/* Subroutine */ int sgglse_(integer *m, integer *n, integer *p, real *a,
integer *lda, real *b, integer *ldb, real *c__, real *d__, real *x,
real *work, integer *lwork, integer *info)
{
/* -- LAPACK driver routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
June 30, 1999
Purpose
=======
SGGLSE solves the linear equality-constrained least squares (LSE)
problem:
minimize || c - A*x ||_2 subject to B*x = d
where A is an M-by-N matrix, B is a P-by-N matrix, c is a given
M-vector, and d is a given P-vector. It is assumed that
P <= N <= M+P, and
rank(B) = P and rank( ( A ) ) = N.
( ( B ) )
These conditions ensure that the LSE problem has a unique solution,
which is obtained using a GRQ factorization of the matrices B and A.
Arguments
=========
M (input) INTEGER
The number of rows of the matrix A. M >= 0.
N (input) INTEGER
The number of columns of the matrices A and B. N >= 0.
P (input) INTEGER
The number of rows of the matrix B. 0 <= P <= N <= M+P.
A (input/output) REAL array, dimension (LDA,N)
On entry, the M-by-N matrix A.
On exit, A is destroyed.
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 is destroyed.
LDB (input) INTEGER
The leading dimension of the array B. LDB >= max(1,P).
C (input/output) REAL array, dimension (M)
On entry, C contains the right hand side vector for the
least squares part of the LSE problem.
On exit, the residual sum of squares for the solution
is given by the sum of squares of elements N-P+1 to M of
vector C.
D (input/output) REAL array, dimension (P)
On entry, D contains the right hand side vector for the
constrained equation.
On exit, D is destroyed.
X (output) REAL array, dimension (N)
On exit, X is the solution of the LSE problem.
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+N+P).
For optimum performance LWORK >= P+min(M,N)+max(M,N)*NB,
where NB is an upper bound for the optimal blocksizes for
SGEQRF, SGERQF, SORMQR and SORMRQ.
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.
=====================================================================
Test the input parameters
Parameter adjustments */
/* Table of constant values */
static integer c__1 = 1;
static integer c_n1 = -1;
static real c_b29 = -1.f;
static real c_b31 = 1.f;
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, i__1, i__2;
/* Local variables */
static integer lopt;
extern /* Subroutine */ int sgemv_(char *, integer *, integer *, real *,
real *, integer *, real *, integer *, real *, real *, integer *), scopy_(integer *, real *, integer *, real *, integer *),
saxpy_(integer *, real *, real *, integer *, real *, integer *),
strmv_(char *, char *, char *, integer *, real *, integer *, real
*, integer *), strsv_(char *, char *,
char *, integer *, real *, integer *, real *, integer *);
static integer nb, mn, nr;
extern /* Subroutine */ int xerbla_(char *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *, ftnlen, ftnlen);
extern /* Subroutine */ int sggrqf_(integer *, integer *, integer *, real
*, integer *, real *, real *, integer *, real *, real *, integer *
, integer *);
static integer nb1, nb2, nb3, nb4, lwkopt;
static logical lquery;
extern /* Subroutine */ int sormqr_(char *, char *, integer *, integer *,
integer *, real *, integer *, real *, real *, integer *, real *,
integer *, integer *), sormrq_(char *, char *,
integer *, integer *, integer *, real *, integer *, real *, real *
, integer *, real *, integer *, 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;
--c__;
--d__;
--x;
--work;
/* Function Body */
*info = 0;
mn = min(*m,*n);
nb1 = ilaenv_(&c__1, "SGEQRF", " ", m, n, &c_n1, &c_n1, (ftnlen)6, (
ftnlen)1);
nb2 = ilaenv_(&c__1, "SGERQF", " ", m, n, &c_n1, &c_n1, (ftnlen)6, (
ftnlen)1);
nb3 = ilaenv_(&c__1, "SORMQR", " ", m, n, p, &c_n1, (ftnlen)6, (ftnlen)1);
nb4 = ilaenv_(&c__1, "SORMRQ", " ", m, n, p, &c_n1, (ftnlen)6, (ftnlen)1);
/* Computing MAX */
i__1 = max(nb1,nb2), i__1 = max(i__1,nb3);
nb = max(i__1,nb4);
lwkopt = *p + mn + max(*m,*n) * nb;
work[1] = (real) lwkopt;
lquery = *lwork == -1;
if (*m < 0) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*p < 0 || *p > *n || *p < *n - *m) {
*info = -3;
} else if (*lda < max(1,*m)) {
*info = -5;
} else if (*ldb < max(1,*p)) {
*info = -7;
} else /* if(complicated condition) */ {
/* Computing MAX */
i__1 = 1, i__2 = *m + *n + *p;
if (*lwork < max(i__1,i__2) && ! lquery) {
*info = -12;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SGGLSE", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
/* Compute the GRQ factorization of matrices B and A:
B*Q' = ( 0 T12 ) P Z'*A*Q' = ( R11 R12 ) N-P
N-P P ( 0 R22 ) M+P-N
N-P P
where T12 and R11 are upper triangular, and Q and Z are
orthogonal. */
i__1 = *lwork - *p - mn;
sggrqf_(p, m, n, &b[b_offset], ldb, &work[1], &a[a_offset], lda, &work[*p
+ 1], &work[*p + mn + 1], &i__1, info);
lopt = work[*p + mn + 1];
/* Update c = Z'*c = ( c1 ) N-P
( c2 ) M+P-N */
i__1 = max(1,*m);
i__2 = *lwork - *p - mn;
sormqr_("Left", "Transpose", m, &c__1, &mn, &a[a_offset], lda, &work[*p +
1], &c__[1], &i__1, &work[*p + mn + 1], &i__2, info);
/* Computing MAX */
i__1 = lopt, i__2 = (integer) work[*p + mn + 1];
lopt = max(i__1,i__2);
/* Solve T12*x2 = d for x2 */
strsv_("Upper", "No transpose", "Non unit", p, &b_ref(1, *n - *p + 1),
ldb, &d__[1], &c__1);
/* Update c1 */
i__1 = *n - *p;
sgemv_("No transpose", &i__1, p, &c_b29, &a_ref(1, *n - *p + 1), lda, &
d__[1], &c__1, &c_b31, &c__[1], &c__1);
/* Sovle R11*x1 = c1 for x1 */
i__1 = *n - *p;
strsv_("Upper", "No transpose", "Non unit", &i__1, &a[a_offset], lda, &
c__[1], &c__1);
/* Put the solutions in X */
i__1 = *n - *p;
scopy_(&i__1, &c__[1], &c__1, &x[1], &c__1);
scopy_(p, &d__[1], &c__1, &x[*n - *p + 1], &c__1);
/* Compute the residual vector: */
if (*m < *n) {
nr = *m + *p - *n;
i__1 = *n - *m;
sgemv_("No transpose", &nr, &i__1, &c_b29, &a_ref(*n - *p + 1, *m + 1)
, lda, &d__[nr + 1], &c__1, &c_b31, &c__[*n - *p + 1], &c__1);
} else {
nr = *p;
}
strmv_("Upper", "No transpose", "Non unit", &nr, &a_ref(*n - *p + 1, *n -
*p + 1), lda, &d__[1], &c__1);
saxpy_(&nr, &c_b29, &d__[1], &c__1, &c__[*n - *p + 1], &c__1);
/* Backward transformation x = Q'*x */
i__1 = *lwork - *p - mn;
sormrq_("Left", "Transpose", n, &c__1, p, &b[b_offset], ldb, &work[1], &x[
1], n, &work[*p + mn + 1], &i__1, info);
/* Computing MAX */
i__1 = lopt, i__2 = (integer) work[*p + mn + 1];
work[1] = (real) (*p + mn + max(i__1,i__2));
return 0;
/* End of SGGLSE */
