/* Subroutine */ int stgex2_(logical *wantq, logical *wantz, integer *n, real
*a, integer *lda, real *b, integer *ldb, real *q, integer *ldq, real *
z__, integer *ldz, integer *j1, integer *n1, integer *n2, real *work,
integer *lwork, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, q_dim1, q_offset, z_dim1,
z_offset, i__1, i__2;
real r__1;
/* Builtin functions */
double sqrt(doublereal);
/* Local variables */
real f, g;
integer i__, m;
real s[16] /* was [4][4] */, t[16] /* was [4][4] */, be[2], ai[2], ar[2],
sa, sb, li[16] /* was [4][4] */, ir[16] /* was [4][4]
*/, ss, ws, eps;
logical weak;
real ddum;
integer idum;
real taul[4], dsum, taur[4], scpy[16] /* was [4][4] */, tcpy[16]
/* was [4][4] */;
extern /* Subroutine */ int srot_(integer *, real *, integer *, real *,
integer *, real *, real *);
real scale, bqra21, brqa21;
extern /* Subroutine */ int sscal_(integer *, real *, real *, integer *);
real licop[16] /* was [4][4] */;
integer linfo;
extern /* Subroutine */ int sgemm_(char *, char *, integer *, integer *,
integer *, real *, real *, integer *, real *, integer *, real *,
real *, integer *);
real ircop[16] /* was [4][4] */, dnorm;
integer iwork[4];
extern /* Subroutine */ int slagv2_(real *, integer *, real *, integer *,
real *, real *, real *, real *, real *, real *, real *), sgeqr2_(
integer *, integer *, real *, integer *, real *, real *, integer *
), sgerq2_(integer *, integer *, real *, integer *, real *, real *
, integer *), sorg2r_(integer *, integer *, integer *, real *,
integer *, real *, real *, integer *), sorgr2_(integer *, integer
*, integer *, real *, integer *, real *, real *, integer *),
sorm2r_(char *, char *, integer *, integer *, integer *, real *,
integer *, real *, real *, integer *, real *, integer *), sormr2_(char *, char *, integer *, integer *, integer *,
real *, integer *, real *, real *, integer *, real *, integer *);
real dscale;
extern /* Subroutine */ int stgsy2_(char *, integer *, integer *, integer
*, real *, integer *, real *, integer *, real *, integer *, real *
, integer *, real *, integer *, real *, integer *, real *, real *,
real *, integer *, integer *, integer *);
extern doublereal slamch_(char *);
extern /* Subroutine */ int slacpy_(char *, integer *, integer *, real *,
integer *, real *, integer *), slartg_(real *, real *,
real *, real *, real *);
real thresh;
extern /* Subroutine */ int slaset_(char *, integer *, integer *, real *,
real *, real *, integer *), slassq_(integer *, real *,
integer *, real *, real *);
real smlnum;
logical strong;
/* -- LAPACK auxiliary routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* STGEX2 swaps adjacent diagonal blocks (A11, B11) and (A22, B22) */
/* of size 1-by-1 or 2-by-2 in an upper (quasi) triangular matrix pair */
/* (A, B) by an orthogonal equivalence transformation. */
/* (A, B) must be in generalized real Schur canonical form (as returned */
/* by SGGES), i.e. A is block upper triangular with 1-by-1 and 2-by-2 */
/* diagonal blocks. B is upper triangular. */
/* Optionally, the matrices Q and Z of generalized Schur vectors are */
/* updated. */
/* Q(in) * A(in) * Z(in)' = Q(out) * A(out) * Z(out)' */
/* Q(in) * B(in) * Z(in)' = Q(out) * B(out) * Z(out)' */
/* Arguments */
/* ========= */
/* WANTQ (input) LOGICAL */
/* .TRUE. : update the left transformation matrix Q; */
/* .FALSE.: do not update Q. */
/* WANTZ (input) LOGICAL */
/* .TRUE. : update the right transformation matrix Z; */
/* .FALSE.: do not update Z. */
/* N (input) INTEGER */
/* The order of the matrices A and B. N >= 0. */
/* A (input/output) REAL arrays, dimensions (LDA,N) */
/* On entry, the matrix A in the pair (A, B). */
/* On exit, the updated matrix A. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1,N). */
/* B (input/output) REAL arrays, dimensions (LDB,N) */
/* On entry, the matrix B in the pair (A, B). */
/* On exit, the updated matrix B. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= max(1,N). */
/* Q (input/output) REAL array, dimension (LDZ,N) */
/* On entry, if WANTQ = .TRUE., the orthogonal matrix Q. */
/* On exit, the updated matrix Q. */
/* Not referenced if WANTQ = .FALSE.. */
/* LDQ (input) INTEGER */
/* The leading dimension of the array Q. LDQ >= 1. */
/* If WANTQ = .TRUE., LDQ >= N. */
/* Z (input/output) REAL array, dimension (LDZ,N) */
/* On entry, if WANTZ =.TRUE., the orthogonal matrix Z. */
/* On exit, the updated matrix Z. */
/* Not referenced if WANTZ = .FALSE.. */
/* LDZ (input) INTEGER */
/* The leading dimension of the array Z. LDZ >= 1. */
/* If WANTZ = .TRUE., LDZ >= N. */
/* J1 (input) INTEGER */
/* The index to the first block (A11, B11). 1 <= J1 <= N. */
/* N1 (input) INTEGER */
/* The order of the first block (A11, B11). N1 = 0, 1 or 2. */
/* N2 (input) INTEGER */
/* The order of the second block (A22, B22). N2 = 0, 1 or 2. */
/* WORK (workspace) REAL array, dimension (MAX(1,LWORK)). */
/* LWORK (input) INTEGER */
/* The dimension of the array WORK. */
/* LWORK >= MAX( N*(N2+N1), (N2+N1)*(N2+N1)*2 ) */
/* INFO (output) INTEGER */
/* =0: Successful exit */
/* >0: If INFO = 1, the transformed matrix (A, B) would be */
/* too far from generalized Schur form; the blocks are */
/* not swapped and (A, B) and (Q, Z) are unchanged. */
/* The problem of swapping is too ill-conditioned. */
/* <0: If INFO = -16: LWORK is too small. Appropriate value */
/* for LWORK is returned in WORK(1). */
/* Further Details */
/* =============== */
/* Based on contributions by */
/* Bo Kagstrom and Peter Poromaa, Department of Computing Science, */
/* Umea University, S-901 87 Umea, Sweden. */
/* In the current code both weak and strong stability tests are */
/* performed. The user can omit the strong stability test by changing */
/* the internal logical parameter WANDS to .FALSE.. See ref. [2] for */
/* details. */
/* [1] B. Kagstrom; A Direct Method for Reordering Eigenvalues in the */
/* Generalized Real Schur Form of a Regular Matrix Pair (A, B), in */
/* M.S. Moonen et al (eds), Linear Algebra for Large Scale and */
/* Real-Time Applications, Kluwer Academic Publ. 1993, pp 195-218. */
/* [2] B. Kagstrom and P. Poromaa; Computing Eigenspaces with Specified */
/* Eigenvalues of a Regular Matrix Pair (A, B) and Condition */
/* Estimation: Theory, Algorithms and Software, */
/* Report UMINF - 94.04, Department of Computing Science, Umea */
/* University, S-901 87 Umea, Sweden, 1994. Also as LAPACK Working */
/* Note 87. To appear in Numerical Algorithms, 1996. */
/* ===================================================================== */
/* Replaced various illegal calls to SCOPY by calls to SLASET, or by DO */
/* loops. Sven Hammarling, 1/5/02. */
/* .. 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;
q_dim1 = *ldq;
q_offset = 1 + q_dim1;
q -= q_offset;
z_dim1 = *ldz;
z_offset = 1 + z_dim1;
z__ -= z_offset;
--work;
/* Function Body */
*info = 0;
/* Quick return if possible */
if (*n <= 1 || *n1 <= 0 || *n2 <= 0) {
return 0;
}
if (*n1 > *n || *j1 + *n1 > *n) {
return 0;
}
m = *n1 + *n2;
/* Computing MAX */
i__1 = *n * m, i__2 = m * m << 1;
if (*lwork < max(i__1,i__2)) {
*info = -16;
/* Computing MAX */
i__1 = *n * m, i__2 = m * m << 1;
work[1] = (real) max(i__1,i__2);
return 0;
}
weak = FALSE_;
strong = FALSE_;
/* Make a local copy of selected block */
slaset_("Full", &c__4, &c__4, &c_b5, &c_b5, li, &c__4);
slaset_("Full", &c__4, &c__4, &c_b5, &c_b5, ir, &c__4);
slacpy_("Full", &m, &m, &a[*j1 + *j1 * a_dim1], lda, s, &c__4);
slacpy_("Full", &m, &m, &b[*j1 + *j1 * b_dim1], ldb, t, &c__4);
/* Compute threshold for testing acceptance of swapping. */
eps = slamch_("P");
smlnum = slamch_("S") / eps;
dscale = 0.f;
dsum = 1.f;
slacpy_("Full", &m, &m, s, &c__4, &work[1], &m);
i__1 = m * m;
slassq_(&i__1, &work[1], &c__1, &dscale, &dsum);
slacpy_("Full", &m, &m, t, &c__4, &work[1], &m);
i__1 = m * m;
slassq_(&i__1, &work[1], &c__1, &dscale, &dsum);
dnorm = dscale * sqrt(dsum);
/* Computing MAX */
r__1 = eps * 10.f * dnorm;
thresh = dmax(r__1,smlnum);
if (m == 2) {
/* CASE 1: Swap 1-by-1 and 1-by-1 blocks. */
/* Compute orthogonal QL and RQ that swap 1-by-1 and 1-by-1 blocks */
/* using Givens rotations and perform the swap tentatively. */
f = s[5] * t[0] - t[5] * s[0];
g = s[5] * t[4] - t[5] * s[4];
sb = dabs(t[5]);
sa = dabs(s[5]);
slartg_(&f, &g, &ir[4], ir, &ddum);
ir[1] = -ir[4];
ir[5] = ir[0];
srot_(&c__2, s, &c__1, &s[4], &c__1, ir, &ir[1]);
srot_(&c__2, t, &c__1, &t[4], &c__1, ir, &ir[1]);
if (sa >= sb) {
slartg_(s, &s[1], li, &li[1], &ddum);
} else {
slartg_(t, &t[1], li, &li[1], &ddum);
}
srot_(&c__2, s, &c__4, &s[1], &c__4, li, &li[1]);
srot_(&c__2, t, &c__4, &t[1], &c__4, li, &li[1]);
li[5] = li[0];
li[4] = -li[1];
/* Weak stability test: */
/* |S21| + |T21| <= O(EPS * F-norm((S, T))) */
ws = dabs(s[1]) + dabs(t[1]);
weak = ws <= thresh;
if (! weak) {
goto L70;
}
if (TRUE_) {
/* Strong stability test: */
/* F-norm((A-QL'*S*QR, B-QL'*T*QR)) <= O(EPS*F-norm((A,B))) */
slacpy_("Full", &m, &m, &a[*j1 + *j1 * a_dim1], lda, &work[m * m
+ 1], &m);
sgemm_("N", "N", &m, &m, &m, &c_b42, li, &c__4, s, &c__4, &c_b5, &
work[1], &m);
sgemm_("N", "T", &m, &m, &m, &c_b48, &work[1], &m, ir, &c__4, &
c_b42, &work[m * m + 1], &m);
dscale = 0.f;
dsum = 1.f;
i__1 = m * m;
slassq_(&i__1, &work[m * m + 1], &c__1, &dscale, &dsum);
slacpy_("Full", &m, &m, &b[*j1 + *j1 * b_dim1], ldb, &work[m * m
+ 1], &m);
sgemm_("N", "N", &m, &m, &m, &c_b42, li, &c__4, t, &c__4, &c_b5, &
work[1], &m);
sgemm_("N", "T", &m, &m, &m, &c_b48, &work[1], &m, ir, &c__4, &
c_b42, &work[m * m + 1], &m);
i__1 = m * m;
slassq_(&i__1, &work[m * m + 1], &c__1, &dscale, &dsum);
ss = dscale * sqrt(dsum);
strong = ss <= thresh;
if (! strong) {
goto L70;
}
}
/* Update (A(J1:J1+M-1, M+J1:N), B(J1:J1+M-1, M+J1:N)) and */
/* (A(1:J1-1, J1:J1+M), B(1:J1-1, J1:J1+M)). */
i__1 = *j1 + 1;
srot_(&i__1, &a[*j1 * a_dim1 + 1], &c__1, &a[(*j1 + 1) * a_dim1 + 1],
&c__1, ir, &ir[1]);
i__1 = *j1 + 1;
srot_(&i__1, &b[*j1 * b_dim1 + 1], &c__1, &b[(*j1 + 1) * b_dim1 + 1],
&c__1, ir, &ir[1]);
i__1 = *n - *j1 + 1;
srot_(&i__1, &a[*j1 + *j1 * a_dim1], lda, &a[*j1 + 1 + *j1 * a_dim1],
lda, li, &li[1]);
i__1 = *n - *j1 + 1;
srot_(&i__1, &b[*j1 + *j1 * b_dim1], ldb, &b[*j1 + 1 + *j1 * b_dim1],
ldb, li, &li[1]);
/* Set N1-by-N2 (2,1) - blocks to ZERO. */
a[*j1 + 1 + *j1 * a_dim1] = 0.f;
b[*j1 + 1 + *j1 * b_dim1] = 0.f;
/* Accumulate transformations into Q and Z if requested. */
if (*wantz) {
srot_(n, &z__[*j1 * z_dim1 + 1], &c__1, &z__[(*j1 + 1) * z_dim1 +
1], &c__1, ir, &ir[1]);
}
if (*wantq) {
srot_(n, &q[*j1 * q_dim1 + 1], &c__1, &q[(*j1 + 1) * q_dim1 + 1],
&c__1, li, &li[1]);
}
/* Exit with INFO = 0 if swap was successfully performed. */
return 0;
} else {
/* CASE 2: Swap 1-by-1 and 2-by-2 blocks, or 2-by-2 */
/* and 2-by-2 blocks. */
/* Solve the generalized Sylvester equation */
/* S11 * R - L * S22 = SCALE * S12 */
/* T11 * R - L * T22 = SCALE * T12 */
