/* Subroutine */ int cgrqts_(integer *m, integer *p, integer *n, complex *a,
complex *af, complex *q, complex *r__, integer *lda, complex *taua,
complex *b, complex *bf, complex *z__, complex *t, complex *bwk,
integer *ldb, complex *taub, complex *work, integer *lwork, real *
rwork, real *result)
{
/* System generated locals */
integer a_dim1, a_offset, af_dim1, af_offset, r_dim1, r_offset, q_dim1,
q_offset, b_dim1, b_offset, bf_dim1, bf_offset, t_dim1, t_offset,
z_dim1, z_offset, bwk_dim1, bwk_offset, i__1, i__2;
real r__1;
complex q__1;
/* Local variables */
real ulp;
integer info;
real unfl;
extern /* Subroutine */ int cgemm_(char *, char *, integer *, integer *,
integer *, complex *, complex *, integer *, complex *, integer *,
complex *, complex *, integer *), cherk_(char *,
char *, integer *, integer *, real *, complex *, integer *, real *
, complex *, integer *);
real resid, anorm, bnorm;
extern doublereal clange_(char *, integer *, integer *, complex *,
integer *, real *), clanhe_(char *, char *, integer *,
complex *, integer *, real *), slamch_(char *);
extern /* Subroutine */ int cggrqf_(integer *, integer *, integer *,
complex *, integer *, complex *, complex *, integer *, complex *,
complex *, integer *, integer *), clacpy_(char *, integer *,
integer *, complex *, integer *, complex *, integer *),
claset_(char *, integer *, integer *, complex *, complex *,
complex *, integer *), cungqr_(integer *, integer *,
integer *, complex *, integer *, complex *, complex *, integer *,
integer *), cungrq_(integer *, integer *, integer *, complex *,
integer *, complex *, complex *, integer *, integer *);
/* -- LAPACK test routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* CGRQTS tests CGGRQF, which computes the GRQ factorization of an */
/* M-by-N matrix A and a P-by-N matrix B: A = R*Q and B = Z*T*Q. */
/* Arguments */
/* ========= */
/* 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) COMPLEX array, dimension (LDA,N) */
/* The M-by-N matrix A. */
/* AF (output) COMPLEX array, dimension (LDA,N) */
/* Details of the GRQ factorization of A and B, as returned */
/* by CGGRQF, see CGGRQF for further details. */
/* Q (output) COMPLEX array, dimension (LDA,N) */
/* The N-by-N unitary matrix Q. */
/* R (workspace) COMPLEX array, dimension (LDA,MAX(M,N)) */
/* LDA (input) INTEGER */
/* The leading dimension of the arrays A, AF, R and Q. */
/* LDA >= max(M,N). */
/* TAUA (output) COMPLEX array, dimension (min(M,N)) */
/* The scalar factors of the elementary reflectors, as returned */
/* by SGGQRC. */
/* B (input) COMPLEX array, dimension (LDB,N) */
/* On entry, the P-by-N matrix A. */
/* BF (output) COMPLEX array, dimension (LDB,N) */
/* Details of the GQR factorization of A and B, as returned */
/* by CGGRQF, see CGGRQF for further details. */
/* Z (output) REAL array, dimension (LDB,P) */
/* The P-by-P unitary matrix Z. */
/* T (workspace) COMPLEX array, dimension (LDB,max(P,N)) */
/* BWK (workspace) COMPLEX array, dimension (LDB,N) */
/* LDB (input) INTEGER */
/* The leading dimension of the arrays B, BF, Z and T. */
/* LDB >= max(P,N). */
/* TAUB (output) COMPLEX array, dimension (min(P,N)) */
/* The scalar factors of the elementary reflectors, as returned */
/* by SGGRQF. */
/* WORK (workspace) COMPLEX array, dimension (LWORK) */
/* LWORK (input) INTEGER */
/* The dimension of the array WORK, LWORK >= max(M,P,N)**2. */
/* RWORK (workspace) REAL array, dimension (M) */