} /* sgglse_ */
/* Subroutine */ int sggglm_(integer *n, integer *m, integer *p, real *a,
integer *lda, real *b, integer *ldb, real *d__, real *x, real *y,
real *work, integer *lwork, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, i__1, i__2, i__3, i__4;
/* Local variables */
integer i__, nb, np, nb1, nb2, nb3, nb4, lopt;
extern /* Subroutine */ int sgemv_(char *, integer *, integer *, real *,
real *, integer *, real *, integer *, real *, real *, integer *), scopy_(integer *, real *, integer *, real *, integer *),
xerbla_(char *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *);
extern /* Subroutine */ int sggqrf_(integer *, integer *, integer *, real
*, integer *, real *, real *, integer *, real *, real *, integer *
, integer *);
integer lwkmin, lwkopt;
logical lquery;
extern /* Subroutine */ int sormqr_(char *, char *, integer *, integer *,
integer *, real *, integer *, real *, real *, integer *, real *,
integer *, integer *), sormrq_(char *, char *,
integer *, integer *, integer *, real *, integer *, real *, real *
, integer *, real *, integer *, integer *),
strtrs_(char *, char *, char *, integer *, integer *, real *,
integer *, real *, integer *, integer *);
/* -- LAPACK driver routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SGGGLM solves a general Gauss-Markov linear model (GLM) problem: */
/* minimize || y ||_2 subject to d = A*x + B*y */
/* x */
/* where A is an N-by-M matrix, B is an N-by-P matrix, and d is a */
/* given N-vector. It is assumed that M <= N <= M+P, and */
/* rank(A) = M and rank( A B ) = N. */
/* Under these assumptions, the constrained equation is always */
/* consistent, and there is a unique solution x and a minimal 2-norm */
/* solution y, which is obtained using a generalized QR factorization */
/* of the matrices (A, B) given by */
/* A = Q*(R), B = Q*T*Z. */
/* (0) */
/* In particular, if matrix B is square nonsingular, then the problem */
/* GLM is equivalent to the following weighted linear least squares */
/* problem */
/* minimize || inv(B)*(d-A*x) ||_2 */
/* x */
/* where inv(B) denotes the inverse of B. */
/* Arguments */
/* ========= */
/* N (input) INTEGER */
/* The number of rows of the matrices A and B. N >= 0. */
/* M (input) INTEGER */
/* The number of columns of the matrix A. 0 <= M <= N. */
/* P (input) INTEGER */
/* The number of columns of the matrix B. P >= N-M. */
/* A (input/output) REAL array, dimension (LDA,M) */
/* On entry, the N-by-M matrix A. */
/* On exit, the upper triangular part of the array A contains */
/* the M-by-M upper triangular matrix R. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1,N). */
/* B (input/output) REAL array, dimension (LDB,P) */
/* On entry, the N-by-P matrix B. */
/* On exit, if N <= P, the upper triangle of the subarray */
/* B(1:N,P-N+1:P) contains the N-by-N upper triangular matrix T; */
/* if N > P, the elements on and above the (N-P)th subdiagonal */
/* contain the N-by-P upper trapezoidal matrix T. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= max(1,N). */
/* D (input/output) REAL array, dimension (N) */
/* On entry, D is the left hand side of the GLM equation. */
/* On exit, D is destroyed. */
/* X (output) REAL array, dimension (M) */
/* Y (output) REAL array, dimension (P) */
/* On exit, X and Y are the solutions of the GLM problem. */
/* 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,N+M+P). */
/* For optimum performance, LWORK >= M+min(N,P)+max(N,P)*NB, */
/* where NB is an upper bound for the optimal blocksizes for */
/* SGEQRF, SGERQF, SORMQR and SORMRQ. */
/* 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. */
/* = 1: the upper triangular factor R associated with A in the */
/* generalized QR factorization of the pair (A, B) is */
/* singular, so that rank(A) < M; the least squares */
/* solution could not be computed. */
/* = 2: the bottom (N-M) by (N-M) part of the upper trapezoidal */
/* factor T associated with B in the generalized QR */
/* factorization of the pair (A, B) is singular, so that */
/* rank( A B ) < N; the least squares solution could not */
/* be computed. */
/* =================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. 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;
--d__;
--x;
--y;
--work;
/* Function Body */
*info = 0;
np = min(*n,*p);
lquery = *lwork == -1;
if (*n < 0) {
*info = -1;
} else if (*m < 0 || *m > *n) {
*info = -2;
} else if (*p < 0 || *p < *n - *m) {
*info = -3;
} else if (*lda < max(1,*n)) {
*info = -5;
} else if (*ldb < max(1,*n)) {
*info = -7;
}
/* Calculate workspace */
if (*info == 0) {
if (*n == 0) {
lwkmin = 1;
lwkopt = 1;
} else {
nb1 = ilaenv_(&c__1, "SGEQRF", " ", n, m, &c_n1, &c_n1);
nb2 = ilaenv_(&c__1, "SGERQF", " ", n, m, &c_n1, &c_n1);
nb3 = ilaenv_(&c__1, "SORMQR", " ", n, m, p, &c_n1);
nb4 = ilaenv_(&c__1, "SORMRQ", " ", n, m, p, &c_n1);
/* Computing MAX */
i__1 = max(nb1,nb2), i__1 = max(i__1,nb3);
nb = max(i__1,nb4);
lwkmin = *m + *n + *p;
lwkopt = *m + np + max(*n,*p) * nb;
}
work[1] = (real) lwkopt;
if (*lwork < lwkmin && ! lquery) {
*info = -12;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SGGGLM", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
/* Compute the GQR factorization of matrices A and B: */
/* Q'*A = ( R11 ) M, Q'*B*Z' = ( T11 T12 ) M */
/* ( 0 ) N-M ( 0 T22 ) N-M */
/* M M+P-N N-M */
/* where R11 and T22 are upper triangular, and Q and Z are */
/* orthogonal. */
i__1 = *lwork - *m - np;
sggqrf_(n, m, p, &a[a_offset], lda, &work[1], &b[b_offset], ldb, &work[*m
+ 1], &work[*m + np + 1], &i__1, info);
lopt = work[*m + np + 1];
/* Update left-hand-side vector d = Q'*d = ( d1 ) M */
/* ( d2 ) N-M */
i__1 = max(1,*n);
i__2 = *lwork - *m - np;
sormqr_("Left", "Transpose", n, &c__1, m, &a[a_offset], lda, &work[1], &
d__[1], &i__1, &work[*m + np + 1], &i__2, info);
/* Computing MAX */
i__1 = lopt, i__2 = (integer) work[*m + np + 1];
lopt = max(i__1,i__2);
/* Solve T22*y2 = d2 for y2 */
if (*n > *m) {
i__1 = *n - *m;
i__2 = *n - *m;
strtrs_("Upper", "No transpose", "Non unit", &i__1, &c__1, &b[*m + 1
+ (*m + *p - *n + 1) * b_dim1], ldb, &d__[*m + 1], &i__2,
info);
if (*info > 0) {
*info = 1;
return 0;
}
i__1 = *n - *m;
scopy_(&i__1, &d__[*m + 1], &c__1, &y[*m + *p - *n + 1], &c__1);
}
/* Set y1 = 0 */
i__1 = *m + *p - *n;
for (i__ = 1; i__ <= i__1; ++i__) {
y[i__] = 0.f;
/* L10: */
}
/* Update d1 = d1 - T12*y2 */
i__1 = *n - *m;