/* for R and L. Solutions in LI and IR. */
slacpy_("Full", n1, n2, &t[(*n1 + 1 << 2) - 4], &c__4, li, &c__4);
slacpy_("Full", n1, n2, &s[(*n1 + 1 << 2) - 4], &c__4, &ir[*n2 + 1 + (
*n1 + 1 << 2) - 5], &c__4);
stgsy2_("N", &c__0, n1, n2, s, &c__4, &s[*n1 + 1 + (*n1 + 1 << 2) - 5]
, &c__4, &ir[*n2 + 1 + (*n1 + 1 << 2) - 5], &c__4, t, &c__4, &
t[*n1 + 1 + (*n1 + 1 << 2) - 5], &c__4, li, &c__4, &scale, &
dsum, &dscale, iwork, &idum, &linfo);
/* Compute orthogonal matrix QL: */
/* QL' * LI = [ TL ] */
/* [ 0 ] */
/* where */
/* LI = [ -L ] */
/* [ SCALE * identity(N2) ] */
i__1 = *n2;
for (i__ = 1; i__ <= i__1; ++i__) {
sscal_(n1, &c_b48, &li[(i__ << 2) - 4], &c__1);
li[*n1 + i__ + (i__ << 2) - 5] = scale;
/* L10: */
}
sgeqr2_(&m, n2, li, &c__4, taul, &work[1], &linfo);
if (linfo != 0) {
goto L70;
}
sorg2r_(&m, &m, n2, li, &c__4, taul, &work[1], &linfo);
if (linfo != 0) {
goto L70;
}
/* Compute orthogonal matrix RQ: */
/* IR * RQ' = [ 0 TR], */
/* where IR = [ SCALE * identity(N1), R ] */
i__1 = *n1;
for (i__ = 1; i__ <= i__1; ++i__) {
ir[*n2 + i__ + (i__ << 2) - 5] = scale;
/* L20: */
}
sgerq2_(n1, &m, &ir[*n2], &c__4, taur, &work[1], &linfo);
if (linfo != 0) {
goto L70;
}
sorgr2_(&m, &m, n1, ir, &c__4, taur, &work[1], &linfo);
if (linfo != 0) {
goto L70;
}
/* Perform the swapping tentatively: */
sgemm_("T", "N", &m, &m, &m, &c_b42, li, &c__4, s, &c__4, &c_b5, &
work[1], &m);
sgemm_("N", "T", &m, &m, &m, &c_b42, &work[1], &m, ir, &c__4, &c_b5,
s, &c__4);
sgemm_("T", "N", &m, &m, &m, &c_b42, li, &c__4, t, &c__4, &c_b5, &
work[1], &m);
sgemm_("N", "T", &m, &m, &m, &c_b42, &work[1], &m, ir, &c__4, &c_b5,
t, &c__4);
slacpy_("F", &m, &m, s, &c__4, scpy, &c__4);
slacpy_("F", &m, &m, t, &c__4, tcpy, &c__4);
slacpy_("F", &m, &m, ir, &c__4, ircop, &c__4);
slacpy_("F", &m, &m, li, &c__4, licop, &c__4);
/* Triangularize the B-part by an RQ factorization. */
/* Apply transformation (from left) to A-part, giving S. */
sgerq2_(&m, &m, t, &c__4, taur, &work[1], &linfo);
if (linfo != 0) {
goto L70;
}
sormr2_("R", "T", &m, &m, &m, t, &c__4, taur, s, &c__4, &work[1], &
linfo);
if (linfo != 0) {
goto L70;
}
sormr2_("L", "N", &m, &m, &m, t, &c__4, taur, ir, &c__4, &work[1], &
linfo);
if (linfo != 0) {
goto L70;
}
/* Compute F-norm(S21) in BRQA21. (T21 is 0.) */
dscale = 0.f;
dsum = 1.f;
i__1 = *n2;
for (i__ = 1; i__ <= i__1; ++i__) {
slassq_(n1, &s[*n2 + 1 + (i__ << 2) - 5], &c__1, &dscale, &dsum);
/* L30: */
}
brqa21 = dscale * sqrt(dsum);
/* Triangularize the B-part by a QR factorization. */
/* Apply transformation (from right) to A-part, giving S. */
sgeqr2_(&m, &m, tcpy, &c__4, taul, &work[1], &linfo);
if (linfo != 0) {
goto L70;
}
sorm2r_("L", "T", &m, &m, &m, tcpy, &c__4, taul, scpy, &c__4, &work[1]
, info);
sorm2r_("R", "N", &m, &m, &m, tcpy, &c__4, taul, licop, &c__4, &work[
1], info);
if (linfo != 0) {
goto L70;
}
/* Compute F-norm(S21) in BQRA21. (T21 is 0.) */
dscale = 0.f;
dsum = 1.f;
i__1 = *n2;
for (i__ = 1; i__ <= i__1; ++i__) {
slassq_(n1, &scpy[*n2 + 1 + (i__ << 2) - 5], &c__1, &dscale, &
dsum);
/* L40: */
}
bqra21 = dscale * sqrt(dsum);
/* Decide which method to use. */
/* Weak stability test: */
/* F-norm(S21) <= O(EPS * F-norm((S, T))) */
if (bqra21 <= brqa21 && bqra21 <= thresh) {
slacpy_("F", &m, &m, scpy, &c__4, s, &c__4);
slacpy_("F", &m, &m, tcpy, &c__4, t, &c__4);
slacpy_("F", &m, &m, ircop, &c__4, ir, &c__4);
slacpy_("F", &m, &m, licop, &c__4, li, &c__4);
} else if (brqa21 >= thresh) {
goto L70;
}
/* Set lower triangle of B-part to zero */
i__1 = m - 1;
i__2 = m - 1;
slaset_("Lower", &i__1, &i__2, &c_b5, &c_b5, &t[1], &c__4);
if (TRUE_) {
/* Strong stability test: */
/* F-norm((A-QL*S*QR', B-QL*T*QR')) <= O(EPS*F-norm((A,B))) */
slacpy_("Full", &m, &m, &a[*j1 + *j1 * a_dim1], lda, &work[m * m
+ 1], &m);
sgemm_("N", "N", &m, &m, &m, &c_b42, li, &c__4, s, &c__4, &c_b5, &
work[1], &m);
sgemm_("N", "N", &m, &m, &m, &c_b48, &work[1], &m, ir, &c__4, &
c_b42, &work[m * m + 1], &m);
dscale = 0.f;
dsum = 1.f;
i__1 = m * m;
slassq_(&i__1, &work[m * m + 1], &c__1, &dscale, &dsum);
slacpy_("Full", &m, &m, &b[*j1 + *j1 * b_dim1], ldb, &work[m * m
+ 1], &m);
sgemm_("N", "N", &m, &m, &m, &c_b42, li, &c__4, t, &c__4, &c_b5, &
work[1], &m);
sgemm_("N", "N", &m, &m, &m, &c_b48, &work[1], &m, ir, &c__4, &
c_b42, &work[m * m + 1], &m);
i__1 = m * m;
slassq_(&i__1, &work[m * m + 1], &c__1, &dscale, &dsum);
ss = dscale * sqrt(dsum);
strong = ss <= thresh;
if (! strong) {
goto L70;
}
}
/* If the swap is accepted ("weakly" and "strongly"), apply the */
/* transformations and set N1-by-N2 (2,1)-block to zero. */
slaset_("Full", n1, n2, &c_b5, &c_b5, &s[*n2], &c__4);
/* copy back M-by-M diagonal block starting at index J1 of (A, B) */
slacpy_("F", &m, &m, s, &c__4, &a[*j1 + *j1 * a_dim1], lda)
;
slacpy_("F", &m, &m, t, &c__4, &b[*j1 + *j1 * b_dim1], ldb)
;
slaset_("Full", &c__4, &c__4, &c_b5, &c_b5, t, &c__4);
/* Standardize existing 2-by-2 blocks. */
i__1 = m * m;
for (i__ = 1; i__ <= i__1; ++i__) {
work[i__] = 0.f;
/* L50: */
}
work[1] = 1.f;
t[0] = 1.f;
idum = *lwork - m * m - 2;
if (*n2 > 1) {
slagv2_(&a[*j1 + *j1 * a_dim1], lda, &b[*j1 + *j1 * b_dim1], ldb,
ar, ai, be, &work[1], &work[2], t, &t[1]);
work[m + 1] = -work[2];
work[m + 2] = work[1];
t[*n2 + (*n2 << 2) - 5] = t[0];
t[4] = -t[1];
}
work[m * m] = 1.f;
t[m + (m << 2) - 5] = 1.f;
if (*n1 > 1) {
slagv2_(&a[*j1 + *n2 + (*j1 + *n2) * a_dim1], lda, &b[*j1 + *n2 +
(*j1 + *n2) * b_dim1], ldb, taur, taul, &work[m * m + 1],
&work[*n2 * m + *n2 + 1], &work[*n2 * m + *n2 + 2], &t[*
n2 + 1 + (*n2 + 1 << 2) - 5], &t[m + (m - 1 << 2) - 5]);
work[m * m] = work[*n2 * m + *n2 + 1];
work[m * m - 1] = -work[*n2 * m + *n2 + 2];
t[m + (m << 2) - 5] = t[*n2 + 1 + (*n2 + 1 << 2) - 5];
t[m - 1 + (m << 2) - 5] = -t[m + (m - 1 << 2) - 5];
}
sgemm_("T", "N", n2, n1, n2, &c_b42, &work[1], &m, &a[*j1 + (*j1 + *
n2) * a_dim1], lda, &c_b5, &work[m * m + 1], n2);
slacpy_("Full", n2, n1, &work[m * m + 1], n2, &a[*j1 + (*j1 + *n2) *
a_dim1], lda);
sgemm_("T", "N", n2, n1, n2, &c_b42, &work[1], &m, &b[*j1 + (*j1 + *
n2) * b_dim1], ldb, &c_b5, &work[m * m + 1], n2);
slacpy_("Full", n2, n1, &work[m * m + 1], n2, &b[*j1 + (*j1 + *n2) *
b_dim1], ldb);
sgemm_("N", "N", &m, &m, &m, &c_b42, li, &c__4, &work[1], &m, &c_b5, &
work[m * m + 1], &m);
slacpy_("Full", &m, &m, &work[m * m + 1], &m, li, &c__4);
sgemm_("N", "N", n2, n1, n1, &c_b42, &a[*j1 + (*j1 + *n2) * a_dim1],
lda, &t[*n2 + 1 + (*n2 + 1 << 2) - 5], &c__4, &c_b5, &work[1],
n2);
slacpy_("Full", n2, n1, &work[1], n2, &a[*j1 + (*j1 + *n2) * a_dim1],
lda);
sgemm_("N", "N", n2, n1, n1, &c_b42, &b[*j1 + (*j1 + *n2) * b_dim1],
ldb, &t[*n2 + 1 + (*n2 + 1 << 2) - 5], &c__4, &c_b5, &work[1],
n2);
slacpy_("Full", n2, n1, &work[1], n2, &b[*j1 + (*j1 + *n2) * b_dim1],
ldb);
sgemm_("T", "N", &m, &m, &m, &c_b42, ir, &c__4, t, &c__4, &c_b5, &
work[1], &m);
slacpy_("Full", &m, &m, &work[1], &m, ir, &c__4);
/* Accumulate transformations into Q and Z if requested. */
if (*wantq) {
sgemm_("N", "N", n, &m, &m, &c_b42, &q[*j1 * q_dim1 + 1], ldq, li,
&c__4, &c_b5, &work[1], n);
slacpy_("Full", n, &m, &work[1], n, &q[*j1 * q_dim1 + 1], ldq);
}
if (*wantz) {
sgemm_("N", "N", n, &m, &m, &c_b42, &z__[*j1 * z_dim1 + 1], ldz,
ir, &c__4, &c_b5, &work[1], n);
slacpy_("Full", n, &m, &work[1], n, &z__[*j1 * z_dim1 + 1], ldz);
}
/* Update (A(J1:J1+M-1, M+J1:N), B(J1:J1+M-1, M+J1:N)) and */
/* (A(1:J1-1, J1:J1+M), B(1:J1-1, J1:J1+M)). */
i__ = *j1 + m;
if (i__ <= *n) {
i__1 = *n - i__ + 1;
sgemm_("T", "N", &m, &i__1, &m, &c_b42, li, &c__4, &a[*j1 + i__ *
a_dim1], lda, &c_b5, &work[1], &m);
i__1 = *n - i__ + 1;
slacpy_("Full", &m, &i__1, &work[1], &m, &a[*j1 + i__ * a_dim1],
lda);
i__1 = *n - i__ + 1;
sgemm_("T", "N", &m, &i__1, &m, &c_b42, li, &c__4, &b[*j1 + i__ *
b_dim1], ldb, &c_b5, &work[1], &m);
i__1 = *n - i__ + 1;
slacpy_("Full", &m, &i__1, &work[1], &m, &b[*j1 + i__ * b_dim1],
ldb);
}
i__ = *j1 - 1;
if (i__ > 0) {
sgemm_("N", "N", &i__, &m, &m, &c_b42, &a[*j1 * a_dim1 + 1], lda,
ir, &c__4, &c_b5, &work[1], &i__);
slacpy_("Full", &i__, &m, &work[1], &i__, &a[*j1 * a_dim1 + 1],
lda);
sgemm_("N", "N", &i__, &m, &m, &c_b42, &b[*j1 * b_dim1 + 1], ldb,
ir, &c__4, &c_b5, &work[1], &i__);
slacpy_("Full", &i__, &m, &work[1], &i__, &b[*j1 * b_dim1 + 1],
ldb);
}
/* Exit with INFO = 0 if swap was successfully performed. */
return 0;
}
/* Exit with INFO = 1 if swap was rejected. */
L70:
*info = 1;
return 0;
/* End of STGEX2 */
} /* stgex2_ */
/* Subroutine */ int stgex2_(logical *wantq, logical *wantz, integer *n, real
*a, integer *lda, real *b, integer *ldb, real *q, integer *ldq, real *
z__, integer *ldz, integer *j1, integer *n1, integer *n2, real *work,
integer *lwork, integer *info)
{
/* -- LAPACK auxiliary routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
June 30, 1999
Purpose
=======
STGEX2 swaps adjacent diagonal blocks (A11, B11) and (A22, B22)
of size 1-by-1 or 2-by-2 in an upper (quasi) triangular matrix pair
(A, B) by an orthogonal equivalence transformation.
(A, B) must be in generalized real Schur canonical form (as returned
by SGGES), i.e. A is block upper triangular with 1-by-1 and 2-by-2
diagonal blocks. B is upper triangular.
Optionally, the matrices Q and Z of generalized Schur vectors are
updated.
Q(in) * A(in) * Z(in)' = Q(out) * A(out) * Z(out)'
Q(in) * B(in) * Z(in)' = Q(out) * B(out) * Z(out)'
Arguments
=========
WANTQ (input) LOGICAL
.TRUE. : update the left transformation matrix Q;
.FALSE.: do not update Q.
WANTZ (input) LOGICAL
.TRUE. : update the right transformation matrix Z;
.FALSE.: do not update Z.
N (input) INTEGER
The order of the matrices A and B. N >= 0.
A (input/output) REAL arrays, dimensions (LDA,N)
On entry, the matrix A in the pair (A, B).
On exit, the updated matrix A.
LDA (input) INTEGER
The leading dimension of the array A. LDA >= max(1,N).
B (input/output) REAL arrays, dimensions (LDB,N)
On entry, the matrix B in the pair (A, B).
On exit, the updated matrix B.
LDB (input) INTEGER
The leading dimension of the array B. LDB >= max(1,N).
Q (input/output) REAL array, dimension (LDZ,N)
On entry, if WANTQ = .TRUE., the orthogonal matrix Q.
On exit, the updated matrix Q.
Not referenced if WANTQ = .FALSE..
LDQ (input) INTEGER
The leading dimension of the array Q. LDQ >= 1.
If WANTQ = .TRUE., LDQ >= N.
Z (input/output) REAL array, dimension (LDZ,N)
On entry, if WANTZ =.TRUE., the orthogonal matrix Z.
On exit, the updated matrix Z.
Not referenced if WANTZ = .FALSE..
LDZ (input) INTEGER
The leading dimension of the array Z. LDZ >= 1.
If WANTZ = .TRUE., LDZ >= N.
J1 (input) INTEGER
The index to the first block (A11, B11). 1 <= J1 <= N.
N1 (input) INTEGER
The order of the first block (A11, B11). N1 = 0, 1 or 2.
N2 (input) INTEGER
The order of the second block (A22, B22). N2 = 0, 1 or 2.
WORK (workspace) REAL array, dimension (LWORK).
LWORK (input) INTEGER
The dimension of the array WORK.
