
/* Scale. */
/* Computing MAX */
r__1 = dabs(*alpha), r__2 = dabs(*beta);
orgnrm = dmax(r__1,r__2);
d__[*nl + 1] = 0.f;
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
if ((r__1 = d__[i__], dabs(r__1)) > orgnrm) {
orgnrm = (r__1 = d__[i__], dabs(r__1));
}
/* L10: */
}
slascl_("G", &c__0, &c__0, &orgnrm, &c_b7, &n, &c__1, &d__[1], &n, info);
*alpha /= orgnrm;
*beta /= orgnrm;
/* Sort and Deflate singular values. */
slasd7_(icompq, nl, nr, sqre, k, &d__[1], &z__[1], &work[iw], &vf[1], &
work[ivfw], &vl[1], &work[ivlw], alpha, beta, &work[isigma], &
iwork[idx], &iwork[idxp], &idxq[1], &perm[1], givptr, &givcol[
givcol_offset], ldgcol, &givnum[givnum_offset], ldgnum, c__, s,
info);
/* Solve Secular Equation, compute DIFL, DIFR, and update VF, VL. */
slasd8_(icompq, k, &d__[1], &z__[1], &vf[1], &vl[1], &difl[1], &difr[1],
ldgnum, &work[isigma], &work[iw], info);
/* Save the poles if ICOMPQ = 1. */
if (*icompq == 1) {
scopy_(k, &d__[1], &c__1, &poles[poles_dim1 + 1], &c__1);
scopy_(k, &work[isigma], &c__1, &poles[(poles_dim1 << 1) + 1], &c__1);
}
/* Unscale. */
slascl_("G", &c__0, &c__0, &c_b7, &orgnrm, &n, &c__1, &d__[1], &n, info);
/* Prepare the IDXQ sorting permutation. */
n1 = *k;
n2 = n - *k;
slamrg_(&n1, &n2, &d__[1], &c__1, &c_n1, &idxq[1]);
return 0;
/* End of SLASD6 */
} /* slasd6_ */
doublereal sqrt17_(char *trans, integer *iresid, integer *m, integer *n,
integer *nrhs, real *a, integer *lda, real *x, integer *ldx, real *b,
integer *ldb, real *c__, real *work, integer *lwork)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, c_dim1, c_offset, x_dim1,
x_offset, i__1;
real ret_val;
/* Local variables */
static integer iscl, info;
extern logical lsame_(char *, char *);
extern /* Subroutine */ int sgemm_(char *, char *, integer *, integer *,
integer *, real *, real *, integer *, real *, integer *, real *,
real *, integer *);
static real norma, normb;
static integer ncols;
static real normx, rwork[1];
static integer nrows;
extern doublereal slamch_(char *), slange_(char *, integer *,
integer *, real *, integer *, real *);
extern /* Subroutine */ int xerbla_(char *, integer *);
static real bignum;
extern /* Subroutine */ int slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *), slacpy_(char *, integer *, integer *, real *, integer *,
real *, integer *);
static real smlnum, normrs, err;
/* -- LAPACK test 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
=======
SQRT17 computes the ratio
|| R'*op(A) ||/(||A||*alpha*max(M,N,NRHS)*eps)
where R = op(A)*X - B, op(A) is A or A', and
alpha = ||B|| if IRESID = 1 (zero-residual problem)
alpha = ||R|| if IRESID = 2 (otherwise).
Arguments
=========
TRANS (input) CHARACTER*1
Specifies whether or not the transpose of A is used.
= 'N': No transpose, op(A) = A.
= 'T': Transpose, op(A) = A'.
IRESID (input) INTEGER
IRESID = 1 indicates zero-residual problem.
IRESID = 2 indicates non-zero residual.
M (input) INTEGER
The number of rows of the matrix A.
If TRANS = 'N', the number of rows of the matrix B.
If TRANS = 'T', the number of rows of the matrix X.
N (input) INTEGER
The number of columns of the matrix A.
If TRANS = 'N', the number of rows of the matrix X.
If TRANS = 'T', the number of rows of the matrix B.
NRHS (input) INTEGER
The number of columns of the matrices X and B.
A (input) REAL array, dimension (LDA,N)
The m-by-n matrix A.
LDA (input) INTEGER
The leading dimension of the array A. LDA >= M.
X (input) REAL array, dimension (LDX,NRHS)
If TRANS = 'N', the n-by-nrhs matrix X.
If TRANS = 'T', the m-by-nrhs matrix X.
LDX (input) INTEGER
The leading dimension of the array X.
If TRANS = 'N', LDX >= N.
If TRANS = 'T', LDX >= M.
B (input) REAL array, dimension (LDB,NRHS)
If TRANS = 'N', the m-by-nrhs matrix B.
If TRANS = 'T', the n-by-nrhs matrix B.
LDB (input) INTEGER
The leading dimension of the array B.
If TRANS = 'N', LDB >= M.
If TRANS = 'T', LDB >= N.
C (workspace) REAL array, dimension (LDB,NRHS)
WORK (workspace) REAL array, dimension (LWORK)
LWORK (input) INTEGER
The length of the array WORK. LWORK >= NRHS*(M+N).
=====================================================================
Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1 * 1;
a -= a_offset;
x_dim1 = *ldx;
x_offset = 1 + x_dim1 * 1;
x -= x_offset;
c_dim1 = *ldb;
c_offset = 1 + c_dim1 * 1;
c__ -= c_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1 * 1;
b -= b_offset;
--work;
/* Function Body */
ret_val = 0.f;
if (lsame_(trans, "N")) {
nrows = *m;
ncols = *n;
} else if (lsame_(trans, "T")) {
nrows = *n;
ncols = *m;
} else {
xerbla_("SQRT17", &c__1);
return ret_val;
}
if (*lwork < ncols * *nrhs) {
xerbla_("SQRT17", &c__13);
return ret_val;
}
if (*m <= 0 || *n <= 0 || *nrhs <= 0) {
return ret_val;
}
norma = slange_("One-norm", m, n, &a[a_offset], lda, rwork);
smlnum = slamch_("Safe minimum") / slamch_("Precision");
bignum = 1.f / smlnum;
iscl = 0;
/* compute residual and scale it */
slacpy_("All", &nrows, nrhs, &b[b_offset], ldb, &c__[c_offset], ldb);
sgemm_(trans, "No transpose", &nrows, nrhs, &ncols, &c_b13, &a[a_offset],
lda, &x[x_offset], ldx, &c_b14, &c__[c_offset], ldb);
normrs = slange_("Max", &nrows, nrhs, &c__[c_offset], ldb, rwork);
if (normrs > smlnum) {
iscl = 1;
slascl_("General", &c__0, &c__0, &normrs, &c_b14, &nrows, nrhs, &c__[
c_offset], ldb, &info);
}
/* compute R'*A */
sgemm_("Transpose", trans, nrhs, &ncols, &nrows, &c_b14, &c__[c_offset],
ldb, &a[a_offset], lda, &c_b22, &work[1], nrhs);
/* compute and properly scale error */
err = slange_("One-norm", nrhs, &ncols, &work[1], nrhs, rwork);
if (norma != 0.f) {
err /= norma;
}
if (iscl == 1) {
err *= normrs;
}
if (*iresid == 1) {
normb = slange_("One-norm", &nrows, nrhs, &b[b_offset], ldb, rwork);
if (normb != 0.f) {
err /= normb;
}
} else {
normx = slange_("One-norm", &ncols, nrhs, &x[x_offset], ldx, rwork);
if (normx != 0.f) {
err /= normx;
}
}
/* Computing MAX */
i__1 = max(*m,*n);
ret_val = err / (slamch_("Epsilon") * (real) max(i__1,*nrhs));
return ret_val;
/* End of SQRT17 */
} /* sqrt17_ */
/* Subroutine */ int slals0_(integer *icompq, integer *nl, integer *nr,
integer *sqre, integer *nrhs, real *b, integer *ldb, real *bx,
integer *ldbx, integer *perm, integer *givptr, integer *givcol,
integer *ldgcol, real *givnum, integer *ldgnum, real *poles, real *
difl, real *difr, real *z__, integer *k, real *c__, real *s, real *
work, integer *info)
{
/* -- LAPACK routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
December 1, 1999
Purpose
=======
SLALS0 applies back the multiplying factors of either the left or the
right singular vector matrix of a diagonal matrix appended by a row
to the right hand side matrix B in solving the least squares problem
using the divide-and-conquer SVD approach.
For the left singular vector matrix, three types of orthogonal
matrices are involved:
(1L) Givens rotations: the number of such rotations is GIVPTR; the
pairs of columns/rows they were applied to are stored in GIVCOL;
and the C- and S-values of these rotations are stored in GIVNUM.
(2L) Permutation. The (NL+1)-st row of B is to be moved to the first
row, and for J=2:N, PERM(J)-th row of B is to be moved to the
J-th row.
(3L) The left singular vector matrix of the remaining matrix.
For the right singular vector matrix, four types of orthogonal
matrices are involved:
(1R) The right singular vector matrix of the remaining matrix.
(2R) If SQRE = 1, one extra Givens rotation to generate the right
null space.
(3R) The inverse transformation of (2L).
(4R) The inverse transformation of (1L).
Arguments
=========
ICOMPQ (input) INTEGER
Specifies whether singular vectors are to be computed in
factored form:
= 0: Left singular vector matrix.
= 1: Right singular vector matrix.
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 row dimension N = NL + NR + 1,
and column dimension M = N + SQRE.
NRHS (input) INTEGER
The number of columns of B and BX. NRHS must be at least 1.
B (input/output) REAL array, dimension ( LDB, NRHS )
On input, B contains the right hand sides of the least
squares problem in rows 1 through M. On output, B contains
the solution X in rows 1 through N.
LDB (input) INTEGER
The leading dimension of B. LDB must be at least
max(1,MAX( M, N ) ).
BX (workspace) REAL array, dimension ( LDBX, NRHS )
LDBX (input) INTEGER
The leading dimension of BX.
PERM (input) INTEGER array, dimension ( N )
The permutations (from deflation and sorting) applied
to the two blocks.
GIVPTR (input) INTEGER
The number of Givens rotations which took place in this
subproblem.
GIVCOL (input) INTEGER array, dimension ( LDGCOL, 2 )
Each pair of numbers indicates a pair of rows/columns
involved in a Givens rotation.
LDGCOL (input) INTEGER
The leading dimension of GIVCOL, must be at least N.
GIVNUM (input) REAL array, dimension ( LDGNUM, 2 )
Each number indicates the C or S value used in the
corresponding Givens rotation.
LDGNUM (input) INTEGER
The leading dimension of arrays DIFR, POLES and
GIVNUM, must be at least K.
POLES (input) REAL array, dimension ( LDGNUM, 2 )
On entry, POLES(1:K, 1) contains the new singular
values obtained from solving the secular equation, and
POLES(1:K, 2) is an array containing the poles in the secular
equation.
DIFL (input) REAL array, dimension ( K ).
On entry, DIFL(I) is the distance between I-th updated
(undeflated) singular value and the I-th (undeflated) old
singular value.
DIFR (input) REAL array, dimension ( LDGNUM, 2 ).
On entry, DIFR(I, 1) contains the distances between I-th
updated (undeflated) singular value and the I+1-th
(undeflated) old singular value. And DIFR(I, 2) is the
normalizing factor for the I-th right singular vector.
Z (input) REAL array, dimension ( K )
Contain the components of the deflation-adjusted updating row
vector.
K (input) INTEGER
Contains the dimension of the non-deflated matrix,
This is the order of the related secular equation. 1 <= K <=N.
C (input) REAL
C contains garbage if SQRE =0 and the C-value of a Givens
rotation related to the right null space if SQRE = 1.
S (input) REAL
S contains garbage if SQRE =0 and the S-value of a Givens
rotation related to the right null space if SQRE = 1.
WORK (workspace) REAL array, dimension ( K )
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 Ren-Cang Li, Computer Science Division, University of
California at Berkeley, USA
Osni Marques, LBNL/NERSC, USA
=====================================================================
Test the input parameters.
Parameter adjustments */
/* Table of constant values */
static real c_b5 = -1.f;
static integer c__1 = 1;
static real c_b11 = 1.f;
static real c_b13 = 0.f;
static integer c__0 = 0;
/* System generated locals */
integer givcol_dim1, givcol_offset, b_dim1, b_offset, bx_dim1, bx_offset,
difr_dim1, difr_offset, givnum_dim1, givnum_offset, poles_dim1,
poles_offset, i__1, i__2;
real r__1;
/* Local variables */
static real temp;
extern /* Subroutine */ int srot_(integer *, real *, integer *, real *,
integer *, real *, real *);
extern doublereal snrm2_(integer *, real *, integer *);
static integer i__, j, m, n;
static real diflj, difrj, dsigj;
extern /* Subroutine */ int sscal_(integer *, real *, real *, integer *),
sgemv_(char *, integer *, integer *, real *, real *, integer *,
real *, integer *, real *, real *, integer *), scopy_(
integer *, real *, integer *, real *, integer *);
extern doublereal slamc3_(real *, real *);
static real dj;
extern /* Subroutine */ int xerbla_(char *, integer *);
static real dsigjp;
extern /* Subroutine */ int slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *), slacpy_(char *, integer *, integer *, real *, integer *,
real *, integer *);
static integer nlp1;
#define difr_ref(a_1,a_2) difr[(a_2)*difr_dim1 + a_1]
#define b_ref(a_1,a_2) b[(a_2)*b_dim1 + a_1]
#define poles_ref(a_1,a_2) poles[(a_2)*poles_dim1 + a_1]
#define bx_ref(a_1,a_2) bx[(a_2)*bx_dim1 + a_1]
#define givcol_ref(a_1,a_2) givcol[(a_2)*givcol_dim1 + a_1]
#define givnum_ref(a_1,a_2) givnum[(a_2)*givnum_dim1 + a_1]
b_dim1 = *ldb;
b_offset = 1 + b_dim1 * 1;
b -= b_offset;
bx_dim1 = *ldbx;
bx_offset = 1 + bx_dim1 * 1;
bx -= bx_offset;
--perm;
givcol_dim1 = *ldgcol;
givcol_offset = 1 + givcol_dim1 * 1;
givcol -= givcol_offset;
difr_dim1 = *ldgnum;
difr_offset = 1 + difr_dim1 * 1;
difr -= difr_offset;
poles_dim1 = *ldgnum;
poles_offset = 1 + poles_dim1 * 1;
poles -= poles_offset;
givnum_dim1 = *ldgnum;
givnum_offset = 1 + givnum_dim1 * 1;
givnum -= givnum_offset;
--difl;
--z__;
--work;
/* Function Body */
*info = 0;
if (*icompq < 0 || *icompq > 1) {
*info = -1;
} else if (*nl < 1) {
*info = -2;
} else if (*nr < 1) {
*info = -3;
} else if (*sqre < 0 || *sqre > 1) {
*info = -4;
}
n = *nl + *nr + 1;
if (*nrhs < 1) {
*info = -5;
} else if (*ldb < n) {
*info = -7;
} else if (*ldbx < n) {
*info = -9;
} else if (*givptr < 0) {
*info = -11;
} else if (*ldgcol < n) {
*info = -13;
} else if (*ldgnum < n) {
*info = -15;
} else if (*k < 1) {
*info = -20;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SLALS0", &i__1);
return 0;
}
m = n + *sqre;
nlp1 = *nl + 1;
if (*icompq == 0) {
/* Apply back orthogonal transformations from the left.
Step (1L): apply back the Givens rotations performed. */
i__1 = *givptr;
for (i__ = 1; i__ <= i__1; ++i__) {
srot_(nrhs, &b_ref(givcol_ref(i__, 2), 1), ldb, &b_ref(givcol_ref(
i__, 1), 1), ldb, &givnum_ref(i__, 2), &givnum_ref(i__, 1)
);
/* L10: */
}
/* Step (2L): permute rows of B. */
scopy_(nrhs, &b_ref(nlp1, 1), ldb, &bx_ref(1, 1), ldbx);
i__1 = n;
for (i__ = 2; i__ <= i__1; ++i__) {
scopy_(nrhs, &b_ref(perm[i__], 1), ldb, &bx_ref(i__, 1), ldbx);
/* L20: */
}
/* Step (3L): apply the inverse of the left singular vector
matrix to BX. */
if (*k == 1) {
scopy_(nrhs, &bx[bx_offset], ldbx, &b[b_offset], ldb);
if (z__[1] < 0.f) {
sscal_(nrhs, &c_b5, &b[b_offset], ldb);
}
} else {
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
diflj = difl[j];
dj = poles_ref(j, 1);
dsigj = -poles_ref(j, 2);
if (j < *k) {
difrj = -difr_ref(j, 1);
dsigjp = -poles_ref(j + 1, 2);
}
if (z__[j] == 0.f || poles_ref(j, 2) == 0.f) {
work[j] = 0.f;
} else {
work[j] = -poles_ref(j, 2) * z__[j] / diflj / (poles_ref(
j, 2) + dj);
}
i__2 = j - 1;
for (i__ = 1; i__ <= i__2; ++i__) {
if (z__[i__] == 0.f || poles_ref(i__, 2) == 0.f) {
work[i__] = 0.f;
} else {
work[i__] = poles_ref(i__, 2) * z__[i__] / (slamc3_(&
poles_ref(i__, 2), &dsigj) - diflj) / (
poles_ref(i__, 2) + dj);
}
/* L30: */
}
i__2 = *k;
for (i__ = j + 1; i__ <= i__2; ++i__) {
if (z__[i__] == 0.f || poles_ref(i__, 2) == 0.f) {
work[i__] = 0.f;
} else {
work[i__] = poles_ref(i__, 2) * z__[i__] / (slamc3_(&
poles_ref(i__, 2), &dsigjp) + difrj) / (
poles_ref(i__, 2) + dj);
}
/* L40: */
}
work[1] = -1.f;
temp = snrm2_(k, &work[1], &c__1);
sgemv_("T", k, nrhs, &c_b11, &bx[bx_offset], ldbx, &work[1], &
c__1, &c_b13, &b_ref(j, 1), ldb);
slascl_("G", &c__0, &c__0, &temp, &c_b11, &c__1, nrhs, &b_ref(
j, 1), ldb, info);
/* L50: */
}
}
/* Move the deflated rows of BX to B also. */
if (*k < max(m,n)) {
i__1 = n - *k;
slacpy_("A", &i__1, nrhs, &bx_ref(*k + 1, 1), ldbx, &b_ref(*k + 1,
1), ldb);
}
} else {
/* Apply back the right orthogonal transformations.