sgemv_("No transpose", m, &i__1, &c_b32, &b[(*m + *p - *n + 1) * b_dim1 +
1], ldb, &y[*m + *p - *n + 1], &c__1, &c_b34, &d__[1], &c__1);
/* Solve triangular system: R11*x = d1 */
if (*m > 0) {
strtrs_("Upper", "No Transpose", "Non unit", m, &c__1, &a[a_offset],
lda, &d__[1], m, info);
if (*info > 0) {
*info = 2;
return 0;
}
/* Copy D to X */
scopy_(m, &d__[1], &c__1, &x[1], &c__1);
}
/* Backward transformation y = Z'*y */
/* Computing MAX */
i__1 = 1, i__2 = *n - *p + 1;
i__3 = max(1,*p);
i__4 = *lwork - *m - np;
sormrq_("Left", "Transpose", p, &c__1, &np, &b[max(i__1, i__2)+ b_dim1],
ldb, &work[*m + 1], &y[1], &i__3, &work[*m + np + 1], &i__4, info);
/* Computing MAX */
i__1 = lopt, i__2 = (integer) work[*m + np + 1];
work[1] = (real) (*m + np + max(i__1,i__2));
return 0;
/* End of SGGGLM */
} /* sggglm_ */
/* Subroutine */ int stimrq_(char *line, integer *nm, integer *mval, integer *
nval, integer *nk, integer *kval, integer *nnb, integer *nbval,
integer *nxval, integer *nlda, integer *ldaval, real *timmin, real *a,
real *tau, real *b, real *work, real *reslts, integer *ldr1, integer
*ldr2, integer *ldr3, integer *nout, ftnlen line_len)
{
/* Initialized data */
static char subnam[6*3] = "SGERQF" "SORGRQ" "SORMRQ";
static char sides[1*2] = "L" "R";
static char transs[1*2] = "N" "T";
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)";
static char fmt_9996[] = "(5x,\002K = min(M,N)\002,/)";
static char fmt_9995[] = "(/5x,a6,\002 with SIDE = '\002,a1,\002', TRANS"
" = '\002,a1,\002', \002,a1,\002 =\002,i6,/)";
static char fmt_9994[] = "(\002 *** No pairs (M,N) found with M <= N: "
" \002,a6,\002 not timed\002)";
/* System generated locals */
integer reslts_dim1, reslts_dim2, reslts_dim3, reslts_offset, i__1, i__2,
i__3, i__4, i__5, i__6;
/* 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 char labm[1], side[1];
static integer info;
static char path[3];
static real time;
static integer isub, muse[12], nuse[12], i__, k, m, n;
static char cname[6];
static integer iside, itoff, itran, minmn;
extern doublereal sopla_(char *, integer *, integer *, integer *, integer
*, integer *);
extern /* Subroutine */ int icopy_(integer *, integer *, integer *,
integer *, integer *);
static char trans[1];
static integer k1, i4, m1, n1;
static real s1, s2;
static integer ic;
extern /* Subroutine */ int sprtb4_(char *, char *, char *, integer *,
integer *, integer *, integer *, integer *, integer *, integer *,
real *, integer *, integer *, integer *, ftnlen, ftnlen, ftnlen),
sprtb5_(char *, char *, char *, integer *, integer *, integer *,
integer *, integer *, integer *, real *, integer *, integer *,
integer *, ftnlen, ftnlen, ftnlen);
static integer nb, ik, im, lw, nx, reseed[4];
extern /* Subroutine */ int atimck_(integer *, char *, integer *, integer
*, integer *, integer *, integer *, integer *, ftnlen);
extern doublereal second_(void);
extern /* Subroutine */ int atimin_(char *, char *, integer *, char *,
logical *, integer *, integer *, ftnlen, ftnlen, ftnlen), sgerqf_(
integer *, integer *, real *, integer *, real *, real *, integer *
, integer *), slacpy_(char *, integer *, integer *, real *,
integer *, real *, integer *), xlaenv_(integer *, integer
*);
extern doublereal smflop_(real *, real *, integer *);
static real untime;
extern /* Subroutine */ int stimmg_(integer *, integer *, integer *, real
*, integer *, integer *, integer *);
static logical timsub[3];
extern /* Subroutine */ int slatms_(integer *, integer *, char *, integer
*, char *, real *, integer *, real *, real *, integer *, integer *
, char *, real *, integer *, real *, integer *), sorgrq_(integer *, integer *, integer *, real *, integer
*, real *, real *, integer *, integer *), sormrq_(char *, char *,
integer *, integer *, integer *, real *, integer *, real *, real *
, integer *, real *, integer *, integer *);
static integer lda, icl, inb, imx;
static real ops;
/* Fortran I/O blocks */
static cilist io___9 = { 0, 0, 0, fmt_9999, 0 };
static cilist io___29 = { 0, 0, 0, fmt_9998, 0 };
static cilist io___31 = { 0, 0, 0, fmt_9997, 0 };
static cilist io___32 = { 0, 0, 0, 0, 0 };
static cilist io___33 = { 0, 0, 0, fmt_9996, 0 };
static cilist io___34 = { 0, 0, 0, fmt_9999, 0 };
static cilist io___49 = { 0, 0, 0, fmt_9998, 0 };
static cilist io___50 = { 0, 0, 0, fmt_9997, 0 };
static cilist io___51 = { 0, 0, 0, fmt_9995, 0 };
static cilist io___53 = { 0, 0, 0, fmt_9995, 0 };
static cilist io___54 = { 0, 0, 0, fmt_9994, 0 };
#define subnam_ref(a_0,a_1) &subnam[(a_1)*6 + a_0 - 6]
#define reslts_ref(a_1,a_2,a_3,a_4) reslts[(((a_4)*reslts_dim3 + (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
March 31, 1993
Purpose
=======
STIMRQ times the LAPACK routines to perform the RQ factorization of
a REAL general matrix.
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.
NK (input) INTEGER
The number of values of K in the vector KVAL.
KVAL (input) INTEGER array, dimension (NK)
The values of the matrix dimension K, used in SORMRQ.
NNB (input) INTEGER
The number of values of NB and NX contained in the
vectors NBVAL and NXVAL. The blocking parameters are used
in pairs (NB,NX).
NBVAL (input) INTEGER array, dimension (NNB)
The values of the blocksize NB.
NXVAL (input) INTEGER array, dimension (NNB)
The values of the crossover point NX.
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) REAL
The minimum time a subroutine will be timed.
A (workspace) REAL array, dimension (LDAMAX*NMAX)
where LDAMAX and NMAX are the maximum values of LDA and N.
TAU (workspace) REAL array, dimension (min(M,N))
B (workspace) REAL array, dimension (LDAMAX*NMAX)
WORK (workspace) REAL array, dimension (LDAMAX*NBMAX)
where NBMAX is the maximum value of NB.
RESLTS (workspace) REAL array, dimension
(LDR1,LDR2,LDR3,2*NK)
The timing results for each subroutine over the relevant
values of (M,N), (NB,NX), and LDA.
LDR1 (input) INTEGER
The first dimension of RESLTS. LDR1 >= max(1,NNB).
LDR2 (input) INTEGER
The second dimension of RESLTS. LDR2 >= max(1,NM).
LDR3 (input) INTEGER
The third dimension of RESLTS. LDR3 >= max(1,NLDA).
NOUT (input) INTEGER
The unit number for output.
Internal Parameters
===================
MODE INTEGER
The matrix type. MODE = 3 is a geometric distribution of
eigenvalues. See SLATMS for further details.
COND REAL
The condition number of the matrix. The singular values are
set to values from DMAX to DMAX/COND.
DMAX REAL
The magnitude of the largest singular value.