LWORK >= MAX( N*(N2+N1), (N2+N1)*(N2+N1)*2 )
INFO (output) INTEGER
=0: Successful exit
>0: If INFO = 1, the transformed matrix (A, B) would be
too far from generalized Schur form; the blocks are
not swapped and (A, B) and (Q, Z) are unchanged.
The problem of swapping is too ill-conditioned.
<0: If INFO = -16: LWORK is too small. Appropriate value
for LWORK is returned in WORK(1).
Further Details
===============
Based on contributions by
Bo Kagstrom and Peter Poromaa, Department of Computing Science,
Umea University, S-901 87 Umea, Sweden.
In the current code both weak and strong stability tests are
performed. The user can omit the strong stability test by changing
the internal logical parameter WANDS to .FALSE.. See ref. [2] for
details.
[1] B. Kagstrom; A Direct Method for Reordering Eigenvalues in the
Generalized Real Schur Form of a Regular Matrix Pair (A, B), in
M.S. Moonen et al (eds), Linear Algebra for Large Scale and
Real-Time Applications, Kluwer Academic Publ. 1993, pp 195-218.
[2] B. Kagstrom and P. Poromaa; Computing Eigenspaces with Specified
Eigenvalues of a Regular Matrix Pair (A, B) and Condition
Estimation: Theory, Algorithms and Software,
Report UMINF - 94.04, Department of Computing Science, Umea
University, S-901 87 Umea, Sweden, 1994. Also as LAPACK Working
Note 87. To appear in Numerical Algorithms, 1996.
=====================================================================
Parameter adjustments */
/* Table of constant values */
static integer c__16 = 16;
static real c_b3 = 0.f;
static integer c__0 = 0;
static integer c__1 = 1;
static integer c__4 = 4;
static integer c__2 = 2;
static real c_b38 = 1.f;
static real c_b44 = -1.f;
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_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 */
static logical weak;
static real ddum;
static integer idum;
static real taul[4], dsum, taur[4], scpy[16] /* was [4][4] */,
tcpy[16] /* was [4][4] */;
extern /* Subroutine */ int srot_(integer *, real *, integer *, real *,
integer *, real *, real *);
static real f, g;
static integer i__, m;
static real s[16] /* was [4][4] */, t[16] /* was [4][4] */, scale,
bqra21, brqa21;
extern /* Subroutine */ int sscal_(integer *, real *, real *, integer *);
static real licop[16] /* was [4][4] */;
static integer linfo;
extern /* Subroutine */ int sgemm_(char *, char *, integer *, integer *,
integer *, real *, real *, integer *, real *, integer *, real *,
real *, integer *);
static real ircop[16] /* was [4][4] */, dnorm;
static integer iwork[4];
extern /* Subroutine */ int scopy_(integer *, real *, integer *, real *,
integer *), slagv2_(real *, integer *, real *, integer *, real *,
real *, real *, real *, real *, real *, real *), sgeqr2_(integer *
, integer *, real *, integer *, real *, real *, integer *),
sgerq2_(integer *, integer *, real *, integer *, real *, real *,
integer *);
static real be[2], ai[2];
extern /* Subroutine */ int sorg2r_(integer *, integer *, integer *, real
*, integer *, real *, real *, integer *), sorgr2_(integer *,
integer *, integer *, real *, integer *, real *, real *, integer *
);
static real ar[2], sa, sb, li[16] /* was [4][4] */;
extern /* Subroutine */ int sorm2r_(char *, char *, integer *, integer *,
integer *, real *, integer *, real *, real *, integer *, real *,
integer *), sormr2_(char *, char *, integer *,
integer *, integer *, real *, integer *, real *, real *, integer *
, real *, integer *);
static real dscale, ir[16] /* was [4][4] */;
extern /* Subroutine */ int stgsy2_(char *, integer *, integer *, integer
*, real *, integer *, real *, integer *, real *, integer *, real *
, integer *, real *, integer *, real *, integer *, real *, real *,
real *, integer *, integer *, integer *);
static real ss;
extern doublereal slamch_(char *);
static real ws;
extern /* Subroutine */ int slacpy_(char *, integer *, integer *, real *,
integer *, real *, integer *), slartg_(real *, real *,
real *, real *, real *);
static real thresh;
extern /* Subroutine */ int slassq_(integer *, real *, integer *, real *,
real *);
static real smlnum;
static logical strong;
static real eps;
#define scpy_ref(a_1,a_2) scpy[(a_2)*4 + a_1 - 5]
#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 q_ref(a_1,a_2) q[(a_2)*q_dim1 + a_1]
#define s_ref(a_1,a_2) s[(a_2)*4 + a_1 - 5]
#define t_ref(a_1,a_2) t[(a_2)*4 + a_1 - 5]
#define z___ref(a_1,a_2) z__[(a_2)*z_dim1 + a_1]
#define li_ref(a_1,a_2) li[(a_2)*4 + a_1 - 5]
#define ir_ref(a_1,a_2) ir[(a_2)*4 + a_1 - 5]
a_dim1 = *lda;
a_offset = 1 + a_dim1 * 1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1 * 1;
b -= b_offset;
q_dim1 = *ldq;
q_offset = 1 + q_dim1 * 1;
q -= q_offset;
z_dim1 = *ldz;
z_offset = 1 + z_dim1 * 1;
z__ -= z_offset;
--work;
/* Function Body */
*info = 0;
/* Quick return if possible */
if (*n <= 1 || *n1 <= 0 || *n2 <= 0) {
return 0;
}
if (*n1 > *n || *j1 + *n1 > *n) {
return 0;
}
m = *n1 + *n2;
/* Computing MAX */
i__1 = *n * m, i__2 = m * m << 1;
if (*lwork < max(i__1,i__2)) {
*info = -16;
/* Computing MAX */
i__1 = *n * m, i__2 = m * m << 1;
work[1] = (real) max(i__1,i__2);
return 0;
}
weak = FALSE_;
strong = FALSE_;
/* Make a local copy of selected block */
scopy_(&c__16, &c_b3, &c__0, li, &c__1);
scopy_(&c__16, &c_b3, &c__0, ir, &c__1);
slacpy_("Full", &m, &m, &a_ref(*j1, *j1), lda, s, &c__4);
slacpy_("Full", &m, &m, &b_ref(*j1, *j1), ldb, t, &c__4);
/* Compute threshold for testing acceptance of swapping. */
eps = slamch_("P");
smlnum = slamch_("S") / eps;
dscale = 0.f;
dsum = 1.f;
slacpy_("Full", &m, &m, s, &c__4, &work[1], &m);
i__1 = m * m;
slassq_(&i__1, &work[1], &c__1, &dscale, &dsum);
slacpy_("Full", &m, &m, t, &c__4, &work[1], &m);
i__1 = m * m;
slassq_(&i__1, &work[1], &c__1, &dscale, &dsum);
dnorm = dscale * sqrt(dsum);
/* Computing MAX */
r__1 = eps * 10.f * dnorm;
thresh = dmax(r__1,smlnum);
if (m == 2) {
/* CASE 1: Swap 1-by-1 and 1-by-1 blocks.
Compute orthogonal QL and RQ that swap 1-by-1 and 1-by-1 blocks
using Givens rotations and perform the swap tentatively. */
f = s_ref(2, 2) * t_ref(1, 1) - t_ref(2, 2) * s_ref(1, 1);
g = s_ref(2, 2) * t_ref(1, 2) - t_ref(2, 2) * s_ref(1, 2);
sb = (r__1 = t_ref(2, 2), dabs(r__1));
sa = (r__1 = s_ref(2, 2), dabs(r__1));
slartg_(&f, &g, &ir_ref(1, 2), &ir_ref(1, 1), &ddum);
ir_ref(2, 1) = -ir_ref(1, 2);
ir_ref(2, 2) = ir_ref(1, 1);
srot_(&c__2, &s_ref(1, 1), &c__1, &s_ref(1, 2), &c__1, &ir_ref(1, 1),
&ir_ref(2, 1));
srot_(&c__2, &t_ref(1, 1), &c__1, &t_ref(1, 2), &c__1, &ir_ref(1, 1),
&ir_ref(2, 1));
if (sa >= sb) {
slartg_(&s_ref(1, 1), &s_ref(2, 1), &li_ref(1, 1), &li_ref(2, 1),
&ddum);
} else {
slartg_(&t_ref(1, 1), &t_ref(2, 1), &li_ref(1, 1), &li_ref(2, 1),
&ddum);
}
srot_(&c__2, &s_ref(1, 1), &c__4, &s_ref(2, 1), &c__4, &li_ref(1, 1),
&li_ref(2, 1));
srot_(&c__2, &t_ref(1, 1), &c__4, &t_ref(2, 1), &c__4, &li_ref(1, 1),
&li_ref(2, 1));
li_ref(2, 2) = li_ref(1, 1);
li_ref(1, 2) = -li_ref(2, 1);
/* Weak stability test:
|S21| + |T21| <= O(EPS * F-norm((S, T))) */
ws = (r__1 = s_ref(2, 1), dabs(r__1)) + (r__2 = t_ref(2, 1), dabs(
r__2));
weak = ws <= thresh;
if (! weak) {
goto L70;
}
if (TRUE_) {
/* Strong stability test:
F-norm((A-QL'*S*QR, B-QL'*T*QR)) <= O(EPS*F-norm((A,B))) */
slacpy_("Full", &m, &m, &a_ref(*j1, *j1), lda, &work[m * m + 1], &
m);
sgemm_("N", "N", &m, &m, &m, &c_b38, li, &c__4, s, &c__4, &c_b3, &
work[1], &m);
sgemm_("N", "T", &m, &m, &m, &c_b44, &work[1], &m, ir, &c__4, &
c_b38, &work[m * m + 1], &m);
dscale = 0.f;
dsum = 1.f;
i__1 = m * m;
slassq_(&i__1, &work[m * m + 1], &c__1, &dscale, &dsum);
slacpy_("Full", &m, &m, &b_ref(*j1, *j1), ldb, &work[m * m + 1], &
m);
sgemm_("N", "N", &m, &m, &m, &c_b38, li, &c__4, t, &c__4, &c_b3, &
work[1], &m);
sgemm_("N", "T", &m, &m, &m, &c_b44, &work[1], &m, ir, &c__4, &
c_b38, &work[m * m + 1], &m);
i__1 = m * m;
slassq_(&i__1, &work[m * m + 1], &c__1, &dscale, &dsum);
ss = dscale * sqrt(dsum);
strong = ss <= thresh;
if (! strong) {
goto L70;
}
}
/* Update (A(J1:J1+M-1, M+J1:N), B(J1:J1+M-1, M+J1:N)) and
(A(1:J1-1, J1:J1+M), B(1:J1-1, J1:J1+M)). */
i__1 = *j1 + 1;
srot_(&i__1, &a_ref(1, *j1), &c__1, &a_ref(1, *j1 + 1), &c__1, &
ir_ref(1, 1), &ir_ref(2, 1));
i__1 = *j1 + 1;
srot_(&i__1, &b_ref(1, *j1), &c__1, &b_ref(1, *j1 + 1), &c__1, &
ir_ref(1, 1), &ir_ref(2, 1));
i__1 = *n - *j1 + 1;
srot_(&i__1, &a_ref(*j1, *j1), lda, &a_ref(*j1 + 1, *j1), lda, &
li_ref(1, 1), &li_ref(2, 1));
i__1 = *n - *j1 + 1;
srot_(&i__1, &b_ref(*j1, *j1), ldb, &b_ref(*j1 + 1, *j1), ldb, &
li_ref(1, 1), &li_ref(2, 1));
/* Set N1-by-N2 (2,1) - blocks to ZERO. */
a_ref(*j1 + 1, *j1) = 0.f;
b_ref(*j1 + 1, *j1) = 0.f;
/* Accumulate transformations into Q and Z if requested. */
if (*wantz) {
srot_(n, &z___ref(1, *j1), &c__1, &z___ref(1, *j1 + 1), &c__1, &
ir_ref(1, 1), &ir_ref(2, 1));
}
if (*wantq) {
srot_(n, &q_ref(1, *j1), &c__1, &q_ref(1, *j1 + 1), &c__1, &
li_ref(1, 1), &li_ref(2, 1));
}
/* Exit with INFO = 0 if swap was successfully performed. */
return 0;
} else {
/* CASE 2: Swap 1-by-1 and 2-by-2 blocks, or 2-by-2
and 2-by-2 blocks.
Solve the generalized Sylvester equation
S11 * R - L * S22 = SCALE * S12
T11 * R - L * T22 = SCALE * T12
for R and L. Solutions in LI and IR. */
slacpy_("Full", n1, n2, &t_ref(1, *n1 + 1), &c__4, li, &c__4);
slacpy_("Full", n1, n2, &s_ref(1, *n1 + 1), &c__4, &ir_ref(*n2 + 1, *
n1 + 1), &c__4);
stgsy2_("N", &c__0, n1, n2, s, &c__4, &s_ref(*n1 + 1, *n1 + 1), &c__4,
&ir_ref(*n2 + 1, *n1 + 1), &c__4, t, &c__4, &t_ref(*n1 + 1, *
n1 + 1), &c__4, li, &c__4, &scale, &dsum, &dscale, iwork, &
idum, &linfo);
/* Compute orthogonal matrix QL:
QL' * LI = [ TL ]
[ 0 ]
where
LI = [ -L ]
[ SCALE * identity(N2) ] */
i__1 = *n2;
for (i__ = 1; i__ <= i__1; ++i__) {
sscal_(n1, &c_b44, &li_ref(1, i__), &c__1);
li_ref(*n1 + i__, i__) = scale;
/* L10: */
}
sgeqr2_(&m, n2, li, &c__4, taul, &work[1], &linfo);
if (linfo != 0) {
goto L70;
}
sorg2r_(&m, &m, n2, li, &c__4, taul, &work[1], &linfo);
if (linfo != 0) {
goto L70;
}
/* Compute orthogonal matrix RQ:
IR * RQ' = [ 0 TR],
where IR = [ SCALE * identity(N1), R ] */
i__1 = *n1;
for (i__ = 1; i__ <= i__1; ++i__) {
ir_ref(*n2 + i__, i__) = scale;
/* L20: */
}
sgerq2_(n1, &m, &ir_ref(*n2 + 1, 1), &c__4, taur, &work[1], &linfo);
if (linfo != 0) {
goto L70;
}
sorgr2_(&m, &m, n1, ir, &c__4, taur, &work[1], &linfo);
if (linfo != 0) {
goto L70;
}
/* Perform the swapping tentatively: */
sgemm_("T", "N", &m, &m, &m, &c_b38, li, &c__4, s, &c__4, &c_b3, &
work[1], &m);
sgemm_("N", "T", &m, &m, &m, &c_b38, &work[1], &m, ir, &c__4, &c_b3,
s, &c__4);
sgemm_("T", "N", &m, &m, &m, &c_b38, li, &c__4, t, &c__4, &c_b3, &
work[1], &m);
sgemm_("N", "T", &m, &m, &m, &c_b38, &work[1], &m, ir, &c__4, &c_b3,
t, &c__4);
slacpy_("F", &m, &m, s, &c__4, scpy, &c__4);
slacpy_("F", &m, &m, t, &c__4, tcpy, &c__4);
slacpy_("F", &m, &m, ir, &c__4, ircop, &c__4);
slacpy_("F", &m, &m, li, &c__4, licop, &c__4);
/* Triangularize the B-part by an RQ factorization.
