/* Subroutine */ int dtimqp_(char *line, integer *nm, integer *mval, integer *
nval, integer *nlda, integer *ldaval, doublereal *timmin, doublereal *
a, doublereal *copya, doublereal *tau, doublereal *work, integer *
iwork, doublereal *reslts, integer *ldr1, integer *ldr2, integer *
nout, ftnlen line_len)
{
/* Initialized data */
static char subnam[6*1] = "DGEQPF";
static integer modes[2] = { 2,3 };
static integer iseed[4] = { 0,0,0,1 };
/* Format strings */
static char fmt_9999[] = "(1x,a6,\002 timing run not attempted\002,/)";
static char fmt_9998[] = "(/\002 *** Speed of \002,a6,\002 in megaflops "
"***\002)";
static char fmt_9997[] = "(5x,\002line \002,i2,\002 with LDA = \002,i5)";
/* System generated locals */
integer reslts_dim1, reslts_dim2, reslts_offset, i__1, i__2, i__3;
/* Builtin functions
Subroutine */ int s_copy(char *, char *, ftnlen, ftnlen);
integer s_wsfe(cilist *), do_fio(integer *, char *, ftnlen), e_wsfe(void),
s_wsle(cilist *), e_wsle(void);
/* Local variables */
static integer ilda;
static doublereal cond;
static integer mode;
static doublereal dmax__;
static integer info;
static char path[3];
static doublereal time;
static integer i__, m, n;
static char cname[6];
static integer imode;
extern doublereal dopla_(char *, integer *, integer *, integer *, integer
*, integer *);
static integer minmn;
extern /* Subroutine */ int icopy_(integer *, integer *, integer *,
integer *, integer *);
static doublereal s1, s2;
extern /* Subroutine */ int dprtb5_(char *, char *, char *, integer *,
integer *, integer *, integer *, integer *, integer *, doublereal
*, integer *, integer *, integer *, ftnlen, ftnlen, ftnlen);
static integer ic, im;
extern doublereal dlamch_(char *), dsecnd_(void);
extern /* Subroutine */ int dgeqpf_(integer *, integer *, doublereal *,
integer *, integer *, doublereal *, doublereal *, integer *),
atimck_(integer *, char *, integer *, integer *, integer *,
integer *, integer *, integer *, ftnlen), dlacpy_(char *, integer
*, integer *, doublereal *, integer *, doublereal *, integer *);
extern doublereal dmflop_(doublereal *, doublereal *, integer *);
extern /* Subroutine */ int atimin_(char *, char *, integer *, char *,
logical *, integer *, integer *, ftnlen, ftnlen, ftnlen), dlatms_(
integer *, integer *, char *, integer *, char *, doublereal *,
integer *, doublereal *, doublereal *, integer *, integer *, char
*, doublereal *, integer *, doublereal *, integer *);
static doublereal untime;
static logical timsub[1];
static integer lda, icl;
static doublereal ops;
/* Fortran I/O blocks */
static cilist io___8 = { 0, 0, 0, fmt_9999, 0 };
static cilist io___27 = { 0, 0, 0, fmt_9998, 0 };
static cilist io___28 = { 0, 0, 0, fmt_9997, 0 };
static cilist io___29 = { 0, 0, 0, 0, 0 };
#define subnam_ref(a_0,a_1) &subnam[(a_1)*6 + a_0 - 6]
#define reslts_ref(a_1,a_2,a_3) reslts[((a_3)*reslts_dim2 + (a_2))*\
reslts_dim1 + a_1]
/* -- LAPACK timing 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
=======
DTIMQP times the LAPACK routines to perform the QR factorization with
column pivoting of a DOUBLE PRECISION general matrix.
Two matrix types may be used for timing. The number of types is
set in the parameter NMODE and the matrix types are set in the vector
MODES, using the following key:
2. BREAK1 D(1:N-1)=1 and D(N)=1.0/COND in DLATMS
3. GEOM D(I)=COND**(-(I-1)/(N-1)) in DLATMS
These numbers are chosen to correspond with the matrix types in the
test code.
Arguments
=========
LINE (input) CHARACTER*80
The input line that requested this routine. The first six
characters contain either the name of a subroutine or a
generic path name. The remaining characters may be used to
specify the individual routines to be timed. See ATIMIN for
a full description of the format of the input line.
NM (input) INTEGER
The number of values of M and N contained in the vectors
MVAL and NVAL. The matrix sizes are used in pairs (M,N).
MVAL (input) INTEGER array, dimension (NM)
The values of the matrix row dimension M.
NVAL (input) INTEGER array, dimension (NM)
The values of the matrix column dimension N.
NLDA (input) INTEGER
The number of values of LDA contained in the vector LDAVAL.