/* RESULT (output) REAL array, dimension (4) */
/* The test ratios: */
/* RESULT(1) = norm( R - A*Q' ) / ( MAX(M,N)*norm(A)*ULP) */
/* RESULT(2) = norm( T*Q - Z'*B ) / (MAX(P,N)*norm(B)*ULP) */
/* RESULT(3) = norm( I - Q'*Q ) / ( N*ULP ) */
/* RESULT(4) = norm( I - Z'*Z ) / ( P*ULP ) */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Parameter adjustments */
r_dim1 = *lda;
r_offset = 1 + r_dim1;
r__ -= r_offset;
q_dim1 = *lda;
q_offset = 1 + q_dim1;
q -= q_offset;
af_dim1 = *lda;
af_offset = 1 + af_dim1;
af -= af_offset;
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
--taua;
bwk_dim1 = *ldb;
bwk_offset = 1 + bwk_dim1;
bwk -= bwk_offset;
t_dim1 = *ldb;
t_offset = 1 + t_dim1;
t -= t_offset;
z_dim1 = *ldb;
z_offset = 1 + z_dim1;
z__ -= z_offset;
bf_dim1 = *ldb;
bf_offset = 1 + bf_dim1;
bf -= bf_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
--taub;
--work;
--rwork;
--result;
/* Function Body */
ulp = slamch_("Precision");
unfl = slamch_("Safe minimum");
/* Copy the matrix A to the array AF. */
clacpy_("Full", m, n, &a[a_offset], lda, &af[af_offset], lda);
clacpy_("Full", p, n, &b[b_offset], ldb, &bf[bf_offset], ldb);
/* Computing MAX */
r__1 = clange_("1", m, n, &a[a_offset], lda, &rwork[1]);
anorm = dmax(r__1,unfl);
/* Computing MAX */
r__1 = clange_("1", p, n, &b[b_offset], ldb, &rwork[1]);
bnorm = dmax(r__1,unfl);
/* Factorize the matrices A and B in the arrays AF and BF. */
cggrqf_(m, p, n, &af[af_offset], lda, &taua[1], &bf[bf_offset], ldb, &
taub[1], &work[1], lwork, &info);
/* Generate the N-by-N matrix Q */
claset_("Full", n, n, &c_b3, &c_b3, &q[q_offset], lda);
if (*m <= *n) {
if (*m > 0 && *m < *n) {
i__1 = *n - *m;
clacpy_("Full", m, &i__1, &af[af_offset], lda, &q[*n - *m + 1 +
q_dim1], lda);
}
if (*m > 1) {
i__1 = *m - 1;
i__2 = *m - 1;
clacpy_("Lower", &i__1, &i__2, &af[(*n - *m + 1) * af_dim1 + 2],
lda, &q[*n - *m + 2 + (*n - *m + 1) * q_dim1], lda);
}
} else {
if (*n > 1) {
i__1 = *n - 1;
i__2 = *n - 1;
clacpy_("Lower", &i__1, &i__2, &af[*m - *n + 2 + af_dim1], lda, &
q[q_dim1 + 2], lda);
}
}
i__1 = min(*m,*n);
cungrq_(n, n, &i__1, &q[q_offset], lda, &taua[1], &work[1], lwork, &info);
/* Generate the P-by-P matrix Z */
claset_("Full", p, p, &c_b3, &c_b3, &z__[z_offset], ldb);
if (*p > 1) {
i__1 = *p - 1;
clacpy_("Lower", &i__1, n, &bf[bf_dim1 + 2], ldb, &z__[z_dim1 + 2],
ldb);
}
i__1 = min(*p,*n);
cungqr_(p, p, &i__1, &z__[z_offset], ldb, &taub[1], &work[1], lwork, &
info);
/* Copy R */
claset_("Full", m, n, &c_b1, &c_b1, &r__[r_offset], lda);
if (*m <= *n) {
clacpy_("Upper", m, m, &af[(*n - *m + 1) * af_dim1 + 1], lda, &r__[(*
n - *m + 1) * r_dim1 + 1], lda);
} else {
i__1 = *m - *n;
clacpy_("Full", &i__1, n, &af[af_offset], lda, &r__[r_offset], lda);
clacpy_("Upper", n, n, &af[*m - *n + 1 + af_dim1], lda, &r__[*m - *n
+ 1 + r_dim1], lda);
}
/* Copy T */
claset_("Full", p, n, &c_b1, &c_b1, &t[t_offset], ldb);
clacpy_("Upper", p, n, &bf[bf_offset], ldb, &t[t_offset], ldb);
/* Compute R - A*Q' */
q__1.r = -1.f, q__1.i = -0.f;
cgemm_("No transpose", "Conjugate transpose", m, n, n, &q__1, &a[a_offset]
, lda, &q[q_offset], lda, &c_b2, &r__[r_offset], lda);
/* Compute norm( R - A*Q' ) / ( MAX(M,N)*norm(A)*ULP ) . */
resid = clange_("1", m, n, &r__[r_offset], lda, &rwork[1]);
if (anorm > 0.f) {
/* Computing MAX */
i__1 = max(1,*m);
result[1] = resid / (real) max(i__1,*n) / anorm / ulp;
} else {
result[1] = 0.f;
}