Step (1R): apply back the new right singular vector matrix
to B. */
if (*k == 1) {
scopy_(nrhs, &b[b_offset], ldb, &bx[bx_offset], ldbx);
} else {
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
dsigj = poles_ref(j, 2);
if (z__[j] == 0.f) {
work[j] = 0.f;
} else {
work[j] = -z__[j] / difl[j] / (dsigj + poles_ref(j, 1)) /
difr_ref(j, 2);
}
i__2 = j - 1;
for (i__ = 1; i__ <= i__2; ++i__) {
if (z__[j] == 0.f) {
work[i__] = 0.f;
} else {
r__1 = -poles_ref(i__ + 1, 2);
work[i__] = z__[j] / (slamc3_(&dsigj, &r__1) -
difr_ref(i__, 1)) / (dsigj + poles_ref(i__, 1)
) / difr_ref(i__, 2);
}
/* L60: */
}
i__2 = *k;
for (i__ = j + 1; i__ <= i__2; ++i__) {
if (z__[j] == 0.f) {
work[i__] = 0.f;
} else {
r__1 = -poles_ref(i__, 2);
work[i__] = z__[j] / (slamc3_(&dsigj, &r__1) - difl[
i__]) / (dsigj + poles_ref(i__, 1)) /
difr_ref(i__, 2);
}
/* L70: */
}
sgemv_("T", k, nrhs, &c_b11, &b[b_offset], ldb, &work[1], &
c__1, &c_b13, &bx_ref(j, 1), ldbx);
/* L80: */
}
}
/* Step (2R): if SQRE = 1, apply back the rotation that is
related to the right null space of the subproblem. */
if (*sqre == 1) {
scopy_(nrhs, &b_ref(m, 1), ldb, &bx_ref(m, 1), ldbx);
srot_(nrhs, &bx_ref(1, 1), ldbx, &bx_ref(m, 1), ldbx, c__, s);
}
if (*k < max(m,n)) {
i__1 = n - *k;
slacpy_("A", &i__1, nrhs, &b_ref(*k + 1, 1), ldb, &bx_ref(*k + 1,
1), ldbx);
}
/* Step (3R): permute rows of B. */
scopy_(nrhs, &bx_ref(1, 1), ldbx, &b_ref(nlp1, 1), ldb);
if (*sqre == 1) {
scopy_(nrhs, &bx_ref(m, 1), ldbx, &b_ref(m, 1), ldb);
}
i__1 = n;
for (i__ = 2; i__ <= i__1; ++i__) {
scopy_(nrhs, &bx_ref(i__, 1), ldbx, &b_ref(perm[i__], 1), ldb);
/* L90: */
}
/* Step (4R): apply back the Givens rotations performed. */
for (i__ = *givptr; i__ >= 1; --i__) {
r__1 = -givnum_ref(i__, 1);
srot_(nrhs, &b_ref(givcol_ref(i__, 2), 1), ldb, &b_ref(givcol_ref(
i__, 1), 1), ldb, &givnum_ref(i__, 2), &r__1);
/* L100: */
}
}
return 0;
/* End of SLALS0 */
} /* slals0_ */
/* Subroutine */ int sstedc_(char *compz, integer *n, real *d__, real *e,
real *z__, integer *ldz, real *work, integer *lwork, integer *iwork,
integer *liwork, integer *info)
{
/* System generated locals */
integer z_dim1, z_offset, i__1, i__2;
real r__1, r__2;
/* Builtin functions */
double log(doublereal);
integer pow_ii(integer *, integer *);
double sqrt(doublereal);
/* Local variables */
integer i__, j, k, m;
real p;
integer ii, lgn;
real eps, tiny;
extern logical lsame_(char *, char *);
extern /* Subroutine */ int sgemm_(char *, char *, integer *, integer *,
integer *, real *, real *, integer *, real *, integer *, real *,
real *, integer *);
integer lwmin, start;
extern /* Subroutine */ int sswap_(integer *, real *, integer *, real *,
integer *), slaed0_(integer *, integer *, integer *, real *, real
*, real *, integer *, real *, integer *, real *, integer *,
integer *);
extern doublereal slamch_(char *);
extern /* Subroutine */ int xerbla_(char *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *);
integer finish;
extern /* Subroutine */ int slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *), slacpy_(char *, integer *, integer *, real *, integer *,
real *, integer *), slaset_(char *, integer *, integer *,
real *, real *, real *, integer *);
integer liwmin, icompz;
real orgnrm;
extern doublereal slanst_(char *, integer *, real *, real *);
extern /* Subroutine */ int ssterf_(integer *, real *, real *, integer *),
slasrt_(char *, integer *, real *, integer *);
logical lquery;
integer smlsiz;
extern /* Subroutine */ int ssteqr_(char *, integer *, real *, real *,
real *, integer *, real *, integer *);
integer storez, strtrw;
/* -- LAPACK driver routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SSTEDC computes all eigenvalues and, optionally, eigenvectors of a */
/* symmetric tridiagonal matrix using the divide and conquer method. */
/* The eigenvectors of a full or band real symmetric matrix can also be */
/* found if SSYTRD or SSPTRD or SSBTRD has been used to reduce this */
/* matrix to tridiagonal form. */
/* This code makes very mild assumptions about floating point */
/* arithmetic. It will work on machines with a guard digit in */
/* add/subtract, or on those binary machines without guard digits */
/* which subtract like the Cray X-MP, Cray Y-MP, Cray C-90, or Cray-2. */
/* It could conceivably fail on hexadecimal or decimal machines */
/* without guard digits, but we know of none. See SLAED3 for details. */
/* Arguments */
/* ========= */
/* COMPZ (input) CHARACTER*1 */
/* = 'N': Compute eigenvalues only. */
/* = 'I': Compute eigenvectors of tridiagonal matrix also. */
/* = 'V': Compute eigenvectors of original dense symmetric */
/* matrix also. On entry, Z contains the orthogonal */
/* matrix used to reduce the original matrix to */
/* tridiagonal form. */
/* N (input) INTEGER */
/* The dimension of the symmetric tridiagonal matrix. N >= 0. */
/* D (input/output) REAL 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) REAL array, dimension (N-1) */
/* On entry, the subdiagonal elements of the tridiagonal matrix. */
/* On exit, E has been destroyed. */
/* Z (input/output) REAL 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. */
/* If eigenvectors are desired, then LDZ >= max(1,N). */
/* WORK (workspace/output) REAL 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. */
/* If COMPZ = 'N' or N <= 1 then LWORK must be at least 1. */
/* If COMPZ = 'V' and N > 1 then LWORK must be at least */
/* ( 1 + 3*N + 2*N*lg N + 3*N**2 ), */
/* where lg( N ) = smallest integer k such */
/* that 2**k >= N. */
/* If COMPZ = 'I' and N > 1 then LWORK must be at least */
/* ( 1 + 4*N + N**2 ). */
/* Note that for COMPZ = 'I' or 'V', then if N is less than or */
/* equal to the minimum divide size, usually 25, then LWORK need */
/* only be max(1,2*(N-1)). */
/* If LWORK = -1, then a workspace query is assumed; the routine */
/* only calculates the optimal size of the WORK array, returns */
/* this value as the first entry of the WORK array, and no error */
/* message related to LWORK is issued by XERBLA. */
/* IWORK (workspace/output) INTEGER array, dimension (MAX(1,LIWORK)) */
/* On exit, if INFO = 0, IWORK(1) returns the optimal LIWORK. */
/* LIWORK (input) INTEGER */
/* The dimension of the array IWORK. */
/* If COMPZ = 'N' or N <= 1 then LIWORK must be at least 1. */
/* If COMPZ = 'V' and N > 1 then LIWORK must be at least */
/* ( 6 + 6*N + 5*N*lg N ). */
/* If COMPZ = 'I' and N > 1 then LIWORK must be at least */
/* ( 3 + 5*N ). */
/* Note that for COMPZ = 'I' or 'V', then if N is less than or */
/* equal to the minimum divide size, usually 25, then LIWORK */
/* need only be 1. */
/* If LIWORK = -1, then a workspace query is assumed; the */
/* routine only calculates the optimal size of the IWORK array, */
/* returns this value as the first entry of the IWORK array, and */
/* no error message related to LIWORK is issued by XERBLA. */
/* INFO (output) INTEGER */
/* = 0: successful exit. */
/* < 0: if INFO = -i, the i-th argument had an illegal value. */
/* > 0: The algorithm failed to compute an eigenvalue while */
/* working on the submatrix lying in rows and columns */
/* INFO/(N+1) through mod(INFO,N+1). */
/* 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 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;
--iwork;
/* Function Body */
*info = 0;
lquery = *lwork == -1 || *liwork == -1;
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) {
/* Compute the workspace requirements */
smlsiz = ilaenv_(&c__9, "SSTEDC", " ", &c__0, &c__0, &c__0, &c__0);
if (*n <= 1 || icompz == 0) {
liwmin = 1;
lwmin = 1;
} else if (*n <= smlsiz) {
liwmin = 1;
lwmin = *n - 1 << 1;
} else {
lgn = (integer) (log((real) (*n)) / log(2.f));
if (pow_ii(&c__2, &lgn) < *n) {
++lgn;
}
if (pow_ii(&c__2, &lgn) < *n) {
++lgn;
}
if (icompz == 1) {
/* Computing 2nd power */
i__1 = *n;
lwmin = *n * 3 + 1 + (*n << 1) * lgn + i__1 * i__1 * 3;
liwmin = *n * 6 + 6 + *n * 5 * lgn;
} else if (icompz == 2) {
/* Computing 2nd power */
i__1 = *n;
lwmin = (*n << 2) + 1 + i__1 * i__1;
liwmin = *n * 5 + 3;
}
}
work[1] = (real) lwmin;
iwork[1] = liwmin;
if (*lwork < lwmin && ! lquery) {
*info = -8;
} else if (*liwork < liwmin && ! lquery) {
*info = -10;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SSTEDC", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
if (*n == 1) {
if (icompz != 0) {
z__[z_dim1 + 1] = 1.f;
}
return 0;
}
/* If the following conditional clause is removed, then the routine */
/* will use the Divide and Conquer routine to compute only the */
/* eigenvalues, which requires (3N + 3N**2) real workspace and */
/* (2 + 5N + 2N lg(N)) integer workspace. */
/* Since on many architectures SSTERF is much faster than any other */
/* algorithm for finding eigenvalues only, it is used here */
/* as the default. If the conditional clause is removed, then */
/* information on the size of workspace needs to be changed. */
/* If COMPZ = 'N', use SSTERF to compute the eigenvalues. */
if (icompz == 0) {
ssterf_(n, &d__[1], &e[1], info);
goto L50;
}
/* If N is smaller than the minimum divide size (SMLSIZ+1), then */
/* solve the problem with another solver. */
if (*n <= smlsiz) {
ssteqr_(compz, n, &d__[1], &e[1], &z__[z_offset], ldz, &work[1], info);
} else {
/* If COMPZ = 'V', the Z matrix must be stored elsewhere for later */
/* use. */
if (icompz == 1) {
storez = *n * *n + 1;
} else {
storez = 1;
}
if (icompz == 2) {
slaset_("Full", n, n, &c_b17, &c_b18, &z__[z_offset], ldz);
}
/* Scale. */
orgnrm = slanst_("M", n, &d__[1], &e[1]);
if (orgnrm == 0.f) {
goto L50;
}
eps = slamch_("Epsilon");
start = 1;
/* while ( START <= N ) */
L10:
if (start <= *n) {
/* Let FINISH be the position of the next subdiagonal entry */
/* such that E( FINISH ) <= TINY or FINISH = N if no such */
/* subdiagonal exists. The matrix identified by the elements */
/* between START and FINISH constitutes an independent */
/* sub-problem. */
finish = start;
L20:
if (finish < *n) {
tiny = eps * sqrt((r__1 = d__[finish], dabs(r__1))) * sqrt((
r__2 = d__[finish + 1], dabs(r__2)));
if ((r__1 = e[finish], dabs(r__1)) > tiny) {
++finish;
goto L20;
}
}
/* (Sub) Problem determined. Compute its size and solve it. */
m = finish - start + 1;
if (m == 1) {
start = finish + 1;
goto L10;
}
if (m > smlsiz) {
/* Scale. */
orgnrm = slanst_("M", &m, &d__[start], &e[start]);
slascl_("G", &c__0, &c__0, &orgnrm, &c_b18, &m, &c__1, &d__[
start], &m, info);
i__1 = m - 1;
i__2 = m - 1;
slascl_("G", &c__0, &c__0, &orgnrm, &c_b18, &i__1, &c__1, &e[
start], &i__2, info);
if (icompz == 1) {
strtrw = 1;
} else {
strtrw = start;
}
slaed0_(&icompz, n, &m, &d__[start], &e[start], &z__[strtrw +
start * z_dim1], ldz, &work[1], n, &work[storez], &
iwork[1], info);
if (*info != 0) {
*info = (*info / (m + 1) + start - 1) * (*n + 1) + *info %
(m + 1) + start - 1;
goto L50;
}
/* Scale back. */
slascl_("G", &c__0, &c__0, &c_b18, &orgnrm, &m, &c__1, &d__[
start], &m, info);
} else {
if (icompz == 1) {
/* Since QR won't update a Z matrix which is larger than */
/* the length of D, we must solve the sub-problem in a */
/* workspace and then multiply back into Z. */
ssteqr_("I", &m, &d__[start], &e[start], &work[1], &m, &
work[m * m + 1], info);
slacpy_("A", n, &m, &z__[start * z_dim1 + 1], ldz, &work[
storez], n);
sgemm_("N", "N", n, &m, &m, &c_b18, &work[storez], n, &
work[1], &m, &c_b17, &z__[start * z_dim1 + 1],
ldz);
} else if (icompz == 2) {
ssteqr_("I", &m, &d__[start], &e[start], &z__[start +
start * z_dim1], ldz, &work[1], info);
} else {
ssterf_(&m, &d__[start], &e[start], info);
}
if (*info != 0) {
*info = start * (*n + 1) + finish;
goto L50;
}
}
start = finish + 1;
goto L10;
}
/* endwhile */
/* If the problem split any number of times, then the eigenvalues */
/* will not be properly ordered. Here we permute the eigenvalues */
/* (and the associated eigenvectors) into ascending order. */
if (m != *n) {
if (icompz == 0) {
/* Use Quick Sort */
slasrt_("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];
}
/* L30: */
}
if (k != i__) {
d__[k] = d__[i__];
d__[i__] = p;
sswap_(n, &z__[i__ * z_dim1 + 1], &c__1, &z__[k *
z_dim1 + 1], &c__1);
}
/* L40: */
}
}
}
}
L50:
work[1] = (real) lwmin;
iwork[1] = liwmin;
return 0;
/* End of SSTEDC */
} /* sstedc_ */
/* Subroutine */ int slalsd_(char *uplo, integer *smlsiz, integer *n, integer
*nrhs, real *d__, real *e, real *b, integer *ldb, real *rcond,
integer *rank, real *work, integer *iwork, integer *info)
{
/* System generated locals */
integer b_dim1, b_offset, i__1, i__2;
real r__1;
/* Builtin functions */
double log(doublereal), r_sign(real *, real *);
/* Local variables */
static integer difl, difr, perm, nsub, nlvl, sqre, bxst;
extern /* Subroutine */ int srot_(integer *, real *, integer *, real *,
integer *, real *, real *);
static integer c__, i__, j, k;
static real r__;
static integer s, u, z__;
extern /* Subroutine */ int sgemm_(char *, char *, integer *, integer *,
integer *, real *, real *, integer *, real *, integer *, real *,
real *, integer *);
static integer poles, sizei, nsize;
extern /* Subroutine */ int scopy_(integer *, real *, integer *, real *,
integer *);
static integer nwork, icmpq1, icmpq2;
extern doublereal sopbl3_(char *, integer *, integer *, integer *)
;
static real cs;
static integer bx;
static real sn;
static integer st;
extern /* Subroutine */ int slasda_(integer *, integer *, integer *,
integer *, real *, real *, real *, integer *, real *, integer *,
real *, real *, real *, real *, integer *, integer *, integer *,
integer *, real *, real *, real *, real *, integer *, integer *);
extern doublereal slamch_(char *);
static integer vt;
extern /* Subroutine */ int xerbla_(char *, integer *), slalsa_(
integer *, integer *, integer *, integer *, real *, integer *,
real *, integer *, real *, integer *, real *, integer *, real *,
real *, real *, real *, integer *, integer *, integer *, integer *
, real *, real *, real *, real *, integer *, integer *), slascl_(
char *, integer *, integer *, real *, real *, integer *, integer *
, real *, integer *, integer *);
static integer givcol;
extern integer isamax_(integer *, real *, integer *);
extern /* Subroutine */ int slasdq_(char *, integer *, integer *, integer
*, integer *, integer *, real *, real *, real *, integer *, real *
, integer *, real *, integer *, real *, integer *),
slacpy_(char *, integer *, integer *, real *, integer *, real *,
integer *), slartg_(real *, real *, real *, real *, real *
), slaset_(char *, integer *, integer *, real *, real *, real *,
integer *);
static real orgnrm;
static integer givnum;
extern doublereal slanst_(char *, integer *, real *, real *);
extern /* Subroutine */ int slasrt_(char *, integer *, real *, integer *);
static integer givptr, nm1, smlszp, st1;
static real eps;
static integer iwk;
static real tol;
#define b_ref(a_1,a_2) b[(a_2)*b_dim1 + a_1]
/* -- LAPACK routine (instrumented to count ops, version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
October 31, 1999
Purpose
=======
SLALSD uses the singular value decomposition of A to solve the least
squares problem of finding X to minimize the Euclidean norm of each
column of A*X-B, where A is N-by-N upper bidiagonal, and X and B
are N-by-NRHS. The solution X overwrites B.
The singular values of A smaller than RCOND times the largest
singular value are treated as zero in solving the least squares
problem; in this case a minimum norm solution is returned.
The actual singular values are returned in D in ascending order.
This code makes very mild assumptions about floating point
arithmetic. It will work on machines with a guard digit in
add/subtract, or on those binary machines without guard digits
which subtract like the Cray XMP, Cray YMP, Cray C 90, or Cray 2.
It could conceivably fail on hexadecimal or decimal machines
without guard digits, but we know of none.
Arguments
=========
UPLO (input) CHARACTER*1
= 'U': D and E define an upper bidiagonal matrix.
= 'L': D and E define a lower bidiagonal matrix.
SMLSIZ (input) INTEGER
The maximum size of the subproblems at the bottom of the
computation tree.
N (input) INTEGER
The dimension of the bidiagonal matrix. N >= 0.
NRHS (input) INTEGER
The number of columns of B. NRHS must be at least 1.
D (input/output) REAL array, dimension (N)
On entry D contains the main diagonal of the bidiagonal
matrix. On exit, if INFO = 0, D contains its singular values.
E (input) REAL array, dimension (N-1)
Contains the super-diagonal entries of the bidiagonal matrix.
On exit, E has been destroyed.
B (input/output) REAL array, dimension (LDB,NRHS)
On input, B contains the right hand sides of the least
squares problem. On output, B contains the solution X.
LDB (input) INTEGER
The leading dimension of B in the calling subprogram.
LDB must be at least max(1,N).
RCOND (input) REAL
The singular values of A less than or equal to RCOND times
the largest singular value are treated as zero in solving
the least squares problem. If RCOND is negative,
machine precision is used instead.
For example, if diag(S)*X=B were the least squares problem,
where diag(S) is a diagonal matrix of singular values, the
solution would be X(i) = B(i) / S(i) if S(i) is greater than
RCOND*max(S), and X(i) = 0 if S(i) is less than or equal to
RCOND*max(S).
RANK (output) INTEGER
The number of singular values of A greater than RCOND times
the largest singular value.
WORK (workspace) REAL array, dimension at least
(9*N + 2*N*SMLSIZ + 8*N*NLVL + N*NRHS + (SMLSIZ+1)**2),
where NLVL = max(0, INT(log_2 (N/(SMLSIZ+1))) + 1).
IWORK (workspace) INTEGER array, dimension at least
(3 * N * NLVL + 11 * N)
INFO (output) INTEGER
= 0: successful exit.
< 0: if INFO = -i, the i-th argument had an illegal value.
> 0: The algorithm failed to compute an singular value while
working on the submatrix lying in rows and columns
INFO/(N+1) through MOD(INFO,N+1).
=====================================================================
Test the input parameters.
Parameter adjustments */
--d__;
--e;
b_dim1 = *ldb;
b_offset = 1 + b_dim1 * 1;
b -= b_offset;
--work;
--iwork;
/* Function Body */
*info = 0;
if (*n < 0) {
*info = -3;
} else if (*nrhs < 1) {
*info = -4;
} else if (*ldb < 1 || *ldb < *n) {
*info = -8;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SLALSD", &i__1);
return 0;
}
eps = slamch_("Epsilon");
/* Set up the tolerance. */
if (*rcond <= 0.f || *rcond >= 1.f) {
*rcond = eps;
}
*rank = 0;
/* Quick return if possible. */
if (*n == 0) {
return 0;
} else if (*n == 1) {
if (d__[1] == 0.f) {
slaset_("A", &c__1, nrhs, &c_b6, &c_b6, &b[b_offset], ldb);
} else {
*rank = 1;
latime_1.ops += (real) (*nrhs << 1);
slascl_("G", &c__0, &c__0, &d__[1], &c_b11, &c__1, nrhs, &b[
b_offset], ldb, info);
d__[1] = dabs(d__[1]);
}
return 0;
}
/* Rotate the matrix if it is lower bidiagonal. */
if (*(unsigned char *)uplo == 'L') {
latime_1.ops += (real) ((*n - 1) * 6);
i__1 = *n - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
slartg_(&d__[i__], &e[i__], &cs, &sn, &r__);
d__[i__] = r__;
e[i__] = sn * d__[i__ + 1];
d__[i__ + 1] = cs * d__[i__ + 1];
if (*nrhs == 1) {
latime_1.ops += 6.f;
srot_(&c__1, &b_ref(i__, 1), &c__1, &b_ref(i__ + 1, 1), &c__1,
&cs, &sn);
} else {
work[(i__ << 1) - 1] = cs;
work[i__ * 2] = sn;
}
/* L10: */
}
if (*nrhs > 1) {
latime_1.ops += (real) ((*n - 1) * 6 * *nrhs);
i__1 = *nrhs;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = *n - 1;
for (j = 1; j <= i__2; ++j) {
cs = work[(j << 1) - 1];
sn = work[j * 2];
srot_(&c__1, &b_ref(j, i__), &c__1, &b_ref(j + 1, i__), &
c__1, &cs, &sn);
/* L20: */
}
/* L30: */
}
}
}
/* Scale. */
nm1 = *n - 1;
orgnrm = slanst_("M", n, &d__[1], &e[1]);
if (orgnrm == 0.f) {
slaset_("A", n, nrhs, &c_b6, &c_b6, &b[b_offset], ldb);
return 0;
}
latime_1.ops += (real) (*n + nm1);
slascl_("G", &c__0, &c__0, &orgnrm, &c_b11, n, &c__1, &d__[1], n, info);
slascl_("G", &c__0, &c__0, &orgnrm, &c_b11, &nm1, &c__1, &e[1], &nm1,
info);
/* If N is smaller than the minimum divide size SMLSIZ, then solve
the problem with another solver. */
if (*n <= *smlsiz) {
nwork = *n * *n + 1;
slaset_("A", n, n, &c_b6, &c_b11, &work[1], n);
slasdq_("U", &c__0, n, n, &c__0, nrhs, &d__[1], &e[1], &work[1], n, &
work[1], n, &b[b_offset], ldb, &work[nwork], info);
if (*info != 0) {
return 0;
}
latime_1.ops += 1.f;
tol = *rcond * (r__1 = d__[isamax_(n, &d__[1], &c__1)], dabs(r__1));
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
if (d__[i__] <= tol) {
slaset_("A", &c__1, nrhs, &c_b6, &c_b6, &b_ref(i__, 1), ldb);
} else {
latime_1.ops += (real) (*nrhs);
slascl_("G", &c__0, &c__0, &d__[i__], &c_b11, &c__1, nrhs, &
b_ref(i__, 1), ldb, info);
++(*rank);
}
/* L40: */
}
latime_1.ops += sopbl3_("SGEMM ", n, nrhs, n);
sgemm_("T", "N", n, nrhs, n, &c_b11, &work[1], n, &b[b_offset], ldb, &
c_b6, &work[nwork], n);
slacpy_("A", n, nrhs, &work[nwork], n, &b[b_offset], ldb);
/* Unscale. */
latime_1.ops += (real) (*n + *n * *nrhs);
slascl_("G", &c__0, &c__0, &c_b11, &orgnrm, n, &c__1, &d__[1], n,
info);
slasrt_("D", n, &d__[1], info);