LDAVAL (input) INTEGER array, dimension (NLDA)
The values of the leading dimension of the array A.
TIMMIN (input) DOUBLE PRECISION
The minimum time a subroutine will be timed.
A (workspace) DOUBLE PRECISION array, dimension (LDAMAX*NMAX)
where LDAMAX and NMAX are the maximum values of LDA and N.
COPYA (workspace) DOUBLE PRECISION array, dimension (LDAMAX*NMAX)
TAU (workspace) DOUBLE PRECISION array, dimension (min(M,N))
WORK (workspace) DOUBLE PRECISION array, dimension (3*NMAX)
IWORK (workspace) INTEGER array, dimension (2*NMAX)
RESLTS (workspace) DOUBLE PRECISION array, dimension
(LDR1,LDR2,NLDA)
The timing results for each subroutine over the relevant
values of MODE, (M,N), and LDA.
LDR1 (input) INTEGER
The first dimension of RESLTS. LDR1 >= max(1,NM).
LDR2 (input) INTEGER
The second dimension of RESLTS. LDR2 >= max(1,NM).
NOUT (input) INTEGER
The unit number for output.
=====================================================================
Parameter adjustments */
--mval;
--nval;
--ldaval;
--a;
--copya;
--tau;
--work;
--iwork;
reslts_dim1 = *ldr1;
reslts_dim2 = *ldr2;
reslts_offset = 1 + reslts_dim1 * (1 + reslts_dim2 * 1);
reslts -= reslts_offset;
/* Function Body
Extract the timing request from the input line. */
s_copy(path, "Double precision", (ftnlen)1, (ftnlen)16);
s_copy(path + 1, "QP", (ftnlen)2, (ftnlen)2);
atimin_(path, line, &c__1, subnam, timsub, nout, &info, (ftnlen)3, (
ftnlen)80, (ftnlen)6);
if (! timsub[0] || info != 0) {
goto L80;
}
/* Check that M <= LDA for the input values. */
s_copy(cname, line, (ftnlen)6, (ftnlen)6);
atimck_(&c__1, cname, nm, &mval[1], nlda, &ldaval[1], nout, &info, (
ftnlen)6);
if (info > 0) {
io___8.ciunit = *nout;
s_wsfe(&io___8);
do_fio(&c__1, cname, (ftnlen)6);
e_wsfe();
goto L80;
}
/* Set the condition number and scaling factor for the matrices
to be generated. */
dmax__ = 1.;
cond = 1. / dlamch_("Precision");
/* Do for each pair of values (M,N): */
i__1 = *nm;
for (im = 1; im <= i__1; ++im) {
m = mval[im];
n = nval[im];
minmn = min(m,n);
/* Do for each value of LDA: */
i__2 = *nlda;
for (ilda = 1; ilda <= i__2; ++ilda) {
lda = ldaval[ilda];
for (imode = 1; imode <= 2; ++imode) {
mode = modes[imode - 1];
/* Generate a test matrix of size m by n using the
singular value distribution indicated by MODE. */
i__3 = n;
for (i__ = 1; i__ <= i__3; ++i__) {
iwork[n + i__] = 0;
/* L10: */
}
dlatms_(&m, &n, "Uniform", iseed, "Nonsymm", &tau[1], &mode, &
cond, &dmax__, &m, &n, "No packing", ©a[1], &lda,
&work[1], &info);
/* DGEQPF: QR factorization with column pivoting */
dlacpy_("All", &m, &n, ©a[1], &lda, &a[1], &lda);
icopy_(&n, &iwork[n + 1], &c__1, &iwork[1], &c__1);
ic = 0;
s1 = dsecnd_();
L20:
dgeqpf_(&m, &n, &a[1], &lda, &iwork[1], &tau[1], &work[1], &
info);
s2 = dsecnd_();
time = s2 - s1;
++ic;
if (time < *timmin) {
dlacpy_("All", &m, &n, ©a[1], &lda, &a[1], &lda);
icopy_(&n, &iwork[n + 1], &c__1, &iwork[1], &c__1);
goto L20;
}
/* Subtract the time used in DLACPY and ICOPY. */
icl = 1;
s1 = dsecnd_();
L30:
s2 = dsecnd_();
untime = s2 - s1;
++icl;
if (icl <= ic) {
dlacpy_("All", &m, &n, ©a[1], &lda, &a[1], &lda);
icopy_(&n, &iwork[n + 1], &c__1, &iwork[1], &c__1);
goto L30;
}
time = (time - untime) / (doublereal) ic;
ops = dopla_("DGEQPF", &m, &n, &c__0, &c__0, &c__1)
;
reslts_ref(imode, im, ilda) = dmflop_(&ops, &time, &info);
/* L40: */
}
/* L50: */
}
/* L60: */
}
/* Print tables of results */
io___27.ciunit = *nout;
s_wsfe(&io___27);
do_fio(&c__1, subnam_ref(0, 1), (ftnlen)6);
e_wsfe();
if (*nlda > 1) {
i__1 = *nlda;
for (i__ = 1; i__ <= i__1; ++i__) {
io___28.ciunit = *nout;