=====================================================================
Parameter adjustments */
--mval;
--nval;
--kval;
--nbval;
--nxval;
--ldaval;
--a;
--tau;
--b;
--work;
reslts_dim1 = *ldr1;
reslts_dim2 = *ldr2;
reslts_dim3 = *ldr3;
reslts_offset = 1 + reslts_dim1 * (1 + reslts_dim2 * (1 + reslts_dim3 * 1)
);
reslts -= reslts_offset;
/* Function Body
Extract the timing request from the input line. */
s_copy(path, "Single precision", (ftnlen)1, (ftnlen)16);
s_copy(path + 1, "RQ", (ftnlen)2, (ftnlen)2);
atimin_(path, line, &c__3, subnam, timsub, nout, &info, (ftnlen)3, (
ftnlen)80, (ftnlen)6);
if (info != 0) {
goto L230;
}
/* 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___9.ciunit = *nout;
s_wsfe(&io___9);
do_fio(&c__1, cname, (ftnlen)6);
e_wsfe();
goto L230;
}
/* 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);
icopy_(&c__4, iseed, &c__1, reseed, &c__1);
/* Do for each value of LDA: */
i__2 = *nlda;
for (ilda = 1; ilda <= i__2; ++ilda) {
lda = ldaval[ilda];
/* Do for each pair of values (NB, NX) in NBVAL and NXVAL. */
i__3 = *nnb;
for (inb = 1; inb <= i__3; ++inb) {
nb = nbval[inb];
xlaenv_(&c__1, &nb);
nx = nxval[inb];
xlaenv_(&c__3, &nx);
/* Computing MAX */
i__4 = 1, i__5 = m * max(1,nb);
lw = max(i__4,i__5);
/* Generate a test matrix of size M by N. */
icopy_(&c__4, reseed, &c__1, iseed, &c__1);
slatms_(&m, &n, "Uniform", iseed, "Nonsymm", &tau[1], &c__3, &
c_b24, &c_b25, &m, &n, "No packing", &b[1], &lda, &
work[1], &info);
if (timsub[0]) {
/* SGERQF: RQ factorization */
slacpy_("Full", &m, &n, &b[1], &lda, &a[1], &lda);
ic = 0;
s1 = second_();
L10:
sgerqf_(&m, &n, &a[1], &lda, &tau[1], &work[1], &lw, &
info);
s2 = second_();
time = s2 - s1;
++ic;
if (time < *timmin) {
slacpy_("Full", &m, &n, &b[1], &lda, &a[1], &lda);
goto L10;
}
/* Subtract the time used in SLACPY. */
icl = 1;
s1 = second_();
L20:
s2 = second_();
untime = s2 - s1;
++icl;
if (icl <= ic) {
slacpy_("Full", &m, &n, &a[1], &lda, &b[1], &lda);
goto L20;
}
time = (time - untime) / (real) ic;
ops = sopla_("SGERQF", &m, &n, &c__0, &c__0, &nb);
reslts_ref(inb, im, ilda, 1) = smflop_(&ops, &time, &info)
;
} else {
/* If SGERQF was not timed, generate a matrix and factor
it using SGERQF anyway so that the factored form of
the matrix can be used in timing the other routines. */
slacpy_("Full", &m, &n, &b[1], &lda, &a[1], &lda);
sgerqf_(&m, &n, &a[1], &lda, &tau[1], &work[1], &lw, &
info);
}
if (timsub[1]) {
/* SORGRQ: Generate orthogonal matrix Q from the RQ
factorization */
slacpy_("Full", &minmn, &n, &a[1], &lda, &b[1], &lda);
ic = 0;
s1 = second_();
L30:
sorgrq_(&minmn, &n, &minmn, &b[1], &lda, &tau[1], &work[1]
, &lw, &info);
s2 = second_();
time = s2 - s1;
++ic;
if (time < *timmin) {
slacpy_("Full", &minmn, &n, &a[1], &lda, &b[1], &lda);
goto L30;
}
/* Subtract the time used in SLACPY. */
icl = 1;
s1 = second_();
L40:
s2 = second_();
untime = s2 - s1;
++icl;
if (icl <= ic) {
slacpy_("Full", &minmn, &n, &a[1], &lda, &b[1], &lda);
goto L40;
}
time = (time - untime) / (real) ic;
ops = sopla_("SORGRQ", &minmn, &n, &minmn, &c__0, &nb);
reslts_ref(inb, im, ilda, 2) = smflop_(&ops, &time, &info)
;
}
/* L50: */
}
/* L60: */
}
/* L70: */
}
/* Print tables of results */
for (isub = 1; isub <= 2; ++isub) {
if (! timsub[isub - 1]) {
goto L90;
}
io___29.ciunit = *nout;
s_wsfe(&io___29);
do_fio(&c__1, subnam_ref(0, isub), (ftnlen)6);
e_wsfe();
if (*nlda > 1) {
i__1 = *nlda;
for (i__ = 1; i__ <= i__1; ++i__) {
io___31.ciunit = *nout;
s_wsfe(&io___31);
do_fio(&c__1, (char *)&i__, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&ldaval[i__], (ftnlen)sizeof(integer));
e_wsfe();
/* L80: */
}
}
io___32.ciunit = *nout;
s_wsle(&io___32);
e_wsle();
if (isub == 2) {
io___33.ciunit = *nout;
s_wsfe(&io___33);
e_wsfe();
}
sprtb4_("( NB, NX)", "M", "N", nnb, &nbval[1], &nxval[1], nm, &mval[
1], &nval[1], nlda, &reslts_ref(1, 1, 1, isub), ldr1, ldr2,
nout, (ftnlen)11, (ftnlen)1, (ftnlen)1);
L90:
;
}
/* Time SORMRQ separately. Here the starting matrix is M by N, and
K is the free dimension of the matrix multiplied by Q. */
if (timsub[2]) {
/* Check that K <= LDA for the input values. */
atimck_(&c__3, cname, nk, &kval[1], nlda, &ldaval[1], nout, &info, (
ftnlen)6);
if (info > 0) {
io___34.ciunit = *nout;
s_wsfe(&io___34);
do_fio(&c__1, subnam_ref(0, 3), (ftnlen)6);
e_wsfe();
goto L230;
}
/* Use only the pairs (M,N) where M <= N. */
imx = 0;
i__1 = *nm;
for (im = 1; im <= i__1; ++im) {
if (mval[im] <= nval[im]) {
++imx;
muse[imx - 1] = mval[im];
nuse[imx - 1] = nval[im];
}
/* L100: */
}
/* SORMRQ: Multiply by Q stored as a product of elementary
transformations
Do for each pair of values (M,N): */
i__1 = imx;
for (im = 1; im <= i__1; ++im) {
m = muse[im - 1];
n = nuse[im - 1];
/* Do for each value of LDA: */
i__2 = *nlda;
for (ilda = 1; ilda <= i__2; ++ilda) {
lda = ldaval[ilda];
/* Generate an M by N matrix and form its RQ decomposition. */
slatms_(&m, &n, "Uniform", iseed, "Nonsymm", &tau[1], &c__3, &
c_b24, &c_b25, &m, &n, "No packing", &a[1], &lda, &
work[1], &info);
/* Computing MAX */
i__3 = 1, i__4 = m * max(1,nb);
lw = max(i__3,i__4);
sgerqf_(&m, &n, &a[1], &lda, &tau[1], &work[1], &lw, &info);
/* Do first for SIDE = 'L', then for SIDE = 'R' */
i4 = 0;
for (iside = 1; iside <= 2; ++iside) {
*(unsigned char *)side = *(unsigned char *)&sides[iside -
1];
/* Do for each pair of values (NB, NX) in NBVAL and
NXVAL. */
i__3 = *nnb;
for (inb = 1; inb <= i__3; ++inb) {
nb = nbval[inb];
xlaenv_(&c__1, &nb);
nx = nxval[inb];
xlaenv_(&c__3, &nx);
/* Do for each value of K in KVAL */
i__4 = *nk;
for (ik = 1; ik <= i__4; ++ik) {
k = kval[ik];
/* Sort out which variable is which */
if (iside == 1) {
k1 = m;
m1 = n;
n1 = k;
/* Computing MAX */
i__5 = 1, i__6 = n1 * max(1,nb);
lw = max(i__5,i__6);
} else {
k1 = m;
n1 = n;
m1 = k;
/* Computing MAX */
i__5 = 1, i__6 = m1 * max(1,nb);
lw = max(i__5,i__6);
}
/* Do first for TRANS = 'N', then for TRANS = 'T' */
itoff = 0;
for (itran = 1; itran <= 2; ++itran) {
*(unsigned char *)trans = *(unsigned char *)&
transs[itran - 1];
stimmg_(&c__0, &m1, &n1, &b[1], &lda, &c__0, &
c__0);
ic = 0;
s1 = second_();
L110:
sormrq_(side, trans, &m1, &n1, &k1, &a[1], &
lda, &tau[1], &b[1], &lda, &work[1], &
lw, &info);
s2 = second_();
time = s2 - s1;
++ic;
if (time < *timmin) {
stimmg_(&c__0, &m1, &n1, &b[1], &lda, &
c__0, &c__0);
goto L110;
}
/* Subtract the time used in STIMMG. */
icl = 1;
s1 = second_();
L120:
s2 = second_();
untime = s2 - s1;
++icl;
if (icl <= ic) {
stimmg_(&c__0, &m1, &n1, &b[1], &lda, &
c__0, &c__0);
goto L120;
}
time = (time - untime) / (real) ic;
i__5 = iside - 1;
ops = sopla_("SORMRQ", &m1, &n1, &k1, &i__5, &
nb);
reslts_ref(inb, im, ilda, i4 + itoff + ik) =
smflop_(&ops, &time, &info);
itoff = *nk;
/* L130: */
}
/* L140: */
}
/* L150: */
}
i4 = *nk << 1;
/* L160: */
}
/* L170: */
}
/* L180: */
}
/* Print tables of results */
isub = 3;
