/*! return a transposed column vector */
inline _zcovector t(const zrovector& rovec)
{VERBOSE_REPORT;
zcovector covec(rovec.l);
zcopy_(rovec.l, rovec.array, 1, covec.array, 1);
return _(covec);
}
int
f2c_zcopy(integer* N,
doublecomplex* X, integer* incX,
doublecomplex* Y, integer* incY)
{
zcopy_(N, X, incX, Y, incY);
return 0;
}
static PyObject* diag(PyObject *self, PyObject *args)
{
PyObject *F;
matrix *d=NULL;
cholmod_factor *L;
#if PY_MAJOR_VERSION >= 3
const char *descr;
#else
char *descr;
#endif
int k, strt, incx=1, incy, nrows, ncols;
if (!set_options()) return NULL;
if (!PyArg_ParseTuple(args, "O", &F)) return NULL;
#if PY_MAJOR_VERSION >= 3
if (!PyCapsule_CheckExact(F) || !(descr = PyCapsule_GetName(F)))
err_CO("F");
if (strncmp(descr, "CHOLMOD FACTOR", 14))
PY_ERR_TYPE("F is not a CHOLMOD factor");
L = (cholmod_factor *) PyCapsule_GetPointer(F, descr);
#else
if (!PyCObject_Check(F)) err_CO("F");
descr = PyCObject_GetDesc(F);
if (!descr || strncmp(descr, "CHOLMOD FACTOR", 14))
PY_ERR_TYPE("F is not a CHOLMOD factor");
L = (cholmod_factor *) PyCObject_AsVoidPtr(F);
#endif
/* Check factorization */
if (L->xtype == CHOLMOD_PATTERN || L->minor<L->n || !L->is_ll
|| !L->is_super)
PY_ERR(PyExc_ValueError, "F must be a nonsingular supernodal "
"Cholesky factor");
if (!(d = Matrix_New(L->n,1,L->xtype == CHOLMOD_REAL ? DOUBLE :
COMPLEX))) return PyErr_NoMemory();
strt = 0;
for (k=0; k<L->nsuper; k++){
/* x[L->px[k], .... ,L->px[k+1]-1] is a dense lower-triangular
* nrowx times ncols matrix. We copy its diagonal to
* d[strt, ..., strt+ncols-1] */
ncols = (int)((int_t *) L->super)[k+1] -
((int_t *) L->super)[k];
nrows = (int)((int_t *) L->pi)[k+1] - ((int_t *) L->pi)[k];
incy = nrows+1;
if (MAT_ID(d) == DOUBLE)
dcopy_(&ncols, ((double *) L->x) + ((int_t *) L->px)[k],
&incy, MAT_BUFD(d)+strt, &incx);
else
zcopy_(&ncols, ((double complex *) L->x) + ((int_t *) L->px)[k],
&incy, MAT_BUFZ(d)+strt, &incx);
strt += ncols;
}
return (PyObject *)d;
}
/*! zhematrix copy constructor */
inline zhematrix::zhematrix(const zhematrix& mat)
: m(n)
{VERBOSE_REPORT;
//////// initialize ////////
n =mat.n;
array =new comple[n*n];
darray =new comple*[n];
for(int i=0; i<n; i++){ darray[i] =&array[i*n]; }
//////// copy ////////
zcopy_(n*n, mat.array, 1, array, 1);
}
/*! zgematrix copy constructor */
inline zgematrix::zgematrix(const zgematrix& mat)
{VERBOSE_REPORT;
//////// initialize ////////
m =mat.m;
n =mat.n;
array =new comple[m*n];
darray =new comple*[n];
for(int i=0; i<n; i++){ darray[i] =&array[i*m]; }
//////// copy ////////
zcopy_(m*n, mat.array, 1, array, 1);
}
/*! return a transposed column vector */
inline _zcovector t(const zrovector& rovec)
{
#ifdef CPPL_VERBOSE
std::cerr << "# [MARK] t(const zrovector&)"
<< std::endl;
#endif//CPPL_VERBOSE
zcovector covec(rovec.L);
zcopy_(rovec.L, rovec.Array, 1, covec.array, 1);
return _(covec);
}
void parallelStorageDTCCplx::getField()
{
// If process has a part of the field and is stores the outGrid copy relevant field info directly to the out_grid
if(toOutGrid_)
{
for(int kk = 0; kk < toOutGrid_->opSz_[2]; ++kk )
{
for(int jj = 0; jj < toOutGrid_->opSz_[1]; ++jj)
{
zcopy_(toOutGrid_->opSz_[0], &grid_->point(toOutGrid_->loc_[0]+jj*toOutGrid_->addVec1_[0]+kk*toOutGrid_->addVec2_[0], toOutGrid_->loc_[1]+jj*toOutGrid_->addVec1_[1]+kk*toOutGrid_->addVec2_[1], toOutGrid_->loc_[2]+jj*toOutGrid_->addVec1_[2]+kk*toOutGrid_->addVec2_[2]), toOutGrid_->stride_, &outGrid_->point(toOutGrid_->locOutGrid_[0]+jj*toOutGrid_->addVec1_[0]+kk*toOutGrid_->addVec2_[0], toOutGrid_->locOutGrid_[1]+jj*toOutGrid_->addVec1_[1]+kk*toOutGrid_->addVec2_[1], toOutGrid_->locOutGrid_[2]+jj*toOutGrid_->addVec1_[2]+kk*toOutGrid_->addVec2_[2]), toOutGrid_->strideOutGrid_);
}
}
}
// If the process is a slave process not holding the outGrid then copy the field information to a vector and send it to master
if(slave_)
{
for(int kk = 0; kk < slave_->opSz_[2]; ++kk )
for(int jj = 0; jj < slave_->opSz_[1]; ++jj)
zcopy_(slave_->opSz_[0], &grid_->point(slave_->loc_[0]+jj*slave_->addVec1_[0]+kk*slave_->addVec2_[0], slave_->loc_[1]+jj*slave_->addVec1_[1]+kk*slave_->addVec2_[1], slave_->loc_[2]+jj*slave_->addVec1_[2]+kk*slave_->addVec2_[2]), slave_->stride_, &scratch_[ slave_->opSz_[0]*(jj + kk*slave_->opSz_[1]) ], 1);
gridComm_->send(slave_->masterProc_, gridComm_->cantorTagGen(gridComm_->rank(), slave_->masterProc_, 1, 0), scratch_);
}
// If master then for each slave recv the information and copy it to outGrid
if(masterBool_)
{
for(auto & slave : master_)
{
gridComm_->recv(slave->slaveProc_, gridComm_->cantorTagGen(slave->slaveProc_, gridComm_->rank(), 1, 0), scratch_);
for(int kk = 0; kk < slave->sz_[2]; ++kk)
{
for(int jj = 0; jj < slave->sz_[1]; ++jj)
{
zcopy_(slave->sz_[0], &scratch_[(jj + slave->sz_[1] * kk) * slave->sz_[0] ], 1, &outGrid_->point(slave->addVec1_[0]*jj+slave->addVec2_[0]*kk+slave->loc_[0], slave->addVec1_[1]*jj+slave->addVec2_[1]*kk+slave->loc_[1], slave->addVec1_[2]*jj+slave->addVec2_[2]*kk+slave->loc_[2]), slave->stride_);
}
}
}
}
return;
}
/*! zgbmatrix copy constructor */
inline zgbmatrix::zgbmatrix(const zgbmatrix& mat)
{VERBOSE_REPORT;
//////// initialize ////////
m =mat.m;
n =mat.n;
kl =mat.kl;
ku =mat.ku;
array =new comple[(kl+ku+1)*n];
darray =new comple*[n];
for(int i=0; i<n; i++){ darray[i] =&array[i*(kl+ku+1)]; }
//////// copy ////////
zcopy_((kl+ku+1)*n, mat.array, 1, array, 1);
}
/*! make a deep copy of the matrix */
inline void zgbmatrix::copy(const zgbmatrix& mat)
{VERBOSE_REPORT;
m =mat.m;
n =mat.n;
kl =mat.kl;
ku =mat.ku;
delete [] array;
array =new comple[(mat.kl+mat.ku+1)*mat.n];
delete [] darray;
darray =new comple*[n];
for(int i=0; i<n; i++){ darray[i] =&array[i*(kl+ku+1)]; }
zcopy_((mat.kl+mat.ku+1)*mat.n, mat.array, 1, array, 1);
}
void parallelStorageFreqDTCCplx::fieldIn(cplx* fftFact)
{
if(!fieldInFreq_)
return;
// Copy the field information into a vector
for(int jj = 0; jj < fieldInFreq_->sz_[2]; ++jj)
{
for(int ii = 0; ii < fieldInFreq_->sz_[1]; ++ii)
{
zcopy_(fieldInFreq_->sz_[0], &grid_->point(fieldInFreq_->loc_[0]+ii*fieldInFreq_->addVec1_[0]+jj*fieldInFreq_->addVec2_[0], fieldInFreq_->loc_[1]+ii*fieldInFreq_->addVec1_[1]+jj*fieldInFreq_->addVec2_[1], fieldInFreq_->loc_[2]+ii*fieldInFreq_->addVec1_[2]+jj*fieldInFreq_->addVec2_[2]), fieldInFreq_->stride_, &fIn_[ (ii*fieldInFreq_->sz_[2] + jj)*fieldInFreq_->sz_[0]] , 1 );
}
}
// Take an outer product of the prefactor vector and the field vectors to get the discrete Fourier Transform at all points
zgerc_(nfreq_, fieldInFreq_->sz_[0]*fieldInFreq_->sz_[1]*fieldInFreq_->sz_[2], ONE_, fftFact, 1, fIn_.data(), 1, outGrid_->data(), nfreq_);
}
/*! zhematrix copy constructor */
inline zhematrix::zhematrix(const zhematrix& mat)
: n(N), array(Array), darray(Darray)
{
#ifdef CPPL_VERBOSE
std::cerr << "# [MARK] zhematrix::zhematrix(const zhematrix&)"
<< std::endl;
#endif//CPPL_VERBOSE
//////// initialize ////////
N =mat.N;
Array =new std::complex<double>[N*N];
Darray =new std::complex<double>*[N];
for(int i=0; i<N; i++){ Darray[i] =&Array[i*N]; }
//////// copy ////////
zcopy_(N*N, mat.Array, 1, Array, 1);
#ifdef CPPL_DEBUG
std::cerr << "# [NOTE] zhematrix::zhematrix(const zhematrix&) "
<< "A new matrix at " << Array << " has been made." << std::endl;
#endif//CPPL_DEBUG
}
/* Subroutine */ int ztrevc_(char *side, char *howmny, logical *select,
integer *n, doublecomplex *t, integer *ldt, doublecomplex *vl,
integer *ldvl, doublecomplex *vr, integer *ldvr, integer *mm, integer
*m, doublecomplex *work, doublereal *rwork, integer *info)
{
/* -- LAPACK 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
=======
ZTREVC computes some or all of the right and/or left eigenvectors of
a complex upper triangular matrix T.
The right eigenvector x and the left eigenvector y of T corresponding
to an eigenvalue w are defined by:
T*x = w*x, y'*T = w*y'
where y' denotes the conjugate transpose of the vector y.
If all eigenvectors are requested, the routine may either return the
matrices X and/or Y of right or left eigenvectors of T, or the
products Q*X and/or Q*Y, where Q is an input unitary
matrix. If T was obtained from the Schur factorization of an
original matrix A = Q*T*Q', then Q*X and Q*Y are the matrices of
right or left eigenvectors of A.
Arguments
=========
SIDE (input) CHARACTER*1
= 'R': compute right eigenvectors only;
= 'L': compute left eigenvectors only;
= 'B': compute both right and left eigenvectors.
HOWMNY (input) CHARACTER*1
= 'A': compute all right and/or left eigenvectors;
= 'B': compute all right and/or left eigenvectors,
and backtransform them using the input matrices
supplied in VR and/or VL;
= 'S': compute selected right and/or left eigenvectors,
specified by the logical array SELECT.
SELECT (input) LOGICAL array, dimension (N)
If HOWMNY = 'S', SELECT specifies the eigenvectors to be
computed.
If HOWMNY = 'A' or 'B', SELECT is not referenced.
To select the eigenvector corresponding to the j-th
eigenvalue, SELECT(j) must be set to .TRUE..
N (input) INTEGER
The order of the matrix T. N >= 0.
T (input/output) COMPLEX*16 array, dimension (LDT,N)
The upper triangular matrix T. T is modified, but restored
on exit.
LDT (input) INTEGER
The leading dimension of the array T. LDT >= max(1,N).
VL (input/output) COMPLEX*16 array, dimension (LDVL,MM)
On entry, if SIDE = 'L' or 'B' and HOWMNY = 'B', VL must
contain an N-by-N matrix Q (usually the unitary matrix Q of
Schur vectors returned by ZHSEQR).
On exit, if SIDE = 'L' or 'B', VL contains:
if HOWMNY = 'A', the matrix Y of left eigenvectors of T;
VL is lower triangular. The i-th column
VL(i) of VL is the eigenvector corresponding
to T(i,i).
if HOWMNY = 'B', the matrix Q*Y;
if HOWMNY = 'S', the left eigenvectors of T specified by
SELECT, stored consecutively in the columns
of VL, in the same order as their
eigenvalues.
If SIDE = 'R', VL is not referenced.
LDVL (input) INTEGER
The leading dimension of the array VL. LDVL >= max(1,N) if
SIDE = 'L' or 'B'; LDVL >= 1 otherwise.
VR (input/output) COMPLEX*16 array, dimension (LDVR,MM)
On entry, if SIDE = 'R' or 'B' and HOWMNY = 'B', VR must
contain an N-by-N matrix Q (usually the unitary matrix Q of
Schur vectors returned by ZHSEQR).
On exit, if SIDE = 'R' or 'B', VR contains:
if HOWMNY = 'A', the matrix X of right eigenvectors of T;
VR is upper triangular. The i-th column
VR(i) of VR is the eigenvector corresponding
to T(i,i).
if HOWMNY = 'B', the matrix Q*X;
if HOWMNY = 'S', the right eigenvectors of T specified by
SELECT, stored consecutively in the columns
of VR, in the same order as their
eigenvalues.
If SIDE = 'L', VR is not referenced.
LDVR (input) INTEGER
The leading dimension of the array VR. LDVR >= max(1,N) if
SIDE = 'R' or 'B'; LDVR >= 1 otherwise.
MM (input) INTEGER
The number of columns in the arrays VL and/or VR. MM >= M.
M (output) INTEGER
The number of columns in the arrays VL and/or VR actually
used to store the eigenvectors. If HOWMNY = 'A' or 'B', M
is set to N. Each selected eigenvector occupies one
column.
WORK (workspace) COMPLEX*16 array, dimension (2*N)
RWORK (workspace) DOUBLE PRECISION array, dimension (N)
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
Further Details
===============
The algorithm used in this program is basically backward (forward)
substitution, with scaling to make the the code robust against
possible overflow.
Each eigenvector is normalized so that the element of largest
magnitude has magnitude 1; here the magnitude of a complex number
(x,y) is taken to be |x| + |y|.
=====================================================================
Decode and test the input parameters
Parameter adjustments */
/* Table of constant values */
static doublecomplex c_b2 = {1.,0.};
static integer c__1 = 1;
/* System generated locals */
integer t_dim1, t_offset, vl_dim1, vl_offset, vr_dim1, vr_offset, i__1,
i__2, i__3, i__4, i__5;
doublereal d__1, d__2, d__3;
doublecomplex z__1, z__2;
/* Builtin functions */
double d_imag(doublecomplex *);
void d_cnjg(doublecomplex *, doublecomplex *);
/* Local variables */
static logical allv;
static doublereal unfl, ovfl, smin;
static logical over;
static integer i__, j, k;
static doublereal scale;
extern logical lsame_(char *, char *);
static doublereal remax;
static logical leftv, bothv;
extern /* Subroutine */ int zgemv_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublecomplex *, doublecomplex *, integer *);
static logical somev;
extern /* Subroutine */ int zcopy_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *), dlabad_(doublereal *, doublereal *);
static integer ii, ki;
extern doublereal dlamch_(char *);
static integer is;
extern /* Subroutine */ int xerbla_(char *, integer *), zdscal_(
integer *, doublereal *, doublecomplex *, integer *);
extern integer izamax_(integer *, doublecomplex *, integer *);
static logical rightv;
extern doublereal dzasum_(integer *, doublecomplex *, integer *);
static doublereal smlnum;
extern /* Subroutine */ int zlatrs_(char *, char *, char *, char *,
integer *, doublecomplex *, integer *, doublecomplex *,
doublereal *, doublereal *, integer *);
static doublereal ulp;
#define t_subscr(a_1,a_2) (a_2)*t_dim1 + a_1
#define t_ref(a_1,a_2) t[t_subscr(a_1,a_2)]
#define vl_subscr(a_1,a_2) (a_2)*vl_dim1 + a_1
#define vl_ref(a_1,a_2) vl[vl_subscr(a_1,a_2)]
#define vr_subscr(a_1,a_2) (a_2)*vr_dim1 + a_1
#define vr_ref(a_1,a_2) vr[vr_subscr(a_1,a_2)]
--select;
t_dim1 = *ldt;
t_offset = 1 + t_dim1 * 1;
t -= t_offset;
vl_dim1 = *ldvl;
vl_offset = 1 + vl_dim1 * 1;
vl -= vl_offset;
vr_dim1 = *ldvr;
vr_offset = 1 + vr_dim1 * 1;
vr -= vr_offset;
--work;
--rwork;
/* Function Body */
bothv = lsame_(side, "B");
rightv = lsame_(side, "R") || bothv;
leftv = lsame_(side, "L") || bothv;
allv = lsame_(howmny, "A");
over = lsame_(howmny, "B");
somev = lsame_(howmny, "S");
/* Set M to the number of columns required to store the selected
eigenvectors. */
if (somev) {
*m = 0;
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
if (select[j]) {
++(*m);
}
/* L10: */
}
} else {
*m = *n;
}
*info = 0;
if (! rightv && ! leftv) {
*info = -1;
} else if (! allv && ! over && ! somev) {
*info = -2;
} else if (*n < 0) {
*info = -4;
} else if (*ldt < max(1,*n)) {
*info = -6;
} else if (*ldvl < 1 || leftv && *ldvl < *n) {
*info = -8;
} else if (*ldvr < 1 || rightv && *ldvr < *n) {
*info = -10;
} else if (*mm < *m) {
*info = -11;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("ZTREVC", &i__1);
return 0;
}
/* Quick return if possible. */
if (*n == 0) {
return 0;
}
/* Set the constants to control overflow. */
unfl = dlamch_("Safe minimum");
ovfl = 1. / unfl;
dlabad_(&unfl, &ovfl);
ulp = dlamch_("Precision");
smlnum = unfl * (*n / ulp);
/* Store the diagonal elements of T in working array WORK. */
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = i__ + *n;
i__3 = t_subscr(i__, i__);
work[i__2].r = t[i__3].r, work[i__2].i = t[i__3].i;
/* L20: */
}
/* Compute 1-norm of each column of strictly upper triangular
part of T to control overflow in triangular solver. */
rwork[1] = 0.;
i__1 = *n;
for (j = 2; j <= i__1; ++j) {
i__2 = j - 1;
rwork[j] = dzasum_(&i__2, &t_ref(1, j), &c__1);
/* L30: */
}
if (rightv) {
/* Compute right eigenvectors. */
is = *m;
for (ki = *n; ki >= 1; --ki) {
if (somev) {
if (! select[ki]) {
goto L80;
}
}
/* Computing MAX */
i__1 = t_subscr(ki, ki);
d__3 = ulp * ((d__1 = t[i__1].r, abs(d__1)) + (d__2 = d_imag(&
t_ref(ki, ki)), abs(d__2)));
smin = max(d__3,smlnum);
work[1].r = 1., work[1].i = 0.;
/* Form right-hand side. */
i__1 = ki - 1;
for (k = 1; k <= i__1; ++k) {
i__2 = k;
i__3 = t_subscr(k, ki);
z__1.r = -t[i__3].r, z__1.i = -t[i__3].i;
work[i__2].r = z__1.r, work[i__2].i = z__1.i;
/* L40: */
}
/* Solve the triangular system:
(T(1:KI-1,1:KI-1) - T(KI,KI))*X = SCALE*WORK. */
i__1 = ki - 1;
for (k = 1; k <= i__1; ++k) {
i__2 = t_subscr(k, k);
i__3 = t_subscr(k, k);
i__4 = t_subscr(ki, ki);
z__1.r = t[i__3].r - t[i__4].r, z__1.i = t[i__3].i - t[i__4]
.i;
t[i__2].r = z__1.r, t[i__2].i = z__1.i;
i__2 = t_subscr(k, k);
if ((d__1 = t[i__2].r, abs(d__1)) + (d__2 = d_imag(&t_ref(k,
k)), abs(d__2)) < smin) {
i__3 = t_subscr(k, k);
t[i__3].r = smin, t[i__3].i = 0.;
}
/* L50: */
}
if (ki > 1) {
i__1 = ki - 1;
zlatrs_("Upper", "No transpose", "Non-unit", "Y", &i__1, &t[
t_offset], ldt, &work[1], &scale, &rwork[1], info);
i__1 = ki;
work[i__1].r = scale, work[i__1].i = 0.;
}
/* Copy the vector x or Q*x to VR and normalize. */
if (! over) {
zcopy_(&ki, &work[1], &c__1, &vr_ref(1, is), &c__1);
ii = izamax_(&ki, &vr_ref(1, is), &c__1);
i__1 = vr_subscr(ii, is);
remax = 1. / ((d__1 = vr[i__1].r, abs(d__1)) + (d__2 = d_imag(
&vr_ref(ii, is)), abs(d__2)));
zdscal_(&ki, &remax, &vr_ref(1, is), &c__1);
i__1 = *n;
for (k = ki + 1; k <= i__1; ++k) {
i__2 = vr_subscr(k, is);
vr[i__2].r = 0., vr[i__2].i = 0.;
/* L60: */
}
} else {
if (ki > 1) {
i__1 = ki - 1;
z__1.r = scale, z__1.i = 0.;
zgemv_("N", n, &i__1, &c_b2, &vr[vr_offset], ldvr, &work[
1], &c__1, &z__1, &vr_ref(1, ki), &c__1);
}
ii = izamax_(n, &vr_ref(1, ki), &c__1);
i__1 = vr_subscr(ii, ki);
remax = 1. / ((d__1 = vr[i__1].r, abs(d__1)) + (d__2 = d_imag(
&vr_ref(ii, ki)), abs(d__2)));
zdscal_(n, &remax, &vr_ref(1, ki), &c__1);
}
/* Set back the original diagonal elements of T. */
i__1 = ki - 1;
for (k = 1; k <= i__1; ++k) {
i__2 = t_subscr(k, k);
i__3 = k + *n;
t[i__2].r = work[i__3].r, t[i__2].i = work[i__3].i;
/* L70: */
}
--is;
L80:
;
}
}
if (leftv) {
/* Compute left eigenvectors. */
is = 1;
i__1 = *n;
for (ki = 1; ki <= i__1; ++ki) {
if (somev) {
if (! select[ki]) {
goto L130;
}
}
/* Computing MAX */
i__2 = t_subscr(ki, ki);
d__3 = ulp * ((d__1 = t[i__2].r, abs(d__1)) + (d__2 = d_imag(&
t_ref(ki, ki)), abs(d__2)));
smin = max(d__3,smlnum);
i__2 = *n;
work[i__2].r = 1., work[i__2].i = 0.;
/* Form right-hand side. */
i__2 = *n;
for (k = ki + 1; k <= i__2; ++k) {
i__3 = k;
d_cnjg(&z__2, &t_ref(ki, k));
z__1.r = -z__2.r, z__1.i = -z__2.i;
work[i__3].r = z__1.r, work[i__3].i = z__1.i;
/* L90: */
}
/* Solve the triangular system:
(T(KI+1:N,KI+1:N) - T(KI,KI))'*X = SCALE*WORK. */
i__2 = *n;
for (k = ki + 1; k <= i__2; ++k) {
i__3 = t_subscr(k, k);
i__4 = t_subscr(k, k);
i__5 = t_subscr(ki, ki);
z__1.r = t[i__4].r - t[i__5].r, z__1.i = t[i__4].i - t[i__5]
.i;
t[i__3].r = z__1.r, t[i__3].i = z__1.i;
i__3 = t_subscr(k, k);
if ((d__1 = t[i__3].r, abs(d__1)) + (d__2 = d_imag(&t_ref(k,
k)), abs(d__2)) < smin) {
i__4 = t_subscr(k, k);
t[i__4].r = smin, t[i__4].i = 0.;
}
/* L100: */
}
if (ki < *n) {
i__2 = *n - ki;
zlatrs_("Upper", "Conjugate transpose", "Non-unit", "Y", &
i__2, &t_ref(ki + 1, ki + 1), ldt, &work[ki + 1], &
scale, &rwork[1], info);
i__2 = ki;
work[i__2].r = scale, work[i__2].i = 0.;
}
/* Copy the vector x or Q*x to VL and normalize. */
if (! over) {
i__2 = *n - ki + 1;
zcopy_(&i__2, &work[ki], &c__1, &vl_ref(ki, is), &c__1);
i__2 = *n - ki + 1;
ii = izamax_(&i__2, &vl_ref(ki, is), &c__1) + ki - 1;
i__2 = vl_subscr(ii, is);
remax = 1. / ((d__1 = vl[i__2].r, abs(d__1)) + (d__2 = d_imag(
&vl_ref(ii, is)), abs(d__2)));
i__2 = *n - ki + 1;
zdscal_(&i__2, &remax, &vl_ref(ki, is), &c__1);
i__2 = ki - 1;
for (k = 1; k <= i__2; ++k) {
i__3 = vl_subscr(k, is);
vl[i__3].r = 0., vl[i__3].i = 0.;
/* L110: */
}
} else {
if (ki < *n) {
i__2 = *n - ki;
z__1.r = scale, z__1.i = 0.;
zgemv_("N", n, &i__2, &c_b2, &vl_ref(1, ki + 1), ldvl, &
work[ki + 1], &c__1, &z__1, &vl_ref(1, ki), &c__1);
}
ii = izamax_(n, &vl_ref(1, ki), &c__1);
i__2 = vl_subscr(ii, ki);
remax = 1. / ((d__1 = vl[i__2].r, abs(d__1)) + (d__2 = d_imag(
&vl_ref(ii, ki)), abs(d__2)));
zdscal_(n, &remax, &vl_ref(1, ki), &c__1);
}
/* Set back the original diagonal elements of T. */
i__2 = *n;
for (k = ki + 1; k <= i__2; ++k) {
i__3 = t_subscr(k, k);
i__4 = k + *n;
t[i__3].r = work[i__4].r, t[i__3].i = work[i__4].i;
/* L120: */
}
++is;
L130:
;
}
}
return 0;
/* End of ZTREVC */
} /* ztrevc_ */
int zggglm_(int *n, int *m, int *p,
doublecomplex *a, int *lda, doublecomplex *b, int *ldb,
doublecomplex *d__, doublecomplex *x, doublecomplex *y, doublecomplex
*work, int *lwork, int *info)
{
/* System generated locals */
int a_dim1, a_offset, b_dim1, b_offset, i__1, i__2, i__3, i__4;
doublecomplex z__1;
/* Local variables */
int i__, nb, np, nb1, nb2, nb3, nb4, lopt;
extern int zgemv_(char *, int *, int *,
doublecomplex *, doublecomplex *, int *, doublecomplex *,
int *, doublecomplex *, doublecomplex *, int *),
zcopy_(int *, doublecomplex *, int *, doublecomplex *,
int *), xerbla_(char *, int *);
extern int ilaenv_(int *, char *, char *, int *, int *,
int *, int *);
extern int zggqrf_(int *, int *, int *,
doublecomplex *, int *, doublecomplex *, doublecomplex *,
int *, doublecomplex *, doublecomplex *, int *, int *)
;
int lwkmin, lwkopt;
int lquery;
extern int zunmqr_(char *, char *, int *, int *,
int *, doublecomplex *, int *, doublecomplex *,
doublecomplex *, int *, doublecomplex *, int *, int *), zunmrq_(char *, char *, int *, int *,
int *, doublecomplex *, int *, doublecomplex *,
doublecomplex *, int *, doublecomplex *, int *, int *), ztrtrs_(char *, char *, char *, int *,
int *, doublecomplex *, int *, doublecomplex *, int *,
int *);
/* -- LAPACK driver routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* ZGGGLM 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) COMPLEX*16 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) COMPLEX*16 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) COMPLEX*16 array, dimension (N) */
/* On entry, D is the left hand side of the GLM equation. */
/* On exit, D is destroyed. */
/* X (output) COMPLEX*16 array, dimension (M) */
/* Y (output) COMPLEX*16 array, dimension (P) */
/* On exit, X and Y are the solutions of the GLM problem. */
/* WORK (workspace/output) COMPLEX*16 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 */
/* ZGEQRF, ZGERQF, ZUNMQR and ZUNMRQ. */
/* 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, "ZGEQRF", " ", n, m, &c_n1, &c_n1);
nb2 = ilaenv_(&c__1, "ZGERQF", " ", n, m, &c_n1, &c_n1);
nb3 = ilaenv_(&c__1, "ZUNMQR", " ", n, m, p, &c_n1);
nb4 = ilaenv_(&c__1, "ZUNMRQ", " ", 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].r = (double) lwkopt, work[1].i = 0.;
if (*lwork < lwkmin && ! lquery) {
*info = -12;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("ZGGGLM", &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 */
/* unitary. */
i__1 = *lwork - *m - np;
zggqrf_(n, m, p, &a[a_offset], lda, &work[1], &b[b_offset], ldb, &work[*m
+ 1], &work[*m + np + 1], &i__1, info);
i__1 = *m + np + 1;
lopt = (int) work[i__1].r;
/* Update left-hand-side vector d = Q'*d = ( d1 ) M */
/* ( d2 ) N-M */
i__1 = MAX(1,*n);
i__2 = *lwork - *m - np;
zunmqr_("Left", "Conjugate 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__3 = *m + np + 1;
i__1 = lopt, i__2 = (int) work[i__3].r;
lopt = MAX(i__1,i__2);
/* Solve T22*y2 = d2 for y2 */
if (*n > *m) {
i__1 = *n - *m;
i__2 = *n - *m;
ztrtrs_("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;
zcopy_(&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__) {
i__2 = i__;
y[i__2].r = 0., y[i__2].i = 0.;
/* L10: */
}
/* Update d1 = d1 - T12*y2 */
i__1 = *n - *m;
z__1.r = -1., z__1.i = -0.;
zgemv_("No transpose", m, &i__1, &z__1, &b[(*m + *p - *n + 1) * b_dim1 +
1], ldb, &y[*m + *p - *n + 1], &c__1, &c_b2, &d__[1], &c__1);
/* Solve triangular system: R11*x = d1 */
if (*m > 0) {
ztrtrs_("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 */
zcopy_(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;
zunmrq_("Left", "Conjugate 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__4 = *m + np + 1;
i__2 = lopt, i__3 = (int) work[i__4].r;
i__1 = *m + np + MAX(i__2,i__3);
work[1].r = (double) i__1, work[1].i = 0.;
return 0;
/* End of ZGGGLM */
} /* zggglm_ */
/* Subroutine */ int zdrvpt_(logical *dotype, integer *nn, integer *nval,
integer *nrhs, doublereal *thresh, logical *tsterr, doublecomplex *a,
doublereal *d__, doublecomplex *e, doublecomplex *b, doublecomplex *x,
doublecomplex *xact, doublecomplex *work, doublereal *rwork, integer
*nout)
{
/* Initialized data */
static integer iseedy[4] = { 0,0,0,1 };
/* Format strings */
static char fmt_9999[] = "(1x,a6,\002, N =\002,i5,\002, type \002,i2,"
"\002, test \002,i2,\002, ratio = \002,g12.5)";
static char fmt_9998[] = "(1x,a6,\002, FACT='\002,a1,\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, i__5;
doublereal d__1, d__2;
/* Builtin functions */
/* Subroutine */ int s_copy(char *, char *, ftnlen, ftnlen);
double z_abs(doublecomplex *);
integer s_wsfe(cilist *), do_fio(integer *, char *, ftnlen), e_wsfe(void);
/* Local variables */
integer i__, j, k, n;
doublereal z__[3];
integer k1, ia, in, kl, ku, ix, nt, lda;
char fact[1];
doublereal cond;
integer mode;
doublereal dmax__;
integer imat, info;
char path[3], dist[1], type__[1];
integer nrun, ifact;
extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *,
integer *);
integer nfail, iseed[4];
extern doublereal dget06_(doublereal *, doublereal *);
doublereal rcond;
integer nimat;
doublereal anorm;
extern /* Subroutine */ int zget04_(integer *, integer *, doublecomplex *,
integer *, doublecomplex *, integer *, doublereal *, doublereal *
), dcopy_(integer *, doublereal *, integer *, doublereal *,
integer *);
integer izero, nerrs;
extern /* Subroutine */ int zptt01_(integer *, doublereal *,
doublecomplex *, doublereal *, doublecomplex *, doublecomplex *,
doublereal *);
logical zerot;
extern /* Subroutine */ int zcopy_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *), zptt02_(char *, integer *, integer *,
doublereal *, doublecomplex *, doublecomplex *, integer *,
doublecomplex *, integer *, doublereal *), zptt05_(
integer *, integer *, doublereal *, doublecomplex *,
doublecomplex *, integer *, doublecomplex *, integer *,
doublecomplex *, integer *, doublereal *, doublereal *,
doublereal *), zptsv_(integer *, integer *, doublereal *,
doublecomplex *, doublecomplex *, integer *, integer *), zlatb4_(
char *, integer *, integer *, integer *, char *, integer *,
integer *, doublereal *, integer *, doublereal *, char *), aladhd_(integer *, char *), alaerh_(char
*, char *, integer *, integer *, char *, integer *, integer *,
integer *, integer *, integer *, integer *, integer *, integer *,
integer *);
extern integer idamax_(integer *, doublereal *, integer *);
doublereal rcondc;
extern /* Subroutine */ int zdscal_(integer *, doublereal *,
doublecomplex *, integer *), alasvm_(char *, integer *, integer *,
integer *, integer *), dlarnv_(integer *, integer *,
integer *, doublereal *);
doublereal ainvnm;
extern doublereal zlanht_(char *, integer *, doublereal *, doublecomplex *
);
extern /* Subroutine */ int zlacpy_(char *, integer *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *);
extern doublereal dzasum_(integer *, doublecomplex *, integer *);
extern /* Subroutine */ int zlaset_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, doublecomplex *, integer *), zlaptm_(char *, integer *, integer *, doublereal *,
doublereal *, doublecomplex *, doublecomplex *, integer *,
doublereal *, doublecomplex *, integer *), zlatms_(
integer *, integer *, char *, integer *, char *, doublereal *,
integer *, doublereal *, doublereal *, integer *, integer *, char
*, doublecomplex *, integer *, doublecomplex *, integer *), zlarnv_(integer *, integer *, integer *,
doublecomplex *);
doublereal result[6];
extern /* Subroutine */ int zpttrf_(integer *, doublereal *,
doublecomplex *, integer *), zerrvx_(char *, integer *),
zpttrs_(char *, integer *, integer *, doublereal *, doublecomplex
*, doublecomplex *, integer *, integer *), zptsvx_(char *,
integer *, integer *, doublereal *, doublecomplex *, doublereal *
, doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublereal *, doublereal *, doublereal *,
doublecomplex *, doublereal *, integer *);
/* Fortran I/O blocks */
static cilist io___35 = { 0, 0, 0, fmt_9999, 0 };
static cilist io___38 = { 0, 0, 0, fmt_9998, 0 };
/* -- LAPACK test routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* ZDRVPT tests ZPTSV and -SVX. */
/* 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. */
/* 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 dimension N. */
/* NRHS (input) INTEGER */
/* The number of right hand side vectors to be generated for */
/* each linear system. */
/* 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) COMPLEX*16 array, dimension (NMAX*2) */
/* D (workspace) DOUBLE PRECISION array, dimension (NMAX*2) */
/* E (workspace) COMPLEX*16 array, dimension (NMAX*2) */
/* B (workspace) COMPLEX*16 array, dimension (NMAX*NRHS) */
/* X (workspace) COMPLEX*16 array, dimension (NMAX*NRHS) */
/* XACT (workspace) COMPLEX*16 array, dimension (NMAX*NRHS) */
/* WORK (workspace) COMPLEX*16 array, dimension */
/* (NMAX*max(3,NRHS)) */
/* RWORK (workspace) DOUBLE PRECISION array, dimension (NMAX+2*NRHS) */
/* 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 */
--rwork;
--work;
--xact;
--x;
--b;
--e;
--d__;
--a;
--nval;
--dotype;
/* Function Body */
/* .. */
/* .. Executable Statements .. */
s_copy(path, "Zomplex precision", (ftnlen)1, (ftnlen)17);
s_copy(path + 1, "PT", (ftnlen)2, (ftnlen)2);
nrun = 0;
nfail = 0;
nerrs = 0;
for (i__ = 1; i__ <= 4; ++i__) {
iseed[i__ - 1] = iseedy[i__ - 1];
/* L10: */
}
/* Test the error exits */
if (*tsterr) {
zerrvx_(path, nout);
}
infoc_1.infot = 0;
i__1 = *nn;
for (in = 1; in <= i__1; ++in) {
/* Do for each value of N in NVAL. */
n = nval[in];
lda = max(1,n);
nimat = 12;
if (n <= 0) {
nimat = 1;
}
i__2 = nimat;
for (imat = 1; imat <= i__2; ++imat) {
/* Do the tests only if DOTYPE( IMAT ) is true. */
if (n > 0 && ! dotype[imat]) {
goto L110;
}
/* Set up parameters with ZLATB4. */
zlatb4_(path, &imat, &n, &n, type__, &kl, &ku, &anorm, &mode, &
cond, dist);
zerot = imat >= 8 && imat <= 10;
if (imat <= 6) {
/* Type 1-6: generate a symmetric tridiagonal matrix of */
/* known condition number in lower triangular band storage. */
s_copy(srnamc_1.srnamt, "ZLATMS", (ftnlen)6, (ftnlen)6);
zlatms_(&n, &n, dist, iseed, type__, &rwork[1], &mode, &cond,
&anorm, &kl, &ku, "B", &a[1], &c__2, &work[1], &info);
/* Check the error code from ZLATMS. */
if (info != 0) {
alaerh_(path, "ZLATMS", &info, &c__0, " ", &n, &n, &kl, &
ku, &c_n1, &imat, &nfail, &nerrs, nout);
goto L110;
}
izero = 0;
/* Copy the matrix to D and E. */
ia = 1;
i__3 = n - 1;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = i__;
i__5 = ia;
d__[i__4] = a[i__5].r;
i__4 = i__;
i__5 = ia + 1;
e[i__4].r = a[i__5].r, e[i__4].i = a[i__5].i;
ia += 2;
/* L20: */
}
if (n > 0) {
i__3 = n;
i__4 = ia;
d__[i__3] = a[i__4].r;
}
} else {
/* Type 7-12: generate a diagonally dominant matrix with */
/* unknown condition number in the vectors D and E. */
if (! zerot || ! dotype[7]) {
/* Let D and E have values from [-1,1]. */
dlarnv_(&c__2, iseed, &n, &d__[1]);
i__3 = n - 1;
zlarnv_(&c__2, iseed, &i__3, &e[1]);
/* Make the tridiagonal matrix diagonally dominant. */
if (n == 1) {
d__[1] = abs(d__[1]);
} else {
d__[1] = abs(d__[1]) + z_abs(&e[1]);
d__[n] = (d__1 = d__[n], abs(d__1)) + z_abs(&e[n - 1])
;
i__3 = n - 1;
for (i__ = 2; i__ <= i__3; ++i__) {
d__[i__] = (d__1 = d__[i__], abs(d__1)) + z_abs(&
e[i__]) + z_abs(&e[i__ - 1]);
/* L30: */
}
}
/* Scale D and E so the maximum element is ANORM. */
ix = idamax_(&n, &d__[1], &c__1);
dmax__ = d__[ix];
d__1 = anorm / dmax__;
dscal_(&n, &d__1, &d__[1], &c__1);
if (n > 1) {
i__3 = n - 1;
d__1 = anorm / dmax__;
zdscal_(&i__3, &d__1, &e[1], &c__1);
}
} else if (izero > 0) {
/* Reuse the last matrix by copying back the zeroed out */
/* elements. */
if (izero == 1) {
d__[1] = z__[1];
if (n > 1) {
e[1].r = z__[2], e[1].i = 0.;
}
} else if (izero == n) {
i__3 = n - 1;
e[i__3].r = z__[0], e[i__3].i = 0.;
d__[n] = z__[1];
} else {
i__3 = izero - 1;
e[i__3].r = z__[0], e[i__3].i = 0.;
d__[izero] = z__[1];
i__3 = izero;
e[i__3].r = z__[2], e[i__3].i = 0.;
}
}
/* For types 8-10, set one row and column of the matrix to */
/* zero. */
izero = 0;
if (imat == 8) {
izero = 1;
z__[1] = d__[1];
d__[1] = 0.;
if (n > 1) {
z__[2] = e[1].r;
e[1].r = 0., e[1].i = 0.;
}
} else if (imat == 9) {
izero = n;
if (n > 1) {
i__3 = n - 1;
z__[0] = e[i__3].r;
i__3 = n - 1;
e[i__3].r = 0., e[i__3].i = 0.;
}
z__[1] = d__[n];
d__[n] = 0.;
} else if (imat == 10) {
izero = (n + 1) / 2;
if (izero > 1) {
i__3 = izero - 1;
z__[0] = e[i__3].r;
i__3 = izero - 1;
e[i__3].r = 0., e[i__3].i = 0.;
i__3 = izero;
z__[2] = e[i__3].r;
i__3 = izero;
e[i__3].r = 0., e[i__3].i = 0.;
}
z__[1] = d__[izero];
d__[izero] = 0.;
}
}
/* Generate NRHS random solution vectors. */
ix = 1;
i__3 = *nrhs;
for (j = 1; j <= i__3; ++j) {
zlarnv_(&c__2, iseed, &n, &xact[ix]);
ix += lda;
/* L40: */
}
/* Set the right hand side. */
zlaptm_("Lower", &n, nrhs, &c_b24, &d__[1], &e[1], &xact[1], &lda,
&c_b25, &b[1], &lda);
for (ifact = 1; ifact <= 2; ++ifact) {
if (ifact == 1) {
*(unsigned char *)fact = 'F';
} else {
*(unsigned char *)fact = 'N';
}
/* Compute the condition number for comparison with */
/* the value returned by ZPTSVX. */
if (zerot) {
if (ifact == 1) {
goto L100;
}
rcondc = 0.;
} else if (ifact == 1) {
/* Compute the 1-norm of A. */
anorm = zlanht_("1", &n, &d__[1], &e[1]);
dcopy_(&n, &d__[1], &c__1, &d__[n + 1], &c__1);
if (n > 1) {
i__3 = n - 1;
zcopy_(&i__3, &e[1], &c__1, &e[n + 1], &c__1);
}
/* Factor the matrix A. */
zpttrf_(&n, &d__[n + 1], &e[n + 1], &info);
/* Use ZPTTRS to solve for one column at a time of */
/* inv(A), computing the maximum column sum as we go. */
ainvnm = 0.;
i__3 = n;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = n;
for (j = 1; j <= i__4; ++j) {
i__5 = j;
x[i__5].r = 0., x[i__5].i = 0.;
/* L50: */
}
i__4 = i__;
x[i__4].r = 1., x[i__4].i = 0.;
zpttrs_("Lower", &n, &c__1, &d__[n + 1], &e[n + 1], &
x[1], &lda, &info);
/* Computing MAX */
d__1 = ainvnm, d__2 = dzasum_(&n, &x[1], &c__1);
ainvnm = max(d__1,d__2);
/* L60: */
}
/* Compute the 1-norm condition number of A. */
if (anorm <= 0. || ainvnm <= 0.) {
rcondc = 1.;
} else {
rcondc = 1. / anorm / ainvnm;
}
}
if (ifact == 2) {
/* --- Test ZPTSV -- */
dcopy_(&n, &d__[1], &c__1, &d__[n + 1], &c__1);
if (n > 1) {
i__3 = n - 1;
zcopy_(&i__3, &e[1], &c__1, &e[n + 1], &c__1);
}
zlacpy_("Full", &n, nrhs, &b[1], &lda, &x[1], &lda);
/* Factor A as L*D*L' and solve the system A*X = B. */
s_copy(srnamc_1.srnamt, "ZPTSV ", (ftnlen)6, (ftnlen)6);
zptsv_(&n, nrhs, &d__[n + 1], &e[n + 1], &x[1], &lda, &
info);
/* Check error code from ZPTSV . */
if (info != izero) {
alaerh_(path, "ZPTSV ", &info, &izero, " ", &n, &n, &
c__1, &c__1, nrhs, &imat, &nfail, &nerrs,
nout);
}
nt = 0;
if (izero == 0) {
/* Check the factorization by computing the ratio */
/* norm(L*D*L' - A) / (n * norm(A) * EPS ) */
zptt01_(&n, &d__[1], &e[1], &d__[n + 1], &e[n + 1], &
work[1], result);
/* Compute the residual in the solution. */
zlacpy_("Full", &n, nrhs, &b[1], &lda, &work[1], &lda);
zptt02_("Lower", &n, nrhs, &d__[1], &e[1], &x[1], &
lda, &work[1], &lda, &result[1]);
/* Check solution from generated exact solution. */
zget04_(&n, nrhs, &x[1], &lda, &xact[1], &lda, &
rcondc, &result[2]);
nt = 3;
}
/* Print information about the tests that did not pass */
/* the threshold. */
i__3 = nt;
for (k = 1; k <= i__3; ++k) {
if (result[k - 1] >= *thresh) {
if (nfail == 0 && nerrs == 0) {
aladhd_(nout, path);
}
io___35.ciunit = *nout;
s_wsfe(&io___35);
do_fio(&c__1, "ZPTSV ", (ftnlen)6);
do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer))
;
do_fio(&c__1, (char *)&imat, (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;
}
/* L70: */
}
nrun += nt;
}
/* --- Test ZPTSVX --- */
if (ifact > 1) {
/* Initialize D( N+1:2*N ) and E( N+1:2*N ) to zero. */
i__3 = n - 1;
for (i__ = 1; i__ <= i__3; ++i__) {
d__[n + i__] = 0.;
i__4 = n + i__;
e[i__4].r = 0., e[i__4].i = 0.;
/* L80: */
}
if (n > 0) {
d__[n + n] = 0.;
}
}
zlaset_("Full", &n, nrhs, &c_b62, &c_b62, &x[1], &lda);
/* Solve the system and compute the condition number and */
/* error bounds using ZPTSVX. */
s_copy(srnamc_1.srnamt, "ZPTSVX", (ftnlen)6, (ftnlen)6);
zptsvx_(fact, &n, nrhs, &d__[1], &e[1], &d__[n + 1], &e[n + 1]
, &b[1], &lda, &x[1], &lda, &rcond, &rwork[1], &rwork[
*nrhs + 1], &work[1], &rwork[(*nrhs << 1) + 1], &info);
/* Check the error code from ZPTSVX. */
if (info != izero) {
alaerh_(path, "ZPTSVX", &info, &izero, fact, &n, &n, &
c__1, &c__1, nrhs, &imat, &nfail, &nerrs, nout);
}
if (izero == 0) {
if (ifact == 2) {
/* Check the factorization by computing the ratio */
/* norm(L*D*L' - A) / (n * norm(A) * EPS ) */
k1 = 1;
zptt01_(&n, &d__[1], &e[1], &d__[n + 1], &e[n + 1], &
work[1], result);
} else {
k1 = 2;
}
/* Compute the residual in the solution. */
zlacpy_("Full", &n, nrhs, &b[1], &lda, &work[1], &lda);
zptt02_("Lower", &n, nrhs, &d__[1], &e[1], &x[1], &lda, &
work[1], &lda, &result[1]);
/* Check solution from generated exact solution. */
zget04_(&n, nrhs, &x[1], &lda, &xact[1], &lda, &rcondc, &
result[2]);
/* Check error bounds from iterative refinement. */
zptt05_(&n, nrhs, &d__[1], &e[1], &b[1], &lda, &x[1], &
lda, &xact[1], &lda, &rwork[1], &rwork[*nrhs + 1],
&result[3]);
} else {
k1 = 6;
}
/* Check the reciprocal of the condition number. */
result[5] = dget06_(&rcond, &rcondc);
/* Print information about the tests that did not pass */
/* the threshold. */
for (k = k1; k <= 6; ++k) {
if (result[k - 1] >= *thresh) {
if (nfail == 0 && nerrs == 0) {
aladhd_(nout, path);
}
io___38.ciunit = *nout;
s_wsfe(&io___38);
do_fio(&c__1, "ZPTSVX", (ftnlen)6);
do_fio(&c__1, fact, (ftnlen)1);
do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&imat, (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;
}
/* L90: */
}
nrun = nrun + 7 - k1;
L100:
;
}
L110:
;
}
/* L120: */
}
/* Print a summary of the results. */
alasvm_(path, nout, &nfail, &nrun, &nerrs);
return 0;
/* End of ZDRVPT */
} /* zdrvpt_ */
/* Subroutine */ int znaitr_(integer *ido, char *bmat, integer *n, integer *k,
integer *np, integer *nb, doublecomplex *resid, doublereal *rnorm,
doublecomplex *v, integer *ldv, doublecomplex *h__, integer *ldh,
integer *ipntr, doublecomplex *workd, integer *info, ftnlen bmat_len)
{
/* Initialized data */
static logical first = TRUE_;
/* System generated locals */
integer h_dim1, h_offset, v_dim1, v_offset, i__1, i__2, i__3;
doublereal d__1, d__2, d__3, d__4;
doublecomplex z__1;
/* Builtin functions */
double d_imag(doublecomplex *), sqrt(doublereal);
/* Local variables */
static integer i__, j;
static real t0, t1, t2, t3, t4, t5;
static integer jj, ipj, irj, ivj;
static doublereal ulp, tst1;
static integer ierr, iter;
static doublereal unfl, ovfl;
static integer itry;
static doublereal temp1;
static logical orth1, orth2, step3, step4;
static doublereal betaj;
static integer infol;
static doublecomplex cnorm;
extern /* Double Complex */ VOID zdotc_(doublecomplex *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *);
static doublereal rtemp[2];
extern /* Subroutine */ int zgemv_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublecomplex *, doublecomplex *, integer *, ftnlen);
static doublereal wnorm;
extern /* Subroutine */ int dvout_(integer *, integer *, doublereal *,
integer *, char *, ftnlen), zcopy_(integer *, doublecomplex *,
integer *, doublecomplex *, integer *), ivout_(integer *, integer
*, integer *, integer *, char *, ftnlen), zaxpy_(integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *), zmout_(integer *, integer *, integer *, doublecomplex
*, integer *, integer *, char *, ftnlen), zvout_(integer *,
integer *, doublecomplex *, integer *, char *, ftnlen);
extern doublereal dlapy2_(doublereal *, doublereal *);
extern /* Subroutine */ int dlabad_(doublereal *, doublereal *);
extern doublereal dznrm2_(integer *, doublecomplex *, integer *);
static doublereal rnorm1;
extern /* Subroutine */ int zgetv0_(integer *, char *, integer *, logical
*, integer *, integer *, doublecomplex *, integer *,
doublecomplex *, doublereal *, integer *, doublecomplex *,
integer *, ftnlen);
extern doublereal dlamch_(char *, ftnlen);
extern /* Subroutine */ int second_(real *), zdscal_(integer *,
doublereal *, doublecomplex *, integer *);
static logical rstart;
static integer msglvl;
static doublereal smlnum;
extern doublereal zlanhs_(char *, integer *, doublecomplex *, integer *,
doublecomplex *, ftnlen);
extern /* Subroutine */ int zlascl_(char *, integer *, integer *,
doublereal *, doublereal *, integer *, integer *, doublecomplex *,
integer *, integer *, ftnlen);
/* %----------------------------------------------------% */
/* | Include files for debugging and timing information | */
/* %----------------------------------------------------% */
/* \SCCS Information: @(#) */
/* FILE: debug.h SID: 2.3 DATE OF SID: 11/16/95 RELEASE: 2 */
/* %---------------------------------% */
/* | See debug.doc for documentation | */
/* %---------------------------------% */
/* %------------------% */
/* | Scalar Arguments | */
/* %------------------% */
/* %--------------------------------% */
/* | See stat.doc for documentation | */
/* %--------------------------------% */
/* \SCCS Information: @(#) */
/* FILE: stat.h SID: 2.2 DATE OF SID: 11/16/95 RELEASE: 2 */
/* %-----------------% */
/* | Array Arguments | */
/* %-----------------% */
/* %------------% */
/* | Parameters | */
/* %------------% */
/* %--------------% */
/* | Local Arrays | */
/* %--------------% */
/* %---------------% */
/* | Local Scalars | */
/* %---------------% */
/* %----------------------% */
/* | External Subroutines | */
/* %----------------------% */
/* %--------------------% */
/* | External Functions | */
/* %--------------------% */
/* %---------------------% */
/* | Intrinsic Functions | */
/* %---------------------% */
/* %-----------------% */
/* | Data statements | */
/* %-----------------% */
/* Parameter adjustments */
--workd;
--resid;
v_dim1 = *ldv;
v_offset = 1 + v_dim1;
v -= v_offset;
h_dim1 = *ldh;
h_offset = 1 + h_dim1;
h__ -= h_offset;
--ipntr;
/* Function Body */
/* %-----------------------% */
/* | Executable Statements | */
/* %-----------------------% */
if (first) {
/* %-----------------------------------------% */
/* | Set machine-dependent constants for the | */
/* | the splitting and deflation criterion. | */
/* | If norm(H) <= sqrt(OVFL), | */
/* | overflow should not occur. | */
/* | REFERENCE: LAPACK subroutine zlahqr | */
/* %-----------------------------------------% */
unfl = dlamch_("safe minimum", (ftnlen)12);
z__1.r = 1. / unfl, z__1.i = 0. / unfl;
ovfl = z__1.r;
dlabad_(&unfl, &ovfl);
ulp = dlamch_("precision", (ftnlen)9);
smlnum = unfl * (*n / ulp);
first = FALSE_;
}
if (*ido == 0) {
/* %-------------------------------% */
/* | Initialize timing statistics | */
/* | & message level for debugging | */
/* %-------------------------------% */
second_(&t0);
msglvl = debug_1.mcaitr;
/* %------------------------------% */
/* | Initial call to this routine | */
/* %------------------------------% */
*info = 0;
step3 = FALSE_;
step4 = FALSE_;
rstart = FALSE_;
orth1 = FALSE_;
orth2 = FALSE_;
j = *k + 1;
ipj = 1;
irj = ipj + *n;
ivj = irj + *n;
}
/* %-------------------------------------------------% */
/* | When in reverse communication mode one of: | */
/* | STEP3, STEP4, ORTH1, ORTH2, RSTART | */
/* | will be .true. when .... | */
/* | STEP3: return from computing OP*v_{j}. | */
/* | STEP4: return from computing B-norm of OP*v_{j} | */
/* | ORTH1: return from computing B-norm of r_{j+1} | */
/* | ORTH2: return from computing B-norm of | */
/* | correction to the residual vector. | */
/* | RSTART: return from OP computations needed by | */
/* | zgetv0. | */
/* %-------------------------------------------------% */
if (step3) {
goto L50;
}
if (step4) {
goto L60;
}
if (orth1) {
goto L70;
}
if (orth2) {
goto L90;
}
if (rstart) {
goto L30;
}
/* %-----------------------------% */
/* | Else this is the first step | */
/* %-----------------------------% */
/* %--------------------------------------------------------------% */
/* | | */
/* | A R N O L D I I T E R A T I O N L O O P | */
/* | | */
/* | Note: B*r_{j-1} is already in WORKD(1:N)=WORKD(IPJ:IPJ+N-1) | */
/* %--------------------------------------------------------------% */
L1000:
if (msglvl > 1) {
ivout_(&debug_1.logfil, &c__1, &j, &debug_1.ndigit, "_naitr: generat"
"ing Arnoldi vector number", (ftnlen)40);
dvout_(&debug_1.logfil, &c__1, rnorm, &debug_1.ndigit, "_naitr: B-no"
"rm of the current residual is", (ftnlen)41);
}
/* %---------------------------------------------------% */
/* | STEP 1: Check if the B norm of j-th residual | */
/* | vector is zero. Equivalent to determine whether | */
/* | an exact j-step Arnoldi factorization is present. | */
/* %---------------------------------------------------% */
betaj = *rnorm;
if (*rnorm > 0.) {
goto L40;
}
/* %---------------------------------------------------% */
/* | Invariant subspace found, generate a new starting | */
/* | vector which is orthogonal to the current Arnoldi | */
/* | basis and continue the iteration. | */
/* %---------------------------------------------------% */
if (msglvl > 0) {
ivout_(&debug_1.logfil, &c__1, &j, &debug_1.ndigit, "_naitr: ****** "
"RESTART AT STEP ******", (ftnlen)37);
}
/* %---------------------------------------------% */
/* | ITRY is the loop variable that controls the | */
/* | maximum amount of times that a restart is | */
/* | attempted. NRSTRT is used by stat.h | */
/* %---------------------------------------------% */
betaj = 0.;
++timing_1.nrstrt;
itry = 1;
L20:
rstart = TRUE_;
*ido = 0;
L30:
/* %--------------------------------------% */
/* | If in reverse communication mode and | */
/* | RSTART = .true. flow returns here. | */
/* %--------------------------------------% */
zgetv0_(ido, bmat, &itry, &c_false, n, &j, &v[v_offset], ldv, &resid[1],
rnorm, &ipntr[1], &workd[1], &ierr, (ftnlen)1);
if (*ido != 99) {
goto L9000;
}
if (ierr < 0) {
++itry;
if (itry <= 3) {
goto L20;
}
/* %------------------------------------------------% */
/* | Give up after several restart attempts. | */
/* | Set INFO to the size of the invariant subspace | */
/* | which spans OP and exit. | */
/* %------------------------------------------------% */
*info = j - 1;
second_(&t1);
timing_1.tcaitr += t1 - t0;
*ido = 99;
goto L9000;
}
L40:
/* %---------------------------------------------------------% */
/* | STEP 2: v_{j} = r_{j-1}/rnorm and p_{j} = p_{j}/rnorm | */
/* | Note that p_{j} = B*r_{j-1}. In order to avoid overflow | */
/* | when reciprocating a small RNORM, test against lower | */
/* | machine bound. | */
/* %---------------------------------------------------------% */
zcopy_(n, &resid[1], &c__1, &v[j * v_dim1 + 1], &c__1);
if (*rnorm >= unfl) {
temp1 = 1. / *rnorm;
zdscal_(n, &temp1, &v[j * v_dim1 + 1], &c__1);
zdscal_(n, &temp1, &workd[ipj], &c__1);
} else {
/* %-----------------------------------------% */
/* | To scale both v_{j} and p_{j} carefully | */
/* | use LAPACK routine zlascl | */
/* %-----------------------------------------% */
zlascl_("General", &i__, &i__, rnorm, &c_b27, n, &c__1, &v[j * v_dim1
+ 1], n, &infol, (ftnlen)7);
zlascl_("General", &i__, &i__, rnorm, &c_b27, n, &c__1, &workd[ipj],
n, &infol, (ftnlen)7);
}
/* %------------------------------------------------------% */
/* | STEP 3: r_{j} = OP*v_{j}; Note that p_{j} = B*v_{j} | */
/* | Note that this is not quite yet r_{j}. See STEP 4 | */
/* %------------------------------------------------------% */
step3 = TRUE_;
++timing_1.nopx;
second_(&t2);
zcopy_(n, &v[j * v_dim1 + 1], &c__1, &workd[ivj], &c__1);
ipntr[1] = ivj;
ipntr[2] = irj;
ipntr[3] = ipj;
*ido = 1;
/* %-----------------------------------% */
/* | Exit in order to compute OP*v_{j} | */
/* %-----------------------------------% */
goto L9000;
L50:
/* %----------------------------------% */
/* | Back from reverse communication; | */
/* | WORKD(IRJ:IRJ+N-1) := OP*v_{j} | */
/* | if step3 = .true. | */
/* %----------------------------------% */
second_(&t3);
timing_1.tmvopx += t3 - t2;
step3 = FALSE_;
/* %------------------------------------------% */
/* | Put another copy of OP*v_{j} into RESID. | */
/* %------------------------------------------% */
zcopy_(n, &workd[irj], &c__1, &resid[1], &c__1);
/* %---------------------------------------% */
/* | STEP 4: Finish extending the Arnoldi | */
/* | factorization to length j. | */
/* %---------------------------------------% */
second_(&t2);
if (*(unsigned char *)bmat == 'G') {
++timing_1.nbx;
step4 = TRUE_;
ipntr[1] = irj;
ipntr[2] = ipj;
*ido = 2;
/* %-------------------------------------% */
/* | Exit in order to compute B*OP*v_{j} | */
/* %-------------------------------------% */
goto L9000;
} else if (*(unsigned char *)bmat == 'I') {
zcopy_(n, &resid[1], &c__1, &workd[ipj], &c__1);
}
L60:
/* %----------------------------------% */
/* | Back from reverse communication; | */
/* | WORKD(IPJ:IPJ+N-1) := B*OP*v_{j} | */
/* | if step4 = .true. | */
/* %----------------------------------% */
if (*(unsigned char *)bmat == 'G') {
second_(&t3);
timing_1.tmvbx += t3 - t2;
}
step4 = FALSE_;
/* %-------------------------------------% */
/* | The following is needed for STEP 5. | */
/* | Compute the B-norm of OP*v_{j}. | */
/* %-------------------------------------% */
if (*(unsigned char *)bmat == 'G') {
zdotc_(&z__1, n, &resid[1], &c__1, &workd[ipj], &c__1);
cnorm.r = z__1.r, cnorm.i = z__1.i;
d__1 = cnorm.r;
d__2 = d_imag(&cnorm);
wnorm = sqrt(dlapy2_(&d__1, &d__2));
} else if (*(unsigned char *)bmat == 'I') {
wnorm = dznrm2_(n, &resid[1], &c__1);
}
/* %-----------------------------------------% */
/* | Compute the j-th residual corresponding | */
/* | to the j step factorization. | */
/* | Use Classical Gram Schmidt and compute: | */
/* | w_{j} <- V_{j}^T * B * OP * v_{j} | */
/* | r_{j} <- OP*v_{j} - V_{j} * w_{j} | */
/* %-----------------------------------------% */
/* %------------------------------------------% */
/* | Compute the j Fourier coefficients w_{j} | */
/* | WORKD(IPJ:IPJ+N-1) contains B*OP*v_{j}. | */
/* %------------------------------------------% */
zgemv_("C", n, &j, &c_b1, &v[v_offset], ldv, &workd[ipj], &c__1, &c_b2, &
h__[j * h_dim1 + 1], &c__1, (ftnlen)1);
/* %--------------------------------------% */
/* | Orthogonalize r_{j} against V_{j}. | */
/* | RESID contains OP*v_{j}. See STEP 3. | */
/* %--------------------------------------% */
z__1.r = -1., z__1.i = -0.;
zgemv_("N", n, &j, &z__1, &v[v_offset], ldv, &h__[j * h_dim1 + 1], &c__1,
&c_b1, &resid[1], &c__1, (ftnlen)1);
if (j > 1) {
i__1 = j + (j - 1) * h_dim1;
z__1.r = betaj, z__1.i = 0.;
h__[i__1].r = z__1.r, h__[i__1].i = z__1.i;
}
second_(&t4);
orth1 = TRUE_;
second_(&t2);
if (*(unsigned char *)bmat == 'G') {
++timing_1.nbx;
zcopy_(n, &resid[1], &c__1, &workd[irj], &c__1);
ipntr[1] = irj;
ipntr[2] = ipj;
*ido = 2;
/* %----------------------------------% */
/* | Exit in order to compute B*r_{j} | */
/* %----------------------------------% */
goto L9000;
} else if (*(unsigned char *)bmat == 'I') {
zcopy_(n, &resid[1], &c__1, &workd[ipj], &c__1);
}
L70:
/* %---------------------------------------------------% */
/* | Back from reverse communication if ORTH1 = .true. | */
/* | WORKD(IPJ:IPJ+N-1) := B*r_{j}. | */
/* %---------------------------------------------------% */
if (*(unsigned char *)bmat == 'G') {
second_(&t3);
timing_1.tmvbx += t3 - t2;
}
orth1 = FALSE_;
/* %------------------------------% */
/* | Compute the B-norm of r_{j}. | */
/* %------------------------------% */
if (*(unsigned char *)bmat == 'G') {
zdotc_(&z__1, n, &resid[1], &c__1, &workd[ipj], &c__1);
cnorm.r = z__1.r, cnorm.i = z__1.i;
d__1 = cnorm.r;
d__2 = d_imag(&cnorm);
*rnorm = sqrt(dlapy2_(&d__1, &d__2));
} else if (*(unsigned char *)bmat == 'I') {
*rnorm = dznrm2_(n, &resid[1], &c__1);
}
/* %-----------------------------------------------------------% */
/* | STEP 5: Re-orthogonalization / Iterative refinement phase | */
/* | Maximum NITER_ITREF tries. | */
/* | | */
/* | s = V_{j}^T * B * r_{j} | */
/* | r_{j} = r_{j} - V_{j}*s | */
/* | alphaj = alphaj + s_{j} | */
/* | | */
/* | The stopping criteria used for iterative refinement is | */
/* | discussed in Parlett's book SEP, page 107 and in Gragg & | */
/* | Reichel ACM TOMS paper; Algorithm 686, Dec. 1990. | */
/* | Determine if we need to correct the residual. The goal is | */
/* | to enforce ||v(:,1:j)^T * r_{j}|| .le. eps * || r_{j} || | */
/* | The following test determines whether the sine of the | */
/* | angle between OP*x and the computed residual is less | */
/* | than or equal to 0.717. | */
/* %-----------------------------------------------------------% */
if (*rnorm > wnorm * .717f) {
goto L100;
}
iter = 0;
++timing_1.nrorth;
/* %---------------------------------------------------% */
/* | Enter the Iterative refinement phase. If further | */
/* | refinement is necessary, loop back here. The loop | */
/* | variable is ITER. Perform a step of Classical | */
/* | Gram-Schmidt using all the Arnoldi vectors V_{j} | */
/* %---------------------------------------------------% */
L80:
if (msglvl > 2) {
rtemp[0] = wnorm;
rtemp[1] = *rnorm;
dvout_(&debug_1.logfil, &c__2, rtemp, &debug_1.ndigit, "_naitr: re-o"
"rthogonalization; wnorm and rnorm are", (ftnlen)49);
zvout_(&debug_1.logfil, &j, &h__[j * h_dim1 + 1], &debug_1.ndigit,
"_naitr: j-th column of H", (ftnlen)24);
}
/* %----------------------------------------------------% */
/* | Compute V_{j}^T * B * r_{j}. | */
/* | WORKD(IRJ:IRJ+J-1) = v(:,1:J)'*WORKD(IPJ:IPJ+N-1). | */
/* %----------------------------------------------------% */
zgemv_("C", n, &j, &c_b1, &v[v_offset], ldv, &workd[ipj], &c__1, &c_b2, &
workd[irj], &c__1, (ftnlen)1);
/* %---------------------------------------------% */
/* | Compute the correction to the residual: | */
/* | r_{j} = r_{j} - V_{j} * WORKD(IRJ:IRJ+J-1). | */
/* | The correction to H is v(:,1:J)*H(1:J,1:J) | */
/* | + v(:,1:J)*WORKD(IRJ:IRJ+J-1)*e'_j. | */
/* %---------------------------------------------% */
z__1.r = -1., z__1.i = -0.;
zgemv_("N", n, &j, &z__1, &v[v_offset], ldv, &workd[irj], &c__1, &c_b1, &
resid[1], &c__1, (ftnlen)1);
zaxpy_(&j, &c_b1, &workd[irj], &c__1, &h__[j * h_dim1 + 1], &c__1);
orth2 = TRUE_;
second_(&t2);
if (*(unsigned char *)bmat == 'G') {
++timing_1.nbx;
zcopy_(n, &resid[1], &c__1, &workd[irj], &c__1);
ipntr[1] = irj;
ipntr[2] = ipj;
*ido = 2;
/* %-----------------------------------% */
/* | Exit in order to compute B*r_{j}. | */
/* | r_{j} is the corrected residual. | */
/* %-----------------------------------% */
goto L9000;
} else if (*(unsigned char *)bmat == 'I') {
zcopy_(n, &resid[1], &c__1, &workd[ipj], &c__1);
}
L90:
/* %---------------------------------------------------% */
/* | Back from reverse communication if ORTH2 = .true. | */
/* %---------------------------------------------------% */
if (*(unsigned char *)bmat == 'G') {
second_(&t3);
timing_1.tmvbx += t3 - t2;
}
/* %-----------------------------------------------------% */
/* | Compute the B-norm of the corrected residual r_{j}. | */
/* %-----------------------------------------------------% */
if (*(unsigned char *)bmat == 'G') {
zdotc_(&z__1, n, &resid[1], &c__1, &workd[ipj], &c__1);
cnorm.r = z__1.r, cnorm.i = z__1.i;
d__1 = cnorm.r;
d__2 = d_imag(&cnorm);
rnorm1 = sqrt(dlapy2_(&d__1, &d__2));
} else if (*(unsigned char *)bmat == 'I') {
rnorm1 = dznrm2_(n, &resid[1], &c__1);
}
if (msglvl > 0 && iter > 0) {
ivout_(&debug_1.logfil, &c__1, &j, &debug_1.ndigit, "_naitr: Iterati"
"ve refinement for Arnoldi residual", (ftnlen)49);
if (msglvl > 2) {
rtemp[0] = *rnorm;
rtemp[1] = rnorm1;
dvout_(&debug_1.logfil, &c__2, rtemp, &debug_1.ndigit, "_naitr: "
"iterative refinement ; rnorm and rnorm1 are", (ftnlen)51);
}
}
/* %-----------------------------------------% */
/* | Determine if we need to perform another | */
/* | step of re-orthogonalization. | */
/* %-----------------------------------------% */
if (rnorm1 > *rnorm * .717f) {
/* %---------------------------------------% */
/* | No need for further refinement. | */
/* | The cosine of the angle between the | */
/* | corrected residual vector and the old | */
/* | residual vector is greater than 0.717 | */
/* | In other words the corrected residual | */
/* | and the old residual vector share an | */
/* | angle of less than arcCOS(0.717) | */
/* %---------------------------------------% */
*rnorm = rnorm1;
} else {
/* %-------------------------------------------% */
/* | Another step of iterative refinement step | */
/* | is required. NITREF is used by stat.h | */
/* %-------------------------------------------% */
++timing_1.nitref;
*rnorm = rnorm1;
++iter;
if (iter <= 1) {
goto L80;
}
/* %-------------------------------------------------% */
/* | Otherwise RESID is numerically in the span of V | */
/* %-------------------------------------------------% */
i__1 = *n;
for (jj = 1; jj <= i__1; ++jj) {
i__2 = jj;
resid[i__2].r = 0., resid[i__2].i = 0.;
/* L95: */
}
*rnorm = 0.;
}
/* %----------------------------------------------% */
/* | Branch here directly if iterative refinement | */
/* | wasn't necessary or after at most NITER_REF | */
/* | steps of iterative refinement. | */
/* %----------------------------------------------% */
L100:
rstart = FALSE_;
orth2 = FALSE_;
second_(&t5);
timing_1.titref += t5 - t4;
/* %------------------------------------% */
/* | STEP 6: Update j = j+1; Continue | */
/* %------------------------------------% */
++j;
if (j > *k + *np) {
second_(&t1);
timing_1.tcaitr += t1 - t0;
*ido = 99;
i__1 = *k + *np - 1;
for (i__ = max(1,*k); i__ <= i__1; ++i__) {
/* %--------------------------------------------% */
/* | Check for splitting and deflation. | */
/* | Use a standard test as in the QR algorithm | */
/* | REFERENCE: LAPACK subroutine zlahqr | */
/* %--------------------------------------------% */
i__2 = i__ + i__ * h_dim1;
d__1 = h__[i__2].r;
d__2 = d_imag(&h__[i__ + i__ * h_dim1]);
i__3 = i__ + 1 + (i__ + 1) * h_dim1;
d__3 = h__[i__3].r;
d__4 = d_imag(&h__[i__ + 1 + (i__ + 1) * h_dim1]);
tst1 = dlapy2_(&d__1, &d__2) + dlapy2_(&d__3, &d__4);
if (tst1 == 0.) {
i__2 = *k + *np;
tst1 = zlanhs_("1", &i__2, &h__[h_offset], ldh, &workd[*n + 1]
, (ftnlen)1);
}
i__2 = i__ + 1 + i__ * h_dim1;
d__1 = h__[i__2].r;
d__2 = d_imag(&h__[i__ + 1 + i__ * h_dim1]);
/* Computing MAX */
d__3 = ulp * tst1;
if (dlapy2_(&d__1, &d__2) <= max(d__3,smlnum)) {
i__3 = i__ + 1 + i__ * h_dim1;
h__[i__3].r = 0., h__[i__3].i = 0.;
}
/* L110: */
}
if (msglvl > 2) {
i__1 = *k + *np;
i__2 = *k + *np;
zmout_(&debug_1.logfil, &i__1, &i__2, &h__[h_offset], ldh, &
debug_1.ndigit, "_naitr: Final upper Hessenberg matrix H"
" of order K+NP", (ftnlen)53);
}
goto L9000;
}
/* %--------------------------------------------------------% */
/* | Loop back to extend the factorization by another step. | */
/* %--------------------------------------------------------% */
goto L1000;
/* %---------------------------------------------------------------% */
/* | | */
/* | E N D O F M A I N I T E R A T I O N L O O P | */
/* | | */
/* %---------------------------------------------------------------% */
L9000:
return 0;
/* %---------------% */
/* | End of znaitr | */
/* %---------------% */
} /* znaitr_ */
/* Subroutine */ int zlatzm_(char *side, integer *m, integer *n,
doublecomplex *v, integer *incv, doublecomplex *tau, doublecomplex *
c1, doublecomplex *c2, integer *ldc, doublecomplex *work)
{
/* -- LAPACK routine (version 2.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
September 30, 1994
Purpose
=======
ZLATZM applies a Householder matrix generated by ZTZRQF to a matrix.
Let P = I - tau*u*u', u = ( 1 ),
( v )
where v is an (m-1) vector if SIDE = 'L', or a (n-1) vector if
SIDE = 'R'.
If SIDE equals 'L', let
C = [ C1 ] 1
[ C2 ] m-1
n
Then C is overwritten by P*C.
If SIDE equals 'R', let
C = [ C1, C2 ] m
1 n-1
Then C is overwritten by C*P.
Arguments
=========
SIDE (input) CHARACTER*1
= 'L': form P * C
= 'R': form C * P
M (input) INTEGER
The number of rows of the matrix C.
N (input) INTEGER
The number of columns of the matrix C.
V (input) COMPLEX*16 array, dimension
(1 + (M-1)*abs(INCV)) if SIDE = 'L'
(1 + (N-1)*abs(INCV)) if SIDE = 'R'
The vector v in the representation of P. V is not used
if TAU = 0.
INCV (input) INTEGER
The increment between elements of v. INCV <> 0
TAU (input) COMPLEX*16
The value tau in the representation of P.
C1 (input/output) COMPLEX*16 array, dimension
(LDC,N) if SIDE = 'L'
(M,1) if SIDE = 'R'
On entry, the n-vector C1 if SIDE = 'L', or the m-vector C1
if SIDE = 'R'.
On exit, the first row of P*C if SIDE = 'L', or the first
column of C*P if SIDE = 'R'.
C2 (input/output) COMPLEX*16 array, dimension
(LDC, N) if SIDE = 'L'
(LDC, N-1) if SIDE = 'R'
On entry, the (m - 1) x n matrix C2 if SIDE = 'L', or the
m x (n - 1) matrix C2 if SIDE = 'R'.
On exit, rows 2:m of P*C if SIDE = 'L', or columns 2:m of C*P
if SIDE = 'R'.
LDC (input) INTEGER
The leading dimension of the arrays C1 and C2.
LDC >= max(1,M).
WORK (workspace) COMPLEX*16 array, dimension
(N) if SIDE = 'L'
(M) if SIDE = 'R'
=====================================================================
Parameter adjustments
Function Body */
/* Table of constant values */
static doublecomplex c_b1 = {1.,0.};
static integer c__1 = 1;
/* System generated locals */
integer c1_dim1, c1_offset, c2_dim1, c2_offset, i__1;
doublecomplex z__1;
/* Local variables */
extern logical lsame_(char *, char *);
extern /* Subroutine */ int zgerc_(integer *, integer *, doublecomplex *,
doublecomplex *, integer *, doublecomplex *, integer *,
doublecomplex *, integer *), zgemv_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublecomplex *, doublecomplex *, integer *),
zgeru_(integer *, integer *, doublecomplex *, doublecomplex *,
integer *, doublecomplex *, integer *, doublecomplex *, integer *)
, zcopy_(integer *, doublecomplex *, integer *, doublecomplex *,
integer *), zaxpy_(integer *, doublecomplex *, doublecomplex *,
integer *, doublecomplex *, integer *), zlacgv_(integer *,
doublecomplex *, integer *);
#define V(I) v[(I)-1]
#define WORK(I) work[(I)-1]
#define C2(I,J) c2[(I)-1 + ((J)-1)* ( *ldc)]
#define C1(I,J) c1[(I)-1 + ((J)-1)* ( *ldc)]
if (min(*m,*n) == 0 || tau->r == 0. && tau->i == 0.) {
return 0;
}
if (lsame_(side, "L")) {
/* w := conjg( C1 + v' * C2 ) */
zcopy_(n, &C1(1,1), ldc, &WORK(1), &c__1);
zlacgv_(n, &WORK(1), &c__1);
i__1 = *m - 1;
zgemv_("Conjugate transpose", &i__1, n, &c_b1, &C2(1,1), ldc, &
V(1), incv, &c_b1, &WORK(1), &c__1);
/* [ C1 ] := [ C1 ] - tau* [ 1 ] * w'
[ C2 ] [ C2 ] [ v ] */
zlacgv_(n, &WORK(1), &c__1);
z__1.r = -tau->r, z__1.i = -tau->i;
zaxpy_(n, &z__1, &WORK(1), &c__1, &C1(1,1), ldc);
i__1 = *m - 1;
z__1.r = -tau->r, z__1.i = -tau->i;
zgeru_(&i__1, n, &z__1, &V(1), incv, &WORK(1), &c__1, &C2(1,1),
ldc);
} else if (lsame_(side, "R")) {
/* w := C1 + C2 * v */
zcopy_(m, &C1(1,1), &c__1, &WORK(1), &c__1);
i__1 = *n - 1;
zgemv_("No transpose", m, &i__1, &c_b1, &C2(1,1), ldc, &V(1),
incv, &c_b1, &WORK(1), &c__1);
/* [ C1, C2 ] := [ C1, C2 ] - tau* w * [ 1 , v'] */
z__1.r = -tau->r, z__1.i = -tau->i;
zaxpy_(m, &z__1, &WORK(1), &c__1, &C1(1,1), &c__1);
i__1 = *n - 1;
z__1.r = -tau->r, z__1.i = -tau->i;
zgerc_(m, &i__1, &z__1, &WORK(1), &c__1, &V(1), incv, &C2(1,1),
ldc);
}
return 0;
/* End of ZLATZM */
} /* zlatzm_ */
/* Subroutine */ int zdrvpb_(logical *dotype, integer *nn, integer *nval,
integer *nrhs, doublereal *thresh, logical *tsterr, integer *nmax,
doublecomplex *a, doublecomplex *afac, doublecomplex *asav,
doublecomplex *b, doublecomplex *bsav, doublecomplex *x,
doublecomplex *xact, doublereal *s, doublecomplex *work, doublereal *
rwork, integer *nout)
{
/* Initialized data */
static integer iseedy[4] = { 1988,1989,1990,1991 };
static char facts[1*3] = "F" "N" "E";
static char equeds[1*2] = "N" "Y";
/* Format strings */
static char fmt_9999[] = "(1x,a6,\002, UPLO='\002,a1,\002', N =\002,i5"
",\002, KD =\002,i5,\002, type \002,i1,\002, test(\002,i1,\002)"
"=\002,g12.5)";
static char fmt_9997[] = "(1x,a6,\002( '\002,a1,\002', '\002,a1,\002',"
" \002,i5,\002, \002,i5,\002, ... ), EQUED='\002,a1,\002', type"
" \002,i1,\002, test(\002,i1,\002)=\002,g12.5)";
static char fmt_9998[] = "(1x,a6,\002( '\002,a1,\002', '\002,a1,\002',"
" \002,i5,\002, \002,i5,\002, ... ), type \002,i1,\002, test(\002"
",i1,\002)=\002,g12.5)";
/* System generated locals */
address a__1[2];
integer i__1, i__2, i__3, i__4, i__5, i__6, i__7[2];
char ch__1[2];
/* Builtin functions
Subroutine */ int s_copy(char *, char *, ftnlen, ftnlen);
integer s_wsfe(cilist *), do_fio(integer *, char *, ftnlen), e_wsfe(void);
/* Subroutine */ int s_cat(char *, char **, integer *, integer *, ftnlen);
/* Local variables */
static integer ldab;
static char fact[1];
static integer ioff, mode, koff;
static doublereal amax;
static char path[3];
static integer imat, info;
static char dist[1], uplo[1], type__[1];
static integer nrun, i__, k, n, ifact, nfail, iseed[4], nfact;
extern doublereal dget06_(doublereal *, doublereal *);
static integer kdval[4];
extern logical lsame_(char *, char *);
static char equed[1];
static integer nbmin;
static doublereal rcond, roldc, scond;
static integer nimat;
static doublereal anorm;
extern /* Subroutine */ int zget04_(integer *, integer *, doublecomplex *,
integer *, doublecomplex *, integer *, doublereal *, doublereal *
);
static logical equil;
extern /* Subroutine */ int zpbt01_(char *, integer *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *,
doublereal *, doublereal *), zpbt02_(char *, integer *,
integer *, integer *, doublecomplex *, integer *, doublecomplex *,
integer *, doublecomplex *, integer *, doublereal *, doublereal *
), zpbt05_(char *, integer *, integer *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *,
doublereal *, doublereal *, doublereal *);
static integer iuplo, izero, i1, i2, k1, nerrs;
static logical zerot;
extern /* Subroutine */ int zcopy_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *), zpbsv_(char *, integer *, integer *,
integer *, doublecomplex *, integer *, doublecomplex *, integer *,
integer *), zswap_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *);
static char xtype[1];
extern /* Subroutine */ int zlatb4_(char *, integer *, integer *, integer
*, char *, integer *, integer *, doublereal *, integer *,
doublereal *, char *), aladhd_(integer *,
char *);
static integer kd, nb, in, kl;
extern /* Subroutine */ int alaerh_(char *, char *, integer *, integer *,
char *, integer *, integer *, integer *, integer *, integer *,
integer *, integer *, integer *, integer *);
static logical prefac;
static integer iw, ku, nt;
static doublereal rcondc;
static logical nofact;
static char packit[1];
static integer iequed;
extern doublereal zlanhb_(char *, char *, integer *, integer *,
doublecomplex *, integer *, doublereal *),
zlange_(char *, integer *, integer *, doublecomplex *, integer *,
doublereal *);
extern /* Subroutine */ int zlaqhb_(char *, integer *, integer *,
doublecomplex *, integer *, doublereal *, doublereal *,
doublereal *, char *), alasvm_(char *, integer *,
integer *, integer *, integer *);
static doublereal cndnum;
extern /* Subroutine */ int zlaipd_(integer *, doublecomplex *, integer *,
integer *);
static doublereal ainvnm;
extern /* Subroutine */ int xlaenv_(integer *, integer *), zlacpy_(char *,
integer *, integer *, doublecomplex *, integer *, doublecomplex *
, integer *), zlarhs_(char *, char *, char *, char *,
integer *, integer *, integer *, integer *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *,
doublecomplex *, integer *, integer *, integer *), zlaset_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, doublecomplex *, integer *), zpbequ_(char *, integer *, integer *, doublecomplex *,
integer *, doublereal *, doublereal *, doublereal *, integer *), zpbtrf_(char *, integer *, integer *, doublecomplex *,
integer *, integer *), zlatms_(integer *, integer *, char
*, integer *, char *, doublereal *, integer *, doublereal *,
doublereal *, integer *, integer *, char *, doublecomplex *,
integer *, doublecomplex *, integer *);
static doublereal result[6];
extern /* Subroutine */ int zpbtrs_(char *, integer *, integer *, integer
*, doublecomplex *, integer *, doublecomplex *, integer *,
integer *), zpbsvx_(char *, char *, integer *, integer *,
integer *, doublecomplex *, integer *, doublecomplex *, integer *,
char *, doublereal *, doublecomplex *, integer *, doublecomplex *
, integer *, doublereal *, doublereal *, doublereal *,
doublecomplex *, doublereal *, integer *),
zerrvx_(char *, integer *);
static integer lda, ikd, nkd;
/* Fortran I/O blocks */
static cilist io___57 = { 0, 0, 0, fmt_9999, 0 };
static cilist io___60 = { 0, 0, 0, fmt_9997, 0 };
static cilist io___61 = { 0, 0, 0, fmt_9998, 0 };
/* -- LAPACK test routine (version 3.0) --
Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd.,
Courant Institute, Argonne National Lab, and Rice University
June 30, 1999
Purpose
=======
ZDRVPB tests the driver routines ZPBSV and -SVX.
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.
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 dimension N.
NRHS (input) INTEGER
The number of right hand side vectors to be generated for
each linear system.
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.
NMAX (input) INTEGER
The maximum value permitted for N, used in dimensioning the
work arrays.
A (workspace) COMPLEX*16 array, dimension (NMAX*NMAX)
AFAC (workspace) COMPLEX*16 array, dimension (NMAX*NMAX)
ASAV (workspace) COMPLEX*16 array, dimension (NMAX*NMAX)
B (workspace) COMPLEX*16 array, dimension (NMAX*NRHS)
BSAV (workspace) COMPLEX*16 array, dimension (NMAX*NRHS)
X (workspace) COMPLEX*16 array, dimension (NMAX*NRHS)
XACT (workspace) COMPLEX*16 array, dimension (NMAX*NRHS)
S (workspace) DOUBLE PRECISION array, dimension (NMAX)
WORK (workspace) COMPLEX*16 array, dimension
(NMAX*max(3,NRHS))
RWORK (workspace) DOUBLE PRECISION array, dimension (NMAX+2*NRHS)
NOUT (input) INTEGER
The unit number for output.
=====================================================================
Parameter adjustments */
--rwork;
--work;
--s;
--xact;
--x;
--bsav;
--b;
--asav;
--afac;
--a;
--nval;
--dotype;
/* Function Body
Initialize constants and the random number seed. */
s_copy(path, "Zomplex precision", (ftnlen)1, (ftnlen)17);
s_copy(path + 1, "PB", (ftnlen)2, (ftnlen)2);
nrun = 0;
nfail = 0;
nerrs = 0;
for (i__ = 1; i__ <= 4; ++i__) {
iseed[i__ - 1] = iseedy[i__ - 1];
/* L10: */
}
/* Test the error exits */
if (*tsterr) {
zerrvx_(path, nout);
}
infoc_1.infot = 0;
kdval[0] = 0;
/* Set the block size and minimum block size for testing. */
nb = 1;
nbmin = 2;
xlaenv_(&c__1, &nb);
xlaenv_(&c__2, &nbmin);
/* Do for each value of N in NVAL */
i__1 = *nn;
for (in = 1; in <= i__1; ++in) {
n = nval[in];
lda = max(n,1);
*(unsigned char *)xtype = 'N';
/* Set limits on the number of loop iterations.
Computing MAX */
i__2 = 1, i__3 = min(n,4);
nkd = max(i__2,i__3);
nimat = 8;
if (n == 0) {
nimat = 1;
}
kdval[1] = n + (n + 1) / 4;
kdval[2] = (n * 3 - 1) / 4;
kdval[3] = (n + 1) / 4;
i__2 = nkd;
for (ikd = 1; ikd <= i__2; ++ikd) {
/* Do for KD = 0, (5*N+1)/4, (3N-1)/4, and (N+1)/4. This order
makes it easier to skip redundant values for small values
of N. */
kd = kdval[ikd - 1];
ldab = kd + 1;
/* Do first for UPLO = 'U', then for UPLO = 'L' */
for (iuplo = 1; iuplo <= 2; ++iuplo) {
koff = 1;
if (iuplo == 1) {
*(unsigned char *)uplo = 'U';
*(unsigned char *)packit = 'Q';
/* Computing MAX */
i__3 = 1, i__4 = kd + 2 - n;
koff = max(i__3,i__4);
} else {
*(unsigned char *)uplo = 'L';
*(unsigned char *)packit = 'B';
}
i__3 = nimat;
for (imat = 1; imat <= i__3; ++imat) {
/* Do the tests only if DOTYPE( IMAT ) is true. */
if (! dotype[imat]) {
goto L80;
}
/* Skip types 2, 3, or 4 if the matrix size is too small. */
zerot = imat >= 2 && imat <= 4;
if (zerot && n < imat - 1) {
goto L80;
}
if (! zerot || ! dotype[1]) {
/* Set up parameters with ZLATB4 and generate a test
matrix with ZLATMS. */
zlatb4_(path, &imat, &n, &n, type__, &kl, &ku, &anorm,
&mode, &cndnum, dist);
s_copy(srnamc_1.srnamt, "ZLATMS", (ftnlen)6, (ftnlen)
6);
zlatms_(&n, &n, dist, iseed, type__, &rwork[1], &mode,
&cndnum, &anorm, &kd, &kd, packit, &a[koff],
&ldab, &work[1], &info);
/* Check error code from ZLATMS. */
if (info != 0) {
alaerh_(path, "ZLATMS", &info, &c__0, uplo, &n, &
n, &c_n1, &c_n1, &c_n1, &imat, &nfail, &
nerrs, nout);
goto L80;
}
} else if (izero > 0) {
/* Use the same matrix for types 3 and 4 as for type
2 by copying back the zeroed out column, */
iw = (lda << 1) + 1;
if (iuplo == 1) {
ioff = (izero - 1) * ldab + kd + 1;
i__4 = izero - i1;
zcopy_(&i__4, &work[iw], &c__1, &a[ioff - izero +
i1], &c__1);
iw = iw + izero - i1;
i__4 = i2 - izero + 1;
/* Computing MAX */
i__6 = ldab - 1;
i__5 = max(i__6,1);
zcopy_(&i__4, &work[iw], &c__1, &a[ioff], &i__5);
} else {
ioff = (i1 - 1) * ldab + 1;
i__4 = izero - i1;
/* Computing MAX */
i__6 = ldab - 1;
i__5 = max(i__6,1);
zcopy_(&i__4, &work[iw], &c__1, &a[ioff + izero -
i1], &i__5);
ioff = (izero - 1) * ldab + 1;
iw = iw + izero - i1;
i__4 = i2 - izero + 1;
zcopy_(&i__4, &work[iw], &c__1, &a[ioff], &c__1);
}
}
/* For types 2-4, zero one row and column of the matrix
to test that INFO is returned correctly. */
izero = 0;
if (zerot) {
if (imat == 2) {
izero = 1;
} else if (imat == 3) {
izero = n;
} else {
izero = n / 2 + 1;
}
/* Save the zeroed out row and column in WORK(*,3) */
iw = lda << 1;
/* Computing MIN */
i__5 = (kd << 1) + 1;
i__4 = min(i__5,n);
for (i__ = 1; i__ <= i__4; ++i__) {
i__5 = iw + i__;
work[i__5].r = 0., work[i__5].i = 0.;
/* L20: */
}
++iw;
/* Computing MAX */
i__4 = izero - kd;
i1 = max(i__4,1);
/* Computing MIN */
i__4 = izero + kd;
i2 = min(i__4,n);
if (iuplo == 1) {
ioff = (izero - 1) * ldab + kd + 1;
i__4 = izero - i1;
zswap_(&i__4, &a[ioff - izero + i1], &c__1, &work[
iw], &c__1);
iw = iw + izero - i1;
i__4 = i2 - izero + 1;
/* Computing MAX */
i__6 = ldab - 1;
i__5 = max(i__6,1);
zswap_(&i__4, &a[ioff], &i__5, &work[iw], &c__1);
} else {
ioff = (i1 - 1) * ldab + 1;
i__4 = izero - i1;
/* Computing MAX */
i__6 = ldab - 1;
i__5 = max(i__6,1);
zswap_(&i__4, &a[ioff + izero - i1], &i__5, &work[
iw], &c__1);
ioff = (izero - 1) * ldab + 1;
iw = iw + izero - i1;
i__4 = i2 - izero + 1;
zswap_(&i__4, &a[ioff], &c__1, &work[iw], &c__1);
}
}
/* Set the imaginary part of the diagonals. */
if (iuplo == 1) {
zlaipd_(&n, &a[kd + 1], &ldab, &c__0);
} else {
zlaipd_(&n, &a[1], &ldab, &c__0);
}
/* Save a copy of the matrix A in ASAV. */
i__4 = kd + 1;
zlacpy_("Full", &i__4, &n, &a[1], &ldab, &asav[1], &ldab);
for (iequed = 1; iequed <= 2; ++iequed) {
*(unsigned char *)equed = *(unsigned char *)&equeds[
iequed - 1];
if (iequed == 1) {
nfact = 3;
} else {
nfact = 1;
}
i__4 = nfact;
for (ifact = 1; ifact <= i__4; ++ifact) {
*(unsigned char *)fact = *(unsigned char *)&facts[
ifact - 1];
prefac = lsame_(fact, "F");
nofact = lsame_(fact, "N");
equil = lsame_(fact, "E");
if (zerot) {
if (prefac) {
goto L60;
}
rcondc = 0.;
} else if (! lsame_(fact, "N")) {
/* Compute the condition number for comparison
with the value returned by ZPBSVX (FACT =
'N' reuses the condition number from the
previous iteration with FACT = 'F'). */
i__5 = kd + 1;
zlacpy_("Full", &i__5, &n, &asav[1], &ldab, &
afac[1], &ldab);
if (equil || iequed > 1) {
/* Compute row and column scale factors to
equilibrate the matrix A. */
zpbequ_(uplo, &n, &kd, &afac[1], &ldab, &
s[1], &scond, &amax, &info);
if (info == 0 && n > 0) {
if (iequed > 1) {
scond = 0.;
}
/* Equilibrate the matrix. */
zlaqhb_(uplo, &n, &kd, &afac[1], &
ldab, &s[1], &scond, &amax,
equed);
}
}
/* Save the condition number of the
non-equilibrated system for use in ZGET04. */
if (equil) {
roldc = rcondc;
}
/* Compute the 1-norm of A. */
anorm = zlanhb_("1", uplo, &n, &kd, &afac[1],
&ldab, &rwork[1]);
/* Factor the matrix A. */
zpbtrf_(uplo, &n, &kd, &afac[1], &ldab, &info);
/* Form the inverse of A. */
zlaset_("Full", &n, &n, &c_b47, &c_b48, &a[1],
&lda);
s_copy(srnamc_1.srnamt, "ZPBTRS", (ftnlen)6, (
ftnlen)6);
zpbtrs_(uplo, &n, &kd, &n, &afac[1], &ldab, &
a[1], &lda, &info);
/* Compute the 1-norm condition number of A. */
ainvnm = zlange_("1", &n, &n, &a[1], &lda, &
rwork[1]);
if (anorm <= 0. || ainvnm <= 0.) {
rcondc = 1.;
} else {
rcondc = 1. / anorm / ainvnm;
}
}
/* Restore the matrix A. */
i__5 = kd + 1;
zlacpy_("Full", &i__5, &n, &asav[1], &ldab, &a[1],
&ldab);
/* Form an exact solution and set the right hand
side. */
s_copy(srnamc_1.srnamt, "ZLARHS", (ftnlen)6, (
ftnlen)6);
zlarhs_(path, xtype, uplo, " ", &n, &n, &kd, &kd,
nrhs, &a[1], &ldab, &xact[1], &lda, &b[1],
&lda, iseed, &info);
*(unsigned char *)xtype = 'C';
zlacpy_("Full", &n, nrhs, &b[1], &lda, &bsav[1], &
lda);
if (nofact) {
/* --- Test ZPBSV ---
Compute the L*L' or U'*U factorization of the
matrix and solve the system. */
i__5 = kd + 1;
zlacpy_("Full", &i__5, &n, &a[1], &ldab, &
afac[1], &ldab);
zlacpy_("Full", &n, nrhs, &b[1], &lda, &x[1],
&lda);
s_copy(srnamc_1.srnamt, "ZPBSV ", (ftnlen)6, (
ftnlen)6);
zpbsv_(uplo, &n, &kd, nrhs, &afac[1], &ldab, &
x[1], &lda, &info);
/* Check error code from ZPBSV . */
if (info != izero) {
alaerh_(path, "ZPBSV ", &info, &izero,
uplo, &n, &n, &kd, &kd, nrhs, &
imat, &nfail, &nerrs, nout);
goto L40;
} else if (info != 0) {
goto L40;
}
/* Reconstruct matrix from factors and compute
residual. */
zpbt01_(uplo, &n, &kd, &a[1], &ldab, &afac[1],
&ldab, &rwork[1], result);
/* Compute residual of the computed solution. */
zlacpy_("Full", &n, nrhs, &b[1], &lda, &work[
1], &lda);
zpbt02_(uplo, &n, &kd, nrhs, &a[1], &ldab, &x[
1], &lda, &work[1], &lda, &rwork[1], &
result[1]);
/* Check solution from generated exact solution. */
zget04_(&n, nrhs, &x[1], &lda, &xact[1], &lda,
&rcondc, &result[2]);
nt = 3;
/* Print information about the tests that did
not pass the threshold. */
i__5 = nt;
for (k = 1; k <= i__5; ++k) {
if (result[k - 1] >= *thresh) {
if (nfail == 0 && nerrs == 0) {
aladhd_(nout, path);
}
io___57.ciunit = *nout;
s_wsfe(&io___57);
do_fio(&c__1, "ZPBSV ", (ftnlen)6);
do_fio(&c__1, uplo, (ftnlen)1);
do_fio(&c__1, (char *)&n, (ftnlen)
sizeof(integer));
do_fio(&c__1, (char *)&kd, (ftnlen)
sizeof(integer));
do_fio(&c__1, (char *)&imat, (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;
}
/* L30: */
}
nrun += nt;
L40:
;
}
/* --- Test ZPBSVX --- */
if (! prefac) {
i__5 = kd + 1;
zlaset_("Full", &i__5, &n, &c_b47, &c_b47, &
afac[1], &ldab);
}
zlaset_("Full", &n, nrhs, &c_b47, &c_b47, &x[1], &
lda);
if (iequed > 1 && n > 0) {
/* Equilibrate the matrix if FACT='F' and
EQUED='Y' */
zlaqhb_(uplo, &n, &kd, &a[1], &ldab, &s[1], &
scond, &amax, equed);
}
/* Solve the system and compute the condition
number and error bounds using ZPBSVX. */
s_copy(srnamc_1.srnamt, "ZPBSVX", (ftnlen)6, (
ftnlen)6);
zpbsvx_(fact, uplo, &n, &kd, nrhs, &a[1], &ldab, &
afac[1], &ldab, equed, &s[1], &b[1], &lda,
&x[1], &lda, &rcond, &rwork[1], &rwork[*
nrhs + 1], &work[1], &rwork[(*nrhs << 1)
+ 1], &info);
/* Check the error code from ZPBSVX. */
if (info != izero) {
/* Writing concatenation */
i__7[0] = 1, a__1[0] = fact;
i__7[1] = 1, a__1[1] = uplo;
s_cat(ch__1, a__1, i__7, &c__2, (ftnlen)2);
alaerh_(path, "ZPBSVX", &info, &izero, ch__1,
&n, &n, &kd, &kd, nrhs, &imat, &nfail,
&nerrs, nout);
goto L60;
}
if (info == 0) {
if (! prefac) {
/* Reconstruct matrix from factors and
compute residual. */
zpbt01_(uplo, &n, &kd, &a[1], &ldab, &
afac[1], &ldab, &rwork[(*nrhs <<
1) + 1], result);
k1 = 1;
} else {
k1 = 2;
}
/* Compute residual of the computed solution. */
zlacpy_("Full", &n, nrhs, &bsav[1], &lda, &
work[1], &lda);
zpbt02_(uplo, &n, &kd, nrhs, &asav[1], &ldab,
&x[1], &lda, &work[1], &lda, &rwork[(*
nrhs << 1) + 1], &result[1]);
/* Check solution from generated exact solution. */
if (nofact || prefac && lsame_(equed, "N")) {
zget04_(&n, nrhs, &x[1], &lda, &xact[1], &
lda, &rcondc, &result[2]);
} else {
zget04_(&n, nrhs, &x[1], &lda, &xact[1], &
lda, &roldc, &result[2]);
}
/* Check the error bounds from iterative
refinement. */
zpbt05_(uplo, &n, &kd, nrhs, &asav[1], &ldab,
&b[1], &lda, &x[1], &lda, &xact[1], &
lda, &rwork[1], &rwork[*nrhs + 1], &
result[3]);
} else {
k1 = 6;
}
/* Compare RCOND from ZPBSVX with the computed
value in RCONDC. */
result[5] = dget06_(&rcond, &rcondc);
/* Print information about the tests that did not
pass the threshold. */
for (k = k1; k <= 6; ++k) {
if (result[k - 1] >= *thresh) {
if (nfail == 0 && nerrs == 0) {
aladhd_(nout, path);
}
if (prefac) {
io___60.ciunit = *nout;
s_wsfe(&io___60);
do_fio(&c__1, "ZPBSVX", (ftnlen)6);
do_fio(&c__1, fact, (ftnlen)1);
do_fio(&c__1, uplo, (ftnlen)1);
do_fio(&c__1, (char *)&n, (ftnlen)
sizeof(integer));
do_fio(&c__1, (char *)&kd, (ftnlen)
sizeof(integer));
do_fio(&c__1, equed, (ftnlen)1);
do_fio(&c__1, (char *)&imat, (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();
} else {
io___61.ciunit = *nout;
s_wsfe(&io___61);
do_fio(&c__1, "ZPBSVX", (ftnlen)6);
do_fio(&c__1, fact, (ftnlen)1);
do_fio(&c__1, uplo, (ftnlen)1);
do_fio(&c__1, (char *)&n, (ftnlen)
sizeof(integer));
do_fio(&c__1, (char *)&kd, (ftnlen)
sizeof(integer));
do_fio(&c__1, (char *)&imat, (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 = nrun + 7 - k1;
L60:
;
}
/* L70: */
}
L80:
;
}
/* L90: */
}
/* L100: */
}
/* L110: */
}
/* Print a summary of the results. */
alasvm_(path, nout, &nfail, &nrun, &nerrs);
return 0;
/* End of ZDRVPB */
} /* zdrvpb_ */
/* Subroutine */ int zchkgt_(logical *dotype, integer *nn, integer *nval,
integer *nns, integer *nsval, doublereal *thresh, logical *tsterr,
doublecomplex *a, doublecomplex *af, doublecomplex *b, doublecomplex *
x, doublecomplex *xact, doublecomplex *work, doublereal *rwork,
integer *iwork, integer *nout)
{
/* Initialized data */
static integer iseedy[4] = { 0,0,0,1 };
static char transs[1*3] = "N" "T" "C";
/* Format strings */
static char fmt_9999[] = "(12x,\002N =\002,i5,\002,\002,10x,\002 type"
" \002,i2,\002, test(\002,i2,\002) = \002,g12.5)";
static char fmt_9997[] = "(\002 NORM ='\002,a1,\002', N =\002,i5,\002"
",\002,10x,\002 type \002,i2,\002, test(\002,i2,\002) = \002,g12."
"5)";
static char fmt_9998[] = "(\002 TRANS='\002,a1,\002', N =\002,i5,\002, N"
"RHS=\002,i3,\002, type \002,i2,\002, test(\002,i2,\002) = \002,g"
"12.5)";
/* System generated locals */
integer i__1, i__2, i__3, i__4, i__5;
doublereal d__1, d__2;
/* Builtin functions */
/* Subroutine */ int s_copy(char *, char *, ftnlen, ftnlen);
integer s_wsfe(cilist *), do_fio(integer *, char *, ftnlen), e_wsfe(void);
/* Local variables */
integer i__, j, k, m, n;
doublecomplex z__[3];
integer in, kl, ku, ix, lda;
doublereal cond;
integer mode, koff, imat, info;
char path[3], dist[1];
integer irhs, nrhs;
char norm[1], type__[1];
integer nrun;
extern /* Subroutine */ int alahd_(integer *, char *);
integer nfail, iseed[4];
extern doublereal dget06_(doublereal *, doublereal *);
doublereal rcond;
integer nimat;
doublereal anorm;
integer itran;
extern /* Subroutine */ int zget04_(integer *, integer *, doublecomplex *,
integer *, doublecomplex *, integer *, doublereal *, doublereal *
);
char trans[1];
integer izero, nerrs;
extern /* Subroutine */ int zgtt01_(integer *, doublecomplex *,
doublecomplex *, doublecomplex *, doublecomplex *, doublecomplex *
, doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublereal *, doublereal *), zgtt02_(char *, integer *,
integer *, doublecomplex *, doublecomplex *, doublecomplex *,
doublecomplex *, integer *, doublecomplex *, integer *,
doublereal *, doublereal *), zgtt05_(char *, integer *,
integer *, doublecomplex *, doublecomplex *, doublecomplex *,
doublecomplex *, integer *, doublecomplex *, integer *,
doublecomplex *, integer *, doublereal *, doublereal *,
doublereal *);
logical zerot;
extern /* Subroutine */ int zcopy_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *), zlatb4_(char *, integer *, integer *,
integer *, char *, integer *, integer *, doublereal *, integer *,
doublereal *, char *), alaerh_(char *,
char *, integer *, integer *, char *, integer *, integer *,
integer *, integer *, integer *, integer *, integer *, integer *,
integer *);
doublereal rcondc, rcondi;
extern /* Subroutine */ int zdscal_(integer *, doublereal *,
doublecomplex *, integer *), alasum_(char *, integer *, integer *,
integer *, integer *);
doublereal rcondo, ainvnm;
logical trfcon;
extern /* Subroutine */ int zerrge_(char *, integer *);
extern doublereal zlangt_(char *, integer *, doublecomplex *,
doublecomplex *, doublecomplex *);
extern /* Subroutine */ int zlagtm_(char *, integer *, integer *,
doublereal *, doublecomplex *, doublecomplex *, doublecomplex *,
doublecomplex *, integer *, doublereal *, doublecomplex *,
integer *), zlacpy_(char *, integer *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *);
extern doublereal dzasum_(integer *, doublecomplex *, integer *);
extern /* Subroutine */ int zgtcon_(char *, integer *, doublecomplex *,
doublecomplex *, doublecomplex *, doublecomplex *, integer *,
doublereal *, doublereal *, doublecomplex *, integer *),
zlatms_(integer *, integer *, char *, integer *, char *,
doublereal *, integer *, doublereal *, doublereal *, integer *,
integer *, char *, doublecomplex *, integer *, doublecomplex *,
integer *), zlarnv_(integer *, integer *,
integer *, doublecomplex *);
doublereal result[7];
extern /* Subroutine */ int zgtrfs_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, doublecomplex *, doublecomplex *
, doublecomplex *, doublecomplex *, doublecomplex *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *,
doublereal *, doublereal *, doublecomplex *, doublereal *,
integer *), zgttrf_(integer *, doublecomplex *,
doublecomplex *, doublecomplex *, doublecomplex *, integer *,
integer *), zgttrs_(char *, integer *, integer *, doublecomplex *,
doublecomplex *, doublecomplex *, doublecomplex *, integer *,
doublecomplex *, integer *, integer *);
/* Fortran I/O blocks */
static cilist io___29 = { 0, 0, 0, fmt_9999, 0 };
static cilist io___39 = { 0, 0, 0, fmt_9997, 0 };
static cilist io___44 = { 0, 0, 0, fmt_9998, 0 };
/* -- LAPACK test routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* ZCHKGT tests ZGTTRF, -TRS, -RFS, and -CON */
/* 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. */
/* 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 dimension N. */
/* NNS (input) INTEGER */
/* The number of values of NRHS contained in the vector NSVAL. */
/* NSVAL (input) INTEGER array, dimension (NNS) */
/* The values of the number of right hand sides NRHS. */
/* 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) COMPLEX*16 array, dimension (NMAX*4) */
/* AF (workspace) COMPLEX*16 array, dimension (NMAX*4) */
/* B (workspace) COMPLEX*16 array, dimension (NMAX*NSMAX) */
/* where NSMAX is the largest entry in NSVAL. */
/* X (workspace) COMPLEX*16 array, dimension (NMAX*NSMAX) */
/* XACT (workspace) COMPLEX*16 array, dimension (NMAX*NSMAX) */
/* WORK (workspace) COMPLEX*16 array, dimension */
/* (NMAX*max(3,NSMAX)) */
/* RWORK (workspace) DOUBLE PRECISION array, dimension */
/* (max(NMAX)+2*NSMAX) */
/* 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;
--rwork;
--work;
--xact;
--x;
--b;
--af;
--a;
--nsval;
--nval;
--dotype;
/* Function Body */
/* .. */
/* .. Executable Statements .. */
s_copy(path, "Zomplex precision", (ftnlen)1, (ftnlen)17);
s_copy(path + 1, "GT", (ftnlen)2, (ftnlen)2);
nrun = 0;
nfail = 0;
nerrs = 0;
for (i__ = 1; i__ <= 4; ++i__) {
iseed[i__ - 1] = iseedy[i__ - 1];
/* L10: */
}
/* Test the error exits */
if (*tsterr) {
zerrge_(path, nout);
}
infoc_1.infot = 0;
i__1 = *nn;
for (in = 1; in <= i__1; ++in) {
/* Do for each value of N in NVAL. */
n = nval[in];
/* Computing MAX */
i__2 = n - 1;
m = max(i__2,0);
lda = max(1,n);
nimat = 12;
if (n <= 0) {
nimat = 1;
}
i__2 = nimat;
for (imat = 1; imat <= i__2; ++imat) {
/* Do the tests only if DOTYPE( IMAT ) is true. */
if (! dotype[imat]) {
goto L100;
}
/* Set up parameters with ZLATB4. */
zlatb4_(path, &imat, &n, &n, type__, &kl, &ku, &anorm, &mode, &
cond, dist);
zerot = imat >= 8 && imat <= 10;
if (imat <= 6) {
/* Types 1-6: generate matrices of known condition number. */
/* Computing MAX */
i__3 = 2 - ku, i__4 = 3 - max(1,n);
koff = max(i__3,i__4);
s_copy(srnamc_1.srnamt, "ZLATMS", (ftnlen)32, (ftnlen)6);
zlatms_(&n, &n, dist, iseed, type__, &rwork[1], &mode, &cond,
&anorm, &kl, &ku, "Z", &af[koff], &c__3, &work[1], &
info);
/* Check the error code from ZLATMS. */
if (info != 0) {
alaerh_(path, "ZLATMS", &info, &c__0, " ", &n, &n, &kl, &
ku, &c_n1, &imat, &nfail, &nerrs, nout);
goto L100;
}
izero = 0;
if (n > 1) {
i__3 = n - 1;
zcopy_(&i__3, &af[4], &c__3, &a[1], &c__1);
i__3 = n - 1;
zcopy_(&i__3, &af[3], &c__3, &a[n + m + 1], &c__1);
}
zcopy_(&n, &af[2], &c__3, &a[m + 1], &c__1);
} else {
/* Types 7-12: generate tridiagonal matrices with */
/* unknown condition numbers. */
if (! zerot || ! dotype[7]) {
/* Generate a matrix with elements whose real and */
/* imaginary parts are from [-1,1]. */
i__3 = n + (m << 1);
zlarnv_(&c__2, iseed, &i__3, &a[1]);
if (anorm != 1.) {
i__3 = n + (m << 1);
zdscal_(&i__3, &anorm, &a[1], &c__1);
}
} else if (izero > 0) {
/* Reuse the last matrix by copying back the zeroed out */
/* elements. */
if (izero == 1) {
i__3 = n;
a[i__3].r = z__[1].r, a[i__3].i = z__[1].i;
if (n > 1) {
a[1].r = z__[2].r, a[1].i = z__[2].i;
}
} else if (izero == n) {
i__3 = n * 3 - 2;
a[i__3].r = z__[0].r, a[i__3].i = z__[0].i;
i__3 = (n << 1) - 1;
a[i__3].r = z__[1].r, a[i__3].i = z__[1].i;
} else {
i__3 = (n << 1) - 2 + izero;
a[i__3].r = z__[0].r, a[i__3].i = z__[0].i;
i__3 = n - 1 + izero;
a[i__3].r = z__[1].r, a[i__3].i = z__[1].i;
i__3 = izero;
a[i__3].r = z__[2].r, a[i__3].i = z__[2].i;
}
}
/* If IMAT > 7, set one column of the matrix to 0. */
if (! zerot) {
izero = 0;
} else if (imat == 8) {
izero = 1;
i__3 = n;
z__[1].r = a[i__3].r, z__[1].i = a[i__3].i;
i__3 = n;
a[i__3].r = 0., a[i__3].i = 0.;
if (n > 1) {
z__[2].r = a[1].r, z__[2].i = a[1].i;
a[1].r = 0., a[1].i = 0.;
}
} else if (imat == 9) {
izero = n;
i__3 = n * 3 - 2;
z__[0].r = a[i__3].r, z__[0].i = a[i__3].i;
i__3 = (n << 1) - 1;
z__[1].r = a[i__3].r, z__[1].i = a[i__3].i;
i__3 = n * 3 - 2;
a[i__3].r = 0., a[i__3].i = 0.;
i__3 = (n << 1) - 1;
a[i__3].r = 0., a[i__3].i = 0.;
} else {
izero = (n + 1) / 2;
i__3 = n - 1;
for (i__ = izero; i__ <= i__3; ++i__) {
i__4 = (n << 1) - 2 + i__;
a[i__4].r = 0., a[i__4].i = 0.;
i__4 = n - 1 + i__;
a[i__4].r = 0., a[i__4].i = 0.;
i__4 = i__;
a[i__4].r = 0., a[i__4].i = 0.;
/* L20: */
}
i__3 = n * 3 - 2;
a[i__3].r = 0., a[i__3].i = 0.;
i__3 = (n << 1) - 1;
a[i__3].r = 0., a[i__3].i = 0.;
}
}
/* + TEST 1 */
/* Factor A as L*U and compute the ratio */
/* norm(L*U - A) / (n * norm(A) * EPS ) */
i__3 = n + (m << 1);
zcopy_(&i__3, &a[1], &c__1, &af[1], &c__1);
s_copy(srnamc_1.srnamt, "ZGTTRF", (ftnlen)32, (ftnlen)6);
zgttrf_(&n, &af[1], &af[m + 1], &af[n + m + 1], &af[n + (m << 1)
+ 1], &iwork[1], &info);
/* Check error code from ZGTTRF. */
if (info != izero) {
alaerh_(path, "ZGTTRF", &info, &izero, " ", &n, &n, &c__1, &
c__1, &c_n1, &imat, &nfail, &nerrs, nout);
}
trfcon = info != 0;
zgtt01_(&n, &a[1], &a[m + 1], &a[n + m + 1], &af[1], &af[m + 1], &
af[n + m + 1], &af[n + (m << 1) + 1], &iwork[1], &work[1],
&lda, &rwork[1], result);
/* Print the test ratio if it is .GE. THRESH. */
if (result[0] >= *thresh) {
if (nfail == 0 && nerrs == 0) {
alahd_(nout, path);
}
io___29.ciunit = *nout;
s_wsfe(&io___29);
do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&imat, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&c__1, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&result[0], (ftnlen)sizeof(doublereal));
e_wsfe();
++nfail;
}
++nrun;
for (itran = 1; itran <= 2; ++itran) {
*(unsigned char *)trans = *(unsigned char *)&transs[itran - 1]
;
if (itran == 1) {
*(unsigned char *)norm = 'O';
} else {
*(unsigned char *)norm = 'I';
}
anorm = zlangt_(norm, &n, &a[1], &a[m + 1], &a[n + m + 1]);
if (! trfcon) {
/* Use ZGTTRS to solve for one column at a time of */
/* inv(A), computing the maximum column sum as we go. */
ainvnm = 0.;
i__3 = n;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = n;
for (j = 1; j <= i__4; ++j) {
i__5 = j;
x[i__5].r = 0., x[i__5].i = 0.;
/* L30: */
}
i__4 = i__;
x[i__4].r = 1., x[i__4].i = 0.;
zgttrs_(trans, &n, &c__1, &af[1], &af[m + 1], &af[n +
m + 1], &af[n + (m << 1) + 1], &iwork[1], &x[
1], &lda, &info);
/* Computing MAX */
d__1 = ainvnm, d__2 = dzasum_(&n, &x[1], &c__1);
ainvnm = max(d__1,d__2);
/* L40: */
}
/* Compute RCONDC = 1 / (norm(A) * norm(inv(A)) */
if (anorm <= 0. || ainvnm <= 0.) {
rcondc = 1.;
} else {
rcondc = 1. / anorm / ainvnm;
}
if (itran == 1) {
rcondo = rcondc;
} else {
rcondi = rcondc;
}
} else {
rcondc = 0.;
}
/* + TEST 7 */
/* Estimate the reciprocal of the condition number of the */
/* matrix. */
s_copy(srnamc_1.srnamt, "ZGTCON", (ftnlen)32, (ftnlen)6);
zgtcon_(norm, &n, &af[1], &af[m + 1], &af[n + m + 1], &af[n +
(m << 1) + 1], &iwork[1], &anorm, &rcond, &work[1], &
info);
/* Check error code from ZGTCON. */
if (info != 0) {
alaerh_(path, "ZGTCON", &info, &c__0, norm, &n, &n, &c_n1,
&c_n1, &c_n1, &imat, &nfail, &nerrs, nout);
}
result[6] = dget06_(&rcond, &rcondc);
/* Print the test ratio if it is .GE. THRESH. */
if (result[6] >= *thresh) {
if (nfail == 0 && nerrs == 0) {
alahd_(nout, path);
}
io___39.ciunit = *nout;
s_wsfe(&io___39);
do_fio(&c__1, norm, (ftnlen)1);
do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&imat, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&c__7, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&result[6], (ftnlen)sizeof(
doublereal));
e_wsfe();
++nfail;
}
++nrun;
/* L50: */
}
/* Skip the remaining tests if the matrix is singular. */
if (trfcon) {
goto L100;
}
i__3 = *nns;
for (irhs = 1; irhs <= i__3; ++irhs) {
nrhs = nsval[irhs];
/* Generate NRHS random solution vectors. */
ix = 1;
i__4 = nrhs;
for (j = 1; j <= i__4; ++j) {
zlarnv_(&c__2, iseed, &n, &xact[ix]);
ix += lda;
/* L60: */
}
for (itran = 1; itran <= 3; ++itran) {
*(unsigned char *)trans = *(unsigned char *)&transs[itran
- 1];
if (itran == 1) {
rcondc = rcondo;
} else {
rcondc = rcondi;
}
/* Set the right hand side. */
zlagtm_(trans, &n, &nrhs, &c_b63, &a[1], &a[m + 1], &a[n
+ m + 1], &xact[1], &lda, &c_b64, &b[1], &lda);
/* + TEST 2 */
/* Solve op(A) * X = B and compute the residual. */
zlacpy_("Full", &n, &nrhs, &b[1], &lda, &x[1], &lda);
s_copy(srnamc_1.srnamt, "ZGTTRS", (ftnlen)32, (ftnlen)6);
zgttrs_(trans, &n, &nrhs, &af[1], &af[m + 1], &af[n + m +
1], &af[n + (m << 1) + 1], &iwork[1], &x[1], &lda,
&info);
/* Check error code from ZGTTRS. */
if (info != 0) {
alaerh_(path, "ZGTTRS", &info, &c__0, trans, &n, &n, &
c_n1, &c_n1, &nrhs, &imat, &nfail, &nerrs,
nout);
}
zlacpy_("Full", &n, &nrhs, &b[1], &lda, &work[1], &lda);
zgtt02_(trans, &n, &nrhs, &a[1], &a[m + 1], &a[n + m + 1],
&x[1], &lda, &work[1], &lda, &rwork[1], &result[
1]);
/* + TEST 3 */
/* Check solution from generated exact solution. */
zget04_(&n, &nrhs, &x[1], &lda, &xact[1], &lda, &rcondc, &
result[2]);
/* + TESTS 4, 5, and 6 */
/* Use iterative refinement to improve the solution. */
s_copy(srnamc_1.srnamt, "ZGTRFS", (ftnlen)32, (ftnlen)6);
zgtrfs_(trans, &n, &nrhs, &a[1], &a[m + 1], &a[n + m + 1],
&af[1], &af[m + 1], &af[n + m + 1], &af[n + (m <<
1) + 1], &iwork[1], &b[1], &lda, &x[1], &lda, &
rwork[1], &rwork[nrhs + 1], &work[1], &rwork[(
nrhs << 1) + 1], &info);
/* Check error code from ZGTRFS. */
if (info != 0) {
alaerh_(path, "ZGTRFS", &info, &c__0, trans, &n, &n, &
c_n1, &c_n1, &nrhs, &imat, &nfail, &nerrs,
nout);
}
zget04_(&n, &nrhs, &x[1], &lda, &xact[1], &lda, &rcondc, &
result[3]);
zgtt05_(trans, &n, &nrhs, &a[1], &a[m + 1], &a[n + m + 1],
&b[1], &lda, &x[1], &lda, &xact[1], &lda, &rwork[
1], &rwork[nrhs + 1], &result[4]);
/* Print information about the tests that did not pass the */
/* threshold. */
for (k = 2; k <= 6; ++k) {
if (result[k - 1] >= *thresh) {
if (nfail == 0 && nerrs == 0) {
alahd_(nout, path);
}
io___44.ciunit = *nout;
s_wsfe(&io___44);
do_fio(&c__1, trans, (ftnlen)1);
do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer))
;
do_fio(&c__1, (char *)&nrhs, (ftnlen)sizeof(
integer));
do_fio(&c__1, (char *)&imat, (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;
}
/* L70: */
}
nrun += 5;
/* L80: */
}
/* L90: */
}
L100:
;
}
/* L110: */
}
/* Print a summary of the results. */
alasum_(path, nout, &nfail, &nrun, &nerrs);
return 0;
/* End of ZCHKGT */
} /* zchkgt_ */
/* Subroutine */ int zgetv0_(integer *ido, char *bmat, integer *itry, logical
*initv, integer *n, integer *j, doublecomplex *v, integer *ldv,
doublecomplex *resid, doublereal *rnorm, integer *ipntr,
doublecomplex *workd, integer *ierr, ftnlen bmat_len)
{
/* Initialized data */
static logical inits = TRUE_;
/* System generated locals */
integer v_dim1, v_offset, i__1, i__2;
doublereal d__1, d__2;
doublecomplex z__1;
/* Local variables */
static real t0, t1, t2, t3;
static integer jj, iter;
static logical orth;
static integer iseed[4], idist;
static doublecomplex cnorm;
extern /* Double Complex */ void zdotc_(doublecomplex *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *);
static logical first;
extern /* Subroutine */ int zgemv_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublecomplex *, doublecomplex *, integer *, ftnlen),
dvout_(integer *, integer *, doublereal *, integer *, char *,
ftnlen), zcopy_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *), zvout_(integer *, integer *,
doublecomplex *, integer *, char *, ftnlen);
extern doublereal dlapy2_(doublereal *, doublereal *), dznrm2_(integer *,
doublecomplex *, integer *);
static doublereal rnorm0;
extern /* Subroutine */ int arscnd_(real *);
static integer msglvl;
extern /* Subroutine */ int zlarnv_(integer *, integer *, integer *,
doublecomplex *);
/* %----------------------------------------------------% */
/* | Include files for debugging and timing information | */
/* %----------------------------------------------------% */
/* \SCCS Information: @(#) */
/* FILE: debug.h SID: 2.3 DATE OF SID: 11/16/95 RELEASE: 2 */
/* %---------------------------------% */
/* | See debug.doc for documentation | */
/* %---------------------------------% */
/* %------------------% */
/* | Scalar Arguments | */
/* %------------------% */
/* %--------------------------------% */
/* | See stat.doc for documentation | */
/* %--------------------------------% */
/* \SCCS Information: @(#) */
/* FILE: stat.h SID: 2.2 DATE OF SID: 11/16/95 RELEASE: 2 */
/* %-----------------% */
/* | Array Arguments | */
/* %-----------------% */
/* %------------% */
/* | Parameters | */
/* %------------% */
/* %------------------------% */
/* | Local Scalars & Arrays | */
/* %------------------------% */
/* %----------------------% */
/* | External Subroutines | */
/* %----------------------% */
/* %--------------------% */
/* | External Functions | */
/* %--------------------% */
/* %-----------------% */
/* | Data Statements | */
/* %-----------------% */
/* Parameter adjustments */
--workd;
--resid;
v_dim1 = *ldv;
v_offset = 1 + v_dim1;
v -= v_offset;
--ipntr;
/* Function Body */
/* %-----------------------% */
/* | Executable Statements | */
/* %-----------------------% */
/* %-----------------------------------% */
/* | Initialize the seed of the LAPACK | */
/* | random number generator | */
/* %-----------------------------------% */
if (inits) {
iseed[0] = 1;
iseed[1] = 3;
iseed[2] = 5;
iseed[3] = 7;
inits = FALSE_;
}
if (*ido == 0) {
/* %-------------------------------% */
/* | Initialize timing statistics | */
/* | & message level for debugging | */
/* %-------------------------------% */
arscnd_(&t0);
msglvl = debug_1.mgetv0;
*ierr = 0;
iter = 0;
first = FALSE_;
orth = FALSE_;
/* %-----------------------------------------------------% */
/* | Possibly generate a random starting vector in RESID | */
/* | Use a LAPACK random number generator used by the | */
/* | matrix generation routines. | */
/* | idist = 1: uniform (0,1) distribution; | */
/* | idist = 2: uniform (-1,1) distribution; | */
/* | idist = 3: normal (0,1) distribution; | */
/* %-----------------------------------------------------% */
if (! (*initv)) {
idist = 2;
zlarnv_(&idist, iseed, n, &resid[1]);
}
/* %----------------------------------------------------------% */
/* | Force the starting vector into the range of OP to handle | */
/* | the generalized problem when B is possibly (singular). | */
/* %----------------------------------------------------------% */
arscnd_(&t2);
if (*(unsigned char *)bmat == 'G') {
++timing_1.nopx;
ipntr[1] = 1;
ipntr[2] = *n + 1;
zcopy_(n, &resid[1], &c__1, &workd[1], &c__1);
*ido = -1;
goto L9000;
}
}
/* %----------------------------------------% */
/* | Back from computing B*(initial-vector) | */
/* %----------------------------------------% */
if (first) {
goto L20;
}
/* %-----------------------------------------------% */
/* | Back from computing B*(orthogonalized-vector) | */
/* %-----------------------------------------------% */
if (orth) {
goto L40;
}
arscnd_(&t3);
timing_1.tmvopx += t3 - t2;
/* %------------------------------------------------------% */
/* | Starting vector is now in the range of OP; r = OP*r; | */
/* | Compute B-norm of starting vector. | */
/* %------------------------------------------------------% */
arscnd_(&t2);
first = TRUE_;
if (*(unsigned char *)bmat == 'G') {
++timing_1.nbx;
zcopy_(n, &workd[*n + 1], &c__1, &resid[1], &c__1);
ipntr[1] = *n + 1;
ipntr[2] = 1;
*ido = 2;
goto L9000;
} else if (*(unsigned char *)bmat == 'I') {
zcopy_(n, &resid[1], &c__1, &workd[1], &c__1);
}
L20:
if (*(unsigned char *)bmat == 'G') {
arscnd_(&t3);
timing_1.tmvbx += t3 - t2;
}
first = FALSE_;
if (*(unsigned char *)bmat == 'G') {
zdotc_(&z__1, n, &resid[1], &c__1, &workd[1], &c__1);
cnorm.r = z__1.r, cnorm.i = z__1.i;
d__1 = cnorm.r;
d__2 = d_imag(&cnorm);
rnorm0 = sqrt(dlapy2_(&d__1, &d__2));
} else if (*(unsigned char *)bmat == 'I') {
rnorm0 = dznrm2_(n, &resid[1], &c__1);
}
*rnorm = rnorm0;
/* %---------------------------------------------% */
/* | Exit if this is the very first Arnoldi step | */
/* %---------------------------------------------% */
if (*j == 1) {
goto L50;
}
/* %---------------------------------------------------------------- */
/* | Otherwise need to B-orthogonalize the starting vector against | */
/* | the current Arnoldi basis using Gram-Schmidt with iter. ref. | */
/* | This is the case where an invariant subspace is encountered | */
/* | in the middle of the Arnoldi factorization. | */
/* | | */
/* | s = V^{T}*B*r; r = r - V*s; | */
/* | | */
/* | Stopping criteria used for iter. ref. is discussed in | */
/* | Parlett's book, page 107 and in Gragg & Reichel TOMS paper. | */
/* %---------------------------------------------------------------% */
orth = TRUE_;
L30:
i__1 = *j - 1;
zgemv_("C", n, &i__1, &c_b1, &v[v_offset], ldv, &workd[1], &c__1, &c_b2, &
workd[*n + 1], &c__1, (ftnlen)1);
i__1 = *j - 1;
z__1.r = -1., z__1.i = -0.;
zgemv_("N", n, &i__1, &z__1, &v[v_offset], ldv, &workd[*n + 1], &c__1, &
c_b1, &resid[1], &c__1, (ftnlen)1);
/* %----------------------------------------------------------% */
/* | Compute the B-norm of the orthogonalized starting vector | */
/* %----------------------------------------------------------% */
arscnd_(&t2);
if (*(unsigned char *)bmat == 'G') {
++timing_1.nbx;
zcopy_(n, &resid[1], &c__1, &workd[*n + 1], &c__1);
ipntr[1] = *n + 1;
ipntr[2] = 1;
*ido = 2;
goto L9000;
} else if (*(unsigned char *)bmat == 'I') {
zcopy_(n, &resid[1], &c__1, &workd[1], &c__1);
}
L40:
if (*(unsigned char *)bmat == 'G') {
arscnd_(&t3);
timing_1.tmvbx += t3 - t2;
}
if (*(unsigned char *)bmat == 'G') {
zdotc_(&z__1, n, &resid[1], &c__1, &workd[1], &c__1);
cnorm.r = z__1.r, cnorm.i = z__1.i;
d__1 = cnorm.r;
d__2 = d_imag(&cnorm);
*rnorm = sqrt(dlapy2_(&d__1, &d__2));
} else if (*(unsigned char *)bmat == 'I') {
*rnorm = dznrm2_(n, &resid[1], &c__1);
}
/* %--------------------------------------% */
/* | Check for further orthogonalization. | */
/* %--------------------------------------% */
if (msglvl > 2) {
dvout_(&debug_1.logfil, &c__1, &rnorm0, &debug_1.ndigit, "_getv0: re"
"-orthonalization ; rnorm0 is", (ftnlen)38);
dvout_(&debug_1.logfil, &c__1, rnorm, &debug_1.ndigit, "_getv0: re-o"
"rthonalization ; rnorm is", (ftnlen)37);
}
if (*rnorm > rnorm0 * .717f) {
goto L50;
}
++iter;
if (iter <= 1) {
/* %-----------------------------------% */
/* | Perform iterative refinement step | */
/* %-----------------------------------% */
rnorm0 = *rnorm;
goto L30;
} else {
/* %------------------------------------% */
/* | Iterative refinement step "failed" | */
/* %------------------------------------% */
i__1 = *n;
for (jj = 1; jj <= i__1; ++jj) {
i__2 = jj;
resid[i__2].r = 0., resid[i__2].i = 0.;
/* L45: */
}
*rnorm = 0.;
*ierr = -1;
}
L50:
if (msglvl > 0) {
dvout_(&debug_1.logfil, &c__1, rnorm, &debug_1.ndigit, "_getv0: B-no"
"rm of initial / restarted starting vector", (ftnlen)53);
}
if (msglvl > 2) {
zvout_(&debug_1.logfil, n, &resid[1], &debug_1.ndigit, "_getv0: init"
"ial / restarted starting vector", (ftnlen)43);
}
*ido = 99;
arscnd_(&t1);
timing_1.tgetv0 += t1 - t0;
L9000:
return 0;
/* %---------------% */
/* | End of zgetv0 | */
/* %---------------% */
} /* zgetv0_ */
/* Subroutine */ int zptsvx_(char *fact, integer *n, integer *nrhs,
doublereal *d__, doublecomplex *e, doublereal *df, doublecomplex *ef,
doublecomplex *b, integer *ldb, doublecomplex *x, integer *ldx,
doublereal *rcond, doublereal *ferr, doublereal *berr, doublecomplex *
work, doublereal *rwork, integer *info)
{
/* -- LAPACK 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
=======
ZPTSVX uses the factorization A = L*D*L**H to compute the solution
to a complex system of linear equations A*X = B, where A is an
N-by-N Hermitian positive definite tridiagonal matrix and X and B
are N-by-NRHS matrices.
Error bounds on the solution and a condition estimate are also
provided.
Description
===========
The following steps are performed:
1. If FACT = 'N', the matrix A is factored as A = L*D*L**H, where L
is a unit lower bidiagonal matrix and D is diagonal. The
factorization can also be regarded as having the form
A = U**H*D*U.
2. If the leading i-by-i principal minor is not positive definite,
then the routine returns with INFO = i. Otherwise, the factored
form of A is used to estimate the condition number of the matrix
A. If the reciprocal of the condition number is less than machine
precision, INFO = N+1 is returned as a warning, but the routine
still goes on to solve for X and compute error bounds as
described below.
3. The system of equations is solved for X using the factored form
of A.
4. Iterative refinement is applied to improve the computed solution
matrix and calculate error bounds and backward error estimates
for it.
Arguments
=========
FACT (input) CHARACTER*1
Specifies whether or not the factored form of the matrix
A is supplied on entry.
= 'F': On entry, DF and EF contain the factored form of A.
D, E, DF, and EF will not be modified.
= 'N': The matrix A will be copied to DF and EF and
factored.
N (input) INTEGER
The order of the matrix A. N >= 0.
NRHS (input) INTEGER
The number of right hand sides, i.e., the number of columns
of the matrices B and X. NRHS >= 0.
D (input) DOUBLE PRECISION array, dimension (N)
The n diagonal elements of the tridiagonal matrix A.
E (input) COMPLEX*16 array, dimension (N-1)
The (n-1) subdiagonal elements of the tridiagonal matrix A.
DF (input or output) DOUBLE PRECISION array, dimension (N)
If FACT = 'F', then DF is an input argument and on entry
contains the n diagonal elements of the diagonal matrix D
from the L*D*L**H factorization of A.
If FACT = 'N', then DF is an output argument and on exit
contains the n diagonal elements of the diagonal matrix D
from the L*D*L**H factorization of A.
EF (input or output) COMPLEX*16 array, dimension (N-1)
If FACT = 'F', then EF is an input argument and on entry
contains the (n-1) subdiagonal elements of the unit
bidiagonal factor L from the L*D*L**H factorization of A.
If FACT = 'N', then EF is an output argument and on exit
contains the (n-1) subdiagonal elements of the unit
bidiagonal factor L from the L*D*L**H factorization of A.
B (input) COMPLEX*16 array, dimension (LDB,NRHS)
The N-by-NRHS right hand side matrix B.
LDB (input) INTEGER
The leading dimension of the array B. LDB >= max(1,N).
X (output) COMPLEX*16 array, dimension (LDX,NRHS)
If INFO = 0 or INFO = N+1, the N-by-NRHS solution matrix X.
LDX (input) INTEGER
The leading dimension of the array X. LDX >= max(1,N).
RCOND (output) DOUBLE PRECISION
The reciprocal condition number of the matrix A. If RCOND
is less than the machine precision (in particular, if
RCOND = 0), the matrix is singular to working precision.
This condition is indicated by a return code of INFO > 0.
FERR (output) DOUBLE PRECISION array, dimension (NRHS)
The forward error bound for each solution vector
X(j) (the j-th column of the solution matrix X).
If XTRUE is the true solution corresponding to X(j), FERR(j)
is an estimated upper bound for the magnitude of the largest
element in (X(j) - XTRUE) divided by the magnitude of the
largest element in X(j).
BERR (output) DOUBLE PRECISION array, dimension (NRHS)
The componentwise relative backward error of each solution
vector X(j) (i.e., the smallest relative change in any
element of A or B that makes X(j) an exact solution).
WORK (workspace) COMPLEX*16 array, dimension (N)
RWORK (workspace) DOUBLE PRECISION array, dimension (N)
INFO (output) INTEGER
= 0: successful exit
< 0: if INFO = -i, the i-th argument had an illegal value
> 0: if INFO = i, and i is
<= N: the leading minor of order i of A is
not positive definite, so the factorization
could not be completed, and the solution has not
been computed. RCOND = 0 is returned.
= N+1: U is nonsingular, but RCOND is less than machine
precision, meaning that the matrix is singular
to working precision. Nevertheless, the
solution and error bounds are computed because
there are a number of situations where the
computed solution can be more accurate than the
value of RCOND would suggest.
=====================================================================
Test the input parameters.
Parameter adjustments */
/* Table of constant values */
static integer c__1 = 1;
/* System generated locals */
integer b_dim1, b_offset, x_dim1, x_offset, i__1;
/* Local variables */
extern logical lsame_(char *, char *);
static doublereal anorm;
extern /* Subroutine */ int dcopy_(integer *, doublereal *, integer *,
doublereal *, integer *), zcopy_(integer *, doublecomplex *,
integer *, doublecomplex *, integer *);
extern doublereal dlamch_(char *);
static logical nofact;
extern /* Subroutine */ int xerbla_(char *, integer *);
extern doublereal zlanht_(char *, integer *, doublereal *, doublecomplex *
);
extern /* Subroutine */ int zlacpy_(char *, integer *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *),
zptcon_(integer *, doublereal *, doublecomplex *, doublereal *,
doublereal *, doublereal *, integer *), zptrfs_(char *, integer *,
integer *, doublereal *, doublecomplex *, doublereal *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublereal *, doublereal *, doublecomplex *,
doublereal *, integer *), zpttrf_(integer *, doublereal *,
doublecomplex *, integer *), zpttrs_(char *, integer *, integer *
, doublereal *, doublecomplex *, doublecomplex *, integer *,
integer *);
--d__;
--e;
--df;
--ef;
b_dim1 = *ldb;
b_offset = 1 + b_dim1 * 1;
b -= b_offset;
x_dim1 = *ldx;
x_offset = 1 + x_dim1 * 1;
x -= x_offset;
--ferr;
--berr;
--work;
--rwork;
/* Function Body */
*info = 0;
nofact = lsame_(fact, "N");
if (! nofact && ! lsame_(fact, "F")) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*nrhs < 0) {
*info = -3;
} else if (*ldb < max(1,*n)) {
*info = -9;
} else if (*ldx < max(1,*n)) {
*info = -11;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("ZPTSVX", &i__1);
return 0;
}
if (nofact) {
/* Compute the L*D*L' (or U'*D*U) factorization of A. */
dcopy_(n, &d__[1], &c__1, &df[1], &c__1);
if (*n > 1) {
i__1 = *n - 1;
zcopy_(&i__1, &e[1], &c__1, &ef[1], &c__1);
}
zpttrf_(n, &df[1], &ef[1], info);
/* Return if INFO is non-zero. */
if (*info != 0) {
if (*info > 0) {
*rcond = 0.;
}
return 0;
}
}
/* Compute the norm of the matrix A. */
anorm = zlanht_("1", n, &d__[1], &e[1]);
/* Compute the reciprocal of the condition number of A. */
zptcon_(n, &df[1], &ef[1], &anorm, rcond, &rwork[1], info);
/* Set INFO = N+1 if the matrix is singular to working precision. */
if (*rcond < dlamch_("Epsilon")) {
*info = *n + 1;
}
/* Compute the solution vectors X. */
zlacpy_("Full", n, nrhs, &b[b_offset], ldb, &x[x_offset], ldx);
zpttrs_("Lower", n, nrhs, &df[1], &ef[1], &x[x_offset], ldx, info);
/* Use iterative refinement to improve the computed solutions and
compute error bounds and backward error estimates for them. */
zptrfs_("Lower", n, nrhs, &d__[1], &e[1], &df[1], &ef[1], &b[b_offset],
ldb, &x[x_offset], ldx, &ferr[1], &berr[1], &work[1], &rwork[1],
info);
return 0;
/* End of ZPTSVX */
} /* zptsvx_ */
/* Subroutine */ int zlarz_(char *side, integer *m, integer *n, integer *l,
doublecomplex *v, integer *incv, doublecomplex *tau, doublecomplex *
c__, integer *ldc, doublecomplex *work)
{
/* System generated locals */
integer c_dim1, c_offset;
doublecomplex z__1;
/* Local variables */
/* -- LAPACK routine (version 3.2) -- */
/* November 2006 */
/* Purpose */
/* ======= */
/* ZLARZ applies a complex elementary reflector H to a complex */
/* M-by-N matrix C, from either the left or the right. H is represented */
/* in the form */
/* H = I - tau * v * v' */
/* where tau is a complex scalar and v is a complex vector. */
/* If tau = 0, then H is taken to be the unit matrix. */
/* To apply H' (the conjugate transpose of H), supply conjg(tau) instead */
/* tau. */
/* H is a product of k elementary reflectors as returned by ZTZRZF. */
/* Arguments */
/* ========= */
/* SIDE (input) CHARACTER*1 */
/* = 'L': form H * C */
/* = 'R': form C * H */
/* M (input) INTEGER */
/* The number of rows of the matrix C. */
/* N (input) INTEGER */
/* The number of columns of the matrix C. */
/* L (input) INTEGER */
/* The number of entries of the vector V containing */
/* the meaningful part of the Householder vectors. */
/* If SIDE = 'L', M >= L >= 0, if SIDE = 'R', N >= L >= 0. */
/* V (input) COMPLEX*16 array, dimension (1+(L-1)*abs(INCV)) */
/* The vector v in the representation of H as returned by */
/* ZTZRZF. V is not used if TAU = 0. */
/* INCV (input) INTEGER */
/* The increment between elements of v. INCV <> 0. */
/* TAU (input) COMPLEX*16 */
/* The value tau in the representation of H. */
/* C (input/output) COMPLEX*16 array, dimension (LDC,N) */
/* On entry, the M-by-N matrix C. */
/* On exit, C is overwritten by the matrix H * C if SIDE = 'L', */
/* or C * H if SIDE = 'R'. */
/* LDC (input) INTEGER */
/* The leading dimension of the array C. LDC >= max(1,M). */
/* WORK (workspace) COMPLEX*16 array, dimension */
/* (N) if SIDE = 'L' */
/* or (M) if SIDE = 'R' */
/* Further Details */
/* =============== */
/* Based on contributions by */
/* A. Petitet, Computer Science Dept., Univ. of Tenn., Knoxville, USA */
/* ===================================================================== */
/* Parameter adjustments */
--v;
c_dim1 = *ldc;
c_offset = 1 + c_dim1;
c__ -= c_offset;
--work;
/* Function Body */
if (lsame_(side, "L")) {
/* Form H * C */
if (tau->r != 0. || tau->i != 0.) {
/* w( 1:n ) = conjg( C( 1, 1:n ) ) */
zcopy_(n, &c__[c_offset], ldc, &work[1], &c__1);
zlacgv_(n, &work[1], &c__1);
/* w( 1:n ) = conjg( w( 1:n ) + C( m-l+1:m, 1:n )' * v( 1:l ) ) */
zgemv_("Conjugate transpose", l, n, &c_b1, &c__[*m - *l + 1 +
c_dim1], ldc, &v[1], incv, &c_b1, &work[1], &c__1);
zlacgv_(n, &work[1], &c__1);
/* C( 1, 1:n ) = C( 1, 1:n ) - tau * w( 1:n ) */
z__1.r = -tau->r, z__1.i = -tau->i;
zaxpy_(n, &z__1, &work[1], &c__1, &c__[c_offset], ldc);
/* tau * v( 1:l ) * conjg( w( 1:n )' ) */
z__1.r = -tau->r, z__1.i = -tau->i;
zgeru_(l, n, &z__1, &v[1], incv, &work[1], &c__1, &c__[*m - *l +
1 + c_dim1], ldc);
}
} else {
/* Form C * H */
if (tau->r != 0. || tau->i != 0.) {
/* w( 1:m ) = C( 1:m, 1 ) */
zcopy_(m, &c__[c_offset], &c__1, &work[1], &c__1);
/* w( 1:m ) = w( 1:m ) + C( 1:m, n-l+1:n, 1:n ) * v( 1:l ) */
zgemv_("No transpose", m, l, &c_b1, &c__[(*n - *l + 1) * c_dim1 +
1], ldc, &v[1], incv, &c_b1, &work[1], &c__1);
/* C( 1:m, 1 ) = C( 1:m, 1 ) - tau * w( 1:m ) */
z__1.r = -tau->r, z__1.i = -tau->i;
zaxpy_(m, &z__1, &work[1], &c__1, &c__[c_offset], &c__1);
/* tau * w( 1:m ) * v( 1:l )' */
z__1.r = -tau->r, z__1.i = -tau->i;
zgerc_(m, l, &z__1, &work[1], &c__1, &v[1], incv, &c__[(*n - *l +
1) * c_dim1 + 1], ldc);
}
}
return 0;
/* End of ZLARZ */
} /* zlarz_ */
/* Subroutine */ int zgtsvx_(char *fact, char *trans, integer *n, integer *
nrhs, doublecomplex *dl, doublecomplex *d__, doublecomplex *du,
doublecomplex *dlf, doublecomplex *df, doublecomplex *duf,
doublecomplex *du2, integer *ipiv, doublecomplex *b, integer *ldb,
doublecomplex *x, integer *ldx, doublereal *rcond, doublereal *ferr,
doublereal *berr, doublecomplex *work, doublereal *rwork, integer *
info)
{
/* System generated locals */
integer b_dim1, b_offset, x_dim1, x_offset, i__1;
/* Local variables */
char norm[1];
extern logical lsame_(char *, char *);
doublereal anorm;
extern /* Subroutine */ int zcopy_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *);
extern doublereal dlamch_(char *);
logical nofact;
extern /* Subroutine */ int xerbla_(char *, integer *);
extern doublereal zlangt_(char *, integer *, doublecomplex *,
doublecomplex *, doublecomplex *);
logical notran;
extern /* Subroutine */ int zlacpy_(char *, integer *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *),
zgtcon_(char *, integer *, doublecomplex *, doublecomplex *,
doublecomplex *, doublecomplex *, integer *, doublereal *,
doublereal *, doublecomplex *, integer *), zgtrfs_(char *,
integer *, integer *, doublecomplex *, doublecomplex *,
doublecomplex *, doublecomplex *, doublecomplex *, doublecomplex *
, doublecomplex *, integer *, doublecomplex *, integer *,
doublecomplex *, integer *, doublereal *, doublereal *,
doublecomplex *, doublereal *, integer *), zgttrf_(
integer *, doublecomplex *, doublecomplex *, doublecomplex *,
doublecomplex *, integer *, integer *), zgttrs_(char *, integer *,
integer *, doublecomplex *, doublecomplex *, doublecomplex *,
doublecomplex *, integer *, doublecomplex *, integer *, integer *);
/* -- LAPACK routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* ZGTSVX uses the LU factorization to compute the solution to a complex */
/* system of linear equations A * X = B, A**T * X = B, or A**H * X = B, */
/* where A is a tridiagonal matrix of order N and X and B are N-by-NRHS */
/* matrices. */
/* Error bounds on the solution and a condition estimate are also */
/* provided. */
/* Description */
/* =========== */
/* The following steps are performed: */
/* 1. If FACT = 'N', the LU decomposition is used to factor the matrix A */
/* as A = L * U, where L is a product of permutation and unit lower */
/* bidiagonal matrices and U is upper triangular with nonzeros in */
/* only the main diagonal and first two superdiagonals. */
/* 2. If some U(i,i)=0, so that U is exactly singular, then the routine */
/* returns with INFO = i. Otherwise, the factored form of A is used */
/* to estimate the condition number of the matrix A. If the */
/* reciprocal of the condition number is less than machine precision, */
/* INFO = N+1 is returned as a warning, but the routine still goes on */
/* to solve for X and compute error bounds as described below. */
/* 3. The system of equations is solved for X using the factored form */
/* of A. */
/* 4. Iterative refinement is applied to improve the computed solution */
/* matrix and calculate error bounds and backward error estimates */
/* for it. */
/* Arguments */
/* ========= */
/* FACT (input) CHARACTER*1 */
/* Specifies whether or not the factored form of A has been */
/* supplied on entry. */
/* = 'F': DLF, DF, DUF, DU2, and IPIV contain the factored form */
/* of A; DL, D, DU, DLF, DF, DUF, DU2 and IPIV will not */
/* be modified. */
/* = 'N': The matrix will be copied to DLF, DF, and DUF */
/* and factored. */
/* TRANS (input) CHARACTER*1 */
/* Specifies the form of the system of equations: */
/* = 'N': A * X = B (No transpose) */
/* = 'T': A**T * X = B (Transpose) */
/* = 'C': A**H * X = B (Conjugate transpose) */
/* N (input) INTEGER */
/* The order of the matrix A. N >= 0. */
/* NRHS (input) INTEGER */
/* The number of right hand sides, i.e., the number of columns */
/* of the matrix B. NRHS >= 0. */
/* DL (input) COMPLEX*16 array, dimension (N-1) */
/* The (n-1) subdiagonal elements of A. */
/* D (input) COMPLEX*16 array, dimension (N) */
/* The n diagonal elements of A. */
/* DU (input) COMPLEX*16 array, dimension (N-1) */
/* The (n-1) superdiagonal elements of A. */
/* DLF (input or output) COMPLEX*16 array, dimension (N-1) */
/* If FACT = 'F', then DLF is an input argument and on entry */
/* contains the (n-1) multipliers that define the matrix L from */
/* the LU factorization of A as computed by ZGTTRF. */
/* If FACT = 'N', then DLF is an output argument and on exit */
/* contains the (n-1) multipliers that define the matrix L from */
/* the LU factorization of A. */
/* DF (input or output) COMPLEX*16 array, dimension (N) */
/* If FACT = 'F', then DF is an input argument and on entry */
/* contains the n diagonal elements of the upper triangular */
/* matrix U from the LU factorization of A. */
/* If FACT = 'N', then DF is an output argument and on exit */
/* contains the n diagonal elements of the upper triangular */
/* matrix U from the LU factorization of A. */
/* DUF (input or output) COMPLEX*16 array, dimension (N-1) */
/* If FACT = 'F', then DUF is an input argument and on entry */
/* contains the (n-1) elements of the first superdiagonal of U. */
/* If FACT = 'N', then DUF is an output argument and on exit */
/* contains the (n-1) elements of the first superdiagonal of U. */
/* DU2 (input or output) COMPLEX*16 array, dimension (N-2) */
/* If FACT = 'F', then DU2 is an input argument and on entry */
/* contains the (n-2) elements of the second superdiagonal of */
/* U. */
/* If FACT = 'N', then DU2 is an output argument and on exit */
/* contains the (n-2) elements of the second superdiagonal of */
/* U. */
/* IPIV (input or output) INTEGER array, dimension (N) */
/* If FACT = 'F', then IPIV is an input argument and on entry */
/* contains the pivot indices from the LU factorization of A as */
/* computed by ZGTTRF. */
/* If FACT = 'N', then IPIV is an output argument and on exit */
/* contains the pivot indices from the LU factorization of A; */
/* row i of the matrix was interchanged with row IPIV(i). */
/* IPIV(i) will always be either i or i+1; IPIV(i) = i indicates */
/* a row interchange was not required. */
/* B (input) COMPLEX*16 array, dimension (LDB,NRHS) */
/* The N-by-NRHS right hand side matrix B. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= max(1,N). */
/* X (output) COMPLEX*16 array, dimension (LDX,NRHS) */
/* If INFO = 0 or INFO = N+1, the N-by-NRHS solution matrix X. */
/* LDX (input) INTEGER */
/* The leading dimension of the array X. LDX >= max(1,N). */
/* RCOND (output) DOUBLE PRECISION */
/* The estimate of the reciprocal condition number of the matrix */
/* A. If RCOND is less than the machine precision (in */
/* particular, if RCOND = 0), the matrix is singular to working */
/* precision. This condition is indicated by a return code of */
/* INFO > 0. */
/* FERR (output) DOUBLE PRECISION array, dimension (NRHS) */
/* The estimated forward error bound for each solution vector */
/* X(j) (the j-th column of the solution matrix X). */
/* If XTRUE is the true solution corresponding to X(j), FERR(j) */
/* is an estimated upper bound for the magnitude of the largest */
/* element in (X(j) - XTRUE) divided by the magnitude of the */
/* largest element in X(j). The estimate is as reliable as */
/* the estimate for RCOND, and is almost always a slight */
/* overestimate of the true error. */
/* BERR (output) DOUBLE PRECISION array, dimension (NRHS) */
/* The componentwise relative backward error of each solution */
/* vector X(j) (i.e., the smallest relative change in */
/* any element of A or B that makes X(j) an exact solution). */
/* WORK (workspace) COMPLEX*16 array, dimension (2*N) */
/* RWORK (workspace) DOUBLE PRECISION array, dimension (N) */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* > 0: if INFO = i, and i is */
/* <= N: U(i,i) is exactly zero. The factorization */
/* has not been completed unless i = N, but the */
/* factor U is exactly singular, so the solution */
/* and error bounds could not be computed. */
/* RCOND = 0 is returned. */
/* = N+1: U is nonsingular, but RCOND is less than machine */
/* precision, meaning that the matrix is singular */
/* to working precision. Nevertheless, the */
/* solution and error bounds are computed because */
/* there are a number of situations where the */
/* computed solution can be more accurate than the */
/* value of RCOND would suggest. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Parameter adjustments */
--dl;
--d__;
--du;
--dlf;
--df;
--duf;
--du2;
--ipiv;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
x_dim1 = *ldx;
x_offset = 1 + x_dim1;
x -= x_offset;
--ferr;
--berr;
--work;
--rwork;
/* Function Body */
*info = 0;
nofact = lsame_(fact, "N");
notran = lsame_(trans, "N");
if (! nofact && ! lsame_(fact, "F")) {
*info = -1;
} else if (! notran && ! lsame_(trans, "T") && !
lsame_(trans, "C")) {
*info = -2;
} else if (*n < 0) {
*info = -3;
} else if (*nrhs < 0) {
*info = -4;
} else if (*ldb < max(1,*n)) {
*info = -14;
} else if (*ldx < max(1,*n)) {
*info = -16;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("ZGTSVX", &i__1);
return 0;
}
if (nofact) {
/* Compute the LU factorization of A. */
zcopy_(n, &d__[1], &c__1, &df[1], &c__1);
if (*n > 1) {
i__1 = *n - 1;
zcopy_(&i__1, &dl[1], &c__1, &dlf[1], &c__1);
i__1 = *n - 1;
zcopy_(&i__1, &du[1], &c__1, &duf[1], &c__1);
}
zgttrf_(n, &dlf[1], &df[1], &duf[1], &du2[1], &ipiv[1], info);
/* Return if INFO is non-zero. */
if (*info > 0) {
*rcond = 0.;
return 0;
}
}
/* Compute the norm of the matrix A. */
if (notran) {
*(unsigned char *)norm = '1';
} else {
*(unsigned char *)norm = 'I';
}
anorm = zlangt_(norm, n, &dl[1], &d__[1], &du[1]);
/* Compute the reciprocal of the condition number of A. */
zgtcon_(norm, n, &dlf[1], &df[1], &duf[1], &du2[1], &ipiv[1], &anorm,
rcond, &work[1], info);
/* Compute the solution vectors X. */
zlacpy_("Full", n, nrhs, &b[b_offset], ldb, &x[x_offset], ldx);
zgttrs_(trans, n, nrhs, &dlf[1], &df[1], &duf[1], &du2[1], &ipiv[1], &x[
x_offset], ldx, info);
/* Use iterative refinement to improve the computed solutions and */
/* compute error bounds and backward error estimates for them. */
zgtrfs_(trans, n, nrhs, &dl[1], &d__[1], &du[1], &dlf[1], &df[1], &duf[1],
&du2[1], &ipiv[1], &b[b_offset], ldb, &x[x_offset], ldx, &ferr[1]
, &berr[1], &work[1], &rwork[1], info);
/* Set INFO = N+1 if the matrix is singular to working precision. */
if (*rcond < dlamch_("Epsilon")) {
*info = *n + 1;
}
return 0;
/* End of ZGTSVX */
} /* zgtsvx_ */
/* Subroutine */ int zlarfb_(char *side, char *trans, char *direct, char *
storev, integer *m, integer *n, integer *k, doublecomplex *v, integer
*ldv, doublecomplex *t, integer *ldt, doublecomplex *c__, integer *
ldc, doublecomplex *work, integer *ldwork)
{
/* System generated locals */
integer c_dim1, c_offset, t_dim1, t_offset, v_dim1, v_offset, work_dim1,
work_offset, i__1, i__2, i__3, i__4, i__5;
doublecomplex z__1, z__2;
/* Builtin functions */
void d_cnjg(doublecomplex *, doublecomplex *);
/* Local variables */
integer i__, j;
extern logical lsame_(char *, char *);
integer lastc;
extern /* Subroutine */ int zgemm_(char *, char *, integer *, integer *,
integer *, doublecomplex *, doublecomplex *, integer *,
doublecomplex *, integer *, doublecomplex *, doublecomplex *,
integer *);
integer lastv;
extern /* Subroutine */ int zcopy_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *), ztrmm_(char *, char *, char *, char *
, integer *, integer *, doublecomplex *, doublecomplex *, integer
*, doublecomplex *, integer *);
extern integer ilazlc_(integer *, integer *, doublecomplex *, integer *);
extern /* Subroutine */ int zlacgv_(integer *, doublecomplex *, integer *)
;
extern integer ilazlr_(integer *, integer *, doublecomplex *, integer *);
char transt[1];
/* -- LAPACK auxiliary routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* ZLARFB applies a complex block reflector H or its transpose H' to a */
/* complex M-by-N matrix C, from either the left or the right. */
/* Arguments */
/* ========= */
/* SIDE (input) CHARACTER*1 */
/* = 'L': apply H or H' from the Left */
/* = 'R': apply H or H' from the Right */
/* TRANS (input) CHARACTER*1 */
/* = 'N': apply H (No transpose) */
/* = 'C': apply H' (Conjugate transpose) */
/* DIRECT (input) CHARACTER*1 */
/* Indicates how H is formed from a product of elementary */
/* reflectors */
/* = 'F': H = H(1) H(2) . . . H(k) (Forward) */
/* = 'B': H = H(k) . . . H(2) H(1) (Backward) */
/* STOREV (input) CHARACTER*1 */
/* Indicates how the vectors which define the elementary */
/* reflectors are stored: */
/* = 'C': Columnwise */
/* = 'R': Rowwise */
/* M (input) INTEGER */
/* The number of rows of the matrix C. */
/* N (input) INTEGER */
/* The number of columns of the matrix C. */
/* K (input) INTEGER */
/* The order of the matrix T (= the number of elementary */
/* reflectors whose product defines the block reflector). */
/* V (input) COMPLEX*16 array, dimension */
/* (LDV,K) if STOREV = 'C' */
/* (LDV,M) if STOREV = 'R' and SIDE = 'L' */
/* (LDV,N) if STOREV = 'R' and SIDE = 'R' */
/* The matrix V. See further details. */
/* LDV (input) INTEGER */
/* The leading dimension of the array V. */
/* If STOREV = 'C' and SIDE = 'L', LDV >= max(1,M); */
/* if STOREV = 'C' and SIDE = 'R', LDV >= max(1,N); */
/* if STOREV = 'R', LDV >= K. */
/* T (input) COMPLEX*16 array, dimension (LDT,K) */
/* The triangular K-by-K matrix T in the representation of the */
/* block reflector. */
/* LDT (input) INTEGER */
/* The leading dimension of the array T. LDT >= K. */
/* C (input/output) COMPLEX*16 array, dimension (LDC,N) */
/* On entry, the M-by-N matrix C. */
/* On exit, C is overwritten by H*C or H'*C or C*H or C*H'. */
/* LDC (input) INTEGER */
/* The leading dimension of the array C. LDC >= max(1,M). */
/* WORK (workspace) COMPLEX*16 array, dimension (LDWORK,K) */
/* LDWORK (input) INTEGER */
/* The leading dimension of the array WORK. */
/* If SIDE = 'L', LDWORK >= max(1,N); */
/* if SIDE = 'R', LDWORK >= max(1,M). */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Quick return if possible */
/* Parameter adjustments */
v_dim1 = *ldv;
v_offset = 1 + v_dim1;
v -= v_offset;
t_dim1 = *ldt;
t_offset = 1 + t_dim1;
t -= t_offset;
c_dim1 = *ldc;
c_offset = 1 + c_dim1;
c__ -= c_offset;
work_dim1 = *ldwork;
work_offset = 1 + work_dim1;
work -= work_offset;
/* Function Body */
if (*m <= 0 || *n <= 0) {
return 0;
}
if (lsame_(trans, "N")) {
*(unsigned char *)transt = 'C';
} else {
*(unsigned char *)transt = 'N';
}
if (lsame_(storev, "C")) {
if (lsame_(direct, "F")) {
/* Let V = ( V1 ) (first K rows) */
/* ( V2 ) */
/* where V1 is unit lower triangular. */
if (lsame_(side, "L")) {
/* Form H * C or H' * C where C = ( C1 ) */
/* ( C2 ) */
/* Computing MAX */
i__1 = *k, i__2 = ilazlr_(m, k, &v[v_offset], ldv);
lastv = max(i__1,i__2);
lastc = ilazlc_(&lastv, n, &c__[c_offset], ldc);
/* W := C' * V = (C1'*V1 + C2'*V2) (stored in WORK) */
/* W := C1' */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
zcopy_(&lastc, &c__[j + c_dim1], ldc, &work[j * work_dim1
+ 1], &c__1);
zlacgv_(&lastc, &work[j * work_dim1 + 1], &c__1);
/* L10: */
}
/* W := W * V1 */
ztrmm_("Right", "Lower", "No transpose", "Unit", &lastc, k, &
c_b1, &v[v_offset], ldv, &work[work_offset], ldwork);
if (lastv > *k) {
/* W := W + C2'*V2 */
i__1 = lastv - *k;
zgemm_("Conjugate transpose", "No transpose", &lastc, k, &
i__1, &c_b1, &c__[*k + 1 + c_dim1], ldc, &v[*k +
1 + v_dim1], ldv, &c_b1, &work[work_offset],
ldwork);
}
/* W := W * T' or W * T */
ztrmm_("Right", "Upper", transt, "Non-unit", &lastc, k, &c_b1,
&t[t_offset], ldt, &work[work_offset], ldwork);
/* C := C - V * W' */
if (*m > *k) {
/* C2 := C2 - V2 * W' */
i__1 = lastv - *k;
z__1.r = -1., z__1.i = -0.;
zgemm_("No transpose", "Conjugate transpose", &i__1, &
lastc, k, &z__1, &v[*k + 1 + v_dim1], ldv, &work[
work_offset], ldwork, &c_b1, &c__[*k + 1 + c_dim1]
, ldc);
}
/* W := W * V1' */
ztrmm_("Right", "Lower", "Conjugate transpose", "Unit", &
lastc, k, &c_b1, &v[v_offset], ldv, &work[work_offset]
, ldwork)
;
/* C1 := C1 - W' */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
i__2 = lastc;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = j + i__ * c_dim1;
i__4 = j + i__ * c_dim1;
d_cnjg(&z__2, &work[i__ + j * work_dim1]);
z__1.r = c__[i__4].r - z__2.r, z__1.i = c__[i__4].i -
z__2.i;
c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
/* L20: */
}
/* L30: */
}
} else if (lsame_(side, "R")) {
/* Form C * H or C * H' where C = ( C1 C2 ) */
/* Computing MAX */
i__1 = *k, i__2 = ilazlr_(n, k, &v[v_offset], ldv);
lastv = max(i__1,i__2);
lastc = ilazlr_(m, &lastv, &c__[c_offset], ldc);
/* W := C * V = (C1*V1 + C2*V2) (stored in WORK) */
/* W := C1 */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
zcopy_(&lastc, &c__[j * c_dim1 + 1], &c__1, &work[j *
work_dim1 + 1], &c__1);
/* L40: */
}
/* W := W * V1 */
ztrmm_("Right", "Lower", "No transpose", "Unit", &lastc, k, &
c_b1, &v[v_offset], ldv, &work[work_offset], ldwork);
if (lastv > *k) {
/* W := W + C2 * V2 */
i__1 = lastv - *k;
zgemm_("No transpose", "No transpose", &lastc, k, &i__1, &
c_b1, &c__[(*k + 1) * c_dim1 + 1], ldc, &v[*k + 1
+ v_dim1], ldv, &c_b1, &work[work_offset], ldwork);
}
/* W := W * T or W * T' */
ztrmm_("Right", "Upper", trans, "Non-unit", &lastc, k, &c_b1,
&t[t_offset], ldt, &work[work_offset], ldwork);
/* C := C - W * V' */
if (lastv > *k) {
/* C2 := C2 - W * V2' */
i__1 = lastv - *k;
z__1.r = -1., z__1.i = -0.;
zgemm_("No transpose", "Conjugate transpose", &lastc, &
i__1, k, &z__1, &work[work_offset], ldwork, &v[*k
+ 1 + v_dim1], ldv, &c_b1, &c__[(*k + 1) * c_dim1
+ 1], ldc);
}
/* W := W * V1' */
ztrmm_("Right", "Lower", "Conjugate transpose", "Unit", &
lastc, k, &c_b1, &v[v_offset], ldv, &work[work_offset]
, ldwork)
;
/* C1 := C1 - W */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
i__2 = lastc;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__ + j * c_dim1;
i__4 = i__ + j * c_dim1;
i__5 = i__ + j * work_dim1;
z__1.r = c__[i__4].r - work[i__5].r, z__1.i = c__[
i__4].i - work[i__5].i;
c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
/* L50: */
}
/* L60: */
}
}
} else {
/* Let V = ( V1 ) */
/* ( V2 ) (last K rows) */
/* where V2 is unit upper triangular. */
if (lsame_(side, "L")) {
/* Form H * C or H' * C where C = ( C1 ) */
/* ( C2 ) */
/* Computing MAX */
i__1 = *k, i__2 = ilazlr_(m, k, &v[v_offset], ldv);
lastv = max(i__1,i__2);
lastc = ilazlc_(&lastv, n, &c__[c_offset], ldc);
/* W := C' * V = (C1'*V1 + C2'*V2) (stored in WORK) */
/* W := C2' */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
zcopy_(&lastc, &c__[lastv - *k + j + c_dim1], ldc, &work[
j * work_dim1 + 1], &c__1);
zlacgv_(&lastc, &work[j * work_dim1 + 1], &c__1);
/* L70: */
}
/* W := W * V2 */
ztrmm_("Right", "Upper", "No transpose", "Unit", &lastc, k, &
c_b1, &v[lastv - *k + 1 + v_dim1], ldv, &work[
work_offset], ldwork);
if (lastv > *k) {
/* W := W + C1'*V1 */
i__1 = lastv - *k;
zgemm_("Conjugate transpose", "No transpose", &lastc, k, &
i__1, &c_b1, &c__[c_offset], ldc, &v[v_offset],
ldv, &c_b1, &work[work_offset], ldwork);
}
/* W := W * T' or W * T */
ztrmm_("Right", "Lower", transt, "Non-unit", &lastc, k, &c_b1,
&t[t_offset], ldt, &work[work_offset], ldwork);
/* C := C - V * W' */
if (lastv > *k) {
/* C1 := C1 - V1 * W' */
i__1 = lastv - *k;
z__1.r = -1., z__1.i = -0.;
zgemm_("No transpose", "Conjugate transpose", &i__1, &
lastc, k, &z__1, &v[v_offset], ldv, &work[
work_offset], ldwork, &c_b1, &c__[c_offset], ldc);
}
/* W := W * V2' */
ztrmm_("Right", "Upper", "Conjugate transpose", "Unit", &
lastc, k, &c_b1, &v[lastv - *k + 1 + v_dim1], ldv, &
work[work_offset], ldwork);
/* C2 := C2 - W' */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
i__2 = lastc;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = lastv - *k + j + i__ * c_dim1;
i__4 = lastv - *k + j + i__ * c_dim1;
d_cnjg(&z__2, &work[i__ + j * work_dim1]);
z__1.r = c__[i__4].r - z__2.r, z__1.i = c__[i__4].i -
z__2.i;
c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
/* L80: */
}
/* L90: */
}
} else if (lsame_(side, "R")) {
/* Form C * H or C * H' where C = ( C1 C2 ) */
/* Computing MAX */
i__1 = *k, i__2 = ilazlr_(n, k, &v[v_offset], ldv);
lastv = max(i__1,i__2);
lastc = ilazlr_(m, &lastv, &c__[c_offset], ldc);
/* W := C * V = (C1*V1 + C2*V2) (stored in WORK) */
/* W := C2 */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
zcopy_(&lastc, &c__[(lastv - *k + j) * c_dim1 + 1], &c__1,
&work[j * work_dim1 + 1], &c__1);
/* L100: */
}
/* W := W * V2 */
ztrmm_("Right", "Upper", "No transpose", "Unit", &lastc, k, &
c_b1, &v[lastv - *k + 1 + v_dim1], ldv, &work[
work_offset], ldwork);
if (lastv > *k) {
/* W := W + C1 * V1 */
i__1 = lastv - *k;
zgemm_("No transpose", "No transpose", &lastc, k, &i__1, &
c_b1, &c__[c_offset], ldc, &v[v_offset], ldv, &
c_b1, &work[work_offset], ldwork);
}
/* W := W * T or W * T' */
ztrmm_("Right", "Lower", trans, "Non-unit", &lastc, k, &c_b1,
&t[t_offset], ldt, &work[work_offset], ldwork);
/* C := C - W * V' */
if (lastv > *k) {
/* C1 := C1 - W * V1' */
i__1 = lastv - *k;
z__1.r = -1., z__1.i = -0.;
zgemm_("No transpose", "Conjugate transpose", &lastc, &
i__1, k, &z__1, &work[work_offset], ldwork, &v[
v_offset], ldv, &c_b1, &c__[c_offset], ldc);
}
/* W := W * V2' */
ztrmm_("Right", "Upper", "Conjugate transpose", "Unit", &
lastc, k, &c_b1, &v[lastv - *k + 1 + v_dim1], ldv, &
work[work_offset], ldwork);
/* C2 := C2 - W */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
i__2 = lastc;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__ + (lastv - *k + j) * c_dim1;
i__4 = i__ + (lastv - *k + j) * c_dim1;
i__5 = i__ + j * work_dim1;
z__1.r = c__[i__4].r - work[i__5].r, z__1.i = c__[
i__4].i - work[i__5].i;
c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
/* L110: */
}
/* L120: */
}
}
}
} else if (lsame_(storev, "R")) {
if (lsame_(direct, "F")) {
/* Let V = ( V1 V2 ) (V1: first K columns) */
/* where V1 is unit upper triangular. */
if (lsame_(side, "L")) {
/* Form H * C or H' * C where C = ( C1 ) */
/* ( C2 ) */
/* Computing MAX */
i__1 = *k, i__2 = ilazlc_(k, m, &v[v_offset], ldv);
lastv = max(i__1,i__2);
lastc = ilazlc_(&lastv, n, &c__[c_offset], ldc);
/* W := C' * V' = (C1'*V1' + C2'*V2') (stored in WORK) */
/* W := C1' */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
zcopy_(&lastc, &c__[j + c_dim1], ldc, &work[j * work_dim1
+ 1], &c__1);
zlacgv_(&lastc, &work[j * work_dim1 + 1], &c__1);
/* L130: */
}
/* W := W * V1' */
ztrmm_("Right", "Upper", "Conjugate transpose", "Unit", &
lastc, k, &c_b1, &v[v_offset], ldv, &work[work_offset]
, ldwork)
;
if (lastv > *k) {
/* W := W + C2'*V2' */
i__1 = lastv - *k;
zgemm_("Conjugate transpose", "Conjugate transpose", &
lastc, k, &i__1, &c_b1, &c__[*k + 1 + c_dim1],
ldc, &v[(*k + 1) * v_dim1 + 1], ldv, &c_b1, &work[
work_offset], ldwork);
}
/* W := W * T' or W * T */
ztrmm_("Right", "Upper", transt, "Non-unit", &lastc, k, &c_b1,
&t[t_offset], ldt, &work[work_offset], ldwork);
/* C := C - V' * W' */
if (lastv > *k) {
/* C2 := C2 - V2' * W' */
i__1 = lastv - *k;
z__1.r = -1., z__1.i = -0.;
zgemm_("Conjugate transpose", "Conjugate transpose", &
i__1, &lastc, k, &z__1, &v[(*k + 1) * v_dim1 + 1],
ldv, &work[work_offset], ldwork, &c_b1, &c__[*k
+ 1 + c_dim1], ldc);
}
/* W := W * V1 */
ztrmm_("Right", "Upper", "No transpose", "Unit", &lastc, k, &
c_b1, &v[v_offset], ldv, &work[work_offset], ldwork);
/* C1 := C1 - W' */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
i__2 = lastc;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = j + i__ * c_dim1;
i__4 = j + i__ * c_dim1;
d_cnjg(&z__2, &work[i__ + j * work_dim1]);
z__1.r = c__[i__4].r - z__2.r, z__1.i = c__[i__4].i -
z__2.i;
c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
/* L140: */
}
/* L150: */
}
} else if (lsame_(side, "R")) {
/* Form C * H or C * H' where C = ( C1 C2 ) */
/* Computing MAX */
i__1 = *k, i__2 = ilazlc_(k, n, &v[v_offset], ldv);
lastv = max(i__1,i__2);
lastc = ilazlr_(m, &lastv, &c__[c_offset], ldc);
/* W := C * V' = (C1*V1' + C2*V2') (stored in WORK) */
/* W := C1 */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
zcopy_(&lastc, &c__[j * c_dim1 + 1], &c__1, &work[j *
work_dim1 + 1], &c__1);
/* L160: */
}
/* W := W * V1' */
ztrmm_("Right", "Upper", "Conjugate transpose", "Unit", &
lastc, k, &c_b1, &v[v_offset], ldv, &work[work_offset]
, ldwork)
;
if (lastv > *k) {
/* W := W + C2 * V2' */
i__1 = lastv - *k;
zgemm_("No transpose", "Conjugate transpose", &lastc, k, &
i__1, &c_b1, &c__[(*k + 1) * c_dim1 + 1], ldc, &v[
(*k + 1) * v_dim1 + 1], ldv, &c_b1, &work[
work_offset], ldwork);
}
/* W := W * T or W * T' */
ztrmm_("Right", "Upper", trans, "Non-unit", &lastc, k, &c_b1,
&t[t_offset], ldt, &work[work_offset], ldwork);
/* C := C - W * V */
if (lastv > *k) {
/* C2 := C2 - W * V2 */
i__1 = lastv - *k;
z__1.r = -1., z__1.i = -0.;
zgemm_("No transpose", "No transpose", &lastc, &i__1, k, &
z__1, &work[work_offset], ldwork, &v[(*k + 1) *
v_dim1 + 1], ldv, &c_b1, &c__[(*k + 1) * c_dim1 +
1], ldc);
}
/* W := W * V1 */
ztrmm_("Right", "Upper", "No transpose", "Unit", &lastc, k, &
c_b1, &v[v_offset], ldv, &work[work_offset], ldwork);
/* C1 := C1 - W */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
i__2 = lastc;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__ + j * c_dim1;
i__4 = i__ + j * c_dim1;
i__5 = i__ + j * work_dim1;
z__1.r = c__[i__4].r - work[i__5].r, z__1.i = c__[
i__4].i - work[i__5].i;
c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
/* L170: */
}
/* L180: */
}
}
} else {
/* Let V = ( V1 V2 ) (V2: last K columns) */
/* where V2 is unit lower triangular. */
if (lsame_(side, "L")) {
/* Form H * C or H' * C where C = ( C1 ) */
/* ( C2 ) */
/* Computing MAX */
i__1 = *k, i__2 = ilazlc_(k, m, &v[v_offset], ldv);
lastv = max(i__1,i__2);
lastc = ilazlc_(&lastv, n, &c__[c_offset], ldc);
/* W := C' * V' = (C1'*V1' + C2'*V2') (stored in WORK) */
/* W := C2' */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
zcopy_(&lastc, &c__[lastv - *k + j + c_dim1], ldc, &work[
j * work_dim1 + 1], &c__1);
zlacgv_(&lastc, &work[j * work_dim1 + 1], &c__1);
/* L190: */
}
/* W := W * V2' */
ztrmm_("Right", "Lower", "Conjugate transpose", "Unit", &
lastc, k, &c_b1, &v[(lastv - *k + 1) * v_dim1 + 1],
ldv, &work[work_offset], ldwork);
if (lastv > *k) {
/* W := W + C1'*V1' */
i__1 = lastv - *k;
zgemm_("Conjugate transpose", "Conjugate transpose", &
lastc, k, &i__1, &c_b1, &c__[c_offset], ldc, &v[
v_offset], ldv, &c_b1, &work[work_offset], ldwork);
}
/* W := W * T' or W * T */
ztrmm_("Right", "Lower", transt, "Non-unit", &lastc, k, &c_b1,
&t[t_offset], ldt, &work[work_offset], ldwork);
/* C := C - V' * W' */
if (lastv > *k) {
/* C1 := C1 - V1' * W' */
i__1 = lastv - *k;
z__1.r = -1., z__1.i = -0.;
zgemm_("Conjugate transpose", "Conjugate transpose", &
i__1, &lastc, k, &z__1, &v[v_offset], ldv, &work[
work_offset], ldwork, &c_b1, &c__[c_offset], ldc);
}
/* W := W * V2 */
ztrmm_("Right", "Lower", "No transpose", "Unit", &lastc, k, &
c_b1, &v[(lastv - *k + 1) * v_dim1 + 1], ldv, &work[
work_offset], ldwork);
/* C2 := C2 - W' */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
i__2 = lastc;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = lastv - *k + j + i__ * c_dim1;
i__4 = lastv - *k + j + i__ * c_dim1;
d_cnjg(&z__2, &work[i__ + j * work_dim1]);
z__1.r = c__[i__4].r - z__2.r, z__1.i = c__[i__4].i -
z__2.i;
c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
/* L200: */
}
/* L210: */
}
} else if (lsame_(side, "R")) {
/* Form C * H or C * H' where C = ( C1 C2 ) */
/* Computing MAX */
i__1 = *k, i__2 = ilazlc_(k, n, &v[v_offset], ldv);
lastv = max(i__1,i__2);
lastc = ilazlr_(m, &lastv, &c__[c_offset], ldc);
/* W := C * V' = (C1*V1' + C2*V2') (stored in WORK) */
/* W := C2 */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
zcopy_(&lastc, &c__[(lastv - *k + j) * c_dim1 + 1], &c__1,
&work[j * work_dim1 + 1], &c__1);
/* L220: */
}
/* W := W * V2' */
ztrmm_("Right", "Lower", "Conjugate transpose", "Unit", &
lastc, k, &c_b1, &v[(lastv - *k + 1) * v_dim1 + 1],
ldv, &work[work_offset], ldwork);
if (lastv > *k) {
/* W := W + C1 * V1' */
i__1 = lastv - *k;
zgemm_("No transpose", "Conjugate transpose", &lastc, k, &
i__1, &c_b1, &c__[c_offset], ldc, &v[v_offset],
ldv, &c_b1, &work[work_offset], ldwork);
}
/* W := W * T or W * T' */
ztrmm_("Right", "Lower", trans, "Non-unit", &lastc, k, &c_b1,
&t[t_offset], ldt, &work[work_offset], ldwork);
/* C := C - W * V */
if (lastv > *k) {
/* C1 := C1 - W * V1 */
i__1 = lastv - *k;
z__1.r = -1., z__1.i = -0.;
zgemm_("No transpose", "No transpose", &lastc, &i__1, k, &
z__1, &work[work_offset], ldwork, &v[v_offset],
ldv, &c_b1, &c__[c_offset], ldc);
}
/* W := W * V2 */
ztrmm_("Right", "Lower", "No transpose", "Unit", &lastc, k, &
c_b1, &v[(lastv - *k + 1) * v_dim1 + 1], ldv, &work[
work_offset], ldwork);
/* C1 := C1 - W */
i__1 = *k;
for (j = 1; j <= i__1; ++j) {
i__2 = lastc;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__ + (lastv - *k + j) * c_dim1;
i__4 = i__ + (lastv - *k + j) * c_dim1;
i__5 = i__ + j * work_dim1;
z__1.r = c__[i__4].r - work[i__5].r, z__1.i = c__[
i__4].i - work[i__5].i;
c__[i__3].r = z__1.r, c__[i__3].i = z__1.i;
/* L230: */
}
/* L240: */
}
}
}
}
return 0;
/* End of ZLARFB */
} /* zlarfb_ */
/* Subroutine */
int zpbsvx_(char *fact, char *uplo, integer *n, integer *kd, integer *nrhs, doublecomplex *ab, integer *ldab, doublecomplex *afb, integer *ldafb, char *equed, doublereal *s, doublecomplex *b, integer *ldb, doublecomplex *x, integer *ldx, doublereal *rcond, doublereal * ferr, doublereal *berr, doublecomplex *work, doublereal *rwork, integer *info)
{
/* System generated locals */
integer ab_dim1, ab_offset, afb_dim1, afb_offset, b_dim1, b_offset, x_dim1, x_offset, i__1, i__2, i__3, i__4, i__5;
doublereal d__1, d__2;
doublecomplex z__1;
/* Local variables */
integer i__, j, j1, j2;
doublereal amax, smin, smax;
extern logical lsame_(char *, char *);
doublereal scond, anorm;
logical equil, rcequ, upper;
extern /* Subroutine */
int zcopy_(integer *, doublecomplex *, integer *, doublecomplex *, integer *);
extern doublereal dlamch_(char *);
logical nofact;
extern /* Subroutine */
int xerbla_(char *, integer *);
extern doublereal zlanhb_(char *, char *, integer *, integer *, doublecomplex *, integer *, doublereal *);
doublereal bignum;
extern /* Subroutine */
int zlaqhb_(char *, integer *, integer *, doublecomplex *, integer *, doublereal *, doublereal *, doublereal *, char *);
integer infequ;
extern /* Subroutine */
int zpbcon_(char *, integer *, integer *, doublecomplex *, integer *, doublereal *, doublereal *, doublecomplex *, doublereal *, integer *), zlacpy_(char *, integer *, integer *, doublecomplex *, integer *, doublecomplex * , integer *), zpbequ_(char *, integer *, integer *, doublecomplex *, integer *, doublereal *, doublereal *, doublereal *, integer *), zpbrfs_(char *, integer *, integer *, integer *, doublecomplex *, integer *, doublecomplex *, integer *, doublecomplex *, integer *, doublecomplex *, integer * , doublereal *, doublereal *, doublecomplex *, doublereal *, integer *), zpbtrf_(char *, integer *, integer *, doublecomplex *, integer *, integer *);
doublereal smlnum;
extern /* Subroutine */
int zpbtrs_(char *, integer *, integer *, integer *, doublecomplex *, integer *, doublecomplex *, integer *, integer *);
/* -- LAPACK driver routine (version 3.4.1) -- */
/* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
/* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
/* April 2012 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Parameter adjustments */
ab_dim1 = *ldab;
ab_offset = 1 + ab_dim1;
ab -= ab_offset;
afb_dim1 = *ldafb;
afb_offset = 1 + afb_dim1;
afb -= afb_offset;
--s;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
x_dim1 = *ldx;
x_offset = 1 + x_dim1;
x -= x_offset;
--ferr;
--berr;
--work;
--rwork;
/* Function Body */
*info = 0;
nofact = lsame_(fact, "N");
equil = lsame_(fact, "E");
upper = lsame_(uplo, "U");
if (nofact || equil)
{
*(unsigned char *)equed = 'N';
rcequ = FALSE_;
}
else
{
rcequ = lsame_(equed, "Y");
smlnum = dlamch_("Safe minimum");
bignum = 1. / smlnum;
}
/* Test the input parameters. */
if (! nofact && ! equil && ! lsame_(fact, "F"))
{
*info = -1;
}
else if (! upper && ! lsame_(uplo, "L"))
{
*info = -2;
}
else if (*n < 0)
{
*info = -3;
}
else if (*kd < 0)
{
*info = -4;
}
else if (*nrhs < 0)
{
*info = -5;
}
else if (*ldab < *kd + 1)
{
*info = -7;
}
else if (*ldafb < *kd + 1)
{
*info = -9;
}
else if (lsame_(fact, "F") && ! (rcequ || lsame_( equed, "N")))
{
*info = -10;
}
else
{
if (rcequ)
{
smin = bignum;
smax = 0.;
i__1 = *n;
for (j = 1;
j <= i__1;
++j)
{
/* Computing MIN */
d__1 = smin;
d__2 = s[j]; // , expr subst
smin = min(d__1,d__2);
/* Computing MAX */
d__1 = smax;
d__2 = s[j]; // , expr subst
smax = max(d__1,d__2);
/* L10: */
}
if (smin <= 0.)
{
*info = -11;
}
else if (*n > 0)
{
scond = max(smin,smlnum) / min(smax,bignum);
}
else
{
scond = 1.;
}
}
if (*info == 0)
{
if (*ldb < max(1,*n))
{
*info = -13;
}
else if (*ldx < max(1,*n))
{
*info = -15;
}
}
}
if (*info != 0)
{
i__1 = -(*info);
xerbla_("ZPBSVX", &i__1);
return 0;
}
if (equil)
{
/* Compute row and column scalings to equilibrate the matrix A. */
zpbequ_(uplo, n, kd, &ab[ab_offset], ldab, &s[1], &scond, &amax, & infequ);
if (infequ == 0)
{
/* Equilibrate the matrix. */
zlaqhb_(uplo, n, kd, &ab[ab_offset], ldab, &s[1], &scond, &amax, equed);
rcequ = lsame_(equed, "Y");
}
}
/* Scale the right-hand side. */
if (rcequ)
{
i__1 = *nrhs;
for (j = 1;
j <= i__1;
++j)
{
i__2 = *n;
for (i__ = 1;
i__ <= i__2;
++i__)
{
i__3 = i__ + j * b_dim1;
i__4 = i__;
i__5 = i__ + j * b_dim1;
z__1.r = s[i__4] * b[i__5].r;
z__1.i = s[i__4] * b[i__5].i; // , expr subst
b[i__3].r = z__1.r;
b[i__3].i = z__1.i; // , expr subst
/* L20: */
}
/* L30: */
}
}
if (nofact || equil)
{
/* Compute the Cholesky factorization A = U**H *U or A = L*L**H. */
if (upper)
{
i__1 = *n;
for (j = 1;
j <= i__1;
++j)
{
/* Computing MAX */
i__2 = j - *kd;
j1 = max(i__2,1);
i__2 = j - j1 + 1;
zcopy_(&i__2, &ab[*kd + 1 - j + j1 + j * ab_dim1], &c__1, & afb[*kd + 1 - j + j1 + j * afb_dim1], &c__1);
/* L40: */
}
}
else
{
i__1 = *n;
for (j = 1;
j <= i__1;
++j)
{
/* Computing MIN */
i__2 = j + *kd;
j2 = min(i__2,*n);
i__2 = j2 - j + 1;
zcopy_(&i__2, &ab[j * ab_dim1 + 1], &c__1, &afb[j * afb_dim1 + 1], &c__1);
/* L50: */
}
}
zpbtrf_(uplo, n, kd, &afb[afb_offset], ldafb, info);
/* Return if INFO is non-zero. */
if (*info > 0)
{
*rcond = 0.;
return 0;
}
}
/* Compute the norm of the matrix A. */
anorm = zlanhb_("1", uplo, n, kd, &ab[ab_offset], ldab, &rwork[1]);
/* Compute the reciprocal of the condition number of A. */
zpbcon_(uplo, n, kd, &afb[afb_offset], ldafb, &anorm, rcond, &work[1], & rwork[1], info);
/* Compute the solution matrix X. */
zlacpy_("Full", n, nrhs, &b[b_offset], ldb, &x[x_offset], ldx);
zpbtrs_(uplo, n, kd, nrhs, &afb[afb_offset], ldafb, &x[x_offset], ldx, info);
/* Use iterative refinement to improve the computed solution and */
/* compute error bounds and backward error estimates for it. */
zpbrfs_(uplo, n, kd, nrhs, &ab[ab_offset], ldab, &afb[afb_offset], ldafb, &b[b_offset], ldb, &x[x_offset], ldx, &ferr[1], &berr[1], &work[1] , &rwork[1], info);
/* Transform the solution matrix X to a solution of the original */
/* system. */
if (rcequ)
{
i__1 = *nrhs;
for (j = 1;
j <= i__1;
++j)
{
i__2 = *n;
for (i__ = 1;
i__ <= i__2;
++i__)
{
i__3 = i__ + j * x_dim1;
i__4 = i__;
i__5 = i__ + j * x_dim1;
z__1.r = s[i__4] * x[i__5].r;
z__1.i = s[i__4] * x[i__5].i; // , expr subst
x[i__3].r = z__1.r;
x[i__3].i = z__1.i; // , expr subst
/* L60: */
}
/* L70: */
}
i__1 = *nrhs;
for (j = 1;
j <= i__1;
++j)
{
ferr[j] /= scond;
/* L80: */
}
}
/* Set INFO = N+1 if the matrix is singular to working precision. */
if (*rcond < dlamch_("Epsilon"))
{
*info = *n + 1;
}
return 0;
/* End of ZPBSVX */
}
/* Subroutine */ int zggglm_(integer *n, integer *m, integer *p,
doublecomplex *a, integer *lda, doublecomplex *b, integer *ldb,
doublecomplex *d, doublecomplex *x, doublecomplex *y, doublecomplex *
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
=======
ZGGGLM 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) COMPLEX*16 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) COMPLEX*16 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) COMPLEX*16 array, dimension (N)
On entry, D is the left hand side of the GLM equation.
On exit, D is destroyed.
X (output) COMPLEX*16 array, dimension (M)
Y (output) COMPLEX*16 array, dimension (P)
On exit, X and Y are the solutions of the GLM problem.
WORK (workspace/output) COMPLEX*16 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
ZGEQRF, CGERQF, ZUNMQR and CUNMRQ.
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 doublecomplex c_b2 = {1.,0.};
static integer c__1 = 1;
/* System generated locals */
integer a_dim1, a_offset, b_dim1, b_offset, i__1, i__2, i__3, i__4;
doublereal d__1;
doublecomplex z__1;
/* Local variables */
static integer lopt, i;
extern /* Subroutine */ int zgemv_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublecomplex *, doublecomplex *, integer *),
zcopy_(integer *, doublecomplex *, integer *, doublecomplex *,
integer *), ztrsv_(char *, char *, char *, integer *,
doublecomplex *, integer *, doublecomplex *, integer *);
static integer np;
extern /* Subroutine */ int xerbla_(char *, integer *), zggqrf_(
integer *, integer *, integer *, doublecomplex *, integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
doublecomplex *, integer *, integer *), zunmqr_(char *, char *,
integer *, integer *, integer *, doublecomplex *, integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, integer *), zunmrq_(char *, char *,
integer *, integer *, integer *, doublecomplex *, integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, integer *);
#define D(I) d[(I)-1]
#define X(I) x[(I)-1]
#define Y(I) y[(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;
np = min(*n,*p);
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)) {
*info = -12;
}
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("ZGGGLM", &i__1);
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
unitary. */
i__1 = *lwork - *m - np;
zggqrf_(n, m, p, &A(1,1), lda, &WORK(1), &B(1,1), ldb, &WORK(*m
+ 1), &WORK(*m + np + 1), &i__1, info);
i__1 = *m + np + 1;
lopt = (integer) WORK(*m+np+1).r;
/* Update left-hand-side vector d = Q'*d = ( d1 ) M
( d2 ) N-M */
i__1 = max(1,*n);
i__2 = *lwork - *m - np;
zunmqr_("Left", "Conjugate transpose", n, &c__1, m, &A(1,1), lda, &
WORK(1), &D(1), &i__1, &WORK(*m + np + 1), &i__2, info);
/* Computing MAX */
i__3 = *m + np + 1;
i__1 = lopt, i__2 = (integer) WORK(*m+np+1).r;
lopt = max(i__1,i__2);
/* Solve T22*y2 = d2 for y2 */
i__1 = *n - *m;
ztrsv_("Upper", "No transpose", "Non unit", &i__1, &B(*m+1,*m+*p-*n+1), ldb, &D(*m + 1), &c__1);
i__1 = *n - *m;
zcopy_(&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 <= *m+*p-*n; ++i) {
i__2 = i;
Y(i).r = 0., Y(i).i = 0.;
/* L10: */
}
/* Update d1 = d1 - T12*y2 */
i__1 = *n - *m;
z__1.r = -1., z__1.i = 0.;
zgemv_("No transpose", m, &i__1, &z__1, &B(1,*m+*p-*n+1), ldb, &Y(*m + *p - *n + 1), &c__1, &c_b2, &D(1), &c__1);
/* Solve triangular system: R11*x = d1 */
ztrsv_("Upper", "No Transpose", "Non unit", m, &A(1,1), lda, &D(1), &
c__1);
/* Copy D to X */
zcopy_(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;
zunmrq_("Left", "Conjugate transpose", p, &c__1, &np, &B(max(1,*n-*p+1),1), ldb, &WORK(*m + 1), &Y(1), &i__3, &WORK(*m + np + 1), &
i__4, info);
/* Computing MAX */
i__3 = *m + np + 1;
i__1 = lopt, i__2 = (integer) WORK(*m+np+1).r;
d__1 = (doublereal) max(i__1,i__2);
WORK(1).r = d__1, WORK(1).i = 0.;
return 0;
/* End of ZGGGLM */
} /* zggglm_ */
/* Subroutine */ int zlacn2_(integer *n, doublecomplex *v, doublecomplex *x,
doublereal *est, integer *kase, integer *isave)
{
/* System generated locals */
integer i__1, i__2, i__3;
doublereal d__1, d__2;
doublecomplex z__1;
/* Builtin functions */
double z_abs(doublecomplex *), d_imag(doublecomplex *);
/* Local variables */
integer i__;
doublereal temp, absxi;
integer jlast;
extern /* Subroutine */ int zcopy_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *);
extern integer izmax1_(integer *, doublecomplex *, integer *);
extern doublereal dzsum1_(integer *, doublecomplex *, integer *), dlamch_(
char *);
doublereal safmin, altsgn, estold;
/* -- LAPACK auxiliary routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* ZLACN2 estimates the 1-norm of a square, complex matrix A. */
/* Reverse communication is used for evaluating matrix-vector products. */
/* Arguments */
/* ========= */
/* N (input) INTEGER */
/* The order of the matrix. N >= 1. */
/* V (workspace) COMPLEX*16 array, dimension (N) */
/* On the final return, V = A*W, where EST = norm(V)/norm(W) */
/* (W is not returned). */
/* X (input/output) COMPLEX*16 array, dimension (N) */
/* On an intermediate return, X should be overwritten by */
/* A * X, if KASE=1, */
/* A' * X, if KASE=2, */
/* where A' is the conjugate transpose of A, and ZLACN2 must be */
/* re-called with all the other parameters unchanged. */
/* EST (input/output) DOUBLE PRECISION */
/* On entry with KASE = 1 or 2 and ISAVE(1) = 3, EST should be */
/* unchanged from the previous call to ZLACN2. */
/* On exit, EST is an estimate (a lower bound) for norm(A). */
/* KASE (input/output) INTEGER */
/* On the initial call to ZLACN2, KASE should be 0. */
/* On an intermediate return, KASE will be 1 or 2, indicating */
/* whether X should be overwritten by A * X or A' * X. */
/* On the final return from ZLACN2, KASE will again be 0. */
/* ISAVE (input/output) INTEGER array, dimension (3) */
/* ISAVE is used to save variables between calls to ZLACN2 */
/* Further Details */
/* ======= ======= */
/* Contributed by Nick Higham, University of Manchester. */
/* Originally named CONEST, dated March 16, 1988. */
/* Reference: N.J. Higham, "FORTRAN codes for estimating the one-norm