s_wsfe(&io___28);
do_fio(&c__1, (char *)&i__, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&ldaval[i__], (ftnlen)sizeof(integer));
e_wsfe();
/* L70: */
}
}
io___29.ciunit = *nout;
s_wsle(&io___29);
e_wsle();
dprtb5_("Type", "M", "N", &c__2, modes, nm, &mval[1], &nval[1], nlda, &
reslts[reslts_offset], ldr1, ldr2, nout, (ftnlen)4, (ftnlen)1, (
ftnlen)1);
L80:
return 0;
/* End of DTIMQP */
} /* dtimqp_ */
/* Subroutine */ int dchkqp_(logical *dotype, integer *nm, integer *mval,
integer *nn, integer *nval, doublereal *thresh, logical *tsterr,
doublereal *a, doublereal *copya, doublereal *s, doublereal *copys,
doublereal *tau, doublereal *work, integer *iwork, integer *nout)
{
/* Initialized data */
static integer iseedy[4] = { 1988,1989,1990,1991 };
/* Format strings */
static char fmt_9999[] = "(\002 M =\002,i5,\002, N =\002,i5,\002, type"
" \002,i2,\002, test \002,i2,\002, ratio =\002,g12.5)";
/* System generated locals */
integer i__1, i__2, i__3, i__4;
doublereal d__1;
/* Local variables */
integer i__, k, m, n, im, in, lda;
doublereal eps;
integer mode, info;
char path[3];
integer ilow, nrun;
integer ihigh, nfail, iseed[4], imode;
integer mnmin, istep, nerrs, lwork;
doublereal result[3];
/* Fortran I/O blocks */
static cilist io___24 = { 0, 0, 0, fmt_9999, 0 };
/* -- LAPACK test routine (version 3.1.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* January 2007 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* DCHKQP tests DGEQPF. */
/* Arguments */
/* ========= */
/* DOTYPE (input) LOGICAL array, dimension (NTYPES) */
/* The matrix types to be used for testing. Matrices of type j */
/* (for 1 <= j <= NTYPES) are used for testing if DOTYPE(j) = */
/* .TRUE.; if DOTYPE(j) = .FALSE., then type j is not used. */
/* NM (input) INTEGER */
/* The number of values of M contained in the vector MVAL. */
/* MVAL (input) INTEGER array, dimension (NM) */
/* The values of the matrix row dimension M. */
/* NN (input) INTEGER */
/* The number of values of N contained in the vector NVAL. */
/* NVAL (input) INTEGER array, dimension (NN) */
/* The values of the matrix column dimension N. */
/* THRESH (input) DOUBLE PRECISION */
/* The threshold value for the test ratios. A result is */
/* included in the output file if RESULT >= THRESH. To have */
/* every test ratio printed, use THRESH = 0. */
/* TSTERR (input) LOGICAL */
/* Flag that indicates whether error exits are to be tested. */
/* A (workspace) DOUBLE PRECISION array, dimension (MMAX*NMAX) */
/* where MMAX is the maximum value of M in MVAL and NMAX is the */
/* maximum value of N in NVAL. */
/* COPYA (workspace) DOUBLE PRECISION array, dimension (MMAX*NMAX) */
/* S (workspace) DOUBLE PRECISION array, dimension */
/* (min(MMAX,NMAX)) */
/* COPYS (workspace) DOUBLE PRECISION array, dimension */
/* (min(MMAX,NMAX)) */
/* TAU (workspace) DOUBLE PRECISION array, dimension (MMAX) */
/* WORK (workspace) DOUBLE PRECISION array, dimension */
/* (MMAX*NMAX + 4*NMAX + MMAX) */
/* IWORK (workspace) INTEGER array, dimension (NMAX) */
/* NOUT (input) INTEGER */
/* The unit number for output. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. Local Arrays .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Scalars in Common .. */
/* .. */
/* .. Common blocks .. */
/* .. */
/* .. Data statements .. */
/* Parameter adjustments */
--iwork;
--work;
--tau;
--copys;
--s;
--copya;
--a;
--nval;
--mval;
--dotype;
/* Function Body */
/* .. */
/* .. Executable Statements .. */
/* Initialize constants and the random number seed. */
s_copy(path, "Double precision", (ftnlen)1, (ftnlen)16);