/* Compute T*Q - Z'*B */
cgemm_("Conjugate transpose", "No transpose", p, n, p, &c_b2, &z__[
z_offset], ldb, &b[b_offset], ldb, &c_b1, &bwk[bwk_offset], ldb);
q__1.r = -1.f, q__1.i = -0.f;
cgemm_("No transpose", "No transpose", p, n, n, &c_b2, &t[t_offset], ldb,
&q[q_offset], lda, &q__1, &bwk[bwk_offset], ldb);
/* Compute norm( T*Q - Z'*B ) / ( MAX(P,N)*norm(A)*ULP ) . */
resid = clange_("1", p, n, &bwk[bwk_offset], ldb, &rwork[1]);
if (bnorm > 0.f) {
/* Computing MAX */
i__1 = max(1,*p);
result[2] = resid / (real) max(i__1,*m) / bnorm / ulp;
} else {
result[2] = 0.f;
}
/* Compute I - Q*Q' */
claset_("Full", n, n, &c_b1, &c_b2, &r__[r_offset], lda);
cherk_("Upper", "No Transpose", n, n, &c_b34, &q[q_offset], lda, &c_b35, &
r__[r_offset], lda);
/* Compute norm( I - Q'*Q ) / ( N * ULP ) . */
resid = clanhe_("1", "Upper", n, &r__[r_offset], lda, &rwork[1]);
result[3] = resid / (real) max(1,*n) / ulp;
/* Compute I - Z'*Z */
claset_("Full", p, p, &c_b1, &c_b2, &t[t_offset], ldb);
cherk_("Upper", "Conjugate transpose", p, p, &c_b34, &z__[z_offset], ldb,
&c_b35, &t[t_offset], ldb);
/* Compute norm( I - Z'*Z ) / ( P*ULP ) . */
resid = clanhe_("1", "Upper", p, &t[t_offset], ldb, &rwork[1]);
result[4] = resid / (real) max(1,*p) / ulp;
return 0;
/* End of CGRQTS */
} /* cgrqts_ */
/* Subroutine */ int ctimqr_(char *line, integer *nm, integer *mval, integer *
nval, integer *nk, integer *kval, integer *nnb, integer *nbval,
integer *nxval, integer *nlda, integer *ldaval, real *timmin, complex
*a, complex *tau, complex *b, complex *work, real *rwork, real *
reslts, integer *ldr1, integer *ldr2, integer *ldr3, integer *nout,
ftnlen line_len)
{
/* Initialized data */
static char subnam[6*3] = "CGEQRF" "CUNGQR" "CUNMQR";
static char sides[1*2] = "L" "R";
static char transs[1*2] = "N" "C";
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), cgeqrf_(
integer *, integer *, complex *, integer *, complex *, complex *,
integer *, integer *);
extern doublereal second_(void);
extern /* Subroutine */ int clacpy_(char *, integer *, integer *, complex
*, integer *, complex *, integer *), ctimmg_(integer *,
integer *, integer *, complex *, integer *, integer *, integer *),
atimin_(char *, char *, integer *, char *, logical *, integer *,
integer *, ftnlen, ftnlen, ftnlen), clatms_(integer *, integer *,
char *, integer *, char *, real *, integer *, real *, real *,
integer *, integer *, char *, complex *, integer *, complex *,
integer *), xlaenv_(integer *, integer *);
extern doublereal smflop_(real *, real *, integer *);
static real untime;
extern /* Subroutine */ int cungqr_(integer *, integer *, integer *,
complex *, integer *, complex *, complex *, integer *, integer *);
static logical timsub[3];
extern /* Subroutine */ int cunmqr_(char *, char *, integer *, integer *,
integer *, complex *, integer *, complex *, complex *, integer *,
complex *, 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
=======
CTIMQR times the LAPACK routines to perform the QR factorization of
a COMPLEX 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 CUNMQR.
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) COMPLEX array, dimension (LDAMAX*NMAX)
where LDAMAX and NMAX are the maximum values of LDA and N.
TAU (workspace) COMPLEX array, dimension (min(M,N))
B (workspace) COMPLEX array, dimension (LDAMAX*NMAX)
WORK (workspace) COMPLEX array, dimension (LDAMAX*NBMAX)
where NBMAX is the maximum value of NB.