slascl_("G", &c__0, &c__0, &orgnrm, &c_b11, n, nrhs, &b[b_offset],
ldb, info);
return 0;
}
/* Book-keeping and setting up some constants. */
nlvl = (integer) (log((real) (*n) / (real) (*smlsiz + 1)) / log(2.f)) + 1;
smlszp = *smlsiz + 1;
u = 1;
vt = *smlsiz * *n + 1;
difl = vt + smlszp * *n;
difr = difl + nlvl * *n;
z__ = difr + (nlvl * *n << 1);
c__ = z__ + nlvl * *n;
s = c__ + *n;
poles = s + *n;
givnum = poles + (nlvl << 1) * *n;
bx = givnum + (nlvl << 1) * *n;
nwork = bx + *n * *nrhs;
sizei = *n + 1;
k = sizei + *n;
givptr = k + *n;
perm = givptr + *n;
givcol = perm + nlvl * *n;
iwk = givcol + (nlvl * *n << 1);
st = 1;
sqre = 0;
icmpq1 = 1;
icmpq2 = 0;
nsub = 0;
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
if ((r__1 = d__[i__], dabs(r__1)) < eps) {
d__[i__] = r_sign(&eps, &d__[i__]);
}
/* L50: */
}
i__1 = nm1;
for (i__ = 1; i__ <= i__1; ++i__) {
if ((r__1 = e[i__], dabs(r__1)) < eps || i__ == nm1) {
++nsub;
iwork[nsub] = st;
/* Subproblem found. First determine its size and then
apply divide and conquer on it. */
if (i__ < nm1) {
/* A subproblem with E(I) small for I < NM1. */
nsize = i__ - st + 1;
iwork[sizei + nsub - 1] = nsize;
} else if ((r__1 = e[i__], dabs(r__1)) >= eps) {
/* A subproblem with E(NM1) not too small but I = NM1. */
nsize = *n - st + 1;
iwork[sizei + nsub - 1] = nsize;
} else {
/* A subproblem with E(NM1) small. This implies an
1-by-1 subproblem at D(N), which is not solved
explicitly. */
nsize = i__ - st + 1;
iwork[sizei + nsub - 1] = nsize;
++nsub;
iwork[nsub] = *n;
iwork[sizei + nsub - 1] = 1;
scopy_(nrhs, &b_ref(*n, 1), ldb, &work[bx + nm1], n);
}
st1 = st - 1;
if (nsize == 1) {
/* This is a 1-by-1 subproblem and is not solved
explicitly. */
scopy_(nrhs, &b_ref(st, 1), ldb, &work[bx + st1], n);
} else if (nsize <= *smlsiz) {
/* This is a small subproblem and is solved by SLASDQ. */
slaset_("A", &nsize, &nsize, &c_b6, &c_b11, &work[vt + st1],
n);
slasdq_("U", &c__0, &nsize, &nsize, &c__0, nrhs, &d__[st], &e[
st], &work[vt + st1], n, &work[nwork], n, &b_ref(st,
1), ldb, &work[nwork], info);
if (*info != 0) {
return 0;
}
slacpy_("A", &nsize, nrhs, &b_ref(st, 1), ldb, &work[bx + st1]
, n);
} else {
/* A large problem. Solve it using divide and conquer. */
slasda_(&icmpq1, smlsiz, &nsize, &sqre, &d__[st], &e[st], &
work[u + st1], n, &work[vt + st1], &iwork[k + st1], &
work[difl + st1], &work[difr + st1], &work[z__ + st1],
&work[poles + st1], &iwork[givptr + st1], &iwork[
givcol + st1], n, &iwork[perm + st1], &work[givnum +
st1], &work[c__ + st1], &work[s + st1], &work[nwork],
&iwork[iwk], info);
if (*info != 0) {
return 0;
}
bxst = bx + st1;
slalsa_(&icmpq2, smlsiz, &nsize, nrhs, &b_ref(st, 1), ldb, &
work[bxst], n, &work[u + st1], n, &work[vt + st1], &
iwork[k + st1], &work[difl + st1], &work[difr + st1],
&work[z__ + st1], &work[poles + st1], &iwork[givptr +
st1], &iwork[givcol + st1], n, &iwork[perm + st1], &
work[givnum + st1], &work[c__ + st1], &work[s + st1],
&work[nwork], &iwork[iwk], info);
if (*info != 0) {
return 0;
}
}
st = i__ + 1;
}
/* L60: */
}
/* Apply the singular values and treat the tiny ones as zero. */
tol = *rcond * (r__1 = d__[isamax_(n, &d__[1], &c__1)], dabs(r__1));
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
/* Some of the elements in D can be negative because 1-by-1
subproblems were not solved explicitly. */
if ((r__1 = d__[i__], dabs(r__1)) <= tol) {
slaset_("A", &c__1, nrhs, &c_b6, &c_b6, &work[bx + i__ - 1], n);
} else {
++(*rank);
latime_1.ops += (real) (*nrhs);
slascl_("G", &c__0, &c__0, &d__[i__], &c_b11, &c__1, nrhs, &work[
bx + i__ - 1], n, info);
}
d__[i__] = (r__1 = d__[i__], dabs(r__1));
/* L70: */
}
/* Now apply back the right singular vectors. */
icmpq2 = 1;
i__1 = nsub;
for (i__ = 1; i__ <= i__1; ++i__) {
st = iwork[i__];
st1 = st - 1;
nsize = iwork[sizei + i__ - 1];
bxst = bx + st1;
if (nsize == 1) {
scopy_(nrhs, &work[bxst], n, &b_ref(st, 1), ldb);
} else if (nsize <= *smlsiz) {
latime_1.ops += sopbl3_("SGEMM ", &nsize, nrhs, &nsize)
;
sgemm_("T", "N", &nsize, nrhs, &nsize, &c_b11, &work[vt + st1], n,
&work[bxst], n, &c_b6, &b_ref(st, 1), ldb);
} else {
slalsa_(&icmpq2, smlsiz, &nsize, nrhs, &work[bxst], n, &b_ref(st,
1), ldb, &work[u + st1], n, &work[vt + st1], &iwork[k +
st1], &work[difl + st1], &work[difr + st1], &work[z__ +
st1], &work[poles + st1], &iwork[givptr + st1], &iwork[
givcol + st1], n, &iwork[perm + st1], &work[givnum + st1],
&work[c__ + st1], &work[s + st1], &work[nwork], &iwork[
iwk], info);
if (*info != 0) {
return 0;
}
}
/* L80: */
}
/* Unscale and sort the singular values. */
latime_1.ops += (real) (*n + *n * *nrhs);
slascl_("G", &c__0, &c__0, &c_b11, &orgnrm, n, &c__1, &d__[1], n, info);
slasrt_("D", n, &d__[1], info);
slascl_("G", &c__0, &c__0, &orgnrm, &c_b11, n, nrhs, &b[b_offset], ldb,
info);
return 0;
/* End of SLALSD */
} /* slalsd_ */
/* Subroutine */ int snaitr_(integer *ido, char *bmat, integer *n, integer *k,
integer *np, integer *nb, real *resid, real *rnorm, real *v, integer
*ldv, real *h__, integer *ldh, integer *ipntr, real *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;
real r__1, r__2;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
static integer i__, j;
static real t0, t1, t2, t3, t4, t5;
static integer jj, ipj, irj, ivj;
static real ulp, tst1;
static integer ierr, iter;
static real unfl, ovfl;
extern doublereal sdot_(integer *, real *, integer *, real *, integer *);
static integer itry;
static real temp1;
static logical orth1, orth2, step3, step4;
extern doublereal snrm2_(integer *, real *, integer *);
static real betaj;
extern /* Subroutine */ int sscal_(integer *, real *, real *, integer *);
static integer infol;
extern /* Subroutine */ int sgemv_(char *, integer *, integer *, real *,
real *, integer *, real *, integer *, real *, real *, integer *,
ftnlen);
static real xtemp[2];
extern /* Subroutine */ int scopy_(integer *, real *, integer *, real *,
integer *);
static real wnorm;
extern /* Subroutine */ int saxpy_(integer *, real *, real *, integer *,
real *, integer *), ivout_(integer *, integer *, integer *,
integer *, char *, ftnlen), smout_(integer *, integer *, integer *
, real *, integer *, integer *, char *, ftnlen), svout_(integer *,
integer *, real *, integer *, char *, ftnlen), sgetv0_(integer *,
char *, integer *, logical *, integer *, integer *, real *,
integer *, real *, real *, integer *, real *, integer *, ftnlen);
static real rnorm1;
extern /* Subroutine */ int slabad_(real *, real *);
extern doublereal slamch_(char *, ftnlen);
extern /* Subroutine */ int second_(real *), slascl_(char *, integer *,
integer *, real *, real *, integer *, integer *, real *, integer *
, integer *, ftnlen);
static logical rstart;
static integer msglvl;
static real smlnum;
extern doublereal slanhs_(char *, integer *, real *, integer *, real *,
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 | */
/* %---------------% */
/* %-----------------------% */
/* | Local Array Arguments | */
/* %-----------------------% */
/* %----------------------% */
/* | 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 slahqr | */
/* %-----------------------------------------% */
unfl = slamch_("safe minimum", (ftnlen)12);
ovfl = 1.f / unfl;
slabad_(&unfl, &ovfl);
ulp = slamch_("precision", (ftnlen)9);
smlnum = unfl * (*n / ulp);
first = FALSE_;
}
if (*ido == 0) {
/* %-------------------------------% */
/* | Initialize timing statistics | */
/* | & message level for debugging | */
/* %-------------------------------% */
second_(&t0);
msglvl = debug_1.mnaitr;
/* %------------------------------% */
/* | 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 | */
/* | sgetv0. | */
/* %-------------------------------------------------% */
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);
svout_(&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 determing whether | */
/* | an exact j-step Arnoldi factorization is present. | */
/* %---------------------------------------------------% */
betaj = *rnorm;
if (*rnorm > 0.f) {
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.f;
++timing_1.nrstrt;
itry = 1;
L20:
rstart = TRUE_;
*ido = 0;
L30:
/* %--------------------------------------% */
/* | If in reverse communication mode and | */
/* | RSTART = .true. flow returns here. | */
/* %--------------------------------------% */
sgetv0_(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.tnaitr += 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. | */
/* %---------------------------------------------------------% */
scopy_(n, &resid[1], &c__1, &v[j * v_dim1 + 1], &c__1);
if (*rnorm >= unfl) {
temp1 = 1.f / *rnorm;
sscal_(n, &temp1, &v[j * v_dim1 + 1], &c__1);
sscal_(n, &temp1, &workd[ipj], &c__1);
} else {
/* %-----------------------------------------% */
/* | To scale both v_{j} and p_{j} carefully | */
/* | use LAPACK routine SLASCL | */
/* %-----------------------------------------% */
slascl_("General", &i__, &i__, rnorm, &c_b25, n, &c__1, &v[j * v_dim1
+ 1], n, &infol, (ftnlen)7);
slascl_("General", &i__, &i__, rnorm, &c_b25, 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);
scopy_(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. | */
/* %------------------------------------------% */
scopy_(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') {
scopy_(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') {
wnorm = sdot_(n, &resid[1], &c__1, &workd[ipj], &c__1);
wnorm = sqrt((dabs(wnorm)));
} else if (*(unsigned char *)bmat == 'I') {
wnorm = snrm2_(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}. | */
/* %------------------------------------------% */
sgemv_("T", n, &j, &c_b25, &v[v_offset], ldv, &workd[ipj], &c__1, &c_b47,
&h__[j * h_dim1 + 1], &c__1, (ftnlen)1);
/* %--------------------------------------% */
/* | Orthogonalize r_{j} against V_{j}. | */
/* | RESID contains OP*v_{j}. See STEP 3. | */
/* %--------------------------------------% */
sgemv_("N", n, &j, &c_b50, &v[v_offset], ldv, &h__[j * h_dim1 + 1], &c__1,
&c_b25, &resid[1], &c__1, (ftnlen)1);
if (j > 1) {
h__[j + (j - 1) * h_dim1] = betaj;
}
second_(&t4);
orth1 = TRUE_;
second_(&t2);
if (*(unsigned char *)bmat == 'G') {
++timing_1.nbx;
scopy_(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') {
scopy_(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') {
*rnorm = sdot_(n, &resid[1], &c__1, &workd[ipj], &c__1);
*rnorm = sqrt((dabs(*rnorm)));
} else if (*(unsigned char *)bmat == 'I') {
*rnorm = snrm2_(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) {
xtemp[0] = wnorm;
xtemp[1] = *rnorm;
svout_(&debug_1.logfil, &c__2, xtemp, &debug_1.ndigit, "_naitr: re-o"
"rthonalization; wnorm and rnorm are", (ftnlen)47);
svout_(&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). | */
/* %----------------------------------------------------% */
sgemv_("T", n, &j, &c_b25, &v[v_offset], ldv, &workd[ipj], &c__1, &c_b47,
&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. | */
/* %---------------------------------------------% */
sgemv_("N", n, &j, &c_b50, &v[v_offset], ldv, &workd[irj], &c__1, &c_b25,
&resid[1], &c__1, (ftnlen)1);
saxpy_(&j, &c_b25, &workd[irj], &c__1, &h__[j * h_dim1 + 1], &c__1);
orth2 = TRUE_;
second_(&t2);
if (*(unsigned char *)bmat == 'G') {
++timing_1.nbx;
scopy_(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') {
scopy_(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') {
rnorm1 = sdot_(n, &resid[1], &c__1, &workd[ipj], &c__1);
rnorm1 = sqrt((dabs(rnorm1)));
} else if (*(unsigned char *)bmat == 'I') {
rnorm1 = snrm2_(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) {
xtemp[0] = *rnorm;
xtemp[1] = rnorm1;
svout_(&debug_1.logfil, &c__2, xtemp, &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) {
resid[jj] = 0.f;
/* L95: */
}
*rnorm = 0.f;
}
/* %----------------------------------------------% */
/* | 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.tnaitr += 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 slahqr | */
/* %--------------------------------------------% */
tst1 = (r__1 = h__[i__ + i__ * h_dim1], dabs(r__1)) + (r__2 = h__[
i__ + 1 + (i__ + 1) * h_dim1], dabs(r__2));
if (tst1 == 0.f) {
i__2 = *k + *np;
tst1 = slanhs_("1", &i__2, &h__[h_offset], ldh, &workd[*n + 1]
, (ftnlen)1);
}
/* Computing MAX */
r__2 = ulp * tst1;
if ((r__1 = h__[i__ + 1 + i__ * h_dim1], dabs(r__1)) <= dmax(r__2,
smlnum)) {
h__[i__ + 1 + i__ * h_dim1] = 0.f;
}
/* L110: */
}
if (msglvl > 2) {
i__1 = *k + *np;
i__2 = *k + *np;
smout_(&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 snaitr | */
/* %---------------% */
} /* snaitr_ */
/* Subroutine */ int sqrt15_(integer *scale, integer *rksel, integer *m,
integer *n, integer *nrhs, real *a, integer *lda, real *b, integer *
ldb, real *s, integer *rank, real *norma, real *normb, integer *iseed,
real *work, integer *lwork)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, i__1, i__2;
real r__1;
/* Local variables */
static integer info;
static real temp;
extern doublereal snrm2_(integer *, real *, integer *);
static integer j;
extern /* Subroutine */ int sscal_(integer *, real *, real *, integer *),
slarf_(char *, integer *, integer *, real *, integer *, real *,
real *, integer *, real *), sgemm_(char *, char *,
integer *, integer *, integer *, real *, real *, integer *, real *
, integer *, real *, real *, integer *);
extern doublereal sasum_(integer *, real *, integer *);
static real dummy[1];
static integer mn;
extern doublereal slamch_(char *), slange_(char *, integer *,
integer *, real *, integer *, real *);
extern /* Subroutine */ int xerbla_(char *, integer *);
static real bignum;
extern /* Subroutine */ int slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *);
extern doublereal slarnd_(integer *, integer *);
extern /* Subroutine */ int slaord_(char *, integer *, real *, integer *), slaset_(char *, integer *, integer *, real *, real *,
real *, integer *), slaror_(char *, char *, integer *,
integer *, real *, integer *, integer *, real *, integer *), slarnv_(integer *, integer *, integer *, real *);
static real smlnum, eps;
#define a_ref(a_1,a_2) a[(a_2)*a_dim1 + a_1]
/* -- LAPACK test routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
September 30, 1994
Purpose
=======
SQRT15 generates a matrix with full or deficient rank and of various
norms.
Arguments
=========
SCALE (input) INTEGER
SCALE = 1: normally scaled matrix
SCALE = 2: matrix scaled up
SCALE = 3: matrix scaled down
RKSEL (input) INTEGER
RKSEL = 1: full rank matrix
RKSEL = 2: rank-deficient matrix
M (input) INTEGER
The number of rows of the matrix A.
N (input) INTEGER
The number of columns of A.
NRHS (input) INTEGER
The number of columns of B.
A (output) REAL array, dimension (LDA,N)
The M-by-N matrix A.
LDA (input) INTEGER
The leading dimension of the array A.
B (output) REAL array, dimension (LDB, NRHS)
A matrix that is in the range space of matrix A.
LDB (input) INTEGER
The leading dimension of the array B.
S (output) REAL array, dimension MIN(M,N)
Singular values of A.
RANK (output) INTEGER
number of nonzero singular values of A.
NORMA (output) REAL
one-norm of A.
NORMB (output) REAL
one-norm of B.
ISEED (input/output) integer array, dimension (4)
seed for random number generator.
WORK (workspace) REAL array, dimension (LWORK)
LWORK (input) INTEGER
length of work space required.
LWORK >= MAX(M+MIN(M,N),NRHS*MIN(M,N),2*N+M)
=====================================================================
Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1 * 1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1 * 1;
b -= b_offset;
--s;
--iseed;
--work;
/* Function Body */
mn = min(*m,*n);
/* Computing MAX */
i__1 = *m + mn, i__2 = mn * *nrhs, i__1 = max(i__1,i__2), i__2 = (*n << 1)
+ *m;
if (*lwork < max(i__1,i__2)) {
xerbla_("SQRT15", &c__16);
return 0;
}
smlnum = slamch_("Safe minimum");
bignum = 1.f / smlnum;
eps = slamch_("Epsilon");
smlnum = smlnum / eps / eps;
bignum = 1.f / smlnum;
/* Determine rank and (unscaled) singular values */
if (*rksel == 1) {
*rank = mn;
} else if (*rksel == 2) {
*rank = mn * 3 / 4;
i__1 = mn;
for (j = *rank + 1; j <= i__1; ++j) {
s[j] = 0.f;
/* L10: */
}
} else {
xerbla_("SQRT15", &c__2);
}
if (*rank > 0) {
/* Nontrivial case */
s[1] = 1.f;
i__1 = *rank;
for (j = 2; j <= i__1; ++j) {
L20:
temp = slarnd_(&c__1, &iseed[1]);
if (temp > .1f) {
s[j] = dabs(temp);
} else {
goto L20;
}
/* L30: */
}
slaord_("Decreasing", rank, &s[1], &c__1);
/* Generate 'rank' columns of a random orthogonal matrix in A */
slarnv_(&c__2, &iseed[1], m, &work[1]);
r__1 = 1.f / snrm2_(m, &work[1], &c__1);
sscal_(m, &r__1, &work[1], &c__1);
slaset_("Full", m, rank, &c_b18, &c_b19, &a[a_offset], lda)
;
slarf_("Left", m, rank, &work[1], &c__1, &c_b22, &a[a_offset], lda, &
work[*m + 1]);
/* workspace used: m+mn
Generate consistent rhs in the range space of A */
i__1 = *rank * *nrhs;
slarnv_(&c__2, &iseed[1], &i__1, &work[1]);
sgemm_("No transpose", "No transpose", m, nrhs, rank, &c_b19, &a[
a_offset], lda, &work[1], rank, &c_b18, &b[b_offset], ldb);
/* work space used: <= mn *nrhs
generate (unscaled) matrix A */
i__1 = *rank;
for (j = 1; j <= i__1; ++j) {
sscal_(m, &s[j], &a_ref(1, j), &c__1);
/* L40: */
}
if (*rank < *n) {
i__1 = *n - *rank;
slaset_("Full", m, &i__1, &c_b18, &c_b18, &a_ref(1, *rank + 1),
lda);
}
slaror_("Right", "No initialization", m, n, &a[a_offset], lda, &iseed[
1], &work[1], &info);
} else {
/* work space used 2*n+m
Generate null matrix and rhs */
i__1 = mn;
for (j = 1; j <= i__1; ++j) {
s[j] = 0.f;
/* L50: */
}
slaset_("Full", m, n, &c_b18, &c_b18, &a[a_offset], lda);
slaset_("Full", m, nrhs, &c_b18, &c_b18, &b[b_offset], ldb)
;
}
/* Scale the matrix */
if (*scale != 1) {
*norma = slange_("Max", m, n, &a[a_offset], lda, dummy);
if (*norma != 0.f) {
if (*scale == 2) {
/* matrix scaled up */
slascl_("General", &c__0, &c__0, norma, &bignum, m, n, &a[
a_offset], lda, &info);
slascl_("General", &c__0, &c__0, norma, &bignum, &mn, &c__1, &
s[1], &mn, &info);
slascl_("General", &c__0, &c__0, norma, &bignum, m, nrhs, &b[
b_offset], ldb, &info);
} else if (*scale == 3) {
/* matrix scaled down */
slascl_("General", &c__0, &c__0, norma, &smlnum, m, n, &a[
a_offset], lda, &info);
slascl_("General", &c__0, &c__0, norma, &smlnum, &mn, &c__1, &
s[1], &mn, &info);
slascl_("General", &c__0, &c__0, norma, &smlnum, m, nrhs, &b[
b_offset], ldb, &info);
} else {
xerbla_("SQRT15", &c__1);
return 0;
}
}
}
*norma = sasum_(&mn, &s[1], &c__1);
*normb = slange_("One-norm", m, nrhs, &b[b_offset], ldb, dummy)
;
return 0;
/* End of SQRT15 */
} /* sqrt15_ */
/* Subroutine */ int slasd3_(integer *nl, integer *nr, integer *sqre, integer
*k, real *d__, real *q, integer *ldq, real *dsigma, real *u, integer *
ldu, real *u2, integer *ldu2, real *vt, integer *ldvt, real *vt2,
integer *ldvt2, integer *idxc, integer *ctot, real *z__, integer *
info)
{
/* System generated locals */
integer q_dim1, q_offset, u_dim1, u_offset, u2_dim1, u2_offset, vt_dim1,
vt_offset, vt2_dim1, vt2_offset, i__1, i__2;
real r__1, r__2;
/* Builtin functions */
double sqrt(doublereal), r_sign(real *, real *);
/* Local variables */
integer i__, j, m, n, jc;
real rho;
integer nlp1, nlp2, nrp1;
real temp;
extern doublereal snrm2_(integer *, real *, integer *);
integer ctemp;
extern /* Subroutine */ int sgemm_(char *, char *, integer *, integer *,
integer *, real *, real *, integer *, real *, integer *, real *,
real *, integer *);
integer ktemp;
extern /* Subroutine */ int scopy_(integer *, real *, integer *, real *,
integer *);
extern doublereal slamc3_(real *, real *);
extern /* Subroutine */ int slasd4_(integer *, integer *, real *, real *,
real *, real *, real *, real *, integer *), xerbla_(char *,
integer *), slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *), slacpy_(char *, integer *, integer *, real *, integer *,
real *, integer *);
/* -- LAPACK auxiliary routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SLASD3 finds all the square roots of the roots of the secular */
/* equation, as defined by the values in D and Z. It makes the */
/* appropriate calls to SLASD4 and then updates the singular */
/* vectors by matrix multiplication. */
/* This code makes very mild assumptions about floating point */
/* arithmetic. It will work on machines with a guard digit in */
/* add/subtract, or on those binary machines without guard digits */
/* which subtract like the Cray XMP, Cray YMP, Cray C 90, or Cray 2. */
/* It could conceivably fail on hexadecimal or decimal machines */
/* without guard digits, but we know of none. */
/* SLASD3 is called from SLASD1. */
/* 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 (input) INTEGER */
/* The size of the secular equation, 1 =< K = < N. */
/* D (output) REAL array, dimension(K) */
/* On exit the square roots of the roots of the secular equation, */
/* in ascending order. */
/* Q (workspace) REAL array, */
/* dimension at least (LDQ,K). */
/* LDQ (input) INTEGER */
/* The leading dimension of the array Q. LDQ >= K. */
/* DSIGMA (input/output) REAL array, dimension(K) */