s_copy(path + 1, "QP", (ftnlen)2, (ftnlen)2);
nrun = 0;
nfail = 0;
nerrs = 0;
for (i__ = 1; i__ <= 4; ++i__) {
iseed[i__ - 1] = iseedy[i__ - 1];
/* L10: */
}
eps = dlamch_("Epsilon");
/* Test the error exits */
if (*tsterr) {
derrqp_(path, nout);
}
infoc_1.infot = 0;
i__1 = *nm;
for (im = 1; im <= i__1; ++im) {
/* Do for each value of M in MVAL. */
m = mval[im];
lda = max(1,m);
i__2 = *nn;
for (in = 1; in <= i__2; ++in) {
/* Do for each value of N in NVAL. */
n = nval[in];
mnmin = min(m,n);
/* Computing MAX */
i__3 = 1, i__4 = m * max(m,n) + (mnmin << 2) + max(m,n), i__3 =
max(i__3,i__4), i__4 = m * n + (mnmin << 1) + (n << 2);
lwork = max(i__3,i__4);
for (imode = 1; imode <= 6; ++imode) {
if (! dotype[imode]) {
goto L60;
}
/* Do for each type of matrix */
/* 1: zero matrix */
/* 2: one small singular value */
/* 3: geometric distribution of singular values */
/* 4: first n/2 columns fixed */
/* 5: last n/2 columns fixed */
/* 6: every second column fixed */
mode = imode;
if (imode > 3) {
mode = 1;
}
/* Generate test matrix of size m by n using */
/* singular value distribution indicated by `mode'. */
i__3 = n;
for (i__ = 1; i__ <= i__3; ++i__) {
iwork[i__] = 0;
/* L20: */
}
if (imode == 1) {
dlaset_("Full", &m, &n, &c_b11, &c_b11, ©a[1], &lda);
i__3 = mnmin;
for (i__ = 1; i__ <= i__3; ++i__) {
copys[i__] = 0.;
/* L30: */
}
} else {
d__1 = 1. / eps;
dlatms_(&m, &n, "Uniform", iseed, "Nonsymm", ©s[1], &
mode, &d__1, &c_b16, &m, &n, "No packing", ©a[
1], &lda, &work[1], &info);
if (imode >= 4) {
if (imode == 4) {
ilow = 1;
istep = 1;
/* Computing MAX */
i__3 = 1, i__4 = n / 2;
ihigh = max(i__3,i__4);
} else if (imode == 5) {
/* Computing MAX */
i__3 = 1, i__4 = n / 2;
ilow = max(i__3,i__4);
istep = 1;
ihigh = n;
} else if (imode == 6) {
ilow = 1;
istep = 2;
ihigh = n;
}
i__3 = ihigh;
i__4 = istep;
for (i__ = ilow; i__4 < 0 ? i__ >= i__3 : i__ <= i__3;
i__ += i__4) {
iwork[i__] = 1;
/* L40: */
}
}
dlaord_("Decreasing", &mnmin, ©s[1], &c__1);
}
/* Save A and its singular values */
dlacpy_("All", &m, &n, ©a[1], &lda, &a[1], &lda);
/* Compute the QR factorization with pivoting of A */
s_copy(srnamc_1.srnamt, "DGEQPF", (ftnlen)32, (ftnlen)6);
dgeqpf_(&m, &n, &a[1], &lda, &iwork[1], &tau[1], &work[1], &
info);
/* Compute norm(svd(a) - svd(r)) */
result[0] = dqrt12_(&m, &n, &a[1], &lda, ©s[1], &work[1],
&lwork);
/* Compute norm( A*P - Q*R ) */
result[1] = dqpt01_(&m, &n, &mnmin, ©a[1], &a[1], &lda, &
tau[1], &iwork[1], &work[1], &lwork);
/* Compute Q'*Q */
result[2] = dqrt11_(&m, &mnmin, &a[1], &lda, &tau[1], &work[1]
, &lwork);
/* Print information about the tests that did not pass */
/* the threshold. */
for (k = 1; k <= 3; ++k) {
if (result[k - 1] >= *thresh) {
if (nfail == 0 && nerrs == 0) {
alahd_(nout, path);
}
io___24.ciunit = *nout;
s_wsfe(&io___24);
do_fio(&c__1, (char *)&m, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&imode, (ftnlen)sizeof(integer))
;
do_fio(&c__1, (char *)&k, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&result[k - 1], (ftnlen)sizeof(
doublereal));
e_wsfe();
++nfail;
}
/* L50: */
}
nrun += 3;
L60:
;
}
/* L70: */
}
/* L80: */
}
/* Print a summary of the results. */
alasum_(path, nout, &nfail, &nrun, &nerrs);
/* End of DCHKQP */
return 0;
} /* dchkqp_ */
/* Subroutine */ int dgelsx_(integer *m, integer *n, integer *nrhs,
doublereal *a, integer *lda, doublereal *b, integer *ldb, integer *
jpvt, doublereal *rcond, integer *rank, doublereal *work, 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
March 31, 1993
Purpose
=======
This routine is deprecated and has been replaced by routine DGELSY.
DGELSX computes the minimum-norm solution to a real linear least
squares problem:
minimize || A * X - B ||
using a complete orthogonal factorization of A. A is an M-by-N
matrix which may be rank-deficient.
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.
The routine first computes a QR factorization with column pivoting:
A * P = Q * [ R11 R12 ]
[ 0 R22 ]
with R11 defined as the largest leading submatrix whose estimated
condition number is less than 1/RCOND. The order of R11, RANK,
is the effective rank of A.
Then, R22 is considered to be negligible, and R12 is annihilated
by orthogonal transformations from the right, arriving at the
complete orthogonal factorization:
A * P = Q * [ T11 0 ] * Z
[ 0 0 ]
The minimum-norm solution is then
X = P * Z' [ inv(T11)*Q1'*B ]
[ 0 ]
where Q1 consists of the first RANK columns of 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.
NRHS (input) INTEGER
The number of right hand sides, i.e., the number of
columns of matrices B and X. NRHS >= 0.
A (input/output) DOUBLE PRECISION array, dimension (LDA,N)
On entry, the M-by-N matrix A.
On exit, A has been overwritten by details of its
complete orthogonal factorization.
LDA (input) INTEGER
The leading dimension of the array A. LDA >= max(1,M).
B (input/output) DOUBLE PRECISION array, dimension (LDB,NRHS)
On entry, the M-by-NRHS right hand side matrix B.
On exit, the N-by-NRHS solution matrix X.
If m >= n and RANK = n, the residual sum-of-squares for
the solution in the i-th column is given by the sum of
squares of elements N+1:M in that column.
LDB (input) INTEGER
The leading dimension of the array B. LDB >= max(1,M,N).
JPVT (input/output) INTEGER array, dimension (N)
On entry, if JPVT(i) .ne. 0, the i-th column of A is an
initial column, otherwise it is a free column. Before
the QR factorization of A, all initial columns are
permuted to the leading positions; only the remaining
free columns are moved as a result of column pivoting
during the factorization.
On exit, if JPVT(i) = k, then the i-th column of A*P
was the k-th column of A.
RCOND (input) DOUBLE PRECISION
RCOND is used to determine the effective rank of A, which
is defined as the order of the largest leading triangular
submatrix R11 in the QR factorization with pivoting of A,
whose estimated condition number < 1/RCOND.
RANK (output) INTEGER
The effective rank of A, i.e., the order of the submatrix
R11. This is the same as the order of the submatrix T11
in the complete orthogonal factorization of A.
WORK (workspace) DOUBLE PRECISION array, dimension
(max( min(M,N)+3*N, 2*min(M,N)+NRHS )),
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
=====================================================================
Parameter adjustments */
/* Table of constant values */
static integer c__0 = 0;
static doublereal c_b13 = 0.;
static integer c__2 = 2;
static integer c__1 = 1;
static doublereal c_b36 = 1.;
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, i__1, i__2;
doublereal d__1;
/* Local variables */
static doublereal anrm, bnrm, smin, smax;
static integer i__, j, k, iascl, ibscl, ismin, ismax;
static doublereal c1, c2;
extern /* Subroutine */ int dtrsm_(char *, char *, char *, char *,
integer *, integer *, doublereal *, doublereal *, integer *,
doublereal *, integer *), dlaic1_(
integer *, integer *, doublereal *, doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *, doublereal *);
static doublereal s1, s2, t1, t2;
extern /* Subroutine */ int dorm2r_(char *, char *, integer *, integer *,
integer *, doublereal *, integer *, doublereal *, doublereal *,