i4 = 1;
if (imx >= 1) {
for (iside = 1; iside <= 2; ++iside) {
*(unsigned char *)side = *(unsigned char *)&sides[iside - 1];
if (iside == 1) {
io___49.ciunit = *nout;
s_wsfe(&io___49);
do_fio(&c__1, subnam_ref(0, isub), (ftnlen)6);
e_wsfe();
if (*nlda > 1) {
i__1 = *nlda;
for (i__ = 1; i__ <= i__1; ++i__) {
io___50.ciunit = *nout;
s_wsfe(&io___50);
do_fio(&c__1, (char *)&i__, (ftnlen)sizeof(
integer));
do_fio(&c__1, (char *)&ldaval[i__], (ftnlen)
sizeof(integer));
e_wsfe();
/* L190: */
}
}
}
for (itran = 1; itran <= 2; ++itran) {
*(unsigned char *)trans = *(unsigned char *)&transs[itran
- 1];
i__1 = *nk;
for (ik = 1; ik <= i__1; ++ik) {
if (iside == 1) {
n = kval[ik];
io___51.ciunit = *nout;
s_wsfe(&io___51);
do_fio(&c__1, subnam_ref(0, isub), (ftnlen)6);
do_fio(&c__1, side, (ftnlen)1);
do_fio(&c__1, trans, (ftnlen)1);
do_fio(&c__1, "N", (ftnlen)1);
do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer))
;
e_wsfe();
*(unsigned char *)labm = 'M';
} else {
m = kval[ik];
io___53.ciunit = *nout;
s_wsfe(&io___53);
do_fio(&c__1, subnam_ref(0, isub), (ftnlen)6);
do_fio(&c__1, side, (ftnlen)1);
do_fio(&c__1, trans, (ftnlen)1);
do_fio(&c__1, "M", (ftnlen)1);
do_fio(&c__1, (char *)&m, (ftnlen)sizeof(integer))
;
e_wsfe();
*(unsigned char *)labm = 'N';
}
sprtb5_("NB", "K", labm, nnb, &nbval[1], &imx, muse,
nuse, nlda, &reslts_ref(1, 1, 1, i4), ldr1,
ldr2, nout, (ftnlen)2, (ftnlen)1, (ftnlen)1);
++i4;
/* L200: */
}
/* L210: */
}
/* L220: */
}
} else {
io___54.ciunit = *nout;
s_wsfe(&io___54);
do_fio(&c__1, subnam_ref(0, isub), (ftnlen)6);
e_wsfe();
}
}
L230:
return 0;
/* End of STIMRQ */
} /* stimrq_ */
/* Subroutine */ int sgglse_(integer *m, integer *n, integer *p, real *a,
integer *lda, real *b, integer *ldb, real *c, real *d, real *x, real *
work, integer *lwork, integer *info)
{
/* -- LAPACK driver routine (version 2.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
September 30, 1994
Purpose
=======
SGGLSE solves the linear equality-constrained least squares (LSE)
problem:
minimize || c - A*x ||_2 subject to B*x = d
where A is an M-by-N matrix, B is a P-by-N matrix, c is a given
M-vector, and d is a given P-vector. It is assumed that
P <= N <= M+P, and
rank(B) = P and rank( ( A ) ) = N.
( ( B ) )
These conditions ensure that the LSE problem has a unique solution,
which is obtained using a GRQ factorization of the matrices B and A.
Arguments
=========
M (input) INTEGER
The number of rows of the matrix A. M >= 0.
N (input) INTEGER
The number of columns of the matrices A and B. N >= 0.
P (input) INTEGER
The number of rows of the matrix B. 0 <= P <= N <= M+P.
A (input/output) REAL array, dimension (LDA,N)
On entry, the M-by-N matrix A.
On exit, A is destroyed.
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 is destroyed.
LDB (input) INTEGER
The leading dimension of the array B. LDB >= max(1,P).
C (input/output) REAL array, dimension (M)
On entry, C contains the right hand side vector for the
least squares part of the LSE problem.
On exit, the residual sum of squares for the solution
is given by the sum of squares of elements N-P+1 to M of
vector C.
D (input/output) REAL array, dimension (P)
On entry, D contains the right hand side vector for the
constrained equation.
On exit, D is destroyed.
X (output) REAL array, dimension (N)
On exit, X is the solution of the LSE problem.
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+N+P).
For optimum performance LWORK >= P+min(M,N)+max(M,N)*NB,
where NB is an upper bound for the optimal blocksizes for
SGEQRF, SGERQF, SORMQR and SORMRQ.
INFO (output) INTEGER
= 0: successful exit.
< 0: if INFO = -i, the i-th argument had an illegal value.
=====================================================================
Test the input parameters
Parameter adjustments
Function Body */
/* Table of constant values */
static integer c__1 = 1;
static real c_b11 = -1.f;
static real c_b13 = 1.f;
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, i__1, i__2;
/* Local variables */
static integer lopt;
extern /* Subroutine */ int sgemv_(char *, integer *, integer *, real *,
real *, integer *, real *, integer *, real *, real *, integer *), scopy_(integer *, real *, integer *, real *, integer *),
saxpy_(integer *, real *, real *, integer *, real *, integer *),
strmv_(char *, char *, char *, integer *, real *, integer *, real
*, integer *), strsv_(char *, char *,
char *, integer *, real *, integer *, real *, integer *);
static integer mn, nr;
extern /* Subroutine */ int xerbla_(char *, integer *), sggrqf_(
integer *, integer *, integer *, real *, integer *, real *, real *
, integer *, real *, real *, integer *, integer *), sormqr_(char *
, char *, integer *, integer *, integer *, real *, integer *,
real *, real *, integer *, real *, integer *, integer *), sormrq_(char *, char *, integer *, integer *, integer *,
real *, integer *, real *, real *, integer *, real *, integer *,
integer *);
#define C(I) c[(I)-1]
#define D(I) d[(I)-1]
#define X(I) x[(I)-1]
#define WORK(I) work[(I)-1]
#define A(I,J) a[(I)-1 + ((J)-1)* ( *lda)]
#define B(I,J) b[(I)-1 + ((J)-1)* ( *ldb)]
*info = 0;
mn = min(*m,*n);
if (*m < 0) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*p < 0 || *p > *n || *p < *n - *m) {
*info = -3;
} else if (*lda < max(1,*m)) {
*info = -5;
} else if (*ldb < max(1,*p)) {
*info = -7;
} else /* if(complicated condition) */ {
/* Computing MAX */
i__1 = 1, i__2 = *m + *n + *p;
if (*lwork < max(i__1,i__2)) {
*info = -12;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SGGLSE", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
/* Compute the GRQ factorization of matrices B and A:
B*Q' = ( 0 T12 ) P Z'*A*Q' = ( R11 R12 ) N-P
N-P P ( 0 R22 ) M+P-N
N-P P
where T12 and R11 are upper triangular, and Q and Z are
orthogonal. */
i__1 = *lwork - *p - mn;
sggrqf_(p, m, n, &B(1,1), ldb, &WORK(1), &A(1,1), lda, &WORK(*p
+ 1), &WORK(*p + mn + 1), &i__1, info);
lopt = WORK(*p + mn + 1);
/* Update c = Z'*c = ( c1 ) N-P
( c2 ) M+P-N */
i__1 = max(1,*m);
i__2 = *lwork - *p - mn;
sormqr_("Left", "Transpose", m, &c__1, &mn, &A(1,1), lda, &WORK(*p +
1), &C(1), &i__1, &WORK(*p + mn + 1), &i__2, info);
/* Computing MAX */
i__1 = lopt, i__2 = (integer) WORK(*p + mn + 1);
lopt = max(i__1,i__2);
/* Solve T12*x2 = d for x2 */
strsv_("Upper", "No transpose", "Non unit", p, &B(1,*n-*p+1), ldb, &D(1), &c__1);
/* Update c1 */
i__1 = *n - *p;
sgemv_("No transpose", &i__1, p, &c_b11, &A(1,*n-*p+1),
lda, &D(1), &c__1, &c_b13, &C(1), &c__1);
/* Sovle R11*x1 = c1 for x1 */
i__1 = *n - *p;
strsv_("Upper", "No transpose", "Non unit", &i__1, &A(1,1), lda, &C(
1), &c__1);
/* Put the solutions in X */
i__1 = *n - *p;
scopy_(&i__1, &C(1), &c__1, &X(1), &c__1);
scopy_(p, &D(1), &c__1, &X(*n - *p + 1), &c__1);
/* Compute the residual vector: */
if (*m < *n) {
nr = *m + *p - *n;
i__1 = *n - *m;
sgemv_("No transpose", &nr, &i__1, &c_b11, &A(*n-*p+1,*m+1), lda, &D(nr + 1), &c__1, &c_b13, &C(*n - *p + 1), &