RWORK (workspace) REAL array, dimension
(min(MMAX,NMAX))
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 CLATMS 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;
--rwork;
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, "Complex precision", (ftnlen)1, (ftnlen)17);
s_copy(path + 1, "QR", (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 = n * max(1,nb);
lw = max(i__4,i__5);
icopy_(&c__4, reseed, &c__1, iseed, &c__1);
/* Generate a test matrix of size M by N. */
clatms_(&m, &n, "Uniform", iseed, "Nonsymm", &rwork[1], &c__3,
&c_b24, &c_b25, &m, &n, "No packing", &b[1], &lda, &
work[1], &info);
if (timsub[0]) {
/* CGEQRF: QR factorization */
clacpy_("Full", &m, &n, &b[1], &lda, &a[1], &lda);
ic = 0;
s1 = second_();
L10:
cgeqrf_(&m, &n, &a[1], &lda, &tau[1], &work[1], &lw, &
info);
s2 = second_();
time = s2 - s1;
++ic;
if (time < *timmin) {
clacpy_("Full", &m, &n, &b[1], &lda, &a[1], &lda);
goto L10;
}
/* Subtract the time used in CLACPY. */
icl = 1;
s1 = second_();
L20:
s2 = second_();
untime = s2 - s1;
++icl;
if (icl <= ic) {
clacpy_("Full", &m, &n, &a[1], &lda, &b[1], &lda);
goto L20;
}
time = (time - untime) / (real) ic;
ops = sopla_("CGEQRF", &m, &n, &c__0, &c__0, &nb);
reslts_ref(inb, im, ilda, 1) = smflop_(&ops, &time, &info)
;
} else {
/* If CGEQRF was not timed, generate a matrix and factor
it using CGEQRF anyway so that the factored form of
the matrix can be used in timing the other routines. */
clacpy_("Full", &m, &n, &b[1], &lda, &a[1], &lda);
cgeqrf_(&m, &n, &a[1], &lda, &tau[1], &work[1], &lw, &
info);
}
if (timsub[1]) {
/* CUNGQR: Generate orthogonal matrix Q from the QR
factorization */
clacpy_("Full", &m, &minmn, &a[1], &lda, &b[1], &lda);
ic = 0;
s1 = second_();
L30:
cungqr_(&m, &minmn, &minmn, &b[1], &lda, &tau[1], &work[1]
, &lw, &info);
s2 = second_();
time = s2 - s1;
++ic;
if (time < *timmin) {
clacpy_("Full", &m, &minmn, &a[1], &lda, &b[1], &lda);
goto L30;
}
/* Subtract the time used in CLACPY. */
icl = 1;
s1 = second_();
L40:
s2 = second_();
untime = s2 - s1;
++icl;
if (icl <= ic) {
clacpy_("Full", &m, &minmn, &a[1], &lda, &b[1], &lda);
goto L40;
}
time = (time - untime) / (real) ic;
ops = sopla_("CUNGQR", &m, &minmn, &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 CUNMQR 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: */
}
/* CUNMQR: 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 QR decomposition. */
clatms_(&m, &n, "Uniform", iseed, "Nonsymm", &rwork[1], &c__3,
&c_b24, &c_b25, &m, &n, "No packing", &a[1], &lda, &
work[1], &info);
/* Computing MAX */
i__3 = 1, i__4 = n * max(1,nb);
lw = max(i__3,i__4);
cgeqrf_(&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) {
m1 = m;
k1 = n;
n1 = k;
/* Computing MAX */
i__5 = 1, i__6 = n1 * max(1,nb);
lw = max(i__5,i__6);
} else {
n1 = m;
k1 = 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];
ctimmg_(&c__0, &m1, &n1, &b[1], &lda, &c__0, &
c__0);
ic = 0;
s1 = second_();
L110:
cunmqr_(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) {
ctimmg_(&c__0, &m1, &n1, &b[1], &lda, &
c__0, &c__0);
goto L110;
}
/* Subtract the time used in CTIMMG. */
icl = 1;
s1 = second_();
L120:
s2 = second_();
untime = s2 - s1;
++icl;
if (icl <= ic) {
ctimmg_(&c__0, &m1, &n1, &b[1], &lda, &
c__0, &c__0);
goto L120;
}
time = (time - untime) / (real) ic;
i__5 = iside - 1;
ops = sopla_("CUNMQR", &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", labm, "K", 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 CTIMQR */
} /* ctimqr_ */