integer *, doublereal *, integer *), dlabad_(
doublereal *, doublereal *);
extern doublereal dlamch_(char *), dlange_(char *, integer *,
integer *, doublereal *, integer *, doublereal *);
static integer mn;
extern /* Subroutine */ int dlascl_(char *, integer *, integer *,
doublereal *, doublereal *, integer *, integer *, doublereal *,
integer *, integer *), dgeqpf_(integer *, integer *,
doublereal *, integer *, integer *, doublereal *, doublereal *,
integer *), dlaset_(char *, integer *, integer *, doublereal *,
doublereal *, doublereal *, integer *), xerbla_(char *,
integer *);
static doublereal bignum;
extern /* Subroutine */ int dlatzm_(char *, integer *, integer *,
doublereal *, integer *, doublereal *, doublereal *, doublereal *,
integer *, doublereal *);
static doublereal sminpr, smaxpr, smlnum;
extern /* Subroutine */ int dtzrqf_(integer *, integer *, doublereal *,
integer *, doublereal *, integer *);
#define a_ref(a_1,a_2) a[(a_2)*a_dim1 + a_1]
#define b_ref(a_1,a_2) b[(a_2)*b_dim1 + a_1]
a_dim1 = *lda;
a_offset = 1 + a_dim1 * 1;
a -= a_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1 * 1;
b -= b_offset;
--jpvt;
--work;
/* Function Body */
mn = min(*m,*n);
ismin = mn + 1;
ismax = (mn << 1) + 1;
/* Test the input arguments. */
*info = 0;
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(complicated condition) */ {
/* Computing MAX */
i__1 = max(1,*m);
if (*ldb < max(i__1,*n)) {
*info = -7;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("DGELSX", &i__1);
return 0;
}
/* Quick return if possible
Computing MIN */
i__1 = min(*m,*n);
if (min(i__1,*nrhs) == 0) {
*rank = 0;
return 0;
}
/* Get machine parameters */
smlnum = dlamch_("S") / dlamch_("P");
bignum = 1. / smlnum;
dlabad_(&smlnum, &bignum);
/* Scale A, B if max elements outside range [SMLNUM,BIGNUM] */
anrm = dlange_("M", m, n, &a[a_offset], lda, &work[1]);
iascl = 0;
if (anrm > 0. && anrm < smlnum) {
/* Scale matrix norm up to SMLNUM */
dlascl_("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 */
dlascl_("G", &c__0, &c__0, &anrm, &bignum, m, n, &a[a_offset], lda,
info);
iascl = 2;
} else if (anrm == 0.) {
/* Matrix all zero. Return zero solution. */
i__1 = max(*m,*n);
dlaset_("F", &i__1, nrhs, &c_b13, &c_b13, &b[b_offset], ldb);
*rank = 0;
goto L100;
}
bnrm = dlange_("M", m, nrhs, &b[b_offset], ldb, &work[1]);
ibscl = 0;
if (bnrm > 0. && bnrm < smlnum) {
/* Scale matrix norm up to SMLNUM */
dlascl_("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 */
dlascl_("G", &c__0, &c__0, &bnrm, &bignum, m, nrhs, &b[b_offset], ldb,
info);
ibscl = 2;
}
/* Compute QR factorization with column pivoting of A:
A * P = Q * R */
dgeqpf_(m, n, &a[a_offset], lda, &jpvt[1], &work[1], &work[mn + 1], info);
/* workspace 3*N. Details of Householder rotations stored
in WORK(1:MN).
Determine RANK using incremental condition estimation */
work[ismin] = 1.;
work[ismax] = 1.;
smax = (d__1 = a_ref(1, 1), abs(d__1));
smin = smax;
if ((d__1 = a_ref(1, 1), abs(d__1)) == 0.) {
*rank = 0;
i__1 = max(*m,*n);
dlaset_("F", &i__1, nrhs, &c_b13, &c_b13, &b[b_offset], ldb);
goto L100;
} else {
*rank = 1;
}
L10:
if (*rank < mn) {
i__ = *rank + 1;
dlaic1_(&c__2, rank, &work[ismin], &smin, &a_ref(1, i__), &a_ref(i__,
i__), &sminpr, &s1, &c1);
dlaic1_(&c__1, rank, &work[ismax], &smax, &a_ref(1, i__), &a_ref(i__,
i__), &smaxpr, &s2, &c2);
if (smaxpr * *rcond <= sminpr) {
i__1 = *rank;
for (i__ = 1; i__ <= i__1; ++i__) {
work[ismin + i__ - 1] = s1 * work[ismin + i__ - 1];
work[ismax + i__ - 1] = s2 * work[ismax + i__ - 1];
/* L20: */
}
work[ismin + *rank] = c1;