Apply transformation (from left) to A-part, giving S. */
sgerq2_(&m, &m, t, &c__4, taur, &work[1], &linfo);
if (linfo != 0) {
goto L70;
}
sormr2_("R", "T", &m, &m, &m, t, &c__4, taur, s, &c__4, &work[1], &
linfo);
if (linfo != 0) {
goto L70;
}
sormr2_("L", "N", &m, &m, &m, t, &c__4, taur, ir, &c__4, &work[1], &
linfo);
if (linfo != 0) {
goto L70;
}
/* Compute F-norm(S21) in BRQA21. (T21 is 0.) */
dscale = 0.f;
dsum = 1.f;
i__1 = *n2;
for (i__ = 1; i__ <= i__1; ++i__) {
slassq_(n1, &s_ref(*n2 + 1, i__), &c__1, &dscale, &dsum);
/* L30: */
}
brqa21 = dscale * sqrt(dsum);
/* Triangularize the B-part by a QR factorization.
Apply transformation (from right) to A-part, giving S. */
sgeqr2_(&m, &m, tcpy, &c__4, taul, &work[1], &linfo);
if (linfo != 0) {
goto L70;
}
sorm2r_("L", "T", &m, &m, &m, tcpy, &c__4, taul, scpy, &c__4, &work[1]
, info);
sorm2r_("R", "N", &m, &m, &m, tcpy, &c__4, taul, licop, &c__4, &work[
1], info);
if (linfo != 0) {
goto L70;
}
/* Compute F-norm(S21) in BQRA21. (T21 is 0.) */
dscale = 0.f;
dsum = 1.f;
i__1 = *n2;
for (i__ = 1; i__ <= i__1; ++i__) {
slassq_(n1, &scpy_ref(*n2 + 1, i__), &c__1, &dscale, &dsum);
/* L40: */
}
bqra21 = dscale * sqrt(dsum);
/* Decide which method to use.
Weak stability test:
F-norm(S21) <= O(EPS * F-norm((S, T))) */
if (bqra21 <= brqa21 && bqra21 <= thresh) {
slacpy_("F", &m, &m, scpy, &c__4, s, &c__4);
slacpy_("F", &m, &m, tcpy, &c__4, t, &c__4);
slacpy_("F", &m, &m, ircop, &c__4, ir, &c__4);
slacpy_("F", &m, &m, licop, &c__4, li, &c__4);
} else if (brqa21 >= thresh) {
goto L70;
}
/* Set lower triangle of B-part to zero */
i__1 = m;
for (i__ = 2; i__ <= i__1; ++i__) {
i__2 = m - i__ + 1;
scopy_(&i__2, &c_b3, &c__0, &t_ref(i__, i__ - 1), &c__1);
/* L50: */
}
if (TRUE_) {
/* Strong stability test:
F-norm((A-QL*S*QR', B-QL*T*QR')) <= O(EPS*F-norm((A,B))) */
slacpy_("Full", &m, &m, &a_ref(*j1, *j1), lda, &work[m * m + 1], &
m);
sgemm_("N", "N", &m, &m, &m, &c_b38, li, &c__4, s, &c__4, &c_b3, &
work[1], &m);
sgemm_("N", "N", &m, &m, &m, &c_b44, &work[1], &m, ir, &c__4, &
c_b38, &work[m * m + 1], &m);
dscale = 0.f;
dsum = 1.f;
i__1 = m * m;
slassq_(&i__1, &work[m * m + 1], &c__1, &dscale, &dsum);
slacpy_("Full", &m, &m, &b_ref(*j1, *j1), ldb, &work[m * m + 1], &
m);
sgemm_("N", "N", &m, &m, &m, &c_b38, li, &c__4, t, &c__4, &c_b3, &
work[1], &m);
sgemm_("N", "N", &m, &m, &m, &c_b44, &work[1], &m, ir, &c__4, &
c_b38, &work[m * m + 1], &m);
i__1 = m * m;
slassq_(&i__1, &work[m * m + 1], &c__1, &dscale, &dsum);
ss = dscale * sqrt(dsum);
strong = ss <= thresh;
if (! strong) {
goto L70;
}
}
/* If the swap is accepted ("weakly" and "strongly"), apply the
transformations and set N1-by-N2 (2,1)-block to zero. */
i__1 = *n2;
for (i__ = 1; i__ <= i__1; ++i__) {
scopy_(n1, &c_b3, &c__0, &s_ref(*n2 + 1, i__), &c__1);
/* L60: */
}
/* copy back M-by-M diagonal block starting at index J1 of (A, B) */
slacpy_("F", &m, &m, s, &c__4, &a_ref(*j1, *j1), lda);
slacpy_("F", &m, &m, t, &c__4, &b_ref(*j1, *j1), ldb);
scopy_(&c__16, &c_b3, &c__0, t, &c__1);
/* Standardize existing 2-by-2 blocks. */
i__1 = m * m;
scopy_(&i__1, &c_b3, &c__0, &work[1], &c__1);
work[1] = 1.f;
t_ref(1, 1) = 1.f;
idum = *lwork - m * m - 2;
if (*n2 > 1) {
slagv2_(&a_ref(*j1, *j1), lda, &b_ref(*j1, *j1), ldb, ar, ai, be,
&work[1], &work[2], &t_ref(1, 1), &t_ref(2, 1));
work[m + 1] = -work[2];
work[m + 2] = work[1];
t_ref(*n2, *n2) = t_ref(1, 1);
t_ref(1, 2) = -t_ref(2, 1);
}
work[m * m] = 1.f;
t_ref(m, m) = 1.f;
if (*n1 > 1) {
slagv2_(&a_ref(*j1 + *n2, *j1 + *n2), lda, &b_ref(*j1 + *n2, *j1
+ *n2), ldb, taur, taul, &work[m * m + 1], &work[*n2 * m
+ *n2 + 1], &work[*n2 * m + *n2 + 2], &t_ref(*n2 + 1, *n2
+ 1), &t_ref(m, m - 1));
work[m * m] = work[*n2 * m + *n2 + 1];
work[m * m - 1] = -work[*n2 * m + *n2 + 2];
t_ref(m, m) = t_ref(*n2 + 1, *n2 + 1);
t_ref(m - 1, m) = -t_ref(m, m - 1);
}
sgemm_("T", "N", n2, n1, n2, &c_b38, &work[1], &m, &a_ref(*j1, *j1 + *
n2), lda, &c_b3, &work[m * m + 1], n2);
slacpy_("Full", n2, n1, &work[m * m + 1], n2, &a_ref(*j1, *j1 + *n2),
lda);
sgemm_("T", "N", n2, n1, n2, &c_b38, &work[1], &m, &b_ref(*j1, *j1 + *
n2), ldb, &c_b3, &work[m * m + 1], n2);
slacpy_("Full", n2, n1, &work[m * m + 1], n2, &b_ref(*j1, *j1 + *n2),
ldb);
sgemm_("N", "N", &m, &m, &m, &c_b38, li, &c__4, &work[1], &m, &c_b3, &
work[m * m + 1], &m);
slacpy_("Full", &m, &m, &work[m * m + 1], &m, li, &c__4);
sgemm_("N", "N", n2, n1, n1, &c_b38, &a_ref(*j1, *j1 + *n2), lda, &
t_ref(*n2 + 1, *n2 + 1), &c__4, &c_b3, &work[1], n2);
slacpy_("Full", n2, n1, &work[1], n2, &a_ref(*j1, *j1 + *n2), lda);
sgemm_("N", "N", n2, n1, n1, &c_b38, &b_ref(*j1, *j1 + *n2), lda, &
t_ref(*n2 + 1, *n2 + 1), &c__4, &c_b3, &work[1], n2);
slacpy_("Full", n2, n1, &work[1], n2, &b_ref(*j1, *j1 + *n2), ldb);
sgemm_("T", "N", &m, &m, &m, &c_b38, ir, &c__4, t, &c__4, &c_b3, &
work[1], &m);
slacpy_("Full", &m, &m, &work[1], &m, ir, &c__4);
/* Accumulate transformations into Q and Z if requested. */
if (*wantq) {
sgemm_("N", "N", n, &m, &m, &c_b38, &q_ref(1, *j1), ldq, li, &
c__4, &c_b3, &work[1], n);
slacpy_("Full", n, &m, &work[1], n, &q_ref(1, *j1), ldq);
}
if (*wantz) {
sgemm_("N", "N", n, &m, &m, &c_b38, &z___ref(1, *j1), ldz, ir, &
c__4, &c_b3, &work[1], n);
slacpy_("Full", n, &m, &work[1], n, &z___ref(1, *j1), ldz);
}
/* Update (A(J1:J1+M-1, M+J1:N), B(J1:J1+M-1, M+J1:N)) and
(A(1:J1-1, J1:J1+M), B(1:J1-1, J1:J1+M)). */
i__ = *j1 + m;
if (i__ <= *n) {
i__1 = *n - i__ + 1;
sgemm_("T", "N", &m, &i__1, &m, &c_b38, li, &c__4, &a_ref(*j1,
i__), lda, &c_b3, &work[1], &m);
i__1 = *n - i__ + 1;
slacpy_("Full", &m, &i__1, &work[1], &m, &a_ref(*j1, i__), lda);
i__1 = *n - i__ + 1;
sgemm_("T", "N", &m, &i__1, &m, &c_b38, li, &c__4, &b_ref(*j1,
i__), lda, &c_b3, &work[1], &m);
i__1 = *n - i__ + 1;
slacpy_("Full", &m, &i__1, &work[1], &m, &b_ref(*j1, i__), lda);
}
i__ = *j1 - 1;
if (i__ > 0) {
sgemm_("N", "N", &i__, &m, &m, &c_b38, &a_ref(1, *j1), lda, ir, &
c__4, &c_b3, &work[1], &i__);
slacpy_("Full", &i__, &m, &work[1], &i__, &a_ref(1, *j1), lda);
sgemm_("N", "N", &i__, &m, &m, &c_b38, &b_ref(1, *j1), ldb, ir, &
c__4, &c_b3, &work[1], &i__);
slacpy_("Full", &i__, &m, &work[1], &i__, &b_ref(1, *j1), ldb);
}
/* Exit with INFO = 0 if swap was successfully performed. */
return 0;
}
/* Exit with INFO = 1 if swap was rejected. */
L70:
*info = 1;
return 0;
/* End of STGEX2 */
} /* stgex2_ */
/* Subroutine */ int sggsvp_(char *jobu, char *jobv, char *jobq, integer *m,
integer *p, integer *n, real *a, integer *lda, real *b, integer *ldb,
real *tola, real *tolb, integer *k, integer *l, real *u, integer *ldu,
real *v, integer *ldv, real *q, integer *ldq, integer *iwork, real *
tau, real *work, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, q_dim1, q_offset, u_dim1,
u_offset, v_dim1, v_offset, i__1, i__2, i__3;
real r__1;
/* Local variables */
integer i__, j;
extern logical lsame_(char *, char *);
logical wantq, wantu, wantv;
extern /* Subroutine */ int sgeqr2_(integer *, integer *, real *, integer
*, real *, real *, integer *), sgerq2_(integer *, integer *, real
*, integer *, real *, real *, integer *), sorg2r_(integer *,
integer *, integer *, real *, integer *, real *, real *, integer *
), sorm2r_(char *, char *, integer *, integer *, integer *, real *
, integer *, real *, real *, integer *, real *, integer *), sormr2_(char *, char *, integer *, integer *, integer *,
real *, integer *, real *, real *, integer *, real *, integer *), xerbla_(char *, integer *), sgeqpf_(
integer *, integer *, real *, integer *, integer *, real *, real *
, integer *), slacpy_(char *, integer *, integer *, real *,
integer *, real *, integer *), slaset_(char *, integer *,
integer *, real *, real *, real *, integer *), slapmt_(
logical *, integer *, integer *, real *, integer *, integer *);
logical forwrd;
/* -- LAPACK routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SGGSVP computes orthogonal matrices U, V and Q such that */
/* N-K-L K L */
/* U'*A*Q = K ( 0 A12 A13 ) if M-K-L >= 0; */
/* L ( 0 0 A23 ) */
/* M-K-L ( 0 0 0 ) */
/* N-K-L K L */
/* = K ( 0 A12 A13 ) if M-K-L < 0; */
/* M-K ( 0 0 A23 ) */
/* N-K-L K L */
/* V'*B*Q = L ( 0 0 B13 ) */
/* P-L ( 0 0 0 ) */
/* where the K-by-K matrix A12 and L-by-L matrix B13 are nonsingular */
/* upper triangular; A23 is L-by-L upper triangular if M-K-L >= 0, */
/* otherwise A23 is (M-K)-by-L upper trapezoidal. K+L = the effective */
/* numerical rank of the (M+P)-by-N matrix (A',B')'. Z' denotes the */
/* transpose of Z. */
/* This decomposition is the preprocessing step for computing the */
/* Generalized Singular Value Decomposition (GSVD), see subroutine */
/* SGGSVD. */
/* Arguments */
/* ========= */
/* JOBU (input) CHARACTER*1 */
/* = 'U': Orthogonal matrix U is computed; */
/* = 'N': U is not computed. */
/* JOBV (input) CHARACTER*1 */
/* = 'V': Orthogonal matrix V is computed; */
/* = 'N': V is not computed. */
/* JOBQ (input) CHARACTER*1 */
/* = 'Q': Orthogonal matrix Q is computed; */
/* = 'N': Q is not computed. */
/* M (input) INTEGER */
/* The number of rows of the matrix A. M >= 0. */
/* P (input) INTEGER */
/* The number of rows of the matrix B. P >= 0. */
/* N (input) INTEGER */
/* The number of columns of the matrices A and B. N >= 0. */
/* A (input/output) REAL array, dimension (LDA,N) */
/* On entry, the M-by-N matrix A. */
/* On exit, A contains the triangular (or trapezoidal) matrix */
/* described in the Purpose section. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1,M). */
/* B (input/output) REAL array, dimension (LDB,N) */
/* On entry, the P-by-N matrix B. */
/* On exit, B contains the triangular matrix described in */
/* the Purpose section. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= max(1,P). */
/* TOLA (input) REAL */
/* TOLB (input) REAL */
/* TOLA and TOLB are the thresholds to determine the effective */
/* numerical rank of matrix B and a subblock of A. Generally, */
/* they are set to */
/* TOLA = MAX(M,N)*norm(A)*MACHEPS, */
/* TOLB = MAX(P,N)*norm(B)*MACHEPS. */
/* The size of TOLA and TOLB may affect the size of backward */
/* errors of the decomposition. */
/* K (output) INTEGER */
/* L (output) INTEGER */
/* On exit, K and L specify the dimension of the subblocks */
/* described in Purpose. */
/* K + L = effective numerical rank of (A',B')'. */
/* U (output) REAL array, dimension (LDU,M) */
/* If JOBU = 'U', U contains the orthogonal matrix U. */
/* If JOBU = 'N', U is not referenced. */
/* LDU (input) INTEGER */
/* The leading dimension of the array U. LDU >= max(1,M) if */
/* JOBU = 'U'; LDU >= 1 otherwise. */
/* V (output) REAL array, dimension (LDV,M) */
/* If JOBV = 'V', V contains the orthogonal matrix V. */
/* If JOBV = 'N', V is not referenced. */
/* LDV (input) INTEGER */
/* The leading dimension of the array V. LDV >= max(1,P) if */
/* JOBV = 'V'; LDV >= 1 otherwise. */
/* Q (output) REAL array, dimension (LDQ,N) */
/* If JOBQ = 'Q', Q contains the orthogonal matrix Q. */
/* If JOBQ = 'N', Q is not referenced. */
/* LDQ (input) INTEGER */
/* The leading dimension of the array Q. LDQ >= max(1,N) if */
/* JOBQ = 'Q'; LDQ >= 1 otherwise. */
/* IWORK (workspace) INTEGER array, dimension (N) */
/* TAU (workspace) REAL array, dimension (N) */