c__1);
} else {
nr = *p;
}
strmv_("Upper", "No transpose", "Non unit", &nr, &A(*n-*p+1,*n-*p+1), lda, &D(1), &c__1);
saxpy_(&nr, &c_b11, &D(1), &c__1, &C(*n - *p + 1), &c__1);
/* Backward transformation x = Q'*x */
i__1 = *lwork - *p - mn;
sormrq_("Left", "Transpose", n, &c__1, p, &B(1,1), ldb, &WORK(1), &X(
1), n, &WORK(*p + mn + 1), &i__1, info);
/* Computing MAX */
i__1 = lopt, i__2 = (integer) WORK(*p + mn + 1);
WORK(1) = (real) (*p + mn + max(i__1,i__2));
return 0;
/* End of SGGLSE */
} /* sgglse_ */
/* Subroutine */ int sgerqs_(integer *m, integer *n, integer *nrhs, real *a,
integer *lda, real *tau, real *b, integer *ldb, real *work, integer *
lwork, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, i__1;
/* Local variables */
extern /* Subroutine */ int strsm_(char *, char *, char *, char *,
integer *, integer *, real *, real *, integer *, real *, integer *
), xerbla_(char *, integer *), slaset_(char *, integer *, integer *, real *, real *,
real *, integer *), sormrq_(char *, char *, integer *,
integer *, integer *, real *, integer *, real *, real *, integer *
, real *, integer *, 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]
/* -- 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
=======
Compute a minimum-norm solution
min || A*X - B ||
using the RQ factorization
A = R*Q
computed by SGERQF.
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 >= M >= 0.
NRHS (input) INTEGER
The number of columns of B. NRHS >= 0.
A (input) REAL array, dimension (LDA,N)
Details of the RQ factorization of the original matrix A as
returned by SGERQF.
LDA (input) INTEGER
The leading dimension of the array A. LDA >= M.
TAU (input) REAL array, dimension (M)
Details of the orthogonal matrix Q.
B (input/output) REAL array, dimension (LDB,NRHS)
On entry, the right hand side vectors for the linear system.
On exit, the solution vectors X. Each solution vector
is contained in rows 1:N of a column of B.
LDB (input) INTEGER
The leading dimension of the array B. LDB >= max(1,N).
WORK (workspace) REAL array, dimension (LWORK)
LWORK (input) INTEGER
The length of the array WORK. LWORK must be at least NRHS,
and should be at least NRHS*NB, where NB is the block size
for this environment.
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
=====================================================================
Test the input parameters.
Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1 * 1;
a -= a_offset;
--tau;
b_dim1 = *ldb;
b_offset = 1 + b_dim1 * 1;
b -= b_offset;
--work;
/* Function Body */
*info = 0;
if (*m < 0) {
*info = -1;
} else if (*n < 0 || *m > *n) {
*info = -2;
} else if (*nrhs < 0) {
*info = -3;
} else if (*lda < max(1,*m)) {
*info = -5;
} else if (*ldb < max(1,*n)) {
*info = -8;
} else if (*lwork < 1 || *lwork < *nrhs && *m > 0 && *n > 0) {
*info = -10;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SGERQS", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0 || *nrhs == 0 || *m == 0) {
return 0;
}
/* Solve R*X = B(n-m+1:n,:) */
strsm_("Left", "Upper", "No transpose", "Non-unit", m, nrhs, &c_b7, &
a_ref(1, *n - *m + 1), lda, &b_ref(*n - *m + 1, 1), ldb);
/* Set B(1:n-m,:) to zero */
i__1 = *n - *m;
slaset_("Full", &i__1, nrhs, &c_b9, &c_b9, &b[b_offset], ldb);
/* B := Q' * B */
sormrq_("Left", "Transpose", n, nrhs, m, &a[a_offset], lda, &tau[1], &b[
b_offset], ldb, &work[1], lwork, info);
return 0;
/* End of SGERQS */
} /* sgerqs_ */
/* Subroutine */ int sgglse_(integer *m, integer *n, integer *p, real *a,
integer *lda, real *b, integer *ldb, real *c__, real *d__, real *x,
real *work, integer *lwork, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, i__1, i__2;
/* Local variables */
integer nb, mn, nr, nb1, nb2, nb3, nb4, lopt;
extern /* Subroutine */ int sgemv_(char *, integer *, integer *, real *,
real *, integer *, real *, integer *, real *, real *, integer *), scopy_(integer *, real *, integer *, real *, integer *),
saxpy_(integer *, real *, real *, integer *, real *, integer *),
strmv_(char *, char *, char *, integer *, real *, integer *, real
*, integer *), xerbla_(char *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *);
extern /* Subroutine */ int sggrqf_(integer *, integer *, integer *, real
*, integer *, real *, real *, integer *, real *, real *, integer *
, integer *);
integer lwkmin, lwkopt;
logical lquery;
extern /* Subroutine */ int sormqr_(char *, char *, integer *, integer *,
integer *, real *, integer *, real *, real *, integer *, real *,
integer *, integer *), sormrq_(char *, char *,
integer *, integer *, integer *, real *, integer *, real *, real *
, integer *, real *, integer *, integer *),
strtrs_(char *, char *, char *, integer *, integer *, real *,
integer *, real *, integer *, integer *);
/* -- LAPACK driver routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* SGGLSE solves the linear equality-constrained least squares (LSE) */
/* problem: */
/* minimize || c - A*x ||_2 subject to B*x = d */
/* where A is an M-by-N matrix, B is a P-by-N matrix, c is a given */
/* M-vector, and d is a given P-vector. It is assumed that */
/* P <= N <= M+P, and */
/* rank(B) = P and rank( (A) ) = N. */
/* ( (B) ) */
/* These conditions ensure that the LSE problem has a unique solution, */
/* which is obtained using a generalized RQ factorization of the */
/* matrices (B, A) given by */
/* B = (0 R)*Q, A = Z*T*Q. */
/* Arguments */
/* ========= */
/* M (input) INTEGER */
/* The number of rows of the matrix A. M >= 0. */
/* N (input) INTEGER */
/* The number of columns of the matrices A and B. N >= 0. */
/* P (input) INTEGER */
/* The number of rows of the matrix B. 0 <= P <= N <= M+P. */
/* A (input/output) REAL array, dimension (LDA,N) */
/* On entry, the M-by-N matrix A. */
/* On exit, the elements on and above the diagonal of the array */
/* contain the min(M,N)-by-N upper trapezoidal matrix T. */
/* 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, the upper triangle of the subarray B(1:P,N-P+1:N) */
/* contains the P-by-P upper triangular matrix R. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= max(1,P). */
/* C (input/output) REAL array, dimension (M) */
/* On entry, C contains the right hand side vector for the */
/* least squares part of the LSE problem. */
/* On exit, the residual sum of squares for the solution */
/* is given by the sum of squares of elements N-P+1 to M of */
/* vector C. */
/* D (input/output) REAL array, dimension (P) */
/* On entry, D contains the right hand side vector for the */
/* constrained equation. */
/* On exit, D is destroyed. */
/* X (output) REAL array, dimension (N) */
/* On exit, X is the solution of the LSE problem. */
/* 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+N+P). */
/* For optimum performance LWORK >= P+min(M,N)+max(M,N)*NB, */
/* where NB is an upper bound for the optimal blocksizes for */