/* The first K elements of this array contain the old roots */
/* of the deflated updating problem. These are the poles */
/* of the secular equation. */
/* U (output) REAL array, dimension (LDU, N) */
/* The last N - K columns of this matrix contain the deflated */
/* left singular vectors. */
/* LDU (input) INTEGER */
/* The leading dimension of the array U. LDU >= N. */
/* U2 (input) REAL array, dimension (LDU2, N) */
/* The first K columns of this matrix contain the non-deflated */
/* left singular vectors for the split problem. */
/* LDU2 (input) INTEGER */
/* The leading dimension of the array U2. LDU2 >= N. */
/* VT (output) REAL array, dimension (LDVT, M) */
/* The last M - K columns of VT' contain the deflated */
/* right singular vectors. */
/* LDVT (input) INTEGER */
/* The leading dimension of the array VT. LDVT >= N. */
/* VT2 (input/output) REAL array, dimension (LDVT2, N) */
/* The first K columns of VT2' contain the non-deflated */
/* right singular vectors for the split problem. */
/* LDVT2 (input) INTEGER */
/* The leading dimension of the array VT2. LDVT2 >= N. */
/* IDXC (input) INTEGER array, dimension (N) */
/* The permutation used to arrange the columns of U (and rows of */
/* VT) into three groups: the first group contains non-zero */
/* entries only at and above (or before) NL +1; the second */
/* contains non-zero entries only at and below (or after) NL+2; */
/* and the third is dense. The first column of U and the row of */
/* VT are treated separately, however. */
/* The rows of the singular vectors found by SLASD4 */
/* must be likewise permuted before the matrix multiplies can */
/* take place. */
/* CTOT (input) INTEGER array, dimension (4) */
/* A count of the total number of the various types of columns */
/* in U (or rows in VT), as described in IDXC. The fourth column */
/* type is any column which has been deflated. */
/* Z (input/output) REAL array, dimension (K) */
/* The first K elements of this array contain the components */
/* of the deflation-adjusted updating row vector. */
/* INFO (output) INTEGER */
/* = 0: successful exit. */
/* < 0: if INFO = -i, the i-th argument had an illegal value. */
/* > 0: if INFO = 1, an singular value did not converge */
/* Further Details */
/* =============== */
/* Based on contributions by */
/* Ming Gu and Huan Ren, Computer Science Division, University of */
/* California at Berkeley, USA */
/* ===================================================================== */
/* .. 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;
--dsigma;
u_dim1 = *ldu;
u_offset = 1 + u_dim1;
u -= u_offset;
u2_dim1 = *ldu2;
u2_offset = 1 + u2_dim1;
u2 -= u2_offset;
vt_dim1 = *ldvt;
vt_offset = 1 + vt_dim1;
vt -= vt_offset;
vt2_dim1 = *ldvt2;
vt2_offset = 1 + vt2_dim1;
vt2 -= vt2_offset;
--idxc;
--ctot;
--z__;
/* 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;
nlp1 = *nl + 1;
nlp2 = *nl + 2;
if (*k < 1 || *k > n) {
*info = -4;
} else if (*ldq < *k) {
*info = -7;
} else if (*ldu < n) {
*info = -10;
} else if (*ldu2 < n) {
*info = -12;
} else if (*ldvt < m) {
*info = -14;
} else if (*ldvt2 < m) {
*info = -16;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SLASD3", &i__1);
return 0;
}
/* Quick return if possible */
if (*k == 1) {
d__[1] = dabs(z__[1]);
scopy_(&m, &vt2[vt2_dim1 + 1], ldvt2, &vt[vt_dim1 + 1], ldvt);
if (z__[1] > 0.f) {
scopy_(&n, &u2[u2_dim1 + 1], &c__1, &u[u_dim1 + 1], &c__1);
} else {
i__1 = n;
for (i__ = 1; i__ <= i__1; ++i__) {
u[i__ + u_dim1] = -u2[i__ + u2_dim1];
/* L10: */
}
}
return 0;
}
/* Modify values DSIGMA(i) to make sure all DSIGMA(i)-DSIGMA(j) can */
/* be computed with high relative accuracy (barring over/underflow). */
/* This is a problem on machines without a guard digit in */
/* add/subtract (Cray XMP, Cray YMP, Cray C 90 and Cray 2). */
/* The following code replaces DSIGMA(I) by 2*DSIGMA(I)-DSIGMA(I), */
/* which on any of these machines zeros out the bottommost */
/* bit of DSIGMA(I) if it is 1; this makes the subsequent */
/* subtractions DSIGMA(I)-DSIGMA(J) unproblematic when cancellation */
/* occurs. On binary machines with a guard digit (almost all */
/* machines) it does not change DSIGMA(I) at all. On hexadecimal */
/* and decimal machines with a guard digit, it slightly */
/* changes the bottommost bits of DSIGMA(I). It does not account */
/* for hexadecimal or decimal machines without guard digits */
/* (we know of none). We use a subroutine call to compute */
/* 2*DSIGMA(I) to prevent optimizing compilers from eliminating */
/* this code. */
i__1 = *k;
for (i__ = 1; i__ <= i__1; ++i__) {
dsigma[i__] = slamc3_(&dsigma[i__], &dsigma[i__]) - dsigma[i__];
/* L20: */
}
/* Keep a copy of Z. */
scopy_(k, &z__[1], &c__1, &q[q_offset], &c__1);
/* Normalize Z. */
rho = snrm2_(k, &z__[1], &c__1);
slascl_("G", &c__0, &c__0, &rho, &c_b13, k, &c__1, &z__[1], k, info);
rho *= rho;
/* Find the new singular values. */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
slasd4_(k, &j, &dsigma[1], &z__[1], &u[j * u_dim1 + 1], &rho, &d__[j],
&vt[j * vt_dim1 + 1], info);
/* If the zero finder fails, the computation is terminated. */
if (*info != 0) {
return 0;
}
/* L30: */
}
/* Compute updated Z. */
i__1 = *k;
for (i__ = 1; i__ <= i__1; ++i__) {
z__[i__] = u[i__ + *k * u_dim1] * vt[i__ + *k * vt_dim1];
i__2 = i__ - 1;
for (j = 1; j <= i__2; ++j) {
z__[i__] *= u[i__ + j * u_dim1] * vt[i__ + j * vt_dim1] / (dsigma[
i__] - dsigma[j]) / (dsigma[i__] + dsigma[j]);
/* L40: */
}
i__2 = *k - 1;
for (j = i__; j <= i__2; ++j) {
z__[i__] *= u[i__ + j * u_dim1] * vt[i__ + j * vt_dim1] / (dsigma[
i__] - dsigma[j + 1]) / (dsigma[i__] + dsigma[j + 1]);
/* L50: */
}
r__2 = sqrt((r__1 = z__[i__], dabs(r__1)));
z__[i__] = r_sign(&r__2, &q[i__ + q_dim1]);
/* L60: */
}
/* Compute left singular vectors of the modified diagonal matrix, */
/* and store related information for the right singular vectors. */
i__1 = *k;
for (i__ = 1; i__ <= i__1; ++i__) {
vt[i__ * vt_dim1 + 1] = z__[1] / u[i__ * u_dim1 + 1] / vt[i__ *
vt_dim1 + 1];
u[i__ * u_dim1 + 1] = -1.f;
i__2 = *k;
for (j = 2; j <= i__2; ++j) {
vt[j + i__ * vt_dim1] = z__[j] / u[j + i__ * u_dim1] / vt[j + i__
* vt_dim1];
u[j + i__ * u_dim1] = dsigma[j] * vt[j + i__ * vt_dim1];
/* L70: */
}
temp = snrm2_(k, &u[i__ * u_dim1 + 1], &c__1);
q[i__ * q_dim1 + 1] = u[i__ * u_dim1 + 1] / temp;
i__2 = *k;
for (j = 2; j <= i__2; ++j) {
jc = idxc[j];
q[j + i__ * q_dim1] = u[jc + i__ * u_dim1] / temp;
/* L80: */
}
/* L90: */
}
/* Update the left singular vector matrix. */
if (*k == 2) {
sgemm_("N", "N", &n, k, k, &c_b13, &u2[u2_offset], ldu2, &q[q_offset],
ldq, &c_b26, &u[u_offset], ldu);
goto L100;
}
if (ctot[1] > 0) {
sgemm_("N", "N", nl, k, &ctot[1], &c_b13, &u2[(u2_dim1 << 1) + 1],
ldu2, &q[q_dim1 + 2], ldq, &c_b26, &u[u_dim1 + 1], ldu);
if (ctot[3] > 0) {
ktemp = ctot[1] + 2 + ctot[2];
sgemm_("N", "N", nl, k, &ctot[3], &c_b13, &u2[ktemp * u2_dim1 + 1]
, ldu2, &q[ktemp + q_dim1], ldq, &c_b13, &u[u_dim1 + 1],
ldu);
}
} else if (ctot[3] > 0) {
ktemp = ctot[1] + 2 + ctot[2];
sgemm_("N", "N", nl, k, &ctot[3], &c_b13, &u2[ktemp * u2_dim1 + 1],
ldu2, &q[ktemp + q_dim1], ldq, &c_b26, &u[u_dim1 + 1], ldu);
} else {
slacpy_("F", nl, k, &u2[u2_offset], ldu2, &u[u_offset], ldu);
}
scopy_(k, &q[q_dim1 + 1], ldq, &u[nlp1 + u_dim1], ldu);
ktemp = ctot[1] + 2;
ctemp = ctot[2] + ctot[3];
sgemm_("N", "N", nr, k, &ctemp, &c_b13, &u2[nlp2 + ktemp * u2_dim1], ldu2,
&q[ktemp + q_dim1], ldq, &c_b26, &u[nlp2 + u_dim1], ldu);
/* Generate the right singular vectors. */
L100:
i__1 = *k;
for (i__ = 1; i__ <= i__1; ++i__) {
temp = snrm2_(k, &vt[i__ * vt_dim1 + 1], &c__1);
q[i__ + q_dim1] = vt[i__ * vt_dim1 + 1] / temp;
i__2 = *k;
for (j = 2; j <= i__2; ++j) {
jc = idxc[j];
q[i__ + j * q_dim1] = vt[jc + i__ * vt_dim1] / temp;
/* L110: */
}
/* L120: */
}
/* Update the right singular vector matrix. */
if (*k == 2) {
sgemm_("N", "N", k, &m, k, &c_b13, &q[q_offset], ldq, &vt2[vt2_offset]
, ldvt2, &c_b26, &vt[vt_offset], ldvt);
return 0;
}
ktemp = ctot[1] + 1;
sgemm_("N", "N", k, &nlp1, &ktemp, &c_b13, &q[q_dim1 + 1], ldq, &vt2[
vt2_dim1 + 1], ldvt2, &c_b26, &vt[vt_dim1 + 1], ldvt);
ktemp = ctot[1] + 2 + ctot[2];
if (ktemp <= *ldvt2) {
sgemm_("N", "N", k, &nlp1, &ctot[3], &c_b13, &q[ktemp * q_dim1 + 1],
ldq, &vt2[ktemp + vt2_dim1], ldvt2, &c_b13, &vt[vt_dim1 + 1],
ldvt);
}
ktemp = ctot[1] + 1;
nrp1 = *nr + *sqre;
if (ktemp > 1) {
i__1 = *k;
for (i__ = 1; i__ <= i__1; ++i__) {
q[i__ + ktemp * q_dim1] = q[i__ + q_dim1];
/* L130: */
}
i__1 = m;
for (i__ = nlp2; i__ <= i__1; ++i__) {
vt2[ktemp + i__ * vt2_dim1] = vt2[i__ * vt2_dim1 + 1];
/* L140: */
}
}
ctemp = ctot[2] + 1 + ctot[3];
sgemm_("N", "N", k, &nrp1, &ctemp, &c_b13, &q[ktemp * q_dim1 + 1], ldq, &
vt2[ktemp + nlp2 * vt2_dim1], ldvt2, &c_b26, &vt[nlp2 * vt_dim1 +
1], ldvt);
return 0;
/* End of SLASD3 */
} /* slasd3_ */
/* Subroutine */ int cggevx_(char *balanc, char *jobvl, char *jobvr, char *
sense, integer *n, complex *a, integer *lda, complex *b, integer *ldb,
complex *alpha, complex *beta, complex *vl, integer *ldvl, complex *
vr, integer *ldvr, integer *ilo, integer *ihi, real *lscale, real *
rscale, real *abnrm, real *bbnrm, real *rconde, real *rcondv, complex
*work, integer *lwork, real *rwork, integer *iwork, logical *bwork,
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, i__3, i__4;
real r__1, r__2, r__3, r__4;
complex q__1;
/* Builtin functions */
double sqrt(doublereal), r_imag(complex *);
/* Local variables */
integer i__, j, m, jc, in, jr;
real eps;
logical ilv;
real anrm, bnrm;
integer ierr, itau;
real temp;
logical ilvl, ilvr;
integer iwrk, iwrk1;
extern logical lsame_(char *, char *);
integer icols;
logical noscl;
integer irows;
extern /* Subroutine */ int cggbak_(char *, char *, integer *, integer *,
integer *, real *, real *, integer *, complex *, integer *,
integer *), cggbal_(char *, integer *, complex *,
integer *, complex *, integer *, integer *, integer *, real *,
real *, real *, integer *), slabad_(real *, real *);
extern doublereal clange_(char *, integer *, integer *, complex *,
integer *, real *);
extern /* Subroutine */ int cgghrd_(char *, char *, integer *, integer *,
integer *, complex *, integer *, complex *, integer *, complex *,
integer *, complex *, integer *, integer *),
clascl_(char *, integer *, integer *, real *, real *, integer *,
integer *, complex *, integer *, integer *);
logical ilascl, ilbscl;
extern /* Subroutine */ int cgeqrf_(integer *, integer *, complex *,
integer *, complex *, complex *, integer *, integer *), clacpy_(
char *, integer *, integer *, complex *, integer *, complex *,
integer *), claset_(char *, integer *, integer *, complex
*, complex *, complex *, integer *), ctgevc_(char *, char
*, logical *, integer *, complex *, integer *, complex *, integer
*, complex *, integer *, complex *, integer *, integer *, integer
*, complex *, real *, integer *);
logical ldumma[1];
char chtemp[1];
real bignum;
extern /* Subroutine */ int chgeqz_(char *, char *, char *, integer *,
integer *, integer *, complex *, integer *, complex *, integer *,
complex *, complex *, complex *, integer *, complex *, integer *,
complex *, integer *, real *, integer *),
ctgsna_(char *, char *, logical *, integer *, complex *, integer *
, complex *, integer *, complex *, integer *, complex *, integer *
, real *, real *, integer *, integer *, complex *, integer *,
integer *, integer *);
integer ijobvl;
extern /* Subroutine */ int slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *), xerbla_(char *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *);
extern doublereal slamch_(char *);
integer ijobvr;
logical wantsb;
extern /* Subroutine */ int cungqr_(integer *, integer *, integer *,
complex *, integer *, complex *, complex *, integer *, integer *);
real anrmto;
logical wantse;
real bnrmto;
extern /* Subroutine */ int cunmqr_(char *, char *, integer *, integer *,
integer *, complex *, integer *, complex *, complex *, integer *,
complex *, integer *, integer *);
integer minwrk, maxwrk;
logical wantsn;
real smlnum;
logical lquery, wantsv;
/* -- LAPACK driver routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* CGGEVX computes for a pair of N-by-N complex nonsymmetric matrices */
/* (A,B) the generalized eigenvalues, and optionally, the left and/or */
/* right generalized eigenvectors. */
/* Optionally, it also computes a balancing transformation to improve */
/* the conditioning of the eigenvalues and eigenvectors (ILO, IHI, */
/* LSCALE, RSCALE, ABNRM, and BBNRM), reciprocal condition numbers for */
/* the eigenvalues (RCONDE), and reciprocal condition numbers for the */
/* right eigenvectors (RCONDV). */
/* A generalized eigenvalue for a pair of matrices (A,B) is a scalar */
/* lambda or a ratio alpha/beta = lambda, such that A - lambda*B is */
/* singular. It is usually represented as the pair (alpha,beta), as */
/* there is a reasonable interpretation for beta=0, and even for both */
/* being zero. */
/* The right eigenvector v(j) corresponding to the eigenvalue lambda(j) */
/* of (A,B) satisfies */
/* A * v(j) = lambda(j) * B * v(j) . */
/* The left eigenvector u(j) corresponding to the eigenvalue lambda(j) */
/* of (A,B) satisfies */
/* u(j)**H * A = lambda(j) * u(j)**H * B. */
/* where u(j)**H is the conjugate-transpose of u(j). */
/* Arguments */
/* ========= */
/* BALANC (input) CHARACTER*1 */
/* Specifies the balance option to be performed: */
/* = 'N': do not diagonally scale or permute; */
/* = 'P': permute only; */
/* = 'S': scale only; */
/* = 'B': both permute and scale. */
/* Computed reciprocal condition numbers will be for the */
/* matrices after permuting and/or balancing. Permuting does */
/* not change condition numbers (in exact arithmetic), but */
/* balancing does. */
/* JOBVL (input) CHARACTER*1 */
/* = 'N': do not compute the left generalized eigenvectors; */
/* = 'V': compute the left generalized eigenvectors. */
/* JOBVR (input) CHARACTER*1 */
/* = 'N': do not compute the right generalized eigenvectors; */
/* = 'V': compute the right generalized eigenvectors. */
/* SENSE (input) CHARACTER*1 */
/* Determines which reciprocal condition numbers are computed. */
/* = 'N': none are computed; */
/* = 'E': computed for eigenvalues only; */
/* = 'V': computed for eigenvectors only; */
/* = 'B': computed for eigenvalues and eigenvectors. */
/* N (input) INTEGER */
/* The order of the matrices A, B, VL, and VR. N >= 0. */
/* A (input/output) COMPLEX array, dimension (LDA, N) */
/* On entry, the matrix A in the pair (A,B). */
/* On exit, A has been overwritten. If JOBVL='V' or JOBVR='V' */
/* or both, then A contains the first part of the complex Schur */
/* form of the "balanced" versions of the input A and B. */
/* LDA (input) INTEGER */
/* The leading dimension of A. LDA >= max(1,N). */
/* B (input/output) COMPLEX array, dimension (LDB, N) */
/* On entry, the matrix B in the pair (A,B). */
/* On exit, B has been overwritten. If JOBVL='V' or JOBVR='V' */
/* or both, then B contains the second part of the complex */
/* Schur form of the "balanced" versions of the input A and B. */
/* LDB (input) INTEGER */
/* The leading dimension of B. LDB >= max(1,N). */
/* ALPHA (output) COMPLEX array, dimension (N) */
/* BETA (output) COMPLEX array, dimension (N) */
/* On exit, ALPHA(j)/BETA(j), j=1,...,N, will be the generalized */
/* eigenvalues. */
/* Note: the quotient ALPHA(j)/BETA(j) ) may easily over- or */
/* underflow, and BETA(j) may even be zero. Thus, the user */
/* should avoid naively computing the ratio ALPHA/BETA. */
/* However, ALPHA will be always less than and usually */
/* comparable with norm(A) in magnitude, and BETA always less */
/* than and usually comparable with norm(B). */
/* VL (output) COMPLEX array, dimension (LDVL,N) */
/* If JOBVL = 'V', the left generalized eigenvectors u(j) are */
/* stored one after another in the columns of VL, in the same */
/* order as their eigenvalues. */
/* Each eigenvector will be scaled so the largest component */
/* will have abs(real part) + abs(imag. part) = 1. */
/* Not referenced if JOBVL = 'N'. */
/* LDVL (input) INTEGER */
/* The leading dimension of the matrix VL. LDVL >= 1, and */
/* if JOBVL = 'V', LDVL >= N. */
/* VR (output) COMPLEX array, dimension (LDVR,N) */
/* If JOBVR = 'V', the right generalized eigenvectors v(j) are */
/* stored one after another in the columns of VR, in the same */
/* order as their eigenvalues. */
/* Each eigenvector will be scaled so the largest component */
/* will have abs(real part) + abs(imag. part) = 1. */
/* Not referenced if JOBVR = 'N'. */
/* LDVR (input) INTEGER */
/* The leading dimension of the matrix VR. LDVR >= 1, and */
/* if JOBVR = 'V', LDVR >= N. */
/* ILO (output) INTEGER */
/* IHI (output) INTEGER */
/* ILO and IHI are integer values such that on exit */
/* A(i,j) = 0 and B(i,j) = 0 if i > j and */
/* j = 1,...,ILO-1 or i = IHI+1,...,N. */
/* If BALANC = 'N' or 'S', ILO = 1 and IHI = N. */
/* LSCALE (output) REAL array, dimension (N) */
/* Details of the permutations and scaling factors applied */
/* to the left side of A and B. If PL(j) is the index of the */
/* row interchanged with row j, and DL(j) is the scaling */
/* factor applied to row j, then */
/* LSCALE(j) = PL(j) for j = 1,...,ILO-1 */
/* = DL(j) for j = ILO,...,IHI */
/* = PL(j) for j = IHI+1,...,N. */
/* The order in which the interchanges are made is N to IHI+1, */
/* then 1 to ILO-1. */
/* RSCALE (output) REAL array, dimension (N) */
/* Details of the permutations and scaling factors applied */
/* to the right side of A and B. If PR(j) is the index of the */
/* column interchanged with column j, and DR(j) is the scaling */
/* factor applied to column j, then */
/* RSCALE(j) = PR(j) for j = 1,...,ILO-1 */
/* = DR(j) for j = ILO,...,IHI */
/* = PR(j) for j = IHI+1,...,N */
/* The order in which the interchanges are made is N to IHI+1, */
/* then 1 to ILO-1. */
/* ABNRM (output) REAL */
/* The one-norm of the balanced matrix A. */
/* BBNRM (output) REAL */
/* The one-norm of the balanced matrix B. */
/* RCONDE (output) REAL array, dimension (N) */
/* If SENSE = 'E' or 'B', the reciprocal condition numbers of */
/* the eigenvalues, stored in consecutive elements of the array. */
/* If SENSE = 'N' or 'V', RCONDE is not referenced. */
/* RCONDV (output) REAL array, dimension (N) */
/* If SENSE = 'V' or 'B', the estimated reciprocal condition */
/* numbers of the eigenvectors, stored in consecutive elements */
/* of the array. If the eigenvalues cannot be reordered to */
/* compute RCONDV(j), RCONDV(j) is set to 0; this can only occur */
/* when the true value would be very small anyway. */
/* If SENSE = 'N' or 'E', RCONDV is not referenced. */
/* WORK (workspace/output) COMPLEX 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(1,2*N). */
/* If SENSE = 'E', LWORK >= max(1,4*N). */
/* If SENSE = 'V' or 'B', LWORK >= max(1,2*N*N+2*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. */
/* RWORK (workspace) REAL array, dimension (lrwork) */
/* lrwork must be at least max(1,6*N) if BALANC = 'S' or 'B', */
/* and at least max(1,2*N) otherwise. */
/* Real workspace. */
/* IWORK (workspace) INTEGER array, dimension (N+2) */
/* If SENSE = 'E', IWORK is not referenced. */
/* BWORK (workspace) LOGICAL array, dimension (N) */
/* If SENSE = 'N', BWORK is not referenced. */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value. */
/* = 1,...,N: */
/* The QZ iteration failed. No eigenvectors have been */
/* calculated, but ALPHA(j) and BETA(j) should be correct */
/* for j=INFO+1,...,N. */
/* > N: =N+1: other than QZ iteration failed in CHGEQZ. */
/* =N+2: error return from CTGEVC. */
/* Further Details */
/* =============== */
/* Balancing a matrix pair (A,B) includes, first, permuting rows and */
/* columns to isolate eigenvalues, second, applying diagonal similarity */
/* transformation to the rows and columns to make the rows and columns */
/* as close in norm as possible. The computed reciprocal condition */
/* numbers correspond to the balanced matrix. Permuting rows and columns */
/* will not change the condition numbers (in exact arithmetic) but */
/* diagonal scaling will. For further explanation of balancing, see */
/* section 4.11.1.2 of LAPACK Users' Guide. */
/* 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(ABNRM, BBNRM) / RCONDE(I) */
/* An approximate error bound for the angle between the i-th computed */
/* eigenvector VL(i) or VR(i) is given by */
/* EPS * norm(ABNRM, BBNRM) / DIF(i). */
/* For further explanation of the reciprocal condition numbers RCONDE */
/* and RCONDV, see section 4.11 of LAPACK User's Guide. */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. Local Arrays .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Statement Functions .. */
/* .. */
/* .. Statement Function definitions .. */
/* .. */
/* .. Executable Statements .. */
/* Decode the input arguments */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
--alpha;
--beta;
vl_dim1 = *ldvl;