/* WORK (workspace) REAL array, dimension (max(3*N,M,P)) */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value. */
/* Further Details */
/* =============== */
/* The subroutine uses LAPACK subroutine SGEQPF for the QR factorization */
/* with column pivoting to detect the effective numerical rank of the */
/* a matrix. It may be replaced by a better rank determination strategy. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
u_dim1 = *ldu;
u_offset = 1 + u_dim1;
u -= u_offset;
v_dim1 = *ldv;
v_offset = 1 + v_dim1;
v -= v_offset;
q_dim1 = *ldq;
q_offset = 1 + q_dim1;
q -= q_offset;
--iwork;
--tau;
--work;
/* Function Body */
wantu = lsame_(jobu, "U");
wantv = lsame_(jobv, "V");
wantq = lsame_(jobq, "Q");
forwrd = TRUE_;
*info = 0;
if (! (wantu || lsame_(jobu, "N"))) {
*info = -1;
} else if (! (wantv || lsame_(jobv, "N"))) {
*info = -2;
} else if (! (wantq || lsame_(jobq, "N"))) {
*info = -3;
} else if (*m < 0) {
*info = -4;
} else if (*p < 0) {
*info = -5;
} else if (*n < 0) {
*info = -6;
} else if (*lda < max(1,*m)) {
*info = -8;
} else if (*ldb < max(1,*p)) {
*info = -10;
} else if (*ldu < 1 || wantu && *ldu < *m) {
*info = -16;
} else if (*ldv < 1 || wantv && *ldv < *p) {
*info = -18;
} else if (*ldq < 1 || wantq && *ldq < *n) {
*info = -20;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SGGSVP", &i__1);
return 0;
}
/* QR with column pivoting of B: B*P = V*( S11 S12 ) */
/* ( 0 0 ) */
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
iwork[i__] = 0;
/* L10: */
}
sgeqpf_(p, n, &b[b_offset], ldb, &iwork[1], &tau[1], &work[1], info);
/* Update A := A*P */
slapmt_(&forwrd, m, n, &a[a_offset], lda, &iwork[1]);
/* Determine the effective rank of matrix B. */
*l = 0;
i__1 = min(*p,*n);
for (i__ = 1; i__ <= i__1; ++i__) {
if ((r__1 = b[i__ + i__ * b_dim1], dabs(r__1)) > *tolb) {
++(*l);
}
/* L20: */
}
if (wantv) {
/* Copy the details of V, and form V. */
slaset_("Full", p, p, &c_b12, &c_b12, &v[v_offset], ldv);
if (*p > 1) {
i__1 = *p - 1;
slacpy_("Lower", &i__1, n, &b[b_dim1 + 2], ldb, &v[v_dim1 + 2],
ldv);
}
i__1 = min(*p,*n);
sorg2r_(p, p, &i__1, &v[v_offset], ldv, &tau[1], &work[1], info);
}
/* Clean up B */
i__1 = *l - 1;
for (j = 1; j <= i__1; ++j) {
i__2 = *l;
for (i__ = j + 1; i__ <= i__2; ++i__) {
b[i__ + j * b_dim1] = 0.f;
/* L30: */
}
/* L40: */
}
if (*p > *l) {
i__1 = *p - *l;
slaset_("Full", &i__1, n, &c_b12, &c_b12, &b[*l + 1 + b_dim1], ldb);
}
if (wantq) {
/* Set Q = I and Update Q := Q*P */
slaset_("Full", n, n, &c_b12, &c_b22, &q[q_offset], ldq);
slapmt_(&forwrd, n, n, &q[q_offset], ldq, &iwork[1]);
}
if (*p >= *l && *n != *l) {
/* RQ factorization of (S11 S12): ( S11 S12 ) = ( 0 S12 )*Z */
sgerq2_(l, n, &b[b_offset], ldb, &tau[1], &work[1], info);
/* Update A := A*Z' */
sormr2_("Right", "Transpose", m, n, l, &b[b_offset], ldb, &tau[1], &a[
a_offset], lda, &work[1], info);
if (wantq) {
/* Update Q := Q*Z' */
sormr2_("Right", "Transpose", n, n, l, &b[b_offset], ldb, &tau[1],
&q[q_offset], ldq, &work[1], info);
}
/* Clean up B */
i__1 = *n - *l;
slaset_("Full", l, &i__1, &c_b12, &c_b12, &b[b_offset], ldb);
i__1 = *n;
for (j = *n - *l + 1; j <= i__1; ++j) {
i__2 = *l;
for (i__ = j - *n + *l + 1; i__ <= i__2; ++i__) {
b[i__ + j * b_dim1] = 0.f;
/* L50: */
}
/* L60: */
}
}
/* Let N-L L */
/* A = ( A11 A12 ) M, */
/* then the following does the complete QR decomposition of A11: */
/* A11 = U*( 0 T12 )*P1' */
/* ( 0 0 ) */
i__1 = *n - *l;
for (i__ = 1; i__ <= i__1; ++i__) {
iwork[i__] = 0;
/* L70: */
}
i__1 = *n - *l;
sgeqpf_(m, &i__1, &a[a_offset], lda, &iwork[1], &tau[1], &work[1], info);
/* Determine the effective rank of A11 */
*k = 0;
/* Computing MIN */
i__2 = *m, i__3 = *n - *l;
i__1 = min(i__2,i__3);
for (i__ = 1; i__ <= i__1; ++i__) {
if ((r__1 = a[i__ + i__ * a_dim1], dabs(r__1)) > *tola) {
++(*k);
}
/* L80: */
}
/* Update A12 := U'*A12, where A12 = A( 1:M, N-L+1:N ) */
/* Computing MIN */
i__2 = *m, i__3 = *n - *l;
i__1 = min(i__2,i__3);
sorm2r_("Left", "Transpose", m, l, &i__1, &a[a_offset], lda, &tau[1], &a[(
*n - *l + 1) * a_dim1 + 1], lda, &work[1], info);
if (wantu) {
/* Copy the details of U, and form U */
slaset_("Full", m, m, &c_b12, &c_b12, &u[u_offset], ldu);
if (*m > 1) {
i__1 = *m - 1;
i__2 = *n - *l;
slacpy_("Lower", &i__1, &i__2, &a[a_dim1 + 2], lda, &u[u_dim1 + 2]
, ldu);
}
/* Computing MIN */
i__2 = *m, i__3 = *n - *l;
i__1 = min(i__2,i__3);
sorg2r_(m, m, &i__1, &u[u_offset], ldu, &tau[1], &work[1], info);
}
if (wantq) {
/* Update Q( 1:N, 1:N-L ) = Q( 1:N, 1:N-L )*P1 */
i__1 = *n - *l;
slapmt_(&forwrd, n, &i__1, &q[q_offset], ldq, &iwork[1]);
}
/* Clean up A: set the strictly lower triangular part of */
/* A(1:K, 1:K) = 0, and A( K+1:M, 1:N-L ) = 0. */
i__1 = *k - 1;
for (j = 1; j <= i__1; ++j) {
i__2 = *k;
for (i__ = j + 1; i__ <= i__2; ++i__) {
a[i__ + j * a_dim1] = 0.f;
/* L90: */
}
/* L100: */
}
if (*m > *k) {
i__1 = *m - *k;
i__2 = *n - *l;
slaset_("Full", &i__1, &i__2, &c_b12, &c_b12, &a[*k + 1 + a_dim1],
lda);
}
if (*n - *l > *k) {
/* RQ factorization of ( T11 T12 ) = ( 0 T12 )*Z1 */
i__1 = *n - *l;
sgerq2_(k, &i__1, &a[a_offset], lda, &tau[1], &work[1], info);
if (wantq) {
/* Update Q( 1:N,1:N-L ) = Q( 1:N,1:N-L )*Z1' */
i__1 = *n - *l;
sormr2_("Right", "Transpose", n, &i__1, k, &a[a_offset], lda, &
tau[1], &q[q_offset], ldq, &work[1], info);
}
/* Clean up A */
i__1 = *n - *l - *k;
slaset_("Full", k, &i__1, &c_b12, &c_b12, &a[a_offset], lda);
i__1 = *n - *l;
for (j = *n - *l - *k + 1; j <= i__1; ++j) {
i__2 = *k;
for (i__ = j - *n + *l + *k + 1; i__ <= i__2; ++i__) {
a[i__ + j * a_dim1] = 0.f;
/* L110: */
}
/* L120: */
}
}
if (*m > *k) {
/* QR factorization of A( K+1:M,N-L+1:N ) */
i__1 = *m - *k;
sgeqr2_(&i__1, l, &a[*k + 1 + (*n - *l + 1) * a_dim1], lda, &tau[1], &
work[1], info);
if (wantu) {
/* Update U(:,K+1:M) := U(:,K+1:M)*U1 */
i__1 = *m - *k;
/* Computing MIN */
i__3 = *m - *k;
i__2 = min(i__3,*l);
sorm2r_("Right", "No transpose", m, &i__1, &i__2, &a[*k + 1 + (*n
- *l + 1) * a_dim1], lda, &tau[1], &u[(*k + 1) * u_dim1 +
1], ldu, &work[1], info);
}
/* Clean up */
i__1 = *n;
for (j = *n - *l + 1; j <= i__1; ++j) {
i__2 = *m;
for (i__ = j - *n + *k + *l + 1; i__ <= i__2; ++i__) {
a[i__ + j * a_dim1] = 0.f;
/* L130: */
}
/* L140: */
}
}
return 0;
/* End of SGGSVP */
} /* sggsvp_ */
/* Subroutine */
int sgerqf_(integer *m, integer *n, real *a, integer *lda, real *tau, real *work, integer *lwork, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, i__1, i__2, i__3, i__4;
/* Local variables */
integer i__, k, ib, nb, ki, kk, mu, nu, nx, iws, nbmin, iinfo;
extern /* Subroutine */
int sgerq2_(integer *, integer *, real *, integer *, real *, real *, integer *), slarfb_(char *, char *, char *, char *, integer *, integer *, integer *, real *, integer *, real * , integer *, real *, integer *, real *, integer *), xerbla_(char *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *, integer *, integer *);
extern /* Subroutine */
int slarft_(char *, char *, integer *, integer *, real *, integer *, real *, real *, integer *);
integer ldwork, lwkopt;
logical lquery;
/* -- LAPACK computational routine (version 3.4.0) -- */
/* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
/* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
/* November 2011 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* ===================================================================== */
/* .. Local Scalars .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input arguments */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
--tau;
--work;
/* Function Body */
*info = 0;
lquery = *lwork == -1;
if (*m < 0)
{
*info = -1;
}
else if (*n < 0)
{
*info = -2;
}
else if (*lda < max(1,*m))
{
*info = -4;
}
else if (*lwork < max(1,*m) && ! lquery)
{
*info = -7;
}
if (*info == 0)
{
k = min(*m,*n);
if (k == 0)
{
lwkopt = 1;
}
else
{
nb = ilaenv_(&c__1, "SGERQF", " ", m, n, &c_n1, &c_n1);
lwkopt = *m * nb;
work[1] = (real) lwkopt;
}
work[1] = (real) lwkopt;
if (*lwork < max(1,*m) && ! lquery)
{
*info = -7;
}
}
if (*info != 0)
{
i__1 = -(*info);
xerbla_("SGERQF", &i__1);
return 0;
}
else if (lquery)
{
return 0;
}
/* Quick return if possible */
if (k == 0)
{
return 0;
}
nbmin = 2;
nx = 1;
iws = *m;
if (nb > 1 && nb < k)
{
/* Determine when to cross over from blocked to unblocked code. */
/* Computing MAX */
i__1 = 0;
i__2 = ilaenv_(&c__3, "SGERQF", " ", m, n, &c_n1, &c_n1); // , expr subst
nx = max(i__1,i__2);
if (nx < k)
{
/* Determine if workspace is large enough for blocked code. */
ldwork = *m;
iws = ldwork * nb;
if (*lwork < iws)
{
/* Not enough workspace to use optimal NB: reduce NB and */
/* determine the minimum value of NB. */
nb = *lwork / ldwork;
/* Computing MAX */
i__1 = 2;
i__2 = ilaenv_(&c__2, "SGERQF", " ", m, n, &c_n1, & c_n1); // , expr subst
nbmin = max(i__1,i__2);
}
}
}
if (nb >= nbmin && nb < k && nx < k)
{
/* Use blocked code initially. */
/* The last kk rows are handled by the block method. */
ki = (k - nx - 1) / nb * nb;
/* Computing MIN */
i__1 = k;
i__2 = ki + nb; // , expr subst
kk = min(i__1,i__2);
i__1 = k - kk + 1;
i__2 = -nb;
for (i__ = k - kk + ki + 1;
i__2 < 0 ? i__ >= i__1 : i__ <= i__1;
i__ += i__2)
{
/* Computing MIN */
i__3 = k - i__ + 1;
ib = min(i__3,nb);
/* Compute the RQ factorization of the current block */
/* A(m-k+i:m-k+i+ib-1,1:n-k+i+ib-1) */
i__3 = *n - k + i__ + ib - 1;
sgerq2_(&ib, &i__3, &a[*m - k + i__ + a_dim1], lda, &tau[i__], & work[1], &iinfo);
if (*m - k + i__ > 1)
{
/* Form the triangular factor of the block reflector */
/* H = H(i+ib-1) . . . H(i+1) H(i) */
i__3 = *n - k + i__ + ib - 1;
slarft_("Backward", "Rowwise", &i__3, &ib, &a[*m - k + i__ + a_dim1], lda, &tau[i__], &work[1], &ldwork);
/* Apply H to A(1:m-k+i-1,1:n-k+i+ib-1) from the right */
i__3 = *m - k + i__ - 1;
i__4 = *n - k + i__ + ib - 1;
slarfb_("Right", "No transpose", "Backward", "Rowwise", &i__3, &i__4, &ib, &a[*m - k + i__ + a_dim1], lda, &work[1], &ldwork, &a[a_offset], lda, &work[ib + 1], &ldwork);
}
/* L10: */
}
mu = *m - k + i__ + nb - 1;
nu = *n - k + i__ + nb - 1;
}
else
{
mu = *m;
nu = *n;
}
/* Use unblocked code to factor the last or only block */
if (mu > 0 && nu > 0)
{
sgerq2_(&mu, &nu, &a[a_offset], lda, &tau[1], &work[1], &iinfo);
}
work[1] = (real) iws;
return 0;
/* End of SGERQF */
}
/* Subroutine */ int sgerqf_(integer *m, integer *n, real *a, integer *lda,
real *tau, real *work, integer *lwork, integer *info)
{
/* -- LAPACK routine (version 2.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
September 30, 1994
Purpose
=======
SGERQF computes an RQ factorization of a real M-by-N matrix A:
A = R * Q.
Arguments
=========
M (input) INTEGER
The number of rows of the matrix A. M >= 0.
N (input) INTEGER
The number of columns of the matrix A. N >= 0.
A (input/output) REAL array, dimension (LDA,N)
On entry, the M-by-N matrix A.
On exit,
if m <= n, the upper triangle of the subarray
A(1:m,n-m+1:n) contains the M-by-M upper triangular matrix R;
if m >= n, the elements on and above the (m-n)-th subdiagonal
contain the M-by-N upper trapezoidal matrix R;
the remaining elements, with the array TAU, represent the
orthogonal matrix Q as a product of min(m,n) elementary
reflectors (see Further Details).