/* SGEQRF, SGERQF, SORMQR and SORMRQ. */
/* 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. */
/* = 1: the upper triangular factor R associated with B in the */
/* generalized RQ factorization of the pair (B, A) is */
/* singular, so that rank(B) < P; the least squares */
/* solution could not be computed. */
/* = 2: the (N-P) by (N-P) part of the upper trapezoidal factor */
/* T associated with A in the generalized RQ factorization */
/* of the pair (B, A) is singular, so that */
/* rank( (A) ) < N; the least squares solution could not */
/* ( (B) ) */
/* be computed. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. 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;
--c__;
--d__;
--x;
--work;
/* Function Body */
*info = 0;
mn = min(*m,*n);
lquery = *lwork == -1;
if (*m < 0) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*p < 0 || *p > *n || *p < *n - *m) {
*info = -3;
} else if (*lda < max(1,*m)) {
*info = -5;
} else if (*ldb < max(1,*p)) {
*info = -7;
}
/* Calculate workspace */
if (*info == 0) {
if (*n == 0) {
lwkmin = 1;
lwkopt = 1;
} else {
nb1 = ilaenv_(&c__1, "SGEQRF", " ", m, n, &c_n1, &c_n1);
nb2 = ilaenv_(&c__1, "SGERQF", " ", m, n, &c_n1, &c_n1);
nb3 = ilaenv_(&c__1, "SORMQR", " ", m, n, p, &c_n1);
nb4 = ilaenv_(&c__1, "SORMRQ", " ", m, n, p, &c_n1);
/* Computing MAX */
i__1 = max(nb1,nb2), i__1 = max(i__1,nb3);
nb = max(i__1,nb4);
lwkmin = *m + *n + *p;
lwkopt = *p + mn + max(*m,*n) * nb;
}
work[1] = (real) lwkopt;
if (*lwork < lwkmin && ! lquery) {
*info = -12;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SGGLSE", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
/* Compute the GRQ factorization of matrices B and A: */
/* B*Q' = ( 0 T12 ) P Z'*A*Q' = ( R11 R12 ) N-P */
/* N-P P ( 0 R22 ) M+P-N */
/* N-P P */
/* where T12 and R11 are upper triangular, and Q and Z are */
/* orthogonal. */
i__1 = *lwork - *p - mn;
sggrqf_(p, m, n, &b[b_offset], ldb, &work[1], &a[a_offset], lda, &work[*p
+ 1], &work[*p + mn + 1], &i__1, info);
lopt = work[*p + mn + 1];
/* Update c = Z'*c = ( c1 ) N-P */
/* ( c2 ) M+P-N */
i__1 = max(1,*m);
i__2 = *lwork - *p - mn;
sormqr_("Left", "Transpose", m, &c__1, &mn, &a[a_offset], lda, &work[*p +
1], &c__[1], &i__1, &work[*p + mn + 1], &i__2, info);
/* Computing MAX */
i__1 = lopt, i__2 = (integer) work[*p + mn + 1];
lopt = max(i__1,i__2);
/* Solve T12*x2 = d for x2 */
if (*p > 0) {
strtrs_("Upper", "No transpose", "Non-unit", p, &c__1, &b[(*n - *p +
1) * b_dim1 + 1], ldb, &d__[1], p, info);
if (*info > 0) {
*info = 1;
return 0;
}
/* Put the solution in X */
scopy_(p, &d__[1], &c__1, &x[*n - *p + 1], &c__1);
/* Update c1 */
i__1 = *n - *p;
sgemv_("No transpose", &i__1, p, &c_b31, &a[(*n - *p + 1) * a_dim1 +
1], lda, &d__[1], &c__1, &c_b33, &c__[1], &c__1);
}
/* Solve R11*x1 = c1 for x1 */
if (*n > *p) {
i__1 = *n - *p;
i__2 = *n - *p;
strtrs_("Upper", "No transpose", "Non-unit", &i__1, &c__1, &a[
a_offset], lda, &c__[1], &i__2, info);
if (*info > 0) {
*info = 2;
return 0;
}
/* Put the solution in X */
i__1 = *n - *p;
scopy_(&i__1, &c__[1], &c__1, &x[1], &c__1);
}
/* Compute the residual vector: */
if (*m < *n) {
nr = *m + *p - *n;
if (nr > 0) {
i__1 = *n - *m;
sgemv_("No transpose", &nr, &i__1, &c_b31, &a[*n - *p + 1 + (*m +
1) * a_dim1], lda, &d__[nr + 1], &c__1, &c_b33, &c__[*n -
*p + 1], &c__1);
}
} else {
nr = *p;
}
if (nr > 0) {
strmv_("Upper", "No transpose", "Non unit", &nr, &a[*n - *p + 1 + (*n
- *p + 1) * a_dim1], lda, &d__[1], &c__1);
saxpy_(&nr, &c_b31, &d__[1], &c__1, &c__[*n - *p + 1], &c__1);
}
/* Backward transformation x = Q'*x */
i__1 = *lwork - *p - mn;
sormrq_("Left", "Transpose", n, &c__1, p, &b[b_offset], ldb, &work[1], &x[
1], n, &work[*p + mn + 1], &i__1, info);
/* Computing MAX */
i__1 = lopt, i__2 = (integer) work[*p + mn + 1];
work[1] = (real) (*p + mn + max(i__1,i__2));
return 0;
/* End of SGGLSE */
} /* sgglse_ */
/* Subroutine */ int sgerqs_(integer *m, integer *n, integer *nrhs, real *a,
integer *lda, real *tau, real *b, integer *ldb, real *work, integer *
lwork, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, i__1;
/* Local variables */
/* -- LAPACK routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* Compute a minimum-norm solution */
/* min || A*X - B || */
/* using the RQ factorization */
/* A = R*Q */
/* computed by SGERQF. */
/* 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 >= M >= 0. */
/* NRHS (input) INTEGER */
/* The number of columns of B. NRHS >= 0. */
/* A (input) REAL array, dimension (LDA,N) */
/* Details of the RQ factorization of the original matrix A as */
/* returned by SGERQF. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= M. */
/* TAU (input) REAL array, dimension (M) */
/* Details of the orthogonal matrix Q. */
/* B (input/output) REAL array, dimension (LDB,NRHS) */
/* On entry, the right hand side vectors for the linear system. */
/* On exit, the solution vectors X. Each solution vector */
/* is contained in rows 1:N of a column of B. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= max(1,N). */
/* WORK (workspace) REAL array, dimension (LWORK) */
/* LWORK (input) INTEGER */
/* The length of the array WORK. LWORK must be at least NRHS, */
/* and should be at least NRHS*NB, where NB is the block size */
/* for this environment. */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
--tau;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
--work;
/* Function Body */
*info = 0;
if (*m < 0) {
*info = -1;
} else if (*n < 0 || *m > *n) {
*info = -2;
} else if (*nrhs < 0) {
*info = -3;
} else if (*lda < max(1,*m)) {
*info = -5;
} else if (*ldb < max(1,*n)) {
*info = -8;
} else if (*lwork < 1 || *lwork < *nrhs && *m > 0 && *n > 0) {
*info = -10;
}
if (*info != 0) {
i__1 = -(*info);
this_xerbla_("SGERQS", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0 || *nrhs == 0 || *m == 0) {
return 0;
}
/* Solve R*X = B(n-m+1:n,:) */
strsm_("Left", "Upper", "No transpose", "Non-unit", m, nrhs, &c_b7, &a[(*
n - *m + 1) * a_dim1 + 1], lda, &b[*n - *m + 1 + b_dim1], ldb);
/* Set B(1:n-m,:) to zero */
i__1 = *n - *m;
slaset_("Full", &i__1, nrhs, &c_b9, &c_b9, &b[b_offset], ldb);
/* B := Q' * B */
sormrq_("Left", "Transpose", n, nrhs, m, &a[a_offset], lda, &tau[1], &b[
b_offset], ldb, &work[1], lwork, info);
return 0;
/* End of SGERQS */
} /* sgerqs_ */
/* Subroutine */ int sggglm_(integer *n, integer *m, integer *p, real *a,
integer *lda, real *b, integer *ldb, real *d__, real *x, real *y,
real *work, integer *lwork, integer *info)
{
/* -- LAPACK driver routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
June 30, 1999
Purpose
=======
SGGGLM solves a general Gauss-Markov linear model (GLM) problem:
minimize || y ||_2 subject to d = A*x + B*y
x
where A is an N-by-M matrix, B is an N-by-P matrix, and d is a
given N-vector. It is assumed that M <= N <= M+P, and
rank(A) = M and rank( A B ) = N.