vl_offset = 1 + vl_dim1;
vl -= vl_offset;
vr_dim1 = *ldvr;
vr_offset = 1 + vr_dim1;
vr -= vr_offset;
--lscale;
--rscale;
--rconde;
--rcondv;
--work;
--rwork;
--iwork;
--bwork;
/* Function Body */
if (lsame_(jobvl, "N")) {
ijobvl = 1;
ilvl = FALSE_;
} else if (lsame_(jobvl, "V")) {
ijobvl = 2;
ilvl = TRUE_;
} else {
ijobvl = -1;
ilvl = FALSE_;
}
if (lsame_(jobvr, "N")) {
ijobvr = 1;
ilvr = FALSE_;
} else if (lsame_(jobvr, "V")) {
ijobvr = 2;
ilvr = TRUE_;
} else {
ijobvr = -1;
ilvr = FALSE_;
}
ilv = ilvl || ilvr;
noscl = lsame_(balanc, "N") || lsame_(balanc, "P");
wantsn = lsame_(sense, "N");
wantse = lsame_(sense, "E");
wantsv = lsame_(sense, "V");
wantsb = lsame_(sense, "B");
/* Test the input arguments */
*info = 0;
lquery = *lwork == -1;
if (! (noscl || lsame_(balanc, "S") || lsame_(
balanc, "B"))) {
*info = -1;
} else if (ijobvl <= 0) {
*info = -2;
} else if (ijobvr <= 0) {
*info = -3;
} else if (! (wantsn || wantse || wantsb || wantsv)) {
*info = -4;
} else if (*n < 0) {
*info = -5;
} else if (*lda < max(1,*n)) {
*info = -7;
} else if (*ldb < max(1,*n)) {
*info = -9;
} else if (*ldvl < 1 || ilvl && *ldvl < *n) {
*info = -13;
} else if (*ldvr < 1 || ilvr && *ldvr < *n) {
*info = -15;
}
/* 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. The workspace is */
/* computed assuming ILO = 1 and IHI = N, the worst case.) */
if (*info == 0) {
if (*n == 0) {
minwrk = 1;
maxwrk = 1;
} else {
minwrk = *n << 1;
if (wantse) {
minwrk = *n << 2;
} else if (wantsv || wantsb) {
minwrk = (*n << 1) * (*n + 1);
}
maxwrk = minwrk;
/* Computing MAX */
i__1 = maxwrk, i__2 = *n + *n * ilaenv_(&c__1, "CGEQRF", " ", n, &
c__1, n, &c__0);
maxwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk, i__2 = *n + *n * ilaenv_(&c__1, "CUNMQR", " ", n, &
c__1, n, &c__0);
maxwrk = max(i__1,i__2);
if (ilvl) {
/* Computing MAX */
i__1 = maxwrk, i__2 = *n + *n * ilaenv_(&c__1, "CUNGQR",
" ", n, &c__1, n, &c__0);
maxwrk = max(i__1,i__2);
}
}
work[1].r = (real) maxwrk, work[1].i = 0.f;
if (*lwork < minwrk && ! lquery) {
*info = -25;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("CGGEVX", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
/* Get machine constants */
eps = slamch_("P");
smlnum = slamch_("S");
bignum = 1.f / smlnum;
slabad_(&smlnum, &bignum);
smlnum = sqrt(smlnum) / eps;
bignum = 1.f / smlnum;
/* Scale A if max element outside range [SMLNUM,BIGNUM] */
anrm = clange_("M", n, n, &a[a_offset], lda, &rwork[1]);
ilascl = FALSE_;
if (anrm > 0.f && anrm < smlnum) {
anrmto = smlnum;
ilascl = TRUE_;
} else if (anrm > bignum) {
anrmto = bignum;
ilascl = TRUE_;
}
if (ilascl) {
clascl_("G", &c__0, &c__0, &anrm, &anrmto, n, n, &a[a_offset], lda, &
ierr);
}
/* Scale B if max element outside range [SMLNUM,BIGNUM] */
bnrm = clange_("M", n, n, &b[b_offset], ldb, &rwork[1]);
ilbscl = FALSE_;
if (bnrm > 0.f && bnrm < smlnum) {
bnrmto = smlnum;
ilbscl = TRUE_;
} else if (bnrm > bignum) {
bnrmto = bignum;
ilbscl = TRUE_;
}
if (ilbscl) {
clascl_("G", &c__0, &c__0, &bnrm, &bnrmto, n, n, &b[b_offset], ldb, &
ierr);
}
/* Permute and/or balance the matrix pair (A,B) */
/* (Real Workspace: need 6*N if BALANC = 'S' or 'B', 1 otherwise) */
cggbal_(balanc, n, &a[a_offset], lda, &b[b_offset], ldb, ilo, ihi, &
lscale[1], &rscale[1], &rwork[1], &ierr);
/* Compute ABNRM and BBNRM */
*abnrm = clange_("1", n, n, &a[a_offset], lda, &rwork[1]);
if (ilascl) {
rwork[1] = *abnrm;
slascl_("G", &c__0, &c__0, &anrmto, &anrm, &c__1, &c__1, &rwork[1], &
c__1, &ierr);
*abnrm = rwork[1];
}
*bbnrm = clange_("1", n, n, &b[b_offset], ldb, &rwork[1]);
if (ilbscl) {
rwork[1] = *bbnrm;
slascl_("G", &c__0, &c__0, &bnrmto, &bnrm, &c__1, &c__1, &rwork[1], &
c__1, &ierr);
*bbnrm = rwork[1];
}
/* Reduce B to triangular form (QR decomposition of B) */
/* (Complex Workspace: need N, prefer N*NB ) */
irows = *ihi + 1 - *ilo;
if (ilv || ! wantsn) {
icols = *n + 1 - *ilo;
} else {
icols = irows;
}
itau = 1;
iwrk = itau + irows;
i__1 = *lwork + 1 - iwrk;
cgeqrf_(&irows, &icols, &b[*ilo + *ilo * b_dim1], ldb, &work[itau], &work[
iwrk], &i__1, &ierr);
/* Apply the unitary transformation to A */
/* (Complex Workspace: need N, prefer N*NB) */
i__1 = *lwork + 1 - iwrk;
cunmqr_("L", "C", &irows, &icols, &irows, &b[*ilo + *ilo * b_dim1], ldb, &
work[itau], &a[*ilo + *ilo * a_dim1], lda, &work[iwrk], &i__1, &
ierr);
/* Initialize VL and/or VR */
/* (Workspace: need N, prefer N*NB) */
if (ilvl) {
claset_("Full", n, n, &c_b1, &c_b2, &vl[vl_offset], ldvl);
if (irows > 1) {
i__1 = irows - 1;
i__2 = irows - 1;
clacpy_("L", &i__1, &i__2, &b[*ilo + 1 + *ilo * b_dim1], ldb, &vl[
*ilo + 1 + *ilo * vl_dim1], ldvl);
}
i__1 = *lwork + 1 - iwrk;
cungqr_(&irows, &irows, &irows, &vl[*ilo + *ilo * vl_dim1], ldvl, &
work[itau], &work[iwrk], &i__1, &ierr);
}
if (ilvr) {
claset_("Full", n, n, &c_b1, &c_b2, &vr[vr_offset], ldvr);
}
/* Reduce to generalized Hessenberg form */
/* (Workspace: none needed) */
if (ilv || ! wantsn) {
/* Eigenvectors requested -- work on whole matrix. */
cgghrd_(jobvl, jobvr, n, ilo, ihi, &a[a_offset], lda, &b[b_offset],
ldb, &vl[vl_offset], ldvl, &vr[vr_offset], ldvr, &ierr);
} else {
cgghrd_("N", "N", &irows, &c__1, &irows, &a[*ilo + *ilo * a_dim1],
lda, &b[*ilo + *ilo * b_dim1], ldb, &vl[vl_offset], ldvl, &vr[
vr_offset], ldvr, &ierr);
}
/* Perform QZ algorithm (Compute eigenvalues, and optionally, the */
/* Schur forms and Schur vectors) */
/* (Complex Workspace: need N) */
/* (Real Workspace: need N) */
iwrk = itau;
if (ilv || ! wantsn) {
*(unsigned char *)chtemp = 'S';
} else {
*(unsigned char *)chtemp = 'E';
}
i__1 = *lwork + 1 - iwrk;
chgeqz_(chtemp, jobvl, jobvr, n, ilo, ihi, &a[a_offset], lda, &b[b_offset]
, ldb, &alpha[1], &beta[1], &vl[vl_offset], ldvl, &vr[vr_offset],
ldvr, &work[iwrk], &i__1, &rwork[1], &ierr);
if (ierr != 0) {
if (ierr > 0 && ierr <= *n) {
*info = ierr;
} else if (ierr > *n && ierr <= *n << 1) {
*info = ierr - *n;
} else {
*info = *n + 1;
}
goto L90;
}
/* Compute Eigenvectors and estimate condition numbers if desired */
/* CTGEVC: (Complex Workspace: need 2*N ) */
/* (Real Workspace: need 2*N ) */
/* CTGSNA: (Complex Workspace: need 2*N*N if SENSE='V' or 'B') */
/* (Integer Workspace: need N+2 ) */
if (ilv || ! wantsn) {
if (ilv) {
if (ilvl) {
if (ilvr) {
*(unsigned char *)chtemp = 'B';
} else {
*(unsigned char *)chtemp = 'L';
}
} else {
*(unsigned char *)chtemp = 'R';
}
ctgevc_(chtemp, "B", ldumma, n, &a[a_offset], lda, &b[b_offset],
ldb, &vl[vl_offset], ldvl, &vr[vr_offset], ldvr, n, &in, &
work[iwrk], &rwork[1], &ierr);
if (ierr != 0) {
*info = *n + 2;
goto L90;
}
}
if (! wantsn) {
/* compute eigenvectors (STGEVC) and estimate condition */
/* numbers (STGSNA). Note that the definition of the condition */
/* number is not invariant under transformation (u,v) to */
/* (Q*u, Z*v), where (u,v) are eigenvectors of the generalized */
/* Schur form (S,T), Q and Z are orthogonal matrices. In order */
/* to avoid using extra 2*N*N workspace, we have to */
/* re-calculate eigenvectors and estimate the condition numbers */
/* one at a time. */
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = *n;
for (j = 1; j <= i__2; ++j) {
bwork[j] = FALSE_;
/* L10: */
}
bwork[i__] = TRUE_;
iwrk = *n + 1;
iwrk1 = iwrk + *n;
if (wantse || wantsb) {
ctgevc_("B", "S", &bwork[1], n, &a[a_offset], lda, &b[
b_offset], ldb, &work[1], n, &work[iwrk], n, &
c__1, &m, &work[iwrk1], &rwork[1], &ierr);
if (ierr != 0) {
*info = *n + 2;
goto L90;
}
}
i__2 = *lwork - iwrk1 + 1;
ctgsna_(sense, "S", &bwork[1], n, &a[a_offset], lda, &b[
b_offset], ldb, &work[1], n, &work[iwrk], n, &rconde[
i__], &rcondv[i__], &c__1, &m, &work[iwrk1], &i__2, &
iwork[1], &ierr);
/* L20: */
}
}
}
/* Undo balancing on VL and VR and normalization */
/* (Workspace: none needed) */
if (ilvl) {
cggbak_(balanc, "L", n, ilo, ihi, &lscale[1], &rscale[1], n, &vl[
vl_offset], ldvl, &ierr);
i__1 = *n;
for (jc = 1; jc <= i__1; ++jc) {
temp = 0.f;
i__2 = *n;
for (jr = 1; jr <= i__2; ++jr) {
/* Computing MAX */
i__3 = jr + jc * vl_dim1;
r__3 = temp, r__4 = (r__1 = vl[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&vl[jr + jc * vl_dim1]), dabs(r__2));
temp = dmax(r__3,r__4);
/* L30: */
}
if (temp < smlnum) {
goto L50;
}
temp = 1.f / temp;
i__2 = *n;
for (jr = 1; jr <= i__2; ++jr) {
i__3 = jr + jc * vl_dim1;
i__4 = jr + jc * vl_dim1;
q__1.r = temp * vl[i__4].r, q__1.i = temp * vl[i__4].i;
vl[i__3].r = q__1.r, vl[i__3].i = q__1.i;
/* L40: */
}
L50:
;
}
}
if (ilvr) {
cggbak_(balanc, "R", n, ilo, ihi, &lscale[1], &rscale[1], n, &vr[
vr_offset], ldvr, &ierr);
i__1 = *n;
for (jc = 1; jc <= i__1; ++jc) {
temp = 0.f;
i__2 = *n;
for (jr = 1; jr <= i__2; ++jr) {
/* Computing MAX */
i__3 = jr + jc * vr_dim1;
r__3 = temp, r__4 = (r__1 = vr[i__3].r, dabs(r__1)) + (r__2 =
r_imag(&vr[jr + jc * vr_dim1]), dabs(r__2));
temp = dmax(r__3,r__4);
/* L60: */
}
if (temp < smlnum) {
goto L80;
}
temp = 1.f / temp;
i__2 = *n;
for (jr = 1; jr <= i__2; ++jr) {
i__3 = jr + jc * vr_dim1;
i__4 = jr + jc * vr_dim1;
q__1.r = temp * vr[i__4].r, q__1.i = temp * vr[i__4].i;
vr[i__3].r = q__1.r, vr[i__3].i = q__1.i;
/* L70: */
}
L80:
;
}
}
/* Undo scaling if necessary */
if (ilascl) {
clascl_("G", &c__0, &c__0, &anrmto, &anrm, n, &c__1, &alpha[1], n, &
ierr);
}
if (ilbscl) {
clascl_("G", &c__0, &c__0, &bnrmto, &bnrm, n, &c__1, &beta[1], n, &
ierr);
}
L90:
work[1].r = (real) maxwrk, work[1].i = 0.f;
return 0;
/* End of CGGEVX */
} /* cggevx_ */
/* Subroutine */
int cgelsd_(integer *m, integer *n, integer *nrhs, complex * a, integer *lda, complex *b, integer *ldb, real *s, real *rcond, integer *rank, complex *work, integer *lwork, real *rwork, integer * iwork, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, i__1, i__2, i__3, i__4;
/* Builtin functions */
double log(doublereal);
/* Local variables */
integer ie, il, mm;
real eps, anrm, bnrm;
integer itau, nlvl, iascl, ibscl;
real sfmin;
integer minmn, maxmn, itaup, itauq, mnthr, nwork;
extern /* Subroutine */
int cgebrd_(integer *, integer *, complex *, integer *, real *, real *, complex *, complex *, complex *, integer *, integer *), slabad_(real *, real *);
extern real clange_(char *, integer *, integer *, complex *, integer *, real *);
extern /* Subroutine */
int cgelqf_(integer *, integer *, complex *, integer *, complex *, complex *, integer *, integer *), clalsd_( char *, integer *, integer *, integer *, real *, real *, complex * , integer *, real *, integer *, complex *, real *, integer *, integer *), clascl_(char *, integer *, integer *, real *, real *, integer *, integer *, complex *, integer *, integer *), cgeqrf_(integer *, integer *, complex *, integer *, complex *, complex *, integer *, integer *);
extern real slamch_(char *);
extern /* Subroutine */
int clacpy_(char *, integer *, integer *, complex *, integer *, complex *, integer *), claset_(char *, integer *, integer *, complex *, complex *, complex *, integer *), xerbla_(char *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *, integer *, integer *);
real bignum;
extern /* Subroutine */
int slascl_(char *, integer *, integer *, real *, real *, integer *, integer *, real *, integer *, integer *), cunmbr_(char *, char *, char *, integer *, integer *, integer *, complex *, integer *, complex *, complex *, integer *, complex *, integer *, integer *), slaset_( char *, integer *, integer *, real *, real *, real *, integer *), cunmlq_(char *, char *, integer *, integer *, integer *, complex *, integer *, complex *, complex *, integer *, complex *, integer *, integer *);
integer ldwork;
extern /* Subroutine */
int cunmqr_(char *, char *, integer *, integer *, integer *, complex *, integer *, complex *, complex *, integer *, complex *, integer *, integer *);
integer liwork, minwrk, maxwrk;
real smlnum;
integer lrwork;
logical lquery;
integer nrwork, smlsiz;
/* -- LAPACK driver 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 Subroutines .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input arguments. */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
--s;
--work;
--rwork;
--iwork;
/* Function Body */
*info = 0;
minmn = min(*m,*n);
maxmn = max(*m,*n);
lquery = *lwork == -1;
if (*m < 0)
{
*info = -1;
}
else if (*n < 0)
{
*info = -2;
}
else if (*nrhs < 0)
{
*info = -3;
}
else if (*lda < max(1,*m))
{
*info = -5;
}
else if (*ldb < max(1,maxmn))
{
*info = -7;
}
/* 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.) */
if (*info == 0)
{
minwrk = 1;
maxwrk = 1;
liwork = 1;
lrwork = 1;
if (minmn > 0)
{
smlsiz = ilaenv_(&c__9, "CGELSD", " ", &c__0, &c__0, &c__0, &c__0);
mnthr = ilaenv_(&c__6, "CGELSD", " ", m, n, nrhs, &c_n1);
/* Computing MAX */
i__1 = (integer) (log((real) minmn / (real) (smlsiz + 1)) / log( 2.f)) + 1;
nlvl = max(i__1,0);
liwork = minmn * 3 * nlvl + minmn * 11;
mm = *m;
if (*m >= *n && *m >= mnthr)
{
/* Path 1a - overdetermined, with many more rows than */
/* columns. */
mm = *n;
/* Computing MAX */
i__1 = maxwrk;
i__2 = *n * ilaenv_(&c__1, "CGEQRF", " ", m, n, &c_n1, &c_n1); // , expr subst
maxwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk;
i__2 = *nrhs * ilaenv_(&c__1, "CUNMQR", "LC", m, nrhs, n, &c_n1); // , expr subst
maxwrk = max(i__1,i__2);
}
if (*m >= *n)
{
/* Path 1 - overdetermined or exactly determined. */
/* Computing MAX */
/* Computing 2nd power */
i__3 = smlsiz + 1;
i__1 = i__3 * i__3;
i__2 = *n * (*nrhs + 1) + (*nrhs << 1); // , expr subst
lrwork = *n * 10 + (*n << 1) * smlsiz + (*n << 3) * nlvl + smlsiz * 3 * *nrhs + max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk;
i__2 = (*n << 1) + (mm + *n) * ilaenv_(&c__1, "CGEBRD", " ", &mm, n, &c_n1, &c_n1); // , expr subst
maxwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk;
i__2 = (*n << 1) + *nrhs * ilaenv_(&c__1, "CUNMBR", "QLC", &mm, nrhs, n, &c_n1); // , expr subst
maxwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk;
i__2 = (*n << 1) + (*n - 1) * ilaenv_(&c__1, "CUNMBR", "PLN", n, nrhs, n, &c_n1); // , expr subst
maxwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk;
i__2 = (*n << 1) + *n * *nrhs; // , expr subst
maxwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = (*n << 1) + mm;
i__2 = (*n << 1) + *n * *nrhs; // , expr subst
minwrk = max(i__1,i__2);
}
if (*n > *m)
{
/* Computing MAX */
/* Computing 2nd power */
i__3 = smlsiz + 1;
i__1 = i__3 * i__3;
i__2 = *n * (*nrhs + 1) + (*nrhs << 1); // , expr subst
lrwork = *m * 10 + (*m << 1) * smlsiz + (*m << 3) * nlvl + smlsiz * 3 * *nrhs + max(i__1,i__2);
if (*n >= mnthr)
{
/* Path 2a - underdetermined, with many more columns */
/* than rows. */
maxwrk = *m + *m * ilaenv_(&c__1, "CGELQF", " ", m, n, & c_n1, &c_n1);
/* Computing MAX */
i__1 = maxwrk;
i__2 = *m * *m + (*m << 2) + (*m << 1) * ilaenv_(&c__1, "CGEBRD", " ", m, m, &c_n1, &c_n1); // , expr subst
maxwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk;
i__2 = *m * *m + (*m << 2) + *nrhs * ilaenv_(&c__1, "CUNMBR", "QLC", m, nrhs, m, &c_n1); // , expr subst
maxwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk;
i__2 = *m * *m + (*m << 2) + (*m - 1) * ilaenv_(&c__1, "CUNMLQ", "LC", n, nrhs, m, &c_n1); // , expr subst
maxwrk = max(i__1,i__2);
if (*nrhs > 1)
{
/* Computing MAX */
i__1 = maxwrk;
i__2 = *m * *m + *m + *m * *nrhs; // , expr subst
maxwrk = max(i__1,i__2);
}
else
{
/* Computing MAX */
i__1 = maxwrk;
i__2 = *m * *m + (*m << 1); // , expr subst
maxwrk = max(i__1,i__2);
}
/* Computing MAX */
i__1 = maxwrk;
i__2 = *m * *m + (*m << 2) + *m * *nrhs; // , expr subst
maxwrk = max(i__1,i__2);
/* XXX: Ensure the Path 2a case below is triggered. The workspace */
/* calculation should use queries for all routines eventually. */
/* Computing MAX */
/* Computing MAX */
i__3 = *m, i__4 = (*m << 1) - 4, i__3 = max(i__3,i__4);
i__3 = max(i__3,*nrhs);
i__4 = *n - *m * 3; // ; expr subst
i__1 = maxwrk;
i__2 = (*m << 2) + *m * *m + max(i__3,i__4) ; // , expr subst
maxwrk = max(i__1,i__2);
}
else
{
/* Path 2 - underdetermined. */
maxwrk = (*m << 1) + (*n + *m) * ilaenv_(&c__1, "CGEBRD", " ", m, n, &c_n1, &c_n1);
/* Computing MAX */
i__1 = maxwrk;
i__2 = (*m << 1) + *nrhs * ilaenv_(&c__1, "CUNMBR", "QLC", m, nrhs, m, &c_n1); // , expr subst
maxwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk;
i__2 = (*m << 1) + *m * ilaenv_(&c__1, "CUNMBR", "PLN", n, nrhs, m, &c_n1); // , expr subst
maxwrk = max(i__1,i__2);
/* Computing MAX */
i__1 = maxwrk;
i__2 = (*m << 1) + *m * *nrhs; // , expr subst
maxwrk = max(i__1,i__2);
}
/* Computing MAX */
i__1 = (*m << 1) + *n;
i__2 = (*m << 1) + *m * *nrhs; // , expr subst
minwrk = max(i__1,i__2);
}
}
minwrk = min(minwrk,maxwrk);
work[1].r = (real) maxwrk;
work[1].i = 0.f; // , expr subst
iwork[1] = liwork;
rwork[1] = (real) lrwork;
if (*lwork < minwrk && ! lquery)
{
*info = -12;
}
}
if (*info != 0)
{
i__1 = -(*info);
xerbla_("CGELSD", &i__1);
return 0;
}
else if (lquery)
{
return 0;
}
/* Quick return if possible. */
if (*m == 0 || *n == 0)
{
*rank = 0;
return 0;
}
/* Get machine parameters. */
eps = slamch_("P");
sfmin = slamch_("S");
smlnum = sfmin / eps;
bignum = 1.f / smlnum;
slabad_(&smlnum, &bignum);
/* Scale A if max entry outside range [SMLNUM,BIGNUM]. */
anrm = clange_("M", m, n, &a[a_offset], lda, &rwork[1]);
iascl = 0;
if (anrm > 0.f && anrm < smlnum)
{
/* Scale matrix norm up to SMLNUM */
clascl_("G", &c__0, &c__0, &anrm, &smlnum, m, n, &a[a_offset], lda, info);
iascl = 1;
}
else if (anrm > bignum)
{
/* Scale matrix norm down to BIGNUM. */
clascl_("G", &c__0, &c__0, &anrm, &bignum, m, n, &a[a_offset], lda, info);
iascl = 2;
}
else if (anrm == 0.f)
{
/* Matrix all zero. Return zero solution. */
i__1 = max(*m,*n);
claset_("F", &i__1, nrhs, &c_b1, &c_b1, &b[b_offset], ldb);
slaset_("F", &minmn, &c__1, &c_b80, &c_b80, &s[1], &c__1);
*rank = 0;
goto L10;
}
/* Scale B if max entry outside range [SMLNUM,BIGNUM]. */
bnrm = clange_("M", m, nrhs, &b[b_offset], ldb, &rwork[1]);
ibscl = 0;
if (bnrm > 0.f && bnrm < smlnum)
{
/* Scale matrix norm up to SMLNUM. */
clascl_("G", &c__0, &c__0, &bnrm, &smlnum, m, nrhs, &b[b_offset], ldb, info);
ibscl = 1;
}
else if (bnrm > bignum)
{
/* Scale matrix norm down to BIGNUM. */
clascl_("G", &c__0, &c__0, &bnrm, &bignum, m, nrhs, &b[b_offset], ldb, info);
ibscl = 2;
}
/* If M < N make sure B(M+1:N,:) = 0 */
if (*m < *n)
{
i__1 = *n - *m;
claset_("F", &i__1, nrhs, &c_b1, &c_b1, &b[*m + 1 + b_dim1], ldb);
}
/* Overdetermined case. */
if (*m >= *n)
{
/* Path 1 - overdetermined or exactly determined. */
mm = *m;
if (*m >= mnthr)
{
/* Path 1a - overdetermined, with many more rows than columns */
mm = *n;
itau = 1;
nwork = itau + *n;
/* Compute A=Q*R. */
/* (RWorkspace: need N) */
/* (CWorkspace: need N, prefer N*NB) */
i__1 = *lwork - nwork + 1;
cgeqrf_(m, n, &a[a_offset], lda, &work[itau], &work[nwork], &i__1, info);
/* Multiply B by transpose(Q). */
/* (RWorkspace: need N) */
/* (CWorkspace: need NRHS, prefer NRHS*NB) */
i__1 = *lwork - nwork + 1;
cunmqr_("L", "C", m, nrhs, n, &a[a_offset], lda, &work[itau], &b[ b_offset], ldb, &work[nwork], &i__1, info);
/* Zero out below R. */
if (*n > 1)
{
i__1 = *n - 1;
i__2 = *n - 1;
claset_("L", &i__1, &i__2, &c_b1, &c_b1, &a[a_dim1 + 2], lda);
}
}
itauq = 1;
itaup = itauq + *n;
nwork = itaup + *n;
ie = 1;
nrwork = ie + *n;
/* Bidiagonalize R in A. */
/* (RWorkspace: need N) */
/* (CWorkspace: need 2*N+MM, prefer 2*N+(MM+N)*NB) */
i__1 = *lwork - nwork + 1;
cgebrd_(&mm, n, &a[a_offset], lda, &s[1], &rwork[ie], &work[itauq], & work[itaup], &work[nwork], &i__1, info);
/* Multiply B by transpose of left bidiagonalizing vectors of R. */
/* (CWorkspace: need 2*N+NRHS, prefer 2*N+NRHS*NB) */
i__1 = *lwork - nwork + 1;
cunmbr_("Q", "L", "C", &mm, nrhs, n, &a[a_offset], lda, &work[itauq], &b[b_offset], ldb, &work[nwork], &i__1, info);
/* Solve the bidiagonal least squares problem. */
clalsd_("U", &smlsiz, n, nrhs, &s[1], &rwork[ie], &b[b_offset], ldb, rcond, rank, &work[nwork], &rwork[nrwork], &iwork[1], info);
if (*info != 0)
{
goto L10;
}
/* Multiply B by right bidiagonalizing vectors of R. */
i__1 = *lwork - nwork + 1;
cunmbr_("P", "L", "N", n, nrhs, n, &a[a_offset], lda, &work[itaup], & b[b_offset], ldb, &work[nwork], &i__1, info);
}
else /* if(complicated condition) */
{
/* Computing MAX */
i__1 = *m, i__2 = (*m << 1) - 4, i__1 = max(i__1,i__2);
i__1 = max( i__1,*nrhs);
i__2 = *n - *m * 3; // ; expr subst
if (*n >= mnthr && *lwork >= (*m << 2) + *m * *m + max(i__1,i__2))
{
/* Path 2a - underdetermined, with many more columns than rows */
/* and sufficient workspace for an efficient algorithm. */
ldwork = *m;
/* Computing MAX */
/* Computing MAX */
i__3 = *m, i__4 = (*m << 1) - 4, i__3 = max(i__3,i__4);
i__3 = max(i__3,*nrhs);
i__4 = *n - *m * 3; // ; expr subst
i__1 = (*m << 2) + *m * *lda + max(i__3,i__4);
i__2 = *m * *lda + *m + *m * *nrhs; // , expr subst
if (*lwork >= max(i__1,i__2))
{
ldwork = *lda;
}
itau = 1;
nwork = *m + 1;
/* Compute A=L*Q. */
/* (CWorkspace: need 2*M, prefer M+M*NB) */
i__1 = *lwork - nwork + 1;
cgelqf_(m, n, &a[a_offset], lda, &work[itau], &work[nwork], &i__1, info);
il = nwork;