LDA (input) INTEGER
The leading dimension of the array A. LDA >= max(1,M).
TAU (output) REAL array, dimension (min(M,N))
The scalar factors of the elementary reflectors (see Further
Details).
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 >= max(1,M).
For optimum performance LWORK >= M*NB, where NB is
the optimal blocksize.
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
Further Details
===============
The matrix Q is represented as a product of elementary reflectors
Q = H(1) H(2) . . . H(k), where k = min(m,n).
Each H(i) has the form
H(i) = I - tau * v * v'
where tau is a real scalar, and v is a real vector with
v(n-k+i+1:n) = 0 and v(n-k+i) = 1; v(1:n-k+i-1) is stored on exit in
A(m-k+i,1:n-k+i-1), and tau in TAU(i).
=====================================================================
Test the input arguments
Parameter adjustments
Function Body */
/* Table of constant values */
static integer c__1 = 1;
static integer c_n1 = -1;
static integer c__3 = 3;
static integer c__2 = 2;
/* System generated locals */
integer a_dim1, a_offset, i__1, i__2, i__3, i__4;
/* Local variables */
static integer i, k, nbmin, iinfo;
extern /* Subroutine */ int sgerq2_(integer *, integer *, real *, integer
*, real *, real *, integer *);
static integer ib, nb, ki, kk, mu, nu, nx;
extern /* Subroutine */ int slarfb_(char *, char *, char *, char *,
integer *, integer *, integer *, real *, integer *, real *,
integer *, real *, integer *, real *, integer *), xerbla_(char *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *, ftnlen, ftnlen);
extern /* Subroutine */ int slarft_(char *, char *, integer *, integer *,
real *, integer *, real *, real *, integer *);
static integer ldwork, iws;
#define TAU(I) tau[(I)-1]
#define WORK(I) work[(I)-1]
#define A(I,J) a[(I)-1 + ((J)-1)* ( *lda)]
*info = 0;
if (*m < 0) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*lda < max(1,*m)) {
*info = -4;
} else if (*lwork < max(1,*m)) {
*info = -7;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SGERQF", &i__1);
return 0;
}
/* Quick return if possible */
k = min(*m,*n);
if (k == 0) {
WORK(1) = 1.f;
return 0;
}
/* Determine the block size. */
nb = ilaenv_(&c__1, "SGERQF", " ", m, n, &c_n1, &c_n1, 6L, 1L);
nbmin = 2;
nx = 1;
iws = *m;
if (nb > 1 && nb < k) {
/* Determine when to cross over from blocked to unblocked code.
Computing MAX */
i__1 = 0, i__2 = ilaenv_(&c__3, "SGERQF", " ", m, n, &c_n1, &c_n1, 6L,
1L);
nx = max(i__1,i__2);
if (nx < k) {
/* Determine if workspace is large enough for blocked co
de. */
ldwork = *m;
iws = ldwork * nb;
if (*lwork < iws) {
/* Not enough workspace to use optimal NB: reduc
e NB and
determine the minimum value of NB. */
nb = *lwork / ldwork;
/* Computing MAX */
i__1 = 2, i__2 = ilaenv_(&c__2, "SGERQF", " ", m, n, &c_n1, &
c_n1, 6L, 1L);
nbmin = max(i__1,i__2);
}
}
}
if (nb >= nbmin && nb < k && nx < k) {
/* Use blocked code initially.
The last kk rows are handled by the block method. */
ki = (k - nx - 1) / nb * nb;
/* Computing MIN */
i__1 = k, i__2 = ki + nb;
kk = min(i__1,i__2);
i__1 = k - kk + 1;
i__2 = -nb;
for (i = k - kk + ki + 1; -nb < 0 ? i >= k-kk+1 : i <= k-kk+1; i += -nb)
{
/* Computing MIN */
i__3 = k - i + 1;
ib = min(i__3,nb);
/* Compute the RQ factorization of the current block
A(m-k+i:m-k+i+ib-1,1:n-k+i+ib-1) */
i__3 = *n - k + i + ib - 1;
sgerq2_(&ib, &i__3, &A(*m-k+i,1), lda, &TAU(i), &WORK(
1), &iinfo);
if (*m - k + i > 1) {
/* Form the triangular factor of the block reflec
tor
H = H(i+ib-1) . . . H(i+1) H(i) */
i__3 = *n - k + i + ib - 1;
slarft_("Backward", "Rowwise", &i__3, &ib, &A(*m-k+i,1), lda, &TAU(i), &WORK(1), &ldwork);
/* Apply H to A(1:m-k+i-1,1:n-k+i+ib-1) from the
right */
i__3 = *m - k + i - 1;
i__4 = *n - k + i + ib - 1;
slarfb_("Right", "No transpose", "Backward", "Rowwise", &i__3,
&i__4, &ib, &A(*m-k+i,1), lda, &WORK(1), &
ldwork, &A(1,1), lda, &WORK(ib + 1), &ldwork);
}
/* L10: */
}
mu = *m - k + i + nb - 1;
nu = *n - k + i + nb - 1;
} else {
mu = *m;
nu = *n;
}
/* Use unblocked code to factor the last or only block */
if (mu > 0 && nu > 0) {
sgerq2_(&mu, &nu, &A(1,1), lda, &TAU(1), &WORK(1), &iinfo);
}
WORK(1) = (real) iws;
return 0;
/* End of SGERQF */
} /* sgerqf_ */
/*< >*/
/* Subroutine */ int sggsvp_(char *jobu, char *jobv, char *jobq, integer *m,
integer *p, integer *n, real *a, integer *lda, real *b, integer *ldb,
real *tola, real *tolb, integer *k, integer *l, real *u, integer *ldu,
real *v, integer *ldv, real *q, integer *ldq, integer *iwork, real *
tau, real *work, integer *info, ftnlen jobu_len, ftnlen jobv_len,
ftnlen jobq_len)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, q_dim1, q_offset, u_dim1,
u_offset, v_dim1, v_offset, i__1, i__2, i__3;
real r__1;
/* Local variables */
integer i__, j;
extern logical lsame_(char *, char *, ftnlen, ftnlen);
logical wantq, wantu, wantv;
extern /* Subroutine */ int sgeqr2_(integer *, integer *, real *, integer
*, real *, real *, integer *), sgerq2_(integer *, integer *, real
*, integer *, real *, real *, integer *), sorg2r_(integer *,
integer *, integer *, real *, integer *, real *, real *, integer *
), sorm2r_(char *, char *, integer *, integer *, integer *, real *
, integer *, real *, real *, integer *, real *, integer *, ftnlen,
ftnlen), sormr2_(char *, char *, integer *, integer *, integer *,
real *, integer *, real *, real *, integer *, real *, integer *,
ftnlen, ftnlen), xerbla_(char *, integer *, ftnlen), sgeqpf_(
integer *, integer *, real *, integer *, integer *, real *, real *
, integer *), slacpy_(char *, integer *, integer *, real *,
integer *, real *, integer *, ftnlen), slaset_(char *, integer *,
integer *, real *, real *, real *, integer *, ftnlen), slapmt_(
logical *, integer *, integer *, real *, integer *, integer *);
logical forwrd;
(void)jobu_len;
(void)jobv_len;
(void)jobq_len;
/* -- LAPACK routine (version 3.0) -- */
/* Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., */
/* Courant Institute, Argonne National Lab, and Rice University */
/* September 30, 1994 */
/* .. Scalar Arguments .. */
/*< CHARACTER JOBQ, JOBU, JOBV >*/
/*< INTEGER INFO, K, L, LDA, LDB, LDQ, LDU, LDV, M, N, P >*/
/*< REAL TOLA, TOLB >*/
/* .. */
/* .. Array Arguments .. */
/*< INTEGER IWORK( * ) >*/
/*< >*/
/* .. */
/* Purpose */
/* ======= */
/* SGGSVP computes orthogonal matrices U, V and Q such that */
/* N-K-L K L */
/* U'*A*Q = K ( 0 A12 A13 ) if M-K-L >= 0; */
/* L ( 0 0 A23 ) */
/* M-K-L ( 0 0 0 ) */
/* N-K-L K L */
/* = K ( 0 A12 A13 ) if M-K-L < 0; */
/* M-K ( 0 0 A23 ) */
/* N-K-L K L */
/* V'*B*Q = L ( 0 0 B13 ) */
/* P-L ( 0 0 0 ) */
/* where the K-by-K matrix A12 and L-by-L matrix B13 are nonsingular */
/* upper triangular; A23 is L-by-L upper triangular if M-K-L >= 0, */
/* otherwise A23 is (M-K)-by-L upper trapezoidal. K+L = the effective */
/* numerical rank of the (M+P)-by-N matrix (A',B')'. Z' denotes the */
/* transpose of Z. */
/* This decomposition is the preprocessing step for computing the */
/* Generalized Singular Value Decomposition (GSVD), see subroutine */
/* SGGSVD. */
/* Arguments */
/* ========= */
/* JOBU (input) CHARACTER*1 */
/* = 'U': Orthogonal matrix U is computed; */
/* = 'N': U is not computed. */
/* JOBV (input) CHARACTER*1 */
/* = 'V': Orthogonal matrix V is computed; */
/* = 'N': V is not computed. */
/* JOBQ (input) CHARACTER*1 */
/* = 'Q': Orthogonal matrix Q is computed; */
/* = 'N': Q is not computed. */
/* M (input) INTEGER */
/* The number of rows of the matrix A. M >= 0. */
/* P (input) INTEGER */
/* The number of rows of the matrix B. P >= 0. */
/* N (input) INTEGER */
/* The number of columns of the matrices A and B. N >= 0. */
/* A (input/output) REAL array, dimension (LDA,N) */
/* On entry, the M-by-N matrix A. */
/* On exit, A contains the triangular (or trapezoidal) matrix */
/* described in the Purpose section. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1,M). */
/* B (input/output) REAL array, dimension (LDB,N) */
/* On entry, the P-by-N matrix B. */
/* On exit, B contains the triangular matrix described in */
/* the Purpose section. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= max(1,P). */
/* TOLA (input) REAL */
/* TOLB (input) REAL */
/* TOLA and TOLB are the thresholds to determine the effective */
/* numerical rank of matrix B and a subblock of A. Generally, */
/* they are set to */
/* TOLA = MAX(M,N)*norm(A)*MACHEPS, */
/* TOLB = MAX(P,N)*norm(B)*MACHEPS. */
/* The size of TOLA and TOLB may affect the size of backward */
/* errors of the decomposition. */
/* K (output) INTEGER */
/* L (output) INTEGER */
/* On exit, K and L specify the dimension of the subblocks */
/* described in Purpose. */
/* K + L = effective numerical rank of (A',B')'. */
/* U (output) REAL array, dimension (LDU,M) */
/* If JOBU = 'U', U contains the orthogonal matrix U. */
/* If JOBU = 'N', U is not referenced. */
/* LDU (input) INTEGER */
/* The leading dimension of the array U. LDU >= max(1,M) if */
/* JOBU = 'U'; LDU >= 1 otherwise. */
/* V (output) REAL array, dimension (LDV,M) */
/* If JOBV = 'V', V contains the orthogonal matrix V. */
/* If JOBV = 'N', V is not referenced. */
/* LDV (input) INTEGER */
/* The leading dimension of the array V. LDV >= max(1,P) if */
/* JOBV = 'V'; LDV >= 1 otherwise. */
/* Q (output) REAL array, dimension (LDQ,N) */
/* If JOBQ = 'Q', Q contains the orthogonal matrix Q. */
/* If JOBQ = 'N', Q is not referenced. */
/* LDQ (input) INTEGER */
/* The leading dimension of the array Q. LDQ >= max(1,N) if */
/* JOBQ = 'Q'; LDQ >= 1 otherwise. */
/* IWORK (workspace) INTEGER array, dimension (N) */
/* TAU (workspace) REAL array, dimension (N) */
/* WORK (workspace) REAL array, dimension (max(3*N,M,P)) */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value. */
/* Further Details */
/* =============== */
/* The subroutine uses LAPACK subroutine SGEQPF for the QR factorization */
/* with column pivoting to detect the effective numerical rank of the */
/* a matrix. It may be replaced by a better rank determination strategy. */
/* ===================================================================== */
/* .. Parameters .. */
/*< REAL ZERO, ONE >*/
/*< PARAMETER ( ZERO = 0.0E+0, ONE = 1.0E+0 ) >*/
/* .. */
/* .. Local Scalars .. */
/*< LOGICAL FORWRD, WANTQ, WANTU, WANTV >*/
/*< INTEGER I, J >*/
/* .. */
/* .. External Functions .. */
/*< LOGICAL LSAME >*/
/*< EXTERNAL LSAME >*/
/* .. */
/* .. External Subroutines .. */
/*< >*/
/* .. */
/* .. Intrinsic Functions .. */
/*< INTRINSIC ABS, MAX, MIN >*/
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters */
/*< WANTU = LSAME( JOBU, 'U' ) >*/
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
u_dim1 = *ldu;
u_offset = 1 + u_dim1;
u -= u_offset;
v_dim1 = *ldv;
v_offset = 1 + v_dim1;
v -= v_offset;
q_dim1 = *ldq;
q_offset = 1 + q_dim1;
q -= q_offset;
--iwork;
--tau;
--work;
/* Function Body */
wantu = lsame_(jobu, "U", (ftnlen)1, (ftnlen)1);
/*< WANTV = LSAME( JOBV, 'V' ) >*/
wantv = lsame_(jobv, "V", (ftnlen)1, (ftnlen)1);
/*< WANTQ = LSAME( JOBQ, 'Q' ) >*/
wantq = lsame_(jobq, "Q", (ftnlen)1, (ftnlen)1);
/*< FORWRD = .TRUE. >*/