Under these assumptions, the constrained equation is always
consistent, and there is a unique solution x and a minimal 2-norm
solution y, which is obtained using a generalized QR factorization
of A and B.
In particular, if matrix B is square nonsingular, then the problem
GLM is equivalent to the following weighted linear least squares
problem
minimize || inv(B)*(d-A*x) ||_2
x
where inv(B) denotes the inverse of B.
Arguments
=========
N (input) INTEGER
The number of rows of the matrices A and B. N >= 0.
M (input) INTEGER
The number of columns of the matrix A. 0 <= M <= N.
P (input) INTEGER
The number of columns of the matrix B. P >= N-M.
A (input/output) REAL array, dimension (LDA,M)
On entry, the N-by-M matrix A.
On exit, A is destroyed.
LDA (input) INTEGER
The leading dimension of the array A. LDA >= max(1,N).
B (input/output) REAL array, dimension (LDB,P)
On entry, the N-by-P matrix B.
On exit, B is destroyed.
LDB (input) INTEGER
The leading dimension of the array B. LDB >= max(1,N).
D (input/output) REAL array, dimension (N)
On entry, D is the left hand side of the GLM equation.
On exit, D is destroyed.
X (output) REAL array, dimension (M)
Y (output) REAL array, dimension (P)
On exit, X and Y are the solutions of the GLM problem.
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,N+M+P).
For optimum performance, LWORK >= M+min(N,P)+max(N,P)*NB,
where NB is an upper bound for the optimal blocksizes for
SGEQRF, SGERQF, SORMQR and SORMRQ.
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.
===================================================================
Test the input parameters
Parameter adjustments */
/* Table of constant values */
static integer c__1 = 1;
static integer c_n1 = -1;
static real c_b32 = -1.f;
static real c_b34 = 1.f;
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, i__1, i__2, i__3, i__4;
/* Local variables */
static integer lopt, i__;
extern /* Subroutine */ int sgemv_(char *, integer *, integer *, real *,
real *, integer *, real *, integer *, real *, real *, integer *), scopy_(integer *, real *, integer *, real *, integer *),
strsv_(char *, char *, char *, integer *, real *, integer *, real
*, integer *);
static integer nb, np;
extern /* Subroutine */ int xerbla_(char *, integer *);
extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
integer *, integer *, ftnlen, ftnlen);
extern /* Subroutine */ int sggqrf_(integer *, integer *, integer *, real
*, integer *, real *, real *, integer *, real *, real *, integer *
, integer *);
static integer nb1, nb2, nb3, nb4, lwkopt;
static logical lquery;
extern /* Subroutine */ int sormqr_(char *, char *, integer *, integer *,
integer *, real *, integer *, real *, real *, integer *, real *,
integer *, integer *), sormrq_(char *, char *,
integer *, integer *, integer *, real *, integer *, real *, real *
, integer *, real *, integer *, integer *);
#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;
--d__;
--x;
--y;
--work;
/* Function Body */
*info = 0;
np = min(*n,*p);
nb1 = ilaenv_(&c__1, "SGEQRF", " ", n, m, &c_n1, &c_n1, (ftnlen)6, (
ftnlen)1);
nb2 = ilaenv_(&c__1, "SGERQF", " ", n, m, &c_n1, &c_n1, (ftnlen)6, (
ftnlen)1);
nb3 = ilaenv_(&c__1, "SORMQR", " ", n, m, p, &c_n1, (ftnlen)6, (ftnlen)1);
nb4 = ilaenv_(&c__1, "SORMRQ", " ", n, m, p, &c_n1, (ftnlen)6, (ftnlen)1);
/* Computing MAX */
i__1 = max(nb1,nb2), i__1 = max(i__1,nb3);
nb = max(i__1,nb4);
lwkopt = *m + np + max(*n,*p) * nb;
work[1] = (real) lwkopt;
lquery = *lwork == -1;
if (*n < 0) {
*info = -1;
} else if (*m < 0 || *m > *n) {
*info = -2;
} else if (*p < 0 || *p < *n - *m) {
*info = -3;
} else if (*lda < max(1,*n)) {
*info = -5;
} else if (*ldb < max(1,*n)) {
*info = -7;
} else /* if(complicated condition) */ {
/* Computing MAX */
i__1 = 1, i__2 = *n + *m + *p;
if (*lwork < max(i__1,i__2) && ! lquery) {
*info = -12;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("SGGGLM", &i__1);
return 0;
} else if (lquery) {
return 0;
}
/* Quick return if possible */
if (*n == 0) {
return 0;
}
/* Compute the GQR factorization of matrices A and B:
Q'*A = ( R11 ) M, Q'*B*Z' = ( T11 T12 ) M
( 0 ) N-M ( 0 T22 ) N-M
M M+P-N N-M
where R11 and T22 are upper triangular, and Q and Z are
orthogonal. */
i__1 = *lwork - *m - np;
sggqrf_(n, m, p, &a[a_offset], lda, &work[1], &b[b_offset], ldb, &work[*m
+ 1], &work[*m + np + 1], &i__1, info);
lopt = work[*m + np + 1];
/* Update left-hand-side vector d = Q'*d = ( d1 ) M
( d2 ) N-M */
i__1 = max(1,*n);
i__2 = *lwork - *m - np;
sormqr_("Left", "Transpose", n, &c__1, m, &a[a_offset], lda, &work[1], &
d__[1], &i__1, &work[*m + np + 1], &i__2, info);
/* Computing MAX */
i__1 = lopt, i__2 = (integer) work[*m + np + 1];
lopt = max(i__1,i__2);
/* Solve T22*y2 = d2 for y2 */
i__1 = *n - *m;
strsv_("Upper", "No transpose", "Non unit", &i__1, &b_ref(*m + 1, *m + *p
- *n + 1), ldb, &d__[*m + 1], &c__1);
i__1 = *n - *m;
scopy_(&i__1, &d__[*m + 1], &c__1, &y[*m + *p - *n + 1], &c__1);
/* Set y1 = 0 */
i__1 = *m + *p - *n;
for (i__ = 1; i__ <= i__1; ++i__) {
y[i__] = 0.f;
/* L10: */
}
/* Update d1 = d1 - T12*y2 */
i__1 = *n - *m;
sgemv_("No transpose", m, &i__1, &c_b32, &b_ref(1, *m + *p - *n + 1), ldb,
&y[*m + *p - *n + 1], &c__1, &c_b34, &d__[1], &c__1);
/* Solve triangular system: R11*x = d1 */
strsv_("Upper", "No Transpose", "Non unit", m, &a[a_offset], lda, &d__[1],
&c__1);
/* Copy D to X */
scopy_(m, &d__[1], &c__1, &x[1], &c__1);
/* Backward transformation y = Z'*y
Computing MAX */
i__1 = 1, i__2 = *n - *p + 1;
i__3 = max(1,*p);
i__4 = *lwork - *m - np;
sormrq_("Left", "Transpose", p, &c__1, &np, &b_ref(max(i__1,i__2), 1),
ldb, &work[*m + 1], &y[1], &i__3, &work[*m + np + 1], &i__4, info);
/* Computing MAX */
i__1 = lopt, i__2 = (integer) work[*m + np + 1];
work[1] = (real) (*m + np + max(i__1,i__2));
return 0;
/* End of SGGGLM */
} /* sggglm_ */