work[ismax + *rank] = c2;
smin = sminpr;
smax = smaxpr;
++(*rank);
goto L10;
}
}
/* Logically partition R = [ R11 R12 ]
[ 0 R22 ]
where R11 = R(1:RANK,1:RANK)
[R11,R12] = [ T11, 0 ] * Y */
if (*rank < *n) {
dtzrqf_(rank, n, &a[a_offset], lda, &work[mn + 1], info);
}
/* Details of Householder rotations stored in WORK(MN+1:2*MN)
B(1:M,1:NRHS) := Q' * B(1:M,1:NRHS) */
dorm2r_("Left", "Transpose", m, nrhs, &mn, &a[a_offset], lda, &work[1], &
b[b_offset], ldb, &work[(mn << 1) + 1], info);
/* workspace NRHS
B(1:RANK,1:NRHS) := inv(T11) * B(1:RANK,1:NRHS) */
dtrsm_("Left", "Upper", "No transpose", "Non-unit", rank, nrhs, &c_b36, &
a[a_offset], lda, &b[b_offset], ldb);
i__1 = *n;
for (i__ = *rank + 1; i__ <= i__1; ++i__) {
i__2 = *nrhs;
for (j = 1; j <= i__2; ++j) {
b_ref(i__, j) = 0.;
/* L30: */
}
/* L40: */
}
/* B(1:N,1:NRHS) := Y' * B(1:N,1:NRHS) */
if (*rank < *n) {
i__1 = *rank;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = *n - *rank + 1;
dlatzm_("Left", &i__2, nrhs, &a_ref(i__, *rank + 1), lda, &work[
mn + i__], &b_ref(i__, 1), &b_ref(*rank + 1, 1), ldb, &
work[(mn << 1) + 1]);
/* L50: */
}
}
/* workspace NRHS
B(1:N,1:NRHS) := P * B(1:N,1:NRHS) */
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
work[(mn << 1) + i__] = 1.;
/* L60: */
}
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (work[(mn << 1) + i__] == 1.) {
if (jpvt[i__] != i__) {
k = i__;
t1 = b_ref(k, j);
t2 = b_ref(jpvt[k], j);
L70:
b_ref(jpvt[k], j) = t1;
work[(mn << 1) + k] = 0.;
t1 = t2;
k = jpvt[k];
t2 = b_ref(jpvt[k], j);
if (jpvt[k] != i__) {
goto L70;
}
b_ref(i__, j) = t1;
work[(mn << 1) + k] = 0.;
}
}
/* L80: */
}
/* L90: */
}
/* Undo scaling */
if (iascl == 1) {
dlascl_("G", &c__0, &c__0, &anrm, &smlnum, n, nrhs, &b[b_offset], ldb,
info);
dlascl_("U", &c__0, &c__0, &smlnum, &anrm, rank, rank, &a[a_offset],
lda, info);
} else if (iascl == 2) {
dlascl_("G", &c__0, &c__0, &anrm, &bignum, n, nrhs, &b[b_offset], ldb,
info);
dlascl_("U", &c__0, &c__0, &bignum, &anrm, rank, rank, &a[a_offset],
lda, info);
}
if (ibscl == 1) {
dlascl_("G", &c__0, &c__0, &smlnum, &bnrm, n, nrhs, &b[b_offset], ldb,
info);
} else if (ibscl == 2) {
dlascl_("G", &c__0, &c__0, &bignum, &bnrm, n, nrhs, &b[b_offset], ldb,
info);
}
L100:
return 0;
/* End of DGELSX */
} /* dgelsx_ */
/* Subroutine */ int dggsvp_(char *jobu, char *jobv, char *jobq, integer *m,
integer *p, integer *n, doublereal *a, integer *lda, doublereal *b,
integer *ldb, doublereal *tola, doublereal *tolb, integer *k, integer
*l, doublereal *u, integer *ldu, doublereal *v, integer *ldv,
doublereal *q, integer *ldq, integer *iwork, doublereal *tau,
doublereal *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
September 30, 1994
Purpose
=======
DGGSVP 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
DGGSVD.
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) DOUBLE PRECISION 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) DOUBLE PRECISION 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) DOUBLE PRECISION
TOLB (input) DOUBLE PRECISION
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)*MAZHEPS,
TOLB = MAX(P,N)*norm(B)*MAZHEPS.
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) DOUBLE PRECISION 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) DOUBLE PRECISION 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) DOUBLE PRECISION 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) DOUBLE PRECISION array, dimension (N)
WORK (workspace) DOUBLE PRECISION 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 DGEQPF 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.