forwrd = TRUE_;
/*< INFO = 0 >*/
*info = 0;
/*< IF( .NOT.( WANTU .OR. LSAME( JOBU, 'N' ) ) ) THEN >*/
if (! (wantu || lsame_(jobu, "N", (ftnlen)1, (ftnlen)1))) {
/*< INFO = -1 >*/
*info = -1;
/*< ELSE IF( .NOT.( WANTV .OR. LSAME( JOBV, 'N' ) ) ) THEN >*/
} else if (! (wantv || lsame_(jobv, "N", (ftnlen)1, (ftnlen)1))) {
/*< INFO = -2 >*/
*info = -2;
/*< ELSE IF( .NOT.( WANTQ .OR. LSAME( JOBQ, 'N' ) ) ) THEN >*/
} else if (! (wantq || lsame_(jobq, "N", (ftnlen)1, (ftnlen)1))) {
/*< INFO = -3 >*/
*info = -3;
/*< ELSE IF( M.LT.0 ) THEN >*/
} else if (*m < 0) {
/*< INFO = -4 >*/
*info = -4;
/*< ELSE IF( P.LT.0 ) THEN >*/
} else if (*p < 0) {
/*< INFO = -5 >*/
*info = -5;
/*< ELSE IF( N.LT.0 ) THEN >*/
} else if (*n < 0) {
/*< INFO = -6 >*/
*info = -6;
/*< ELSE IF( LDA.LT.MAX( 1, M ) ) THEN >*/
} else if (*lda < max(1,*m)) {
/*< INFO = -8 >*/
*info = -8;
/*< ELSE IF( LDB.LT.MAX( 1, P ) ) THEN >*/
} else if (*ldb < max(1,*p)) {
/*< INFO = -10 >*/
*info = -10;
/*< ELSE IF( LDU.LT.1 .OR. ( WANTU .AND. LDU.LT.M ) ) THEN >*/
} else if (*ldu < 1 || (wantu && *ldu < *m)) {
/*< INFO = -16 >*/
*info = -16;
/*< ELSE IF( LDV.LT.1 .OR. ( WANTV .AND. LDV.LT.P ) ) THEN >*/
} else if (*ldv < 1 || (wantv && *ldv < *p)) {
/*< INFO = -18 >*/
*info = -18;
/*< ELSE IF( LDQ.LT.1 .OR. ( WANTQ .AND. LDQ.LT.N ) ) THEN >*/
} else if (*ldq < 1 || (wantq && *ldq < *n)) {
/*< INFO = -20 >*/
*info = -20;
/*< END IF >*/
}
/*< IF( INFO.NE.0 ) THEN >*/
if (*info != 0) {
/*< CALL XERBLA( 'SGGSVP', -INFO ) >*/
i__1 = -(*info);
xerbla_("SGGSVP", &i__1, (ftnlen)6);
/*< RETURN >*/
return 0;
/*< END IF >*/
}
/* QR with column pivoting of B: B*P = V*( S11 S12 ) */
/* ( 0 0 ) */
/*< DO 10 I = 1, N >*/
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
/*< IWORK( I ) = 0 >*/
iwork[i__] = 0;
/*< 10 CONTINUE >*/
/* L10: */
}
/*< CALL SGEQPF( P, N, B, LDB, IWORK, TAU, WORK, INFO ) >*/
sgeqpf_(p, n, &b[b_offset], ldb, &iwork[1], &tau[1], &work[1], info);
/* Update A := A*P */
/*< CALL SLAPMT( FORWRD, M, N, A, LDA, IWORK ) >*/
slapmt_(&forwrd, m, n, &a[a_offset], lda, &iwork[1]);
/* Determine the effective rank of matrix B. */
/*< L = 0 >*/
*l = 0;
/*< DO 20 I = 1, MIN( P, N ) >*/
i__1 = min(*p,*n);
for (i__ = 1; i__ <= i__1; ++i__) {
/*< >*/
if ((r__1 = b[i__ + i__ * b_dim1], dabs(r__1)) > *tolb) {
++(*l);
}
/*< 20 CONTINUE >*/
/* L20: */
}
/*< IF( WANTV ) THEN >*/
if (wantv) {
/* Copy the details of V, and form V. */
/*< CALL SLASET( 'Full', P, P, ZERO, ZERO, V, LDV ) >*/
slaset_("Full", p, p, &c_b12, &c_b12, &v[v_offset], ldv, (ftnlen)4);
/*< >*/
if (*p > 1) {
i__1 = *p - 1;
slacpy_("Lower", &i__1, n, &b[b_dim1 + 2], ldb, &v[v_dim1 + 2],
ldv, (ftnlen)5);
}
/*< CALL SORG2R( P, P, MIN( P, N ), V, LDV, TAU, WORK, INFO ) >*/
i__1 = min(*p,*n);
sorg2r_(p, p, &i__1, &v[v_offset], ldv, &tau[1], &work[1], info);
/*< END IF >*/
}
/* Clean up B */
/*< DO 40 J = 1, L - 1 >*/
i__1 = *l - 1;
for (j = 1; j <= i__1; ++j) {
/*< DO 30 I = J + 1, L >*/
i__2 = *l;
for (i__ = j + 1; i__ <= i__2; ++i__) {
/*< B( I, J ) = ZERO >*/
b[i__ + j * b_dim1] = (float)0.;
/*< 30 CONTINUE >*/
/* L30: */
}
/*< 40 CONTINUE >*/
/* L40: */
}
/*< >*/
if (*p > *l) {
i__1 = *p - *l;
slaset_("Full", &i__1, n, &c_b12, &c_b12, &b[*l + 1 + b_dim1], ldb, (
ftnlen)4);
}
/*< IF( WANTQ ) THEN >*/
if (wantq) {
/* Set Q = I and Update Q := Q*P */
/*< CALL SLASET( 'Full', N, N, ZERO, ONE, Q, LDQ ) >*/
slaset_("Full", n, n, &c_b12, &c_b22, &q[q_offset], ldq, (ftnlen)4);
/*< CALL SLAPMT( FORWRD, N, N, Q, LDQ, IWORK ) >*/
slapmt_(&forwrd, n, n, &q[q_offset], ldq, &iwork[1]);
/*< END IF >*/
}
/*< IF( P.GE.L .AND. N.NE.L ) THEN >*/
if (*p >= *l && *n != *l) {
/* RQ factorization of (S11 S12): ( S11 S12 ) = ( 0 S12 )*Z */
/*< CALL SGERQ2( L, N, B, LDB, TAU, WORK, INFO ) >*/
sgerq2_(l, n, &b[b_offset], ldb, &tau[1], &work[1], info);
/* Update A := A*Z' */
/*< >*/
sormr2_("Right", "Transpose", m, n, l, &b[b_offset], ldb, &tau[1], &a[
a_offset], lda, &work[1], info, (ftnlen)5, (ftnlen)9);
/*< IF( WANTQ ) THEN >*/
if (wantq) {
/* Update Q := Q*Z' */
/*< >*/
sormr2_("Right", "Transpose", n, n, l, &b[b_offset], ldb, &tau[1],
&q[q_offset], ldq, &work[1], info, (ftnlen)5, (ftnlen)9);
/*< END IF >*/
}
/* Clean up B */
/*< CALL SLASET( 'Full', L, N-L, ZERO, ZERO, B, LDB ) >*/
i__1 = *n - *l;
slaset_("Full", l, &i__1, &c_b12, &c_b12, &b[b_offset], ldb, (ftnlen)
4);
/*< DO 60 J = N - L + 1, N >*/
i__1 = *n;
for (j = *n - *l + 1; j <= i__1; ++j) {
/*< DO 50 I = J - N + L + 1, L >*/
i__2 = *l;
for (i__ = j - *n + *l + 1; i__ <= i__2; ++i__) {
/*< B( I, J ) = ZERO >*/
b[i__ + j * b_dim1] = (float)0.;
/*< 50 CONTINUE >*/
/* L50: */
}
/*< 60 CONTINUE >*/
/* L60: */
}
/*< END IF >*/
}
/* Let N-L L */
/* A = ( A11 A12 ) M, */
/* then the following does the complete QR decomposition of A11: */
/* A11 = U*( 0 T12 )*P1' */
/* ( 0 0 ) */
/*< DO 70 I = 1, N - L >*/
i__1 = *n - *l;
for (i__ = 1; i__ <= i__1; ++i__) {
/*< IWORK( I ) = 0 >*/
iwork[i__] = 0;
/*< 70 CONTINUE >*/
/* L70: */
}
/*< CALL SGEQPF( M, N-L, A, LDA, IWORK, TAU, WORK, INFO ) >*/
i__1 = *n - *l;
sgeqpf_(m, &i__1, &a[a_offset], lda, &iwork[1], &tau[1], &work[1], info);
/* Determine the effective rank of A11 */
/*< K = 0 >*/
*k = 0;
/*< DO 80 I = 1, MIN( M, N-L ) >*/
/* Computing MIN */
i__2 = *m, i__3 = *n - *l;
i__1 = min(i__2,i__3);
for (i__ = 1; i__ <= i__1; ++i__) {
/*< >*/
if ((r__1 = a[i__ + i__ * a_dim1], dabs(r__1)) > *tola) {
++(*k);
}
/*< 80 CONTINUE >*/
/* L80: */
}
/* Update A12 := U'*A12, where A12 = A( 1:M, N-L+1:N ) */
/*< >*/
/* Computing MIN */
i__2 = *m, i__3 = *n - *l;
i__1 = min(i__2,i__3);
sorm2r_("Left", "Transpose", m, l, &i__1, &a[a_offset], lda, &tau[1], &a[(
*n - *l + 1) * a_dim1 + 1], lda, &work[1], info, (ftnlen)4, (
ftnlen)9);
/*< IF( WANTU ) THEN >*/
if (wantu) {
/* Copy the details of U, and form U */
/*< CALL SLASET( 'Full', M, M, ZERO, ZERO, U, LDU ) >*/
slaset_("Full", m, m, &c_b12, &c_b12, &u[u_offset], ldu, (ftnlen)4);
/*< >*/
if (*m > 1) {
i__1 = *m - 1;
i__2 = *n - *l;
slacpy_("Lower", &i__1, &i__2, &a[a_dim1 + 2], lda, &u[u_dim1 + 2]
, ldu, (ftnlen)5);
}
/*< CALL SORG2R( M, M, MIN( M, N-L ), U, LDU, TAU, WORK, INFO ) >*/
/* Computing MIN */
i__2 = *m, i__3 = *n - *l;
i__1 = min(i__2,i__3);
sorg2r_(m, m, &i__1, &u[u_offset], ldu, &tau[1], &work[1], info);
/*< END IF >*/
}
/*< IF( WANTQ ) THEN >*/
if (wantq) {
/* Update Q( 1:N, 1:N-L ) = Q( 1:N, 1:N-L )*P1 */
/*< CALL SLAPMT( FORWRD, N, N-L, Q, LDQ, IWORK ) >*/
i__1 = *n - *l;
slapmt_(&forwrd, n, &i__1, &q[q_offset], ldq, &iwork[1]);
/*< END IF >*/
}
/* Clean up A: set the strictly lower triangular part of */
/* A(1:K, 1:K) = 0, and A( K+1:M, 1:N-L ) = 0. */
/*< DO 100 J = 1, K - 1 >*/
i__1 = *k - 1;
for (j = 1; j <= i__1; ++j) {
/*< DO 90 I = J + 1, K >*/
i__2 = *k;
for (i__ = j + 1; i__ <= i__2; ++i__) {
/*< A( I, J ) = ZERO >*/
a[i__ + j * a_dim1] = (float)0.;
/*< 90 CONTINUE >*/
/* L90: */
}
/*< 100 CONTINUE >*/
/* L100: */
}
/*< >*/
if (*m > *k) {
i__1 = *m - *k;
i__2 = *n - *l;
slaset_("Full", &i__1, &i__2, &c_b12, &c_b12, &a[*k + 1 + a_dim1],
lda, (ftnlen)4);
}
/*< IF( N-L.GT.K ) THEN >*/
if (*n - *l > *k) {
/* RQ factorization of ( T11 T12 ) = ( 0 T12 )*Z1 */
/*< CALL SGERQ2( K, N-L, A, LDA, TAU, WORK, INFO ) >*/
i__1 = *n - *l;
sgerq2_(k, &i__1, &a[a_offset], lda, &tau[1], &work[1], info);
/*< IF( WANTQ ) THEN >*/
if (wantq) {
/* Update Q( 1:N,1:N-L ) = Q( 1:N,1:N-L )*Z1' */
/*< >*/
i__1 = *n - *l;
sormr2_("Right", "Transpose", n, &i__1, k, &a[a_offset], lda, &
tau[1], &q[q_offset], ldq, &work[1], info, (ftnlen)5, (
ftnlen)9);
/*< END IF >*/
}
/* Clean up A */
/*< CALL SLASET( 'Full', K, N-L-K, ZERO, ZERO, A, LDA ) >*/
i__1 = *n - *l - *k;
slaset_("Full", k, &i__1, &c_b12, &c_b12, &a[a_offset], lda, (ftnlen)
4);
/*< DO 120 J = N - L - K + 1, N - L >*/
i__1 = *n - *l;
for (j = *n - *l - *k + 1; j <= i__1; ++j) {
/*< DO 110 I = J - N + L + K + 1, K >*/
i__2 = *k;
for (i__ = j - *n + *l + *k + 1; i__ <= i__2; ++i__) {
/*< A( I, J ) = ZERO >*/
a[i__ + j * a_dim1] = (float)0.;
/*< 110 CONTINUE >*/
/* L110: */
}
/*< 120 CONTINUE >*/
/* L120: */
}
/*< END IF >*/
}
/*< IF( M.GT.K ) THEN >*/
if (*m > *k) {
/* QR factorization of A( K+1:M,N-L+1:N ) */
/*< CALL SGEQR2( M-K, L, A( K+1, N-L+1 ), LDA, TAU, WORK, INFO ) >*/
i__1 = *m - *k;
sgeqr2_(&i__1, l, &a[*k + 1 + (*n - *l + 1) * a_dim1], lda, &tau[1], &
work[1], info);
/*< IF( WANTU ) THEN >*/
if (wantu) {
/* Update U(:,K+1:M) := U(:,K+1:M)*U1 */
/*< >*/
i__1 = *m - *k;
/* Computing MIN */
i__3 = *m - *k;
i__2 = min(i__3,*l);
sorm2r_("Right", "No transpose", m, &i__1, &i__2, &a[*k + 1 + (*n
- *l + 1) * a_dim1], lda, &tau[1], &u[(*k + 1) * u_dim1 +
1], ldu, &work[1], info, (ftnlen)5, (ftnlen)12);
/*< END IF >*/
}
/* Clean up */
/*< DO 140 J = N - L + 1, N >*/
i__1 = *n;
for (j = *n - *l + 1; j <= i__1; ++j) {
/*< DO 130 I = J - N + K + L + 1, M >*/
i__2 = *m;
for (i__ = j - *n + *k + *l + 1; i__ <= i__2; ++i__) {
/*< A( I, J ) = ZERO >*/
a[i__ + j * a_dim1] = (float)0.;
/*< 130 CONTINUE >*/
/* L130: */
}
/*< 140 CONTINUE >*/
/* L140: */
}
/*< END IF >*/
}
/*< RETURN >*/
return 0;
/* End of SGGSVP */
/*< END >*/
} /* sggsvp_ */
/* Subroutine */ int sgerqf_(integer *m, integer *n, real *a, integer *lda,
real *tau, real *work, integer *lwork, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, i__1, i__2, i__3, i__4;
/* Local variables */
integer i__, k, ib, nb, ki, kk, mu, nu, nx, iws, nbmin, iinfo;
extern /* Subroutine */ int sgerq2_(integer *, integer *, real *, integer
*, real *, real *, integer *), slarfb_(char *, char *, char *,
char *, integer *, integer *, integer *, real *, integer *, real *
, integer *, real *, integer *, real *, integer *), xerbla_(char *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *);
extern /* Subroutine */ int slarft_(char *, char *, integer *, integer *,
real *, integer *, real *, real *, integer *);
integer ldwork, lwkopt;
logical lquery;
/* -- LAPACK routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SGERQF computes an RQ factorization of a real M-by-N matrix A: */
/* A = R * Q. */
/* Arguments */
/* ========= */
/* M (input) INTEGER */
/* The number of rows of the matrix A. M >= 0. */
/* N (input) INTEGER */
/* The number of columns of the matrix A. N >= 0. */
/* A (input/output) REAL array, dimension (LDA,N) */
/* On entry, the M-by-N matrix A. */
/* On exit, */
/* if m <= n, the upper triangle of the subarray */
/* A(1:m,n-m+1:n) contains the M-by-M upper triangular matrix R; */
/* if m >= n, the elements on and above the (m-n)-th subdiagonal */
/* contain the M-by-N upper trapezoidal matrix R; */
/* the remaining elements, with the array TAU, represent the */
/* orthogonal matrix Q as a product of min(m,n) elementary */
/* reflectors (see Further Details). */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1,M). */
/* TAU (output) REAL array, dimension (min(M,N)) */
/* The scalar factors of the elementary reflectors (see Further */
/* Details). */
/* 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,M). */
/* For optimum performance LWORK >= M*NB, where NB is */
/* the optimal blocksize. */
/* If LWORK = -1, then a workspace query is assumed; the routine */