/* Copy L to WORK(IL), zeroing out above its diagonal. */
clacpy_("L", m, m, &a[a_offset], lda, &work[il], &ldwork);
i__1 = *m - 1;
i__2 = *m - 1;
claset_("U", &i__1, &i__2, &c_b1, &c_b1, &work[il + ldwork], & ldwork);
itauq = il + ldwork * *m;
itaup = itauq + *m;
nwork = itaup + *m;
ie = 1;
nrwork = ie + *m;
/* Bidiagonalize L in WORK(IL). */
/* (RWorkspace: need M) */
/* (CWorkspace: need M*M+4*M, prefer M*M+4*M+2*M*NB) */
i__1 = *lwork - nwork + 1;
cgebrd_(m, m, &work[il], &ldwork, &s[1], &rwork[ie], &work[itauq], &work[itaup], &work[nwork], &i__1, info);
/* Multiply B by transpose of left bidiagonalizing vectors of L. */
/* (CWorkspace: need M*M+4*M+NRHS, prefer M*M+4*M+NRHS*NB) */
i__1 = *lwork - nwork + 1;
cunmbr_("Q", "L", "C", m, nrhs, m, &work[il], &ldwork, &work[ itauq], &b[b_offset], ldb, &work[nwork], &i__1, info);
/* Solve the bidiagonal least squares problem. */
clalsd_("U", &smlsiz, m, nrhs, &s[1], &rwork[ie], &b[b_offset], ldb, rcond, rank, &work[nwork], &rwork[nrwork], &iwork[1], info);
if (*info != 0)
{
goto L10;
}
/* Multiply B by right bidiagonalizing vectors of L. */
i__1 = *lwork - nwork + 1;
cunmbr_("P", "L", "N", m, nrhs, m, &work[il], &ldwork, &work[ itaup], &b[b_offset], ldb, &work[nwork], &i__1, info);
/* Zero out below first M rows of B. */
i__1 = *n - *m;
claset_("F", &i__1, nrhs, &c_b1, &c_b1, &b[*m + 1 + b_dim1], ldb);
nwork = itau + *m;
/* Multiply transpose(Q) by B. */
/* (CWorkspace: need NRHS, prefer NRHS*NB) */
i__1 = *lwork - nwork + 1;
cunmlq_("L", "C", n, nrhs, m, &a[a_offset], lda, &work[itau], &b[ b_offset], ldb, &work[nwork], &i__1, info);
}
else
{
/* Path 2 - remaining underdetermined cases. */
itauq = 1;
itaup = itauq + *m;
nwork = itaup + *m;
ie = 1;
nrwork = ie + *m;
/* Bidiagonalize A. */
/* (RWorkspace: need M) */
/* (CWorkspace: need 2*M+N, prefer 2*M+(M+N)*NB) */
i__1 = *lwork - nwork + 1;
cgebrd_(m, n, &a[a_offset], lda, &s[1], &rwork[ie], &work[itauq], &work[itaup], &work[nwork], &i__1, info);
/* Multiply B by transpose of left bidiagonalizing vectors. */
/* (CWorkspace: need 2*M+NRHS, prefer 2*M+NRHS*NB) */
i__1 = *lwork - nwork + 1;
cunmbr_("Q", "L", "C", m, nrhs, n, &a[a_offset], lda, &work[itauq] , &b[b_offset], ldb, &work[nwork], &i__1, info);
/* Solve the bidiagonal least squares problem. */
clalsd_("L", &smlsiz, m, nrhs, &s[1], &rwork[ie], &b[b_offset], ldb, rcond, rank, &work[nwork], &rwork[nrwork], &iwork[1], info);
if (*info != 0)
{
goto L10;
}
/* Multiply B by right bidiagonalizing vectors of A. */
i__1 = *lwork - nwork + 1;
cunmbr_("P", "L", "N", n, nrhs, m, &a[a_offset], lda, &work[itaup] , &b[b_offset], ldb, &work[nwork], &i__1, info);
}
}
/* Undo scaling. */
if (iascl == 1)
{
clascl_("G", &c__0, &c__0, &anrm, &smlnum, n, nrhs, &b[b_offset], ldb, info);
slascl_("G", &c__0, &c__0, &smlnum, &anrm, &minmn, &c__1, &s[1], & minmn, info);
}
else if (iascl == 2)
{
clascl_("G", &c__0, &c__0, &anrm, &bignum, n, nrhs, &b[b_offset], ldb, info);
slascl_("G", &c__0, &c__0, &bignum, &anrm, &minmn, &c__1, &s[1], & minmn, info);
}
if (ibscl == 1)
{
clascl_("G", &c__0, &c__0, &smlnum, &bnrm, n, nrhs, &b[b_offset], ldb, info);
}
else if (ibscl == 2)
{
clascl_("G", &c__0, &c__0, &bignum, &bnrm, n, nrhs, &b[b_offset], ldb, info);
}
L10:
work[1].r = (real) maxwrk;
work[1].i = 0.f; // , expr subst
iwork[1] = liwork;
rwork[1] = (real) lrwork;
return 0;
/* End of CGELSD */
}
/* Subroutine */ int sgels_(char *trans, integer *m, integer *n, integer *
nrhs, real *a, integer *lda, real *b, integer *ldb, real *work,
integer *lwork, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, i__1, i__2;
/* Local variables */
integer i__, j, nb, mn;
real anrm, bnrm;
integer brow;
logical tpsd;
integer iascl, ibscl;
extern logical lsame_(char *, char *);
integer wsize;
real rwork[1];
extern /* Subroutine */ int slabad_(real *, real *);
extern doublereal slamch_(char *), slange_(char *, integer *,
integer *, real *, integer *, real *);
extern /* Subroutine */ int xerbla_(char *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *);
integer scllen;
real bignum;
extern /* Subroutine */ int sgelqf_(integer *, integer *, real *, integer
*, real *, real *, integer *, integer *), slascl_(char *, integer
*, integer *, real *, real *, integer *, integer *, real *,
integer *, integer *), sgeqrf_(integer *, integer *, real
*, integer *, real *, real *, integer *, integer *), slaset_(char
*, integer *, integer *, real *, real *, real *, integer *);
real smlnum;
extern /* Subroutine */ int sormlq_(char *, char *, integer *, integer *,
integer *, real *, integer *, real *, real *, integer *, real *,
integer *, integer *);
logical lquery;
extern /* Subroutine */ int sormqr_(char *, char *, integer *, integer *,
integer *, real *, integer *, real *, real *, integer *, real *,
integer *, integer *), strtrs_(char *, char *,
char *, integer *, integer *, real *, integer *, real *, integer *
, integer *);
/* -- LAPACK driver routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SGELS solves overdetermined or underdetermined real linear systems */
/* involving an M-by-N matrix A, or its transpose, using a QR or LQ */
/* factorization of A. It is assumed that A has full rank. */
/* The following options are provided: */
/* 1. If TRANS = 'N' and m >= n: find the least squares solution of */
/* an overdetermined system, i.e., solve the least squares problem */
/* minimize || B - A*X ||. */
/* 2. If TRANS = 'N' and m < n: find the minimum norm solution of */
/* an underdetermined system A * X = B. */
/* 3. If TRANS = 'T' and m >= n: find the minimum norm solution of */
/* an undetermined system A**T * X = B. */
/* 4. If TRANS = 'T' and m < n: find the least squares solution of */
/* an overdetermined system, i.e., solve the least squares problem */
/* minimize || B - A**T * X ||. */
/* Several right hand side vectors b and solution vectors x can be */
/* handled in a single call; they are stored as the columns of the */
/* M-by-NRHS right hand side matrix B and the N-by-NRHS solution */
/* matrix X. */
/* Arguments */
/* ========= */
/* TRANS (input) CHARACTER*1 */
/* = 'N': the linear system involves A; */
/* = 'T': the linear system involves A**T. */
/* M (input) INTEGER */
/* The number of rows of the matrix A. M >= 0. */
/* N (input) INTEGER */
/* The number of columns of the matrix A. N >= 0. */
/* NRHS (input) INTEGER */
/* The number of right hand sides, i.e., the number of */
/* columns of the matrices B and X. NRHS >=0. */
/* A (input/output) REAL array, dimension (LDA,N) */
/* On entry, the M-by-N matrix A. */
/* On exit, */
/* if M >= N, A is overwritten by details of its QR */
/* factorization as returned by SGEQRF; */
/* if M < N, A is overwritten by details of its LQ */
/* factorization as returned by SGELQF. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1,M). */
/* B (input/output) REAL array, dimension (LDB,NRHS) */
/* On entry, the matrix B of right hand side vectors, stored */
/* columnwise; B is M-by-NRHS if TRANS = 'N', or N-by-NRHS */
/* if TRANS = 'T'. */
/* On exit, if INFO = 0, B is overwritten by the solution */
/* vectors, stored columnwise: */
/* if TRANS = 'N' and m >= n, rows 1 to n of B contain the least */
/* squares solution vectors; the residual sum of squares for the */
/* solution in each column is given by the sum of squares of */
/* elements N+1 to M in that column; */
/* if TRANS = 'N' and m < n, rows 1 to N of B contain the */
/* minimum norm solution vectors; */
/* if TRANS = 'T' and m >= n, rows 1 to M of B contain the */
/* minimum norm solution vectors; */
/* if TRANS = 'T' and m < n, rows 1 to M of B contain the */
/* least squares solution vectors; the residual sum of squares */
/* for the solution in each column is given by the sum of */
/* squares of elements M+1 to N in that column. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= MAX(1,M,N). */
/* WORK (workspace/output) REAL 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, MN + max( MN, NRHS ) ). */
/* For optimal performance, */
/* LWORK >= max( 1, MN + max( MN, NRHS )*NB ). */
/* where MN = min(M,N) and NB is the optimum block size. */
/* 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 i-th diagonal element of the */
/* triangular factor of A is zero, so that A does not have */
/* full rank; the least squares solution could not be */
/* computed. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. Local Arrays .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input arguments. */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
--work;
/* Function Body */
*info = 0;
mn = min(*m,*n);
lquery = *lwork == -1;
if (! (lsame_(trans, "N") || lsame_(trans, "T"))) {
*info = -1;
} else if (*m < 0) {
*info = -2;
} else if (*n < 0) {
*info = -3;
} else if (*nrhs < 0) {
*info = -4;
} else if (*lda < max(1,*m)) {
*info = -6;
} else /* if(complicated condition) */ {
/* Computing MAX */
i__1 = max(1,*m);
if (*ldb < max(i__1,*n)) {
*info = -8;
} else /* if(complicated condition) */ {
/* Computing MAX */
i__1 = 1, i__2 = mn + max(mn,*nrhs);
if (*lwork < max(i__1,i__2) && ! lquery) {
*info = -10;
}
}
}
/* Figure out optimal block size */
if (*info == 0 || *info == -10) {
tpsd = TRUE_;
if (lsame_(trans, "N")) {
tpsd = FALSE_;
}
if (*m >= *n) {
nb = ilaenv_(&c__1, "SGEQRF", " ", m, n, &c_n1, &c_n1);
if (tpsd) {
/* Computing MAX */
i__1 = nb, i__2 = ilaenv_(&c__1, "SORMQR", "LN", m, nrhs, n, &
c_n1);
nb = max(i__1,i__2);
} else {
/* Computing MAX */
i__1 = nb, i__2 = ilaenv_(&c__1, "SORMQR", "LT", m, nrhs, n, &
c_n1);
nb = max(i__1,i__2);
}
} else {
nb = ilaenv_(&c__1, "SGELQF", " ", m, n, &c_n1, &c_n1);
if (tpsd) {
/* Computing MAX */
i__1 = nb, i__2 = ilaenv_(&c__1, "SORMLQ", "LT", n, nrhs, m, &
c_n1);
nb = max(i__1,i__2);
} else {
/* Computing MAX */
i__1 = nb, i__2 = ilaenv_(&c__1, "SORMLQ", "LN", n, nrhs, m, &
c_n1);
nb = max(i__1,i__2);
}
}
/* Computing MAX */
i__1 = 1, i__2 = mn + max(mn,*nrhs) * nb;
wsize = max(i__1,i__2);
work[1] = (real) wsize;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SGELS ", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
/* Computing MIN */
i__1 = min(*m,*n);
if (min(i__1,*nrhs) == 0) {
i__1 = max(*m,*n);
slaset_("Full", &i__1, nrhs, &c_b33, &c_b33, &b[b_offset], ldb);
return 0;
}
/* Get machine parameters */
smlnum = slamch_("S") / slamch_("P");
bignum = 1.f / smlnum;
slabad_(&smlnum, &bignum);
/* Scale A, B if max element outside range [SMLNUM,BIGNUM] */
anrm = slange_("M", m, n, &a[a_offset], lda, rwork);
iascl = 0;
if (anrm > 0.f && anrm < smlnum) {
/* Scale matrix norm up to SMLNUM */
slascl_("G", &c__0, &c__0, &anrm, &smlnum, m, n, &a[a_offset], lda,
info);
iascl = 1;
} else if (anrm > bignum) {
/* Scale matrix norm down to BIGNUM */
slascl_("G", &c__0, &c__0, &anrm, &bignum, m, n, &a[a_offset], lda,
info);
iascl = 2;
} else if (anrm == 0.f) {
/* Matrix all zero. Return zero solution. */
i__1 = max(*m,*n);
slaset_("F", &i__1, nrhs, &c_b33, &c_b33, &b[b_offset], ldb);
goto L50;
}
brow = *m;
if (tpsd) {
brow = *n;
}
bnrm = slange_("M", &brow, nrhs, &b[b_offset], ldb, rwork);
ibscl = 0;
if (bnrm > 0.f && bnrm < smlnum) {
/* Scale matrix norm up to SMLNUM */
slascl_("G", &c__0, &c__0, &bnrm, &smlnum, &brow, nrhs, &b[b_offset],
ldb, info);
ibscl = 1;
} else if (bnrm > bignum) {
/* Scale matrix norm down to BIGNUM */
slascl_("G", &c__0, &c__0, &bnrm, &bignum, &brow, nrhs, &b[b_offset],
ldb, info);
ibscl = 2;
}
if (*m >= *n) {
/* compute QR factorization of A */
i__1 = *lwork - mn;
sgeqrf_(m, n, &a[a_offset], lda, &work[1], &work[mn + 1], &i__1, info)
;
/* workspace at least N, optimally N*NB */
if (! tpsd) {
/* Least-Squares Problem min || A * X - B || */
/* B(1:M,1:NRHS) := Q' * B(1:M,1:NRHS) */
i__1 = *lwork - mn;
sormqr_("Left", "Transpose", m, nrhs, n, &a[a_offset], lda, &work[
1], &b[b_offset], ldb, &work[mn + 1], &i__1, info);
/* workspace at least NRHS, optimally NRHS*NB */
/* B(1:N,1:NRHS) := inv(R) * B(1:N,1:NRHS) */
strtrs_("Upper", "No transpose", "Non-unit", n, nrhs, &a[a_offset]
, lda, &b[b_offset], ldb, info);
if (*info > 0) {
return 0;
}
scllen = *n;
} else {
/* Overdetermined system of equations A' * X = B */
/* B(1:N,1:NRHS) := inv(R') * B(1:N,1:NRHS) */
strtrs_("Upper", "Transpose", "Non-unit", n, nrhs, &a[a_offset],
lda, &b[b_offset], ldb, info);
if (*info > 0) {
return 0;
}
/* B(N+1:M,1:NRHS) = ZERO */
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
i__2 = *m;
for (i__ = *n + 1; i__ <= i__2; ++i__) {
b[i__ + j * b_dim1] = 0.f;
/* L10: */
}
/* L20: */
}
/* B(1:M,1:NRHS) := Q(1:N,:) * B(1:N,1:NRHS) */
i__1 = *lwork - mn;
sormqr_("Left", "No transpose", m, nrhs, n, &a[a_offset], lda, &
work[1], &b[b_offset], ldb, &work[mn + 1], &i__1, info);
/* workspace at least NRHS, optimally NRHS*NB */
scllen = *m;
}
} else {
/* Compute LQ factorization of A */
i__1 = *lwork - mn;
sgelqf_(m, n, &a[a_offset], lda, &work[1], &work[mn + 1], &i__1, info)
;
/* workspace at least M, optimally M*NB. */
if (! tpsd) {
/* underdetermined system of equations A * X = B */
/* B(1:M,1:NRHS) := inv(L) * B(1:M,1:NRHS) */
strtrs_("Lower", "No transpose", "Non-unit", m, nrhs, &a[a_offset]
, lda, &b[b_offset], ldb, info);
if (*info > 0) {
return 0;
}
/* B(M+1:N,1:NRHS) = 0 */
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
i__2 = *n;
for (i__ = *m + 1; i__ <= i__2; ++i__) {
b[i__ + j * b_dim1] = 0.f;
/* L30: */
}
/* L40: */
}
/* B(1:N,1:NRHS) := Q(1:N,:)' * B(1:M,1:NRHS) */
i__1 = *lwork - mn;
sormlq_("Left", "Transpose", n, nrhs, m, &a[a_offset], lda, &work[
1], &b[b_offset], ldb, &work[mn + 1], &i__1, info);
/* workspace at least NRHS, optimally NRHS*NB */
scllen = *n;
} else {
/* overdetermined system min || A' * X - B || */
/* B(1:N,1:NRHS) := Q * B(1:N,1:NRHS) */
i__1 = *lwork - mn;
sormlq_("Left", "No transpose", n, nrhs, m, &a[a_offset], lda, &
work[1], &b[b_offset], ldb, &work[mn + 1], &i__1, info);
/* workspace at least NRHS, optimally NRHS*NB */
/* B(1:M,1:NRHS) := inv(L') * B(1:M,1:NRHS) */
strtrs_("Lower", "Transpose", "Non-unit", m, nrhs, &a[a_offset],
lda, &b[b_offset], ldb, info);
if (*info > 0) {
return 0;
}
scllen = *m;
}
}
/* Undo scaling */
if (iascl == 1) {
slascl_("G", &c__0, &c__0, &anrm, &smlnum, &scllen, nrhs, &b[b_offset]
, ldb, info);
} else if (iascl == 2) {
slascl_("G", &c__0, &c__0, &anrm, &bignum, &scllen, nrhs, &b[b_offset]
, ldb, info);
}
if (ibscl == 1) {
slascl_("G", &c__0, &c__0, &smlnum, &bnrm, &scllen, nrhs, &b[b_offset]
, ldb, info);
} else if (ibscl == 2) {
slascl_("G", &c__0, &c__0, &bignum, &bnrm, &scllen, nrhs, &b[b_offset]
, ldb, info);
}
L50:
work[1] = (real) wsize;
return 0;
/* End of SGELS */
} /* sgels_ */
/* Subroutine */ int ssbevx_(char *jobz, char *range, char *uplo, integer *n,
integer *kd, real *ab, integer *ldab, real *q, integer *ldq, real *vl,
real *vu, integer *il, integer *iu, real *abstol, integer *m, real *
w, real *z__, integer *ldz, real *work, integer *iwork, integer *
ifail, integer *info)
{
/* System generated locals */
integer ab_dim1, ab_offset, q_dim1, q_offset, z_dim1, z_offset, i__1,
i__2;
real r__1, r__2;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
integer i__, j, jj;
real eps, vll, vuu, tmp1;
integer indd, inde;
real anrm;
integer imax;
real rmin, rmax;
logical test;
integer itmp1, indee;
real sigma;
extern logical lsame_(char *, char *);
integer iinfo;
extern /* Subroutine */ int sscal_(integer *, real *, real *, integer *);
char order[1];
extern /* Subroutine */ int sgemv_(char *, integer *, integer *, real *,
real *, integer *, real *, integer *, real *, real *, integer *);
logical lower;
extern /* Subroutine */ int scopy_(integer *, real *, integer *, real *,
integer *), sswap_(integer *, real *, integer *, real *, integer *
);
logical wantz, alleig, indeig;
integer iscale, indibl;
logical valeig;
extern doublereal slamch_(char *);
real safmin;
extern /* Subroutine */ int xerbla_(char *, integer *);
real abstll, bignum;
extern doublereal slansb_(char *, char *, integer *, integer *, real *,
integer *, real *);
extern /* Subroutine */ int slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *);
integer indisp, indiwo;
extern /* Subroutine */ int slacpy_(char *, integer *, integer *, real *,
integer *, real *, integer *);
integer indwrk;
extern /* Subroutine */ int ssbtrd_(char *, char *, integer *, integer *,
real *, integer *, real *, real *, real *, integer *, real *,
integer *), sstein_(integer *, real *, real *,
integer *, real *, integer *, integer *, real *, integer *, real *
, integer *, integer *, integer *), ssterf_(integer *, real *,
real *, integer *);
integer nsplit;
real smlnum;
extern /* Subroutine */ int sstebz_(char *, char *, integer *, real *,
real *, integer *, integer *, real *, real *, real *, integer *,
integer *, real *, integer *, integer *, real *, integer *,
integer *), ssteqr_(char *, integer *, real *,
real *, real *, integer *, real *, integer *);
/* -- LAPACK driver routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SSBEVX computes selected eigenvalues and, optionally, eigenvectors */
/* of a real symmetric band matrix A. Eigenvalues and eigenvectors can */
/* be selected by specifying either a range of values or a range of */
/* indices for the desired eigenvalues. */
/* Arguments */
/* ========= */
/* JOBZ (input) CHARACTER*1 */
/* = 'N': Compute eigenvalues only; */
/* = 'V': Compute eigenvalues and eigenvectors. */
/* RANGE (input) CHARACTER*1 */
/* = 'A': all eigenvalues will be found; */
/* = 'V': all eigenvalues in the half-open interval (VL,VU] */
/* will be found; */
/* = 'I': the IL-th through IU-th eigenvalues will be found. */
/* UPLO (input) CHARACTER*1 */
/* = 'U': Upper triangle of A is stored; */
/* = 'L': Lower triangle of A is stored. */
/* N (input) INTEGER */
/* The order of the matrix A. N >= 0. */
/* KD (input) INTEGER */
/* The number of superdiagonals of the matrix A if UPLO = 'U', */
/* or the number of subdiagonals if UPLO = 'L'. KD >= 0. */
/* AB (input/output) REAL array, dimension (LDAB, N) */
/* On entry, the upper or lower triangle of the symmetric band */
/* matrix A, stored in the first KD+1 rows of the array. The */
/* j-th column of A is stored in the j-th column of the array AB */
/* as follows: */
/* if UPLO = 'U', AB(kd+1+i-j,j) = A(i,j) for max(1,j-kd)<=i<=j; */
/* if UPLO = 'L', AB(1+i-j,j) = A(i,j) for j<=i<=min(n,j+kd). */
/* On exit, AB is overwritten by values generated during the */
/* reduction to tridiagonal form. If UPLO = 'U', the first */
/* superdiagonal and the diagonal of the tridiagonal matrix T */
/* are returned in rows KD and KD+1 of AB, and if UPLO = 'L', */
/* the diagonal and first subdiagonal of T are returned in the */
/* first two rows of AB. */
/* LDAB (input) INTEGER */
/* The leading dimension of the array AB. LDAB >= KD + 1. */
/* Q (output) REAL array, dimension (LDQ, N) */
/* If JOBZ = 'V', the N-by-N orthogonal matrix used in the */
/* reduction to tridiagonal form. */
/* If JOBZ = 'N', the array Q is not referenced. */
/* LDQ (input) INTEGER */
/* The leading dimension of the array Q. If JOBZ = 'V', then */
/* LDQ >= max(1,N). */
/* VL (input) REAL */
/* VU (input) REAL */
/* If RANGE='V', the lower and upper bounds of the interval to */
/* be searched for eigenvalues. VL < VU. */
/* Not referenced if RANGE = 'A' or 'I'. */
/* IL (input) INTEGER */
/* IU (input) INTEGER */
/* If RANGE='I', the indices (in ascending order) of the */
/* smallest and largest eigenvalues to be returned. */
/* 1 <= IL <= IU <= N, if N > 0; IL = 1 and IU = 0 if N = 0. */
/* Not referenced if RANGE = 'A' or 'V'. */
/* ABSTOL (input) REAL */
/* The absolute error tolerance for the eigenvalues. */
/* An approximate eigenvalue is accepted as converged */
/* when it is determined to lie in an interval [a,b] */
/* of width less than or equal to */
/* ABSTOL + EPS * max( |a|,|b| ) , */
/* where EPS is the machine precision. If ABSTOL is less than */
/* or equal to zero, then EPS*|T| will be used in its place, */
/* where |T| is the 1-norm of the tridiagonal matrix obtained */
/* by reducing AB to tridiagonal form. */
/* Eigenvalues will be computed most accurately when ABSTOL is */
/* set to twice the underflow threshold 2*SLAMCH('S'), not zero. */
/* If this routine returns with INFO>0, indicating that some */
/* eigenvectors did not converge, try setting ABSTOL to */
/* 2*SLAMCH('S'). */
/* See "Computing Small Singular Values of Bidiagonal Matrices */
/* with Guaranteed High Relative Accuracy," by Demmel and */
/* Kahan, LAPACK Working Note #3. */
/* M (output) INTEGER */
/* The total number of eigenvalues found. 0 <= M <= N. */
/* If RANGE = 'A', M = N, and if RANGE = 'I', M = IU-IL+1. */
/* W (output) REAL array, dimension (N) */
/* The first M elements contain the selected eigenvalues in */
/* ascending order. */
/* Z (output) REAL array, dimension (LDZ, max(1,M)) */
/* If JOBZ = 'V', then if INFO = 0, the first M columns of Z */
/* contain the orthonormal eigenvectors of the matrix A */
/* corresponding to the selected eigenvalues, with the i-th */
/* column of Z holding the eigenvector associated with W(i). */
/* If an eigenvector fails to converge, then that column of Z */
/* contains the latest approximation to the eigenvector, and the */
/* index of the eigenvector is returned in IFAIL. */
/* If JOBZ = 'N', then Z is not referenced. */
/* Note: the user must ensure that at least max(1,M) columns are */
/* supplied in the array Z; if RANGE = 'V', the exact value of M */
/* is not known in advance and an upper bound must be used. */
/* LDZ (input) INTEGER */
/* The leading dimension of the array Z. LDZ >= 1, and if */
/* JOBZ = 'V', LDZ >= max(1,N). */
/* WORK (workspace) REAL array, dimension (7*N) */
/* IWORK (workspace) INTEGER array, dimension (5*N) */
/* IFAIL (output) INTEGER array, dimension (N) */
/* If JOBZ = 'V', then if INFO = 0, the first M elements of */
/* IFAIL are zero. If INFO > 0, then IFAIL contains the */
/* indices of the eigenvectors that failed to converge. */
/* If JOBZ = 'N', then IFAIL is not referenced. */
/* INFO (output) INTEGER */
/* = 0: successful exit. */
/* < 0: if INFO = -i, the i-th argument had an illegal value. */
/* > 0: if INFO = i, then i eigenvectors failed to converge. */
/* Their indices are stored in array IFAIL. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
ab_dim1 = *ldab;
ab_offset = 1 + ab_dim1;
ab -= ab_offset;
q_dim1 = *ldq;
q_offset = 1 + q_dim1;
q -= q_offset;
--w;
z_dim1 = *ldz;
z_offset = 1 + z_dim1;
z__ -= z_offset;
--work;
--iwork;
--ifail;
/* Function Body */
wantz = lsame_(jobz, "V");
alleig = lsame_(range, "A");
valeig = lsame_(range, "V");
indeig = lsame_(range, "I");
lower = lsame_(uplo, "L");
*info = 0;
if (! (wantz || lsame_(jobz, "N"))) {
*info = -1;
} else if (! (alleig || valeig || indeig)) {
*info = -2;
} else if (! (lower || lsame_(uplo, "U"))) {
*info = -3;
} else if (*n < 0) {
*info = -4;
} else if (*kd < 0) {
*info = -5;
} else if (*ldab < *kd + 1) {
*info = -7;
} else if (wantz && *ldq < max(1,*n)) {
*info = -9;
} else {
if (valeig) {
if (*n > 0 && *vu <= *vl) {
*info = -11;
}
} else if (indeig) {
if (*il < 1 || *il > max(1,*n)) {
*info = -12;
} else if (*iu < min(*n,*il) || *iu > *n) {
*info = -13;
}
}
}
if (*info == 0) {
if (*ldz < 1 || wantz && *ldz < *n) {
*info = -18;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SSBEVX", &i__1);
return 0;
}
/* Quick return if possible */
*m = 0;
if (*n == 0) {
return 0;
}
if (*n == 1) {
*m = 1;
if (lower) {
tmp1 = ab[ab_dim1 + 1];
} else {
tmp1 = ab[*kd + 1 + ab_dim1];
}
if (valeig) {
if (! (*vl < tmp1 && *vu >= tmp1)) {
*m = 0;
}
}
if (*m == 1) {
w[1] = tmp1;
if (wantz) {
z__[z_dim1 + 1] = 1.f;
}
}
return 0;
}
/* Get machine constants. */
safmin = slamch_("Safe minimum");
eps = slamch_("Precision");
smlnum = safmin / eps;
bignum = 1.f / smlnum;
rmin = sqrt(smlnum);
/* Computing MIN */
r__1 = sqrt(bignum), r__2 = 1.f / sqrt(sqrt(safmin));
rmax = dmin(r__1,r__2);
/* Scale matrix to allowable range, if necessary. */
iscale = 0;
abstll = *abstol;
if (valeig) {
vll = *vl;
vuu = *vu;
} else {
vll = 0.f;
vuu = 0.f;
}
anrm = slansb_("M", uplo, n, kd, &ab[ab_offset], ldab, &work[1]);
if (anrm > 0.f && anrm < rmin) {
iscale = 1;
sigma = rmin / anrm;
} else if (anrm > rmax) {
iscale = 1;
sigma = rmax / anrm;
}
if (iscale == 1) {
if (lower) {
slascl_("B", kd, kd, &c_b14, &sigma, n, n, &ab[ab_offset], ldab,
info);
} else {
slascl_("Q", kd, kd, &c_b14, &sigma, n, n, &ab[ab_offset], ldab,
info);
}
if (*abstol > 0.f) {
abstll = *abstol * sigma;
}
if (valeig) {
vll = *vl * sigma;
vuu = *vu * sigma;
}
}
/* Call SSBTRD to reduce symmetric band matrix to tridiagonal form. */
indd = 1;
inde = indd + *n;
indwrk = inde + *n;
ssbtrd_(jobz, uplo, n, kd, &ab[ab_offset], ldab, &work[indd], &work[inde],
&q[q_offset], ldq, &work[indwrk], &iinfo);
/* If all eigenvalues are desired and ABSTOL is less than or equal */
/* to zero, then call SSTERF or SSTEQR. If this fails for some */
/* eigenvalue, then try SSTEBZ. */
test = FALSE_;
if (indeig) {
if (*il == 1 && *iu == *n) {
test = TRUE_;
}
}
if ((alleig || test) && *abstol <= 0.f) {
scopy_(n, &work[indd], &c__1, &w[1], &c__1);
indee = indwrk + (*n << 1);
if (! wantz) {
i__1 = *n - 1;
scopy_(&i__1, &work[inde], &c__1, &work[indee], &c__1);
ssterf_(n, &w[1], &work[indee], info);
} else {
slacpy_("A", n, n, &q[q_offset], ldq, &z__[z_offset], ldz);
i__1 = *n - 1;
scopy_(&i__1, &work[inde], &c__1, &work[indee], &c__1);
ssteqr_(jobz, n, &w[1], &work[indee], &z__[z_offset], ldz, &work[
indwrk], info);
if (*info == 0) {
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
ifail[i__] = 0;
/* L10: */
}
}
}
if (*info == 0) {
*m = *n;
goto L30;
}
*info = 0;
}
/* Otherwise, call SSTEBZ and, if eigenvectors are desired, SSTEIN. */
if (wantz) {
*(unsigned char *)order = 'B';
} else {
*(unsigned char *)order = 'E';
}
indibl = 1;
indisp = indibl + *n;
indiwo = indisp + *n;
sstebz_(range, order, n, &vll, &vuu, il, iu, &abstll, &work[indd], &work[
inde], m, &nsplit, &w[1], &iwork[indibl], &iwork[indisp], &work[
indwrk], &iwork[indiwo], info);
if (wantz) {
sstein_(n, &work[indd], &work[inde], m, &w[1], &iwork[indibl], &iwork[
indisp], &z__[z_offset], ldz, &work[indwrk], &iwork[indiwo], &
ifail[1], info);
/* Apply orthogonal matrix used in reduction to tridiagonal */
/* form to eigenvectors returned by SSTEIN. */
i__1 = *m;
for (j = 1; j <= i__1; ++j) {
scopy_(n, &z__[j * z_dim1 + 1], &c__1, &work[1], &c__1);
sgemv_("N", n, n, &c_b14, &q[q_offset], ldq, &work[1], &c__1, &
c_b34, &z__[j * z_dim1 + 1], &c__1);
/* L20: */
}
}
/* If matrix was scaled, then rescale eigenvalues appropriately. */
L30:
if (iscale == 1) {
if (*info == 0) {
imax = *m;
} else {
imax = *info - 1;
}
r__1 = 1.f / sigma;
sscal_(&imax, &r__1, &w[1], &c__1);
}
/* If eigenvalues are not in order, then sort them, along with */
/* eigenvectors. */
if (wantz) {
i__1 = *m - 1;
for (j = 1; j <= i__1; ++j) {
i__ = 0;
tmp1 = w[j];
i__2 = *m;
for (jj = j + 1; jj <= i__2; ++jj) {
if (w[jj] < tmp1) {
i__ = jj;
tmp1 = w[jj];
}
/* L40: */
}
if (i__ != 0) {
itmp1 = iwork[indibl + i__ - 1];
w[i__] = w[j];
iwork[indibl + i__ - 1] = iwork[indibl + j - 1];
w[j] = tmp1;
iwork[indibl + j - 1] = itmp1;
sswap_(n, &z__[i__ * z_dim1 + 1], &c__1, &z__[j * z_dim1 + 1],
&c__1);
if (*info != 0) {
itmp1 = ifail[i__];
ifail[i__] = ifail[j];
ifail[j] = itmp1;
}
}
/* L50: */
}
}
return 0;
/* End of SSBEVX */
} /* ssbevx_ */
/* Subroutine */ int ssteqr_(char *compz, integer *n, real *d__, real *e,
real *z__, integer *ldz, real *work, integer *info)
{
/* System generated locals */
integer z_dim1, z_offset, i__1, i__2;
real r__1, r__2;
/* Local variables */
real b, c__, f, g;
integer i__, j, k, l, m;
real p, r__, s;
integer l1, ii, mm, lm1, mm1, nm1;
real rt1, rt2, eps;
integer lsv;
real tst, eps2;
integer lend, jtot;
real anorm;
integer lendm1, lendp1;
integer iscale;
real safmin;
real safmax;
integer lendsv;
real ssfmin;
integer nmaxit, icompz;
real ssfmax;
/* -- LAPACK routine (version 3.2) -- */
/* November 2006 */
/* Purpose */
/* ======= */
/* SSTEQR 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 SSYTRD or SSPTRD or SSBTRD 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) REAL 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) REAL array, dimension (N-1) */
/* On entry, the (n-1) subdiagonal elements of the tridiagonal */
/* matrix. */
/* On exit, E has been destroyed. */
/* Z (input/output) REAL 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) REAL 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. */
/* ===================================================================== */
/* 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_("SSTEQR", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
if (*n == 1) {
if (icompz == 2) {
z__[z_dim1 + 1] = 1.f;
}
return 0;
}
/* Determine the unit roundoff and over/underflow thresholds. */
eps = slamch_("E");
/* Computing 2nd power */
r__1 = eps;
eps2 = r__1 * r__1;
safmin = slamch_("S");
safmax = 1.f / safmin;
ssfmax = sqrt(safmax) / 3.f;
ssfmin = sqrt(safmin) / eps2;
/* Compute the eigenvalues and eigenvectors of the tridiagonal */
/* matrix. */
if (icompz == 2) {
slaset_("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.f;
}
if (l1 <= nm1) {
i__1 = nm1;
for (m = l1; m <= i__1; ++m) {
tst = (r__1 = e[m], dabs(r__1));
if (tst == 0.f) {
goto L30;
}
if (tst <= sqrt((r__1 = d__[m], dabs(r__1))) * sqrt((r__2 = d__[m
+ 1], dabs(r__2))) * eps) {
e[m] = 0.f;
goto L30;
}
}
}
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 = slanst_("I", &i__1, &d__[l], &e[l]);
iscale = 0;
if (anorm == 0.f) {
goto L10;
}
if (anorm > ssfmax) {
iscale = 1;
i__1 = lend - l + 1;
slascl_("G", &c__0, &c__0, &anorm, &ssfmax, &i__1, &c__1, &d__[l], n,
info);
i__1 = lend - l;
slascl_("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;
slascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &d__[l], n,
info);
i__1 = lend - l;
slascl_("G", &c__0, &c__0, &anorm, &ssfmin, &i__1, &c__1, &e[l], n,
info);
}
/* Choose between QL and QR iteration */
if ((r__1 = d__[lend], dabs(r__1)) < (r__2 = d__[l], dabs(r__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 */
r__2 = (r__1 = e[m], dabs(r__1));
tst = r__2 * r__2;
if (tst <= eps2 * (r__1 = d__[m], dabs(r__1)) * (r__2 = d__[m
+ 1], dabs(r__2)) + safmin) {
goto L60;
}
}
}
m = lend;
L60:
if (m < lend) {
e[m] = 0.f;
}
p = d__[l];
if (m == l) {
goto L80;
}
/* If remaining matrix is 2-by-2, use SLAE2 or SLAEV2 */
/* to compute its eigensystem. */
if (m == l + 1) {
if (icompz > 0) {
slaev2_(&d__[l], &e[l], &d__[l + 1], &rt1, &rt2, &c__, &s);
work[l] = c__;
work[*n - 1 + l] = s;
slasr_("R", "V", "B", n, &c__2, &work[l], &work[*n - 1 + l], &
z__[l * z_dim1 + 1], ldz);
} else {
slae2_(&d__[l], &e[l], &d__[l + 1], &rt1, &rt2);
}
d__[l] = rt1;
d__[l + 1] = rt2;
e[l] = 0.f;
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.f);
r__ = slapy2_(&g, &c_b10);
g = d__[m] - p + e[l] / (g + r_sign(&r__, &g));
s = 1.f;
c__ = 1.f;
p = 0.f;
/* Inner loop */
mm1 = m - 1;
i__1 = l;
for (i__ = mm1; i__ >= i__1; --i__) {
f = s * e[i__];
b = c__ * e[i__];
slartg_(&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.f * 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;
}
}
/* If eigenvectors are desired, then apply saved rotations. */
if (icompz > 0) {
mm = m - l + 1;
slasr_("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 */
r__2 = (r__1 = e[m - 1], dabs(r__1));
tst = r__2 * r__2;
if (tst <= eps2 * (r__1 = d__[m], dabs(r__1)) * (r__2 = d__[m
- 1], dabs(r__2)) + safmin) {
goto L110;
}
}
}
m = lend;
L110:
if (m > lend) {
e[m - 1] = 0.f;
}
p = d__[l];
if (m == l) {
goto L130;
}
/* If remaining matrix is 2-by-2, use SLAE2 or SLAEV2 */
/* to compute its eigensystem. */
if (m == l - 1) {
if (icompz > 0) {
slaev2_(&d__[l - 1], &e[l - 1], &d__[l], &rt1, &rt2, &c__, &s)
;
work[m] = c__;
work[*n - 1 + m] = s;
slasr_("R", "V", "F", n, &c__2, &work[m], &work[*n - 1 + m], &
z__[(l - 1) * z_dim1 + 1], ldz);
} else {
slae2_(&d__[l - 1], &e[l - 1], &d__[l], &rt1, &rt2);
}
d__[l - 1] = rt1;
d__[l] = rt2;
e[l - 1] = 0.f;
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.f);
r__ = slapy2_(&g, &c_b10);
g = d__[m] - p + e[l - 1] / (g + r_sign(&r__, &g));
s = 1.f;
c__ = 1.f;
p = 0.f;
/* Inner loop */
lm1 = l - 1;
i__1 = lm1;
for (i__ = m; i__ <= i__1; ++i__) {
f = s * e[i__];
b = c__ * e[i__];
slartg_(&g, &f, &c__, &s, &r__);
if (i__ != m) {
e[i__ - 1] = r__;
}
g = d__[i__] - p;
r__ = (d__[i__ + 1] - g) * s + c__ * 2.f * 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;
}
}
/* If eigenvectors are desired, then apply saved rotations. */
if (icompz > 0) {
mm = l - m + 1;
slasr_("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;
slascl_("G", &c__0, &c__0, &ssfmax, &anorm, &i__1, &c__1, &d__[lsv],
n, info);
i__1 = lendsv - lsv;
slascl_("G", &c__0, &c__0, &ssfmax, &anorm, &i__1, &c__1, &e[lsv], n,
info);
} else if (iscale == 2) {
i__1 = lendsv - lsv + 1;
slascl_("G", &c__0, &c__0, &ssfmin, &anorm, &i__1, &c__1, &d__[lsv],
n, info);
i__1 = lendsv - lsv;
slascl_("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.f) {
++(*info);
}
}
goto L190;
/* Order eigenvalues and eigenvectors. */
L160:
if (icompz == 0) {
/* Use Quick Sort */
slasrt_("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];
}
}
if (k != i__) {
d__[k] = d__[i__];
d__[i__] = p;
sswap_(n, &z__[i__ * z_dim1 + 1], &c__1, &z__[k * z_dim1 + 1],
&c__1);
}
}
}
L190:
return 0;
/* End of SSTEQR */
} /* ssteqr_ */
doublereal sqrt12_(integer *m, integer *n, real *a, integer *lda, real *s,
real *work, integer *lwork)
{
/* System generated locals */
integer a_dim1, a_offset, i__1, i__2;
real ret_val;
/* Local variables */
integer i__, j, mn, iscl, info;
real anrm;
extern doublereal snrm2_(integer *, real *, integer *), sasum_(integer *,
real *, integer *);
real dummy[1];
extern /* Subroutine */ int saxpy_(integer *, real *, real *, integer *,
real *, integer *), sgebd2_(integer *, integer *, real *, integer
*, real *, real *, real *, real *, real *, integer *), slabad_(
real *, real *);
extern doublereal slamch_(char *), slange_(char *, integer *,
integer *, real *, integer *, real *);
extern /* Subroutine */ int xerbla_(char *, integer *);
real bignum;
extern /* Subroutine */ int slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *), slaset_(char *, integer *, integer *, real *, real *,
real *, integer *), sbdsqr_(char *, integer *, integer *,
integer *, integer *, real *, real *, real *, integer *, real *,
integer *, real *, integer *, real *, integer *);
real smlnum, nrmsvl;
/* -- LAPACK test routine (version 3.1.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* January 2007 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SQRT12 computes the singular values `svlues' of the upper trapezoid */
/* of A(1:M,1:N) and returns the ratio */
/* || s - svlues||/(||svlues||*eps*max(M,N)) */
/* Arguments */
/* ========= */
/* M (input) INTEGER */
/* The number of rows of the matrix A. */
/* N (input) INTEGER */
/* The number of columns of the matrix A. */
/* A (input) REAL array, dimension (LDA,N) */
/* The M-by-N matrix A. Only the upper trapezoid is referenced. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. */
/* S (input) REAL array, dimension (min(M,N)) */
/* The singular values of the matrix A. */
/* WORK (workspace) REAL array, dimension (LWORK) */
/* LWORK (input) INTEGER */
/* The length of the array WORK. LWORK >= max(M*N + 4*min(M,N) + */
/* max(M,N), M*N+2*MIN( M, N )+4*N). */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Local Arrays .. */
/* .. */
/* .. Executable Statements .. */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
--s;
--work;
/* Function Body */
ret_val = 0.f;
/* Test that enough workspace is supplied */
/* Computing MAX */
i__1 = *m * *n + (min(*m,*n) << 2) + max(*m,*n), i__2 = *m * *n + (min(*m,
*n) << 1) + (*n << 2);
if (*lwork < max(i__1,i__2)) {
xerbla_("SQRT12", &c__7);
return ret_val;
}
/* Quick return if possible */
mn = min(*m,*n);
if ((real) mn <= 0.f) {
return ret_val;
}
nrmsvl = snrm2_(&mn, &s[1], &c__1);
/* Copy upper triangle of A into work */
slaset_("Full", m, n, &c_b6, &c_b6, &work[1], m);
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
i__2 = min(j,*m);
for (i__ = 1; i__ <= i__2; ++i__) {
work[(j - 1) * *m + i__] = a[i__ + j * a_dim1];
/* L10: */
}
/* L20: */
}
/* Get machine parameters */
smlnum = slamch_("S") / slamch_("P");
bignum = 1.f / smlnum;
slabad_(&smlnum, &bignum);
/* Scale work if max entry outside range [SMLNUM,BIGNUM] */
anrm = slange_("M", m, n, &work[1], m, dummy);
iscl = 0;
if (anrm > 0.f && anrm < smlnum) {
/* Scale matrix norm up to SMLNUM */
slascl_("G", &c__0, &c__0, &anrm, &smlnum, m, n, &work[1], m, &info);
iscl = 1;
} else if (anrm > bignum) {
/* Scale matrix norm down to BIGNUM */
slascl_("G", &c__0, &c__0, &anrm, &bignum, m, n, &work[1], m, &info);
iscl = 1;
}
if (anrm != 0.f) {
/* Compute SVD of work */
sgebd2_(m, n, &work[1], m, &work[*m * *n + 1], &work[*m * *n + mn + 1]
, &work[*m * *n + (mn << 1) + 1], &work[*m * *n + mn * 3 + 1],
&work[*m * *n + (mn << 2) + 1], &info);
sbdsqr_("Upper", &mn, &c__0, &c__0, &c__0, &work[*m * *n + 1], &work[*
m * *n + mn + 1], dummy, &mn, dummy, &c__1, dummy, &mn, &work[
*m * *n + (mn << 1) + 1], &info);
if (iscl == 1) {
if (anrm > bignum) {
slascl_("G", &c__0, &c__0, &bignum, &anrm, &mn, &c__1, &work[*
m * *n + 1], &mn, &info);
}
if (anrm < smlnum) {
slascl_("G", &c__0, &c__0, &smlnum, &anrm, &mn, &c__1, &work[*
m * *n + 1], &mn, &info);
}
}
} else {
i__1 = mn;
for (i__ = 1; i__ <= i__1; ++i__) {
work[*m * *n + i__] = 0.f;
/* L30: */
}
}
/* Compare s and singular values of work */
saxpy_(&mn, &c_b33, &s[1], &c__1, &work[*m * *n + 1], &c__1);
ret_val = sasum_(&mn, &work[*m * *n + 1], &c__1) / (slamch_("Epsilon") * (real) max(*m,*n));
if (nrmsvl != 0.f) {
ret_val /= nrmsvl;
}
return ret_val;
/* End of SQRT12 */
} /* sqrt12_ */
int sbdsdc_(char *uplo, char *compq, int *n, float *d__,
float *e, float *u, int *ldu, float *vt, int *ldvt, float *q,
int *iq, float *work, int *iwork, int *info)
{
/* System generated locals */
int u_dim1, u_offset, vt_dim1, vt_offset, i__1, i__2;
float r__1;
/* Builtin functions */
double r_sign(float *, float *), log(double);
/* Local variables */
int i__, j, k;
float p, r__;
int z__, ic, ii, kk;
float cs;
int is, iu;
float sn;
int nm1;
float eps;
int ivt, difl, difr, ierr, perm, mlvl, sqre;
extern int lsame_(char *, char *);
int poles;
extern int slasr_(char *, char *, char *, int *,
int *, float *, float *, float *, int *);
int iuplo, nsize, start;
extern int scopy_(int *, float *, int *, float *,
int *), sswap_(int *, float *, int *, float *, int *
), slasd0_(int *, int *, float *, float *, float *, int *
, float *, int *, int *, int *, float *, int *);
extern double slamch_(char *);
extern int slasda_(int *, int *, int *,
int *, float *, float *, float *, int *, float *, int *,
float *, float *, float *, float *, int *, int *, int *,
int *, float *, float *, float *, float *, int *, int *),
xerbla_(char *, int *);
extern int ilaenv_(int *, char *, char *, int *, int *,
int *, int *);
extern int slascl_(char *, int *, int *, float *,
float *, int *, int *, float *, int *, int *);
int givcol;
extern int slasdq_(char *, int *, int *, int
*, int *, int *, float *, float *, float *, int *, float *
, int *, float *, int *, float *, int *);
int icompq;
extern int slaset_(char *, int *, int *, float *,
float *, float *, int *), slartg_(float *, float *, float *
, float *, float *);
float orgnrm;
int givnum;
extern double slanst_(char *, int *, float *, float *);
int givptr, qstart, smlsiz, wstart, smlszp;
/* -- LAPACK routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SBDSDC computes the singular value decomposition (SVD) of a float */
/* N-by-N (upper or lower) bidiagonal matrix B: B = U * S * VT, */
/* using a divide and conquer method, where S is a diagonal matrix */
/* with non-negative diagonal elements (the singular values of B), and */
/* U and VT are orthogonal matrices of left and right singular vectors, */
/* respectively. SBDSDC can be used to compute all singular values, */
/* and optionally, singular vectors or singular vectors in compact form. */
/* This code makes very mild assumptions about floating point */
/* arithmetic. It will work on machines with a guard digit in */
/* add/subtract, or on those binary machines without guard digits */
/* which subtract like the Cray X-MP, Cray Y-MP, Cray C-90, or Cray-2. */
/* It could conceivably fail on hexadecimal or decimal machines */
/* without guard digits, but we know of none. See SLASD3 for details. */
/* The code currently calls SLASDQ if singular values only are desired. */
/* However, it can be slightly modified to compute singular values */
/* using the divide and conquer method. */
/* Arguments */
/* ========= */
/* UPLO (input) CHARACTER*1 */
/* = 'U': B is upper bidiagonal. */
/* = 'L': B is lower bidiagonal. */
/* COMPQ (input) CHARACTER*1 */
/* Specifies whether singular vectors are to be computed */
/* as follows: */
/* = 'N': Compute singular values only; */
/* = 'P': Compute singular values and compute singular */
/* vectors in compact form; */
/* = 'I': Compute singular values and singular vectors. */
/* N (input) INTEGER */
/* The order of the matrix B. N >= 0. */
/* D (input/output) REAL array, dimension (N) */
/* On entry, the n diagonal elements of the bidiagonal matrix B. */
/* On exit, if INFO=0, the singular values of B. */
/* E (input/output) REAL array, dimension (N-1) */
/* On entry, the elements of E contain the offdiagonal */
/* elements of the bidiagonal matrix whose SVD is desired. */
/* On exit, E has been destroyed. */
/* U (output) REAL array, dimension (LDU,N) */
/* If COMPQ = 'I', then: */
/* On exit, if INFO = 0, U contains the left singular vectors */
/* of the bidiagonal matrix. */
/* For other values of COMPQ, U is not referenced. */
/* LDU (input) INTEGER */
/* The leading dimension of the array U. LDU >= 1. */
/* If singular vectors are desired, then LDU >= MAX( 1, N ). */
/* VT (output) REAL array, dimension (LDVT,N) */
/* If COMPQ = 'I', then: */
/* On exit, if INFO = 0, VT' contains the right singular */
/* vectors of the bidiagonal matrix. */
/* For other values of COMPQ, VT is not referenced. */
/* LDVT (input) INTEGER */
/* The leading dimension of the array VT. LDVT >= 1. */
/* If singular vectors are desired, then LDVT >= MAX( 1, N ). */
/* Q (output) REAL array, dimension (LDQ) */
/* If COMPQ = 'P', then: */
/* On exit, if INFO = 0, Q and IQ contain the left */
/* and right singular vectors in a compact form, */
/* requiring O(N log N) space instead of 2*N**2. */
/* In particular, Q contains all the REAL data in */
/* LDQ >= N*(11 + 2*SMLSIZ + 8*INT(LOG_2(N/(SMLSIZ+1)))) */
/* words of memory, where SMLSIZ is returned by ILAENV and */
/* is equal to the maximum size of the subproblems at the */
/* bottom of the computation tree (usually about 25). */
/* For other values of COMPQ, Q is not referenced. */
/* IQ (output) INTEGER array, dimension (LDIQ) */
/* If COMPQ = 'P', then: */
/* On exit, if INFO = 0, Q and IQ contain the left */
/* and right singular vectors in a compact form, */
/* requiring O(N log N) space instead of 2*N**2. */
/* In particular, IQ contains all INTEGER data in */
/* LDIQ >= N*(3 + 3*INT(LOG_2(N/(SMLSIZ+1)))) */
/* words of memory, where SMLSIZ is returned by ILAENV and */
/* is equal to the maximum size of the subproblems at the */
/* bottom of the computation tree (usually about 25). */
/* For other values of COMPQ, IQ is not referenced. */
/* WORK (workspace) REAL array, dimension (MAX(1,LWORK)) */
/* If COMPQ = 'N' then LWORK >= (4 * N). */
/* If COMPQ = 'P' then LWORK >= (6 * N). */
/* If COMPQ = 'I' then LWORK >= (3 * N**2 + 4 * N). */
/* IWORK (workspace) INTEGER array, dimension (8*N) */
/* INFO (output) INTEGER */
/* = 0: successful exit. */
/* < 0: if INFO = -i, the i-th argument had an illegal value. */
/* > 0: The algorithm failed to compute an singular value. */
/* The update process of divide and conquer failed. */
/* Further Details */
/* =============== */
/* Based on contributions by */
/* Ming Gu and Huan Ren, Computer Science Division, University of */
/* California at Berkeley, USA */
/* ===================================================================== */
/* Changed dimension statement in comment describing E from (N) to */
/* (N-1). Sven, 17 Feb 05. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
--d__;
--e;
u_dim1 = *ldu;
u_offset = 1 + u_dim1;
u -= u_offset;
vt_dim1 = *ldvt;
vt_offset = 1 + vt_dim1;
vt -= vt_offset;
--q;
--iq;
--work;
--iwork;
/* Function Body */
*info = 0;
iuplo = 0;
if (lsame_(uplo, "U")) {
iuplo = 1;
}
if (lsame_(uplo, "L")) {
iuplo = 2;
}
if (lsame_(compq, "N")) {
icompq = 0;
} else if (lsame_(compq, "P")) {
icompq = 1;
} else if (lsame_(compq, "I")) {
icompq = 2;
} else {
icompq = -1;
}
if (iuplo == 0) {
*info = -1;
} else if (icompq < 0) {
*info = -2;
} else if (*n < 0) {
*info = -3;
} else if (*ldu < 1 || icompq == 2 && *ldu < *n) {
*info = -7;
} else if (*ldvt < 1 || icompq == 2 && *ldvt < *n) {
*info = -9;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SBDSDC", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
smlsiz = ilaenv_(&c__9, "SBDSDC", " ", &c__0, &c__0, &c__0, &c__0);
if (*n == 1) {
if (icompq == 1) {
q[1] = r_sign(&c_b15, &d__[1]);
q[smlsiz * *n + 1] = 1.f;
} else if (icompq == 2) {
u[u_dim1 + 1] = r_sign(&c_b15, &d__[1]);
vt[vt_dim1 + 1] = 1.f;
}
d__[1] = ABS(d__[1]);
return 0;
}
nm1 = *n - 1;
/* If matrix lower bidiagonal, rotate to be upper bidiagonal */
/* by applying Givens rotations on the left */
wstart = 1;
qstart = 3;
if (icompq == 1) {
scopy_(n, &d__[1], &c__1, &q[1], &c__1);
i__1 = *n - 1;
scopy_(&i__1, &e[1], &c__1, &q[*n + 1], &c__1);
}
if (iuplo == 2) {
qstart = 5;
wstart = (*n << 1) - 1;
i__1 = *n - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
slartg_(&d__[i__], &e[i__], &cs, &sn, &r__);
d__[i__] = r__;
e[i__] = sn * d__[i__ + 1];
d__[i__ + 1] = cs * d__[i__ + 1];
if (icompq == 1) {
q[i__ + (*n << 1)] = cs;
q[i__ + *n * 3] = sn;
} else if (icompq == 2) {
work[i__] = cs;
work[nm1 + i__] = -sn;
}
/* L10: */
}
}
/* If ICOMPQ = 0, use SLASDQ to compute the singular values. */
if (icompq == 0) {
slasdq_("U", &c__0, n, &c__0, &c__0, &c__0, &d__[1], &e[1], &vt[
vt_offset], ldvt, &u[u_offset], ldu, &u[u_offset], ldu, &work[
wstart], info);
goto L40;
}
/* If N is smaller than the minimum divide size SMLSIZ, then solve */
/* the problem with another solver. */
if (*n <= smlsiz) {
if (icompq == 2) {
slaset_("A", n, n, &c_b29, &c_b15, &u[u_offset], ldu);
slaset_("A", n, n, &c_b29, &c_b15, &vt[vt_offset], ldvt);
slasdq_("U", &c__0, n, n, n, &c__0, &d__[1], &e[1], &vt[vt_offset]
, ldvt, &u[u_offset], ldu, &u[u_offset], ldu, &work[
wstart], info);
} else if (icompq == 1) {
iu = 1;
ivt = iu + *n;
slaset_("A", n, n, &c_b29, &c_b15, &q[iu + (qstart - 1) * *n], n);
slaset_("A", n, n, &c_b29, &c_b15, &q[ivt + (qstart - 1) * *n], n);
slasdq_("U", &c__0, n, n, n, &c__0, &d__[1], &e[1], &q[ivt + (
qstart - 1) * *n], n, &q[iu + (qstart - 1) * *n], n, &q[
iu + (qstart - 1) * *n], n, &work[wstart], info);
}
goto L40;
}
if (icompq == 2) {
slaset_("A", n, n, &c_b29, &c_b15, &u[u_offset], ldu);
slaset_("A", n, n, &c_b29, &c_b15, &vt[vt_offset], ldvt);
}
/* Scale. */
orgnrm = slanst_("M", n, &d__[1], &e[1]);
if (orgnrm == 0.f) {
return 0;
}
slascl_("G", &c__0, &c__0, &orgnrm, &c_b15, n, &c__1, &d__[1], n, &ierr);
slascl_("G", &c__0, &c__0, &orgnrm, &c_b15, &nm1, &c__1, &e[1], &nm1, &
ierr);
eps = slamch_("Epsilon");
mlvl = (int) (log((float) (*n) / (float) (smlsiz + 1)) / log(2.f)) + 1;
smlszp = smlsiz + 1;
if (icompq == 1) {
iu = 1;
ivt = smlsiz + 1;
difl = ivt + smlszp;
difr = difl + mlvl;
z__ = difr + (mlvl << 1);
ic = z__ + mlvl;
is = ic + 1;
poles = is + 1;
givnum = poles + (mlvl << 1);
k = 1;
givptr = 2;
perm = 3;
givcol = perm + mlvl;
}
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
if ((r__1 = d__[i__], ABS(r__1)) < eps) {
d__[i__] = r_sign(&eps, &d__[i__]);
}
/* L20: */
}
start = 1;
sqre = 0;
i__1 = nm1;
for (i__ = 1; i__ <= i__1; ++i__) {
if ((r__1 = e[i__], ABS(r__1)) < eps || i__ == nm1) {
/* Subproblem found. First determine its size and then */
/* apply divide and conquer on it. */
if (i__ < nm1) {
/* A subproblem with E(I) small for I < NM1. */
nsize = i__ - start + 1;
} else if ((r__1 = e[i__], ABS(r__1)) >= eps) {
/* A subproblem with E(NM1) not too small but I = NM1. */
nsize = *n - start + 1;
} else {
/* A subproblem with E(NM1) small. This implies an */
/* 1-by-1 subproblem at D(N). Solve this 1-by-1 problem */
/* first. */
nsize = i__ - start + 1;
if (icompq == 2) {
u[*n + *n * u_dim1] = r_sign(&c_b15, &d__[*n]);
vt[*n + *n * vt_dim1] = 1.f;
} else if (icompq == 1) {
q[*n + (qstart - 1) * *n] = r_sign(&c_b15, &d__[*n]);
q[*n + (smlsiz + qstart - 1) * *n] = 1.f;
}
d__[*n] = (r__1 = d__[*n], ABS(r__1));
}
if (icompq == 2) {
slasd0_(&nsize, &sqre, &d__[start], &e[start], &u[start +
start * u_dim1], ldu, &vt[start + start * vt_dim1],
ldvt, &smlsiz, &iwork[1], &work[wstart], info);
} else {
slasda_(&icompq, &smlsiz, &nsize, &sqre, &d__[start], &e[
start], &q[start + (iu + qstart - 2) * *n], n, &q[
start + (ivt + qstart - 2) * *n], &iq[start + k * *n],
&q[start + (difl + qstart - 2) * *n], &q[start + (
difr + qstart - 2) * *n], &q[start + (z__ + qstart -
2) * *n], &q[start + (poles + qstart - 2) * *n], &iq[
start + givptr * *n], &iq[start + givcol * *n], n, &
iq[start + perm * *n], &q[start + (givnum + qstart -
2) * *n], &q[start + (ic + qstart - 2) * *n], &q[
start + (is + qstart - 2) * *n], &work[wstart], &
iwork[1], info);
if (*info != 0) {
return 0;
}
}
start = i__ + 1;
}
/* L30: */
}
/* Unscale */
slascl_("G", &c__0, &c__0, &c_b15, &orgnrm, n, &c__1, &d__[1], n, &ierr);
L40:
/* Use Selection Sort to minimize swaps of singular vectors */
i__1 = *n;
for (ii = 2; ii <= i__1; ++ii) {
i__ = ii - 1;
kk = i__;
p = d__[i__];
i__2 = *n;
for (j = ii; j <= i__2; ++j) {
if (d__[j] > p) {
kk = j;
p = d__[j];
}
/* L50: */
}
if (kk != i__) {
d__[kk] = d__[i__];
d__[i__] = p;
if (icompq == 1) {
iq[i__] = kk;
} else if (icompq == 2) {
sswap_(n, &u[i__ * u_dim1 + 1], &c__1, &u[kk * u_dim1 + 1], &
c__1);
sswap_(n, &vt[i__ + vt_dim1], ldvt, &vt[kk + vt_dim1], ldvt);
}
} else if (icompq == 1) {
iq[i__] = i__;
}
/* L60: */
}
/* If ICOMPQ = 1, use IQ(N,1) as the indicator for UPLO */
if (icompq == 1) {
if (iuplo == 1) {
iq[*n] = 1;
} else {
iq[*n] = 0;
}
}
/* If B is lower bidiagonal, update U by those Givens rotations */
/* which rotated B to be upper bidiagonal */
if (iuplo == 2 && icompq == 2) {
slasr_("L", "V", "B", n, n, &work[1], &work[*n], &u[u_offset], ldu);
}
return 0;
/* End of SBDSDC */
} /* sbdsdc_ */
/* Subroutine */ int slasq1_(integer *n, real *d__, real *e, real *work,
integer *info)
{
/* System generated locals */
integer i__1, i__2;
real r__1, r__2, r__3;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
integer i__;
real eps;
extern /* Subroutine */ int slas2_(real *, real *, real *, real *, real *)
;
real scale;
integer iinfo;
real sigmn, sigmx;
extern /* Subroutine */ int scopy_(integer *, real *, integer *, real *,
integer *), slasq2_(integer *, real *, integer *);
extern doublereal slamch_(char *);
real safmin;
extern /* Subroutine */ int xerbla_(char *, integer *), slascl_(
char *, integer *, integer *, real *, real *, integer *, integer *
, real *, integer *, integer *), slasrt_(char *, integer *
, real *, integer *);
/* -- LAPACK routine (version 3.2) -- */
/* -- Contributed by Osni Marques of the Lawrence Berkeley National -- */
/* -- Laboratory and Beresford Parlett of the Univ. of California at -- */
/* -- Berkeley -- */
/* -- November 2008 -- */
/* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
/* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SLASQ1 computes the singular values of a real N-by-N bidiagonal */
/* matrix with diagonal D and off-diagonal E. The singular values */
/* are computed to high relative accuracy, in the absence of */
/* denormalization, underflow and overflow. The algorithm was first */
/* presented in */
/* "Accurate singular values and differential qd algorithms" by K. V. */
/* Fernando and B. N. Parlett, Numer. Math., Vol-67, No. 2, pp. 191-230, */
/* 1994, */
/* and the present implementation is described in "An implementation of */
/* the dqds Algorithm (Positive Case)", LAPACK Working Note. */
/* Arguments */
/* ========= */
/* N (input) INTEGER */
/* The number of rows and columns in the matrix. N >= 0. */
/* D (input/output) REAL array, dimension (N) */
/* On entry, D contains the diagonal elements of the */
/* bidiagonal matrix whose SVD is desired. On normal exit, */
/* D contains the singular values in decreasing order. */
/* E (input/output) REAL array, dimension (N) */
/* On entry, elements E(1:N-1) contain the off-diagonal elements */
/* of the bidiagonal matrix whose SVD is desired. */
/* On exit, E is overwritten. */
/* WORK (workspace) REAL array, dimension (4*N) */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* > 0: the algorithm failed */
/* = 1, a split was marked by a positive value in E */
/* = 2, current block of Z not diagonalized after 30*N */
/* iterations (in inner while loop) */
/* = 3, termination criterion of outer while loop not met */
/* (program created more than N unreduced blocks) */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Parameter adjustments */
--work;
--e;
--d__;
/* Function Body */
*info = 0;
if (*n < 0) {
*info = -2;
i__1 = -(*info);
xerbla_("SLASQ1", &i__1);
return 0;
} else if (*n == 0) {
return 0;
} else if (*n == 1) {
d__[1] = dabs(d__[1]);
return 0;
} else if (*n == 2) {
slas2_(&d__[1], &e[1], &d__[2], &sigmn, &sigmx);
d__[1] = sigmx;
d__[2] = sigmn;
return 0;
}
/* Estimate the largest singular value. */
sigmx = 0.f;
i__1 = *n - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
d__[i__] = (r__1 = d__[i__], dabs(r__1));
/* Computing MAX */
r__2 = sigmx, r__3 = (r__1 = e[i__], dabs(r__1));
sigmx = dmax(r__2,r__3);
/* L10: */
}
d__[*n] = (r__1 = d__[*n], dabs(r__1));
/* Early return if SIGMX is zero (matrix is already diagonal). */
if (sigmx == 0.f) {
slasrt_("D", n, &d__[1], &iinfo);
return 0;
}
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
/* Computing MAX */
r__1 = sigmx, r__2 = d__[i__];
sigmx = dmax(r__1,r__2);
/* L20: */
}
/* Copy D and E into WORK (in the Z format) and scale (squaring the */
/* input data makes scaling by a power of the radix pointless). */
eps = slamch_("Precision");
safmin = slamch_("Safe minimum");
scale = sqrt(eps / safmin);
scopy_(n, &d__[1], &c__1, &work[1], &c__2);
i__1 = *n - 1;
scopy_(&i__1, &e[1], &c__1, &work[2], &c__2);
i__1 = (*n << 1) - 1;
i__2 = (*n << 1) - 1;
slascl_("G", &c__0, &c__0, &sigmx, &scale, &i__1, &c__1, &work[1], &i__2,
&iinfo);
/* Compute the q's and e's. */
i__1 = (*n << 1) - 1;
for (i__ = 1; i__ <= i__1; ++i__) {
/* Computing 2nd power */
r__1 = work[i__];
work[i__] = r__1 * r__1;
/* L30: */
}
work[*n * 2] = 0.f;
slasq2_(n, &work[1], info);
if (*info == 0) {
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
d__[i__] = sqrt(work[i__]);
/* L40: */
}
slascl_("G", &c__0, &c__0, &scale, &sigmx, n, &c__1, &d__[1], n, &
iinfo);
}
return 0;
/* End of SLASQ1 */
} /* slasq1_ */
/* Subroutine */ int sgeev_(char *jobvl, char *jobvr, integer *n, real *a,
integer *lda, real *wr, real *wi, real *vl, integer *ldvl, real *vr,
integer *ldvr, real *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;
real r__1, r__2;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
integer i__, k;
real r__, cs, sn;
integer ihi;
real scl;
integer ilo;
real dum[1], eps;
integer ibal;
char side[1];
real anrm;
integer ierr, itau, iwrk, nout;
extern /* Subroutine */ int srot_(integer *, real *, integer *, real *,
integer *, real *, real *);
extern doublereal snrm2_(integer *, real *, integer *);
extern logical lsame_(char *, char *);
extern /* Subroutine */ int sscal_(integer *, real *, real *, integer *);
extern doublereal slapy2_(real *, real *);
extern /* Subroutine */ int slabad_(real *, real *);
logical scalea;
real cscale;
extern /* Subroutine */ int sgebak_(char *, char *, integer *, integer *,
integer *, real *, integer *, real *, integer *, integer *), sgebal_(char *, integer *, real *, integer *,
integer *, integer *, real *, integer *);
extern doublereal slamch_(char *), slange_(char *, integer *,
integer *, real *, integer *, real *);
extern /* Subroutine */ int sgehrd_(integer *, integer *, integer *, real
*, integer *, real *, real *, integer *, integer *), xerbla_(char
*, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *);
logical select[1];
real bignum;
extern /* Subroutine */ int slascl_(char *, integer *, integer *, real *,
real *, integer *, integer *, real *, integer *, integer *);
extern integer isamax_(integer *, real *, integer *);
extern /* Subroutine */ int slacpy_(char *, integer *, integer *, real *,
integer *, real *, integer *), slartg_(real *, real *,
real *, real *, real *), sorghr_(integer *, integer *, integer *,
real *, integer *, real *, real *, integer *, integer *), shseqr_(
char *, char *, integer *, integer *, integer *, real *, integer *
, real *, real *, real *, integer *, real *, integer *, integer *), strevc_(char *, char *, logical *, integer *,
real *, integer *, real *, integer *, real *, integer *, integer *
, integer *, real *, integer *);
integer minwrk, maxwrk;
logical wantvl;
real smlnum;
integer hswork;
logical lquery, wantvr;
/* -- LAPACK driver routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SGEEV 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) REAL 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) REAL array, dimension (N) */
/* WI (output) REAL 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) REAL 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) REAL 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) REAL 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. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. Local Arrays .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* 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 SHSEQR, 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, "SGEHRD", " ", 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,
"SORGHR", " ", n, &c__1, n, &c_n1);
maxwrk = max(i__1,i__2);
shseqr_("S", "V", n, &c__1, n, &a[a_offset], lda, &wr[1], &wi[
1], &vl[vl_offset], ldvl, &work[1], &c_n1, info);
hswork = 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,
"SORGHR", " ", n, &c__1, n, &c_n1);
maxwrk = max(i__1,i__2);
shseqr_("S", "V", n, &c__1, n, &a[a_offset], lda, &wr[1], &wi[
1], &vr[vr_offset], ldvr, &work[1], &c_n1, info);
hswork = 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;
shseqr_("E", "N", n, &c__1, n, &a[a_offset], lda, &wr[1], &wi[
1], &vr[vr_offset], ldvr, &work[1], &c_n1, info);
hswork = 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] = (real) maxwrk;
if (*lwork < minwrk && ! lquery) {
*info = -13;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SGEEV ", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
/* Get machine constants */
eps = slamch_("P");
smlnum = slamch_("S");
bignum = 1.f / smlnum;
slabad_(&smlnum, &bignum);
smlnum = sqrt(smlnum) / eps;
bignum = 1.f / smlnum;
/* Scale A if max element outside range [SMLNUM,BIGNUM] */
anrm = slange_("M", n, n, &a[a_offset], lda, dum);
scalea = FALSE_;
if (anrm > 0.f && anrm < smlnum) {
scalea = TRUE_;
cscale = smlnum;
} else if (anrm > bignum) {
scalea = TRUE_;
cscale = bignum;
}
if (scalea) {
slascl_("G", &c__0, &c__0, &anrm, &cscale, n, n, &a[a_offset], lda, &
ierr);
}
/* Balance the matrix */
/* (Workspace: need N) */
ibal = 1;
sgebal_("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;
sgehrd_(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';
slacpy_("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;
sorghr_(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;
shseqr_("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';
slacpy_("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';
slacpy_("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;
sorghr_(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;
shseqr_("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;
shseqr_("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 SHSEQR, then quit */
if (*info > 0) {
goto L50;
}
if (wantvl || wantvr) {
/* Compute left and/or right eigenvectors */
/* (Workspace: need 4*N) */
strevc_(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) */
sgebak_("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.f) {
scl = 1.f / snrm2_(n, &vl[i__ * vl_dim1 + 1], &c__1);
sscal_(n, &scl, &vl[i__ * vl_dim1 + 1], &c__1);
} else if (wi[i__] > 0.f) {
r__1 = snrm2_(n, &vl[i__ * vl_dim1 + 1], &c__1);
r__2 = snrm2_(n, &vl[(i__ + 1) * vl_dim1 + 1], &c__1);
scl = 1.f / slapy2_(&r__1, &r__2);
sscal_(n, &scl, &vl[i__ * vl_dim1 + 1], &c__1);
sscal_(n, &scl, &vl[(i__ + 1) * vl_dim1 + 1], &c__1);
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
/* Computing 2nd power */
r__1 = vl[k + i__ * vl_dim1];
/* Computing 2nd power */
r__2 = vl[k + (i__ + 1) * vl_dim1];
work[iwrk + k - 1] = r__1 * r__1 + r__2 * r__2;
/* L10: */
}
k = isamax_(n, &work[iwrk], &c__1);
slartg_(&vl[k + i__ * vl_dim1], &vl[k + (i__ + 1) * vl_dim1],
&cs, &sn, &r__);
srot_(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.f;
}
/* L20: */
}
}
if (wantvr) {
/* Undo balancing of right eigenvectors */
/* (Workspace: need N) */
sgebak_("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.f) {
scl = 1.f / snrm2_(n, &vr[i__ * vr_dim1 + 1], &c__1);
sscal_(n, &scl, &vr[i__ * vr_dim1 + 1], &c__1);
} else if (wi[i__] > 0.f) {
r__1 = snrm2_(n, &vr[i__ * vr_dim1 + 1], &c__1);
r__2 = snrm2_(n, &vr[(i__ + 1) * vr_dim1 + 1], &c__1);
scl = 1.f / slapy2_(&r__1, &r__2);
sscal_(n, &scl, &vr[i__ * vr_dim1 + 1], &c__1);
sscal_(n, &scl, &vr[(i__ + 1) * vr_dim1 + 1], &c__1);
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
/* Computing 2nd power */
r__1 = vr[k + i__ * vr_dim1];
/* Computing 2nd power */
r__2 = vr[k + (i__ + 1) * vr_dim1];
work[iwrk + k - 1] = r__1 * r__1 + r__2 * r__2;
/* L30: */
}
k = isamax_(n, &work[iwrk], &c__1);
slartg_(&vr[k + i__ * vr_dim1], &vr[k + (i__ + 1) * vr_dim1],
&cs, &sn, &r__);
srot_(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.f;
}
/* L40: */
}
}
/* Undo scaling if necessary */
L50:
if (scalea) {
i__1 = *n - *info;
/* Computing MAX */
i__3 = *n - *info;
i__2 = max(i__3,1);
slascl_("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);
slascl_("G", &c__0, &c__0, &cscale, &anrm, &i__1, &c__1, &wi[*info +
1], &i__2, &ierr);
if (*info > 0) {
i__1 = ilo - 1;
slascl_("G", &c__0, &c__0, &cscale, &anrm, &i__1, &c__1, &wr[1],
n, &ierr);
i__1 = ilo - 1;
slascl_("G", &c__0, &c__0, &cscale, &anrm, &i__1, &c__1, &wi[1],
n, &ierr);
}
}
work[1] = (real) maxwrk;
return 0;
/* End of SGEEV */
} /* sgeev_ */