=====================================================================
Test the input parameters
Parameter adjustments */
/* Table of constant values */
static doublereal c_b12 = 0.;
static doublereal c_b22 = 1.;
/* 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;
doublereal d__1;
/* Local variables */
static integer i__, j;
extern logical lsame_(char *, char *);
static logical wantq, wantu, wantv;
extern /* Subroutine */ int dgeqr2_(integer *, integer *, doublereal *,
integer *, doublereal *, doublereal *, integer *), dgerq2_(
integer *, integer *, doublereal *, integer *, doublereal *,
doublereal *, integer *), dorg2r_(integer *, integer *, integer *,
doublereal *, integer *, doublereal *, doublereal *, integer *),
dorm2r_(char *, char *, integer *, integer *, integer *,
doublereal *, integer *, doublereal *, doublereal *, integer *,
doublereal *, integer *), dormr2_(char *, char *,
integer *, integer *, integer *, doublereal *, integer *,
doublereal *, doublereal *, integer *, doublereal *, integer *), dgeqpf_(integer *, integer *, doublereal *,
integer *, integer *, doublereal *, doublereal *, integer *),
dlacpy_(char *, integer *, integer *, doublereal *, integer *,
doublereal *, integer *), dlaset_(char *, integer *,
integer *, doublereal *, doublereal *, doublereal *, integer *), xerbla_(char *, integer *), dlapmt_(logical *,
integer *, integer *, doublereal *, integer *, integer *);
static logical forwrd;
#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 u_ref(a_1,a_2) u[(a_2)*u_dim1 + a_1]
#define v_ref(a_1,a_2) v[(a_2)*v_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;
u_dim1 = *ldu;
u_offset = 1 + u_dim1 * 1;
u -= u_offset;
v_dim1 = *ldv;
v_offset = 1 + v_dim1 * 1;
v -= v_offset;
q_dim1 = *ldq;
q_offset = 1 + q_dim1 * 1;
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_("DGGSVP", &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: */
}
dgeqpf_(p, n, &b[b_offset], ldb, &iwork[1], &tau[1], &work[1], info);
/* Update A := A*P */
dlapmt_(&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 ((d__1 = b_ref(i__, i__), abs(d__1)) > *tolb) {
++(*l);
}
/* L20: */
}
if (wantv) {
/* Copy the details of V, and form V. */
dlaset_("Full", p, p, &c_b12, &c_b12, &v[v_offset], ldv);
if (*p > 1) {
i__1 = *p - 1;
dlacpy_("Lower", &i__1, n, &b_ref(2, 1), ldb, &v_ref(2, 1), ldv);
}
i__1 = min(*p,*n);
dorg2r_(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_ref(i__, j) = 0.;
/* L30: */
}
/* L40: */
}
if (*p > *l) {
i__1 = *p - *l;
dlaset_("Full", &i__1, n, &c_b12, &c_b12, &b_ref(*l + 1, 1), ldb);
}
if (wantq) {
/* Set Q = I and Update Q := Q*P */
dlaset_("Full", n, n, &c_b12, &c_b22, &q[q_offset], ldq);
dlapmt_(&forwrd, n, n, &q[q_offset], ldq, &iwork[1]);
}
if (*p >= *l && *n != *l) {
/* RQ factorization of (S11 S12): ( S11 S12 ) = ( 0 S12 )*Z */
dgerq2_(l, n, &b[b_offset], ldb, &tau[1], &work[1], info);
/* Update A := A*Z' */
dormr2_("Right", "Transpose", m, n, l, &b[b_offset], ldb, &tau[1], &a[
a_offset], lda, &work[1], info);
if (wantq) {
/* Update Q := Q*Z' */
dormr2_("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;
dlaset_("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_ref(i__, j) = 0.;
/* 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;
dgeqpf_(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 ((d__1 = a_ref(i__, i__), abs(d__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);
dorm2r_("Left", "Transpose", m, l, &i__1, &a[a_offset], lda, &tau[1], &
a_ref(1, *n - *l + 1), lda, &work[1], info);
if (wantu) {
/* Copy the details of U, and form U */
dlaset_("Full", m, m, &c_b12, &c_b12, &u[u_offset], ldu);
if (*m > 1) {
i__1 = *m - 1;
i__2 = *n - *l;
dlacpy_("Lower", &i__1, &i__2, &a_ref(2, 1), lda, &u_ref(2, 1),
ldu);
}
/* Computing MIN */
i__2 = *m, i__3 = *n - *l;
i__1 = min(i__2,i__3);
dorg2r_(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;
dlapmt_(&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_ref(i__, j) = 0.;
/* L90: */
}
/* L100: */
}
if (*m > *k) {
i__1 = *m - *k;
i__2 = *n - *l;
dlaset_("Full", &i__1, &i__2, &c_b12, &c_b12, &a_ref(*k + 1, 1), lda);
}
if (*n - *l > *k) {
/* RQ factorization of ( T11 T12 ) = ( 0 T12 )*Z1 */
i__1 = *n - *l;
dgerq2_(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;
dormr2_("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;
dlaset_("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_ref(i__, j) = 0.;
/* L110: */
}
/* L120: */
}
}
if (*m > *k) {
/* QR factorization of A( K+1:M,N-L+1:N ) */
i__1 = *m - *k;
dgeqr2_(&i__1, l, &a_ref(*k + 1, *n - *l + 1), 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);
dorm2r_("Right", "No transpose", m, &i__1, &i__2, &a_ref(*k + 1, *
n - *l + 1), lda, &tau[1], &u_ref(1, *k + 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_ref(i__, j) = 0.;
/* L130: */
}
/* L140: */
}
}
return 0;
/* End of DGGSVP */
} /* dggsvp_ */