/* only calculates the optimal size of the WORK array, returns */
/* this value as the first entry of the WORK array, and no error */
/* message related to LWORK is issued by XERBLA. */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* Further Details */
/* =============== */
/* The matrix Q is represented as a product of elementary reflectors */
/* Q = H(1) H(2) . . . H(k), where k = min(m,n). */
/* Each H(i) has the form */
/* H(i) = I - tau * v * v' */
/* where tau is a real scalar, and v is a real vector with */
/* v(n-k+i+1:n) = 0 and v(n-k+i) = 1; v(1:n-k+i-1) is stored on exit in */
/* A(m-k+i,1:n-k+i-1), and tau in TAU(i). */
/* ===================================================================== */
/* .. Local Scalars .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input arguments */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
--tau;
--work;
/* Function Body */
*info = 0;
lquery = *lwork == -1;
if (*m < 0) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*lda < max(1,*m)) {
*info = -4;
} else if (*lwork < max(1,*m) && ! lquery) {
*info = -7;
}
if (*info == 0) {
k = min(*m,*n);
if (k == 0) {
lwkopt = 1;
} else {
nb = ilaenv_(&c__1, "SGERQF", " ", m, n, &c_n1, &c_n1);
lwkopt = *m * nb;
work[1] = (real) lwkopt;
}
work[1] = (real) lwkopt;
if (*lwork < max(1,*m) && ! lquery) {
*info = -7;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SGERQF", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (k == 0) {
return 0;
}
nbmin = 2;
nx = 1;
iws = *m;
if (nb > 1 && nb < k) {
/* Determine when to cross over from blocked to unblocked code. */
/* Computing MAX */
i__1 = 0, i__2 = ilaenv_(&c__3, "SGERQF", " ", m, n, &c_n1, &c_n1);
nx = max(i__1,i__2);
if (nx < k) {
/* Determine if workspace is large enough for blocked code. */
ldwork = *m;
iws = ldwork * nb;
if (*lwork < iws) {
/* Not enough workspace to use optimal NB: reduce NB and */
/* determine the minimum value of NB. */
nb = *lwork / ldwork;
/* Computing MAX */
i__1 = 2, i__2 = ilaenv_(&c__2, "SGERQF", " ", m, n, &c_n1, &
c_n1);
nbmin = max(i__1,i__2);
}
}
}
if (nb >= nbmin && nb < k && nx < k) {
/* Use blocked code initially. */
/* The last kk rows are handled by the block method. */
ki = (k - nx - 1) / nb * nb;
/* Computing MIN */
i__1 = k, i__2 = ki + nb;
kk = min(i__1,i__2);
i__1 = k - kk + 1;
i__2 = -nb;
for (i__ = k - kk + ki + 1; i__2 < 0 ? i__ >= i__1 : i__ <= i__1; i__
+= i__2) {
/* Computing MIN */
i__3 = k - i__ + 1;
ib = min(i__3,nb);
/* Compute the RQ factorization of the current block */
/* A(m-k+i:m-k+i+ib-1,1:n-k+i+ib-1) */
i__3 = *n - k + i__ + ib - 1;
sgerq2_(&ib, &i__3, &a[*m - k + i__ + a_dim1], lda, &tau[i__], &
work[1], &iinfo);
if (*m - k + i__ > 1) {
/* Form the triangular factor of the block reflector */
/* H = H(i+ib-1) . . . H(i+1) H(i) */
i__3 = *n - k + i__ + ib - 1;
slarft_("Backward", "Rowwise", &i__3, &ib, &a[*m - k + i__ +
a_dim1], lda, &tau[i__], &work[1], &ldwork);
/* Apply H to A(1:m-k+i-1,1:n-k+i+ib-1) from the right */
i__3 = *m - k + i__ - 1;
i__4 = *n - k + i__ + ib - 1;
slarfb_("Right", "No transpose", "Backward", "Rowwise", &i__3,
&i__4, &ib, &a[*m - k + i__ + a_dim1], lda, &work[1],
&ldwork, &a[a_offset], lda, &work[ib + 1], &ldwork);
}
/* L10: */
}
mu = *m - k + i__ + nb - 1;
nu = *n - k + i__ + nb - 1;
} else {
mu = *m;
nu = *n;
}
/* Use unblocked code to factor the last or only block */
if (mu > 0 && nu > 0) {
sgerq2_(&mu, &nu, &a[a_offset], lda, &tau[1], &work[1], &iinfo);
}
work[1] = (real) iws;
return 0;
/* End of SGERQF */
} /* sgerqf_ */
/* Subroutine */
int sggsvp_(char *jobu, char *jobv, char *jobq, integer *m, integer *p, integer *n, real *a, integer *lda, real *b, integer *ldb, real *tola, real *tolb, integer *k, integer *l, real *u, integer *ldu, real *v, integer *ldv, real *q, integer *ldq, integer *iwork, real * tau, real *work, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, q_dim1, q_offset, u_dim1, u_offset, v_dim1, v_offset, i__1, i__2, i__3;
real r__1;
/* Local variables */
integer i__, j;
extern logical lsame_(char *, char *);
logical wantq, wantu, wantv;
extern /* Subroutine */
int sgeqr2_(integer *, integer *, real *, integer *, real *, real *, integer *), sgerq2_(integer *, integer *, real *, integer *, real *, real *, integer *), sorg2r_(integer *, integer *, integer *, real *, integer *, real *, real *, integer * ), sorm2r_(char *, char *, integer *, integer *, integer *, real * , integer *, real *, real *, integer *, real *, integer *), sormr2_(char *, char *, integer *, integer *, integer *, real *, integer *, real *, real *, integer *, real *, integer *), xerbla_(char *, integer *), sgeqpf_( integer *, integer *, real *, integer *, integer *, real *, real * , integer *), slacpy_(char *, integer *, integer *, real *, integer *, real *, integer *), slaset_(char *, integer *, integer *, real *, real *, real *, integer *), slapmt_( logical *, integer *, integer *, real *, integer *, integer *);
logical forwrd;
/* -- LAPACK computational routine (version 3.4.0) -- */
/* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
/* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
/* November 2011 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
u_dim1 = *ldu;
u_offset = 1 + u_dim1;
u -= u_offset;
v_dim1 = *ldv;
v_offset = 1 + v_dim1;
v -= v_offset;
q_dim1 = *ldq;
q_offset = 1 + q_dim1;
q -= q_offset;
--iwork;
--tau;
--work;
/* Function Body */
wantu = lsame_(jobu, "U");
wantv = lsame_(jobv, "V");
wantq = lsame_(jobq, "Q");
forwrd = TRUE_;
*info = 0;
if (! (wantu || lsame_(jobu, "N")))
{
*info = -1;
}
else if (! (wantv || lsame_(jobv, "N")))
{
*info = -2;
}
else if (! (wantq || lsame_(jobq, "N")))
{
*info = -3;
}
else if (*m < 0)
{
*info = -4;
}
else if (*p < 0)
{
*info = -5;
}
else if (*n < 0)
{
*info = -6;
}
else if (*lda < max(1,*m))
{
*info = -8;
}
else if (*ldb < max(1,*p))
{
*info = -10;
}
else if (*ldu < 1 || wantu && *ldu < *m)
{
*info = -16;
}
else if (*ldv < 1 || wantv && *ldv < *p)
{
*info = -18;
}
else if (*ldq < 1 || wantq && *ldq < *n)
{
*info = -20;
}
if (*info != 0)
{
i__1 = -(*info);
xerbla_("SGGSVP", &i__1);
return 0;
}
/* QR with column pivoting of B: B*P = V*( S11 S12 ) */
/* ( 0 0 ) */
i__1 = *n;
for (i__ = 1;
i__ <= i__1;
++i__)
{
iwork[i__] = 0;
/* L10: */
}
sgeqpf_(p, n, &b[b_offset], ldb, &iwork[1], &tau[1], &work[1], info);
/* Update A := A*P */
slapmt_(&forwrd, m, n, &a[a_offset], lda, &iwork[1]);
/* Determine the effective rank of matrix B. */
*l = 0;
i__1 = min(*p,*n);
for (i__ = 1;
i__ <= i__1;
++i__)
{
if ((r__1 = b[i__ + i__ * b_dim1], f2c_abs(r__1)) > *tolb)
{
++(*l);
}
/* L20: */
}
if (wantv)
{
/* Copy the details of V, and form V. */
slaset_("Full", p, p, &c_b12, &c_b12, &v[v_offset], ldv);
if (*p > 1)
{
i__1 = *p - 1;
slacpy_("Lower", &i__1, n, &b[b_dim1 + 2], ldb, &v[v_dim1 + 2], ldv);
}
i__1 = min(*p,*n);
sorg2r_(p, p, &i__1, &v[v_offset], ldv, &tau[1], &work[1], info);
}
/* Clean up B */
i__1 = *l - 1;
for (j = 1;
j <= i__1;
++j)
{
i__2 = *l;
for (i__ = j + 1;
i__ <= i__2;
++i__)
{
b[i__ + j * b_dim1] = 0.f;
/* L30: */
}
/* L40: */
}
if (*p > *l)
{
i__1 = *p - *l;
slaset_("Full", &i__1, n, &c_b12, &c_b12, &b[*l + 1 + b_dim1], ldb);
}
if (wantq)
{
/* Set Q = I and Update Q := Q*P */
slaset_("Full", n, n, &c_b12, &c_b22, &q[q_offset], ldq);
slapmt_(&forwrd, n, n, &q[q_offset], ldq, &iwork[1]);
}
if (*p >= *l && *n != *l)
{
/* RQ factorization of (S11 S12): ( S11 S12 ) = ( 0 S12 )*Z */
sgerq2_(l, n, &b[b_offset], ldb, &tau[1], &work[1], info);
/* Update A := A*Z**T */
sormr2_("Right", "Transpose", m, n, l, &b[b_offset], ldb, &tau[1], &a[ a_offset], lda, &work[1], info);
if (wantq)
{
/* Update Q := Q*Z**T */
sormr2_("Right", "Transpose", n, n, l, &b[b_offset], ldb, &tau[1], &q[q_offset], ldq, &work[1], info);
}
/* Clean up B */
i__1 = *n - *l;
slaset_("Full", l, &i__1, &c_b12, &c_b12, &b[b_offset], ldb);
i__1 = *n;
for (j = *n - *l + 1;
j <= i__1;
++j)
{
i__2 = *l;
for (i__ = j - *n + *l + 1;
i__ <= i__2;
++i__)
{
b[i__ + j * b_dim1] = 0.f;
/* L50: */
}
/* L60: */
}
}
/* Let N-L L */
/* A = ( A11 A12 ) M, */
/* then the following does the complete QR decomposition of A11: */
/* A11 = U*( 0 T12 )*P1**T */
/* ( 0 0 ) */
i__1 = *n - *l;
for (i__ = 1;
i__ <= i__1;
++i__)
{
iwork[i__] = 0;
/* L70: */
}
i__1 = *n - *l;
sgeqpf_(m, &i__1, &a[a_offset], lda, &iwork[1], &tau[1], &work[1], info);
/* Determine the effective rank of A11 */
*k = 0;
/* Computing MIN */
i__2 = *m;
i__3 = *n - *l; // , expr subst
i__1 = min(i__2,i__3);
for (i__ = 1;
i__ <= i__1;
++i__)
{
if ((r__1 = a[i__ + i__ * a_dim1], f2c_abs(r__1)) > *tola)
{
++(*k);
}
/* L80: */
}
/* Update A12 := U**T*A12, where A12 = A( 1:M, N-L+1:N ) */
/* Computing MIN */
i__2 = *m;
i__3 = *n - *l; // , expr subst
i__1 = min(i__2,i__3);
sorm2r_("Left", "Transpose", m, l, &i__1, &a[a_offset], lda, &tau[1], &a[( *n - *l + 1) * a_dim1 + 1], lda, &work[1], info);
if (wantu)
{
/* Copy the details of U, and form U */
slaset_("Full", m, m, &c_b12, &c_b12, &u[u_offset], ldu);
if (*m > 1)
{
i__1 = *m - 1;
i__2 = *n - *l;
slacpy_("Lower", &i__1, &i__2, &a[a_dim1 + 2], lda, &u[u_dim1 + 2] , ldu);
}
/* Computing MIN */
i__2 = *m;
i__3 = *n - *l; // , expr subst
i__1 = min(i__2,i__3);
sorg2r_(m, m, &i__1, &u[u_offset], ldu, &tau[1], &work[1], info);
}
if (wantq)
{
/* Update Q( 1:N, 1:N-L ) = Q( 1:N, 1:N-L )*P1 */
i__1 = *n - *l;
slapmt_(&forwrd, n, &i__1, &q[q_offset], ldq, &iwork[1]);
}
/* Clean up A: set the strictly lower triangular part of */
/* A(1:K, 1:K) = 0, and A( K+1:M, 1:N-L ) = 0. */
i__1 = *k - 1;
for (j = 1;
j <= i__1;
++j)
{
i__2 = *k;
for (i__ = j + 1;
i__ <= i__2;
++i__)
{
a[i__ + j * a_dim1] = 0.f;
/* L90: */
}
/* L100: */
}
if (*m > *k)
{
i__1 = *m - *k;
i__2 = *n - *l;
slaset_("Full", &i__1, &i__2, &c_b12, &c_b12, &a[*k + 1 + a_dim1], lda);
}
if (*n - *l > *k)
{
/* RQ factorization of ( T11 T12 ) = ( 0 T12 )*Z1 */
i__1 = *n - *l;
sgerq2_(k, &i__1, &a[a_offset], lda, &tau[1], &work[1], info);
if (wantq)
{
/* Update Q( 1:N,1:N-L ) = Q( 1:N,1:N-L )*Z1**T */
i__1 = *n - *l;
sormr2_("Right", "Transpose", n, &i__1, k, &a[a_offset], lda, & tau[1], &q[q_offset], ldq, &work[1], info);
}
/* Clean up A */
i__1 = *n - *l - *k;
slaset_("Full", k, &i__1, &c_b12, &c_b12, &a[a_offset], lda);
i__1 = *n - *l;
for (j = *n - *l - *k + 1;
j <= i__1;
++j)
{
i__2 = *k;
for (i__ = j - *n + *l + *k + 1;
i__ <= i__2;
++i__)
{
a[i__ + j * a_dim1] = 0.f;
/* L110: */
}
/* L120: */
}
}
if (*m > *k)
{
/* QR factorization of A( K+1:M,N-L+1:N ) */
i__1 = *m - *k;
sgeqr2_(&i__1, l, &a[*k + 1 + (*n - *l + 1) * a_dim1], lda, &tau[1], & work[1], info);
if (wantu)
{
/* Update U(:,K+1:M) := U(:,K+1:M)*U1 */
i__1 = *m - *k;
/* Computing MIN */
i__3 = *m - *k;
i__2 = min(i__3,*l);
sorm2r_("Right", "No transpose", m, &i__1, &i__2, &a[*k + 1 + (*n - *l + 1) * a_dim1], lda, &tau[1], &u[(*k + 1) * u_dim1 + 1], ldu, &work[1], info);
}
/* Clean up */
i__1 = *n;
for (j = *n - *l + 1;
j <= i__1;
++j)
{
i__2 = *m;
for (i__ = j - *n + *k + *l + 1;
i__ <= i__2;
++i__)
{
a[i__ + j * a_dim1] = 0.f;
/* L130: */
}
/* L140: */
}
}
return 0;
/* End of SGGSVP */
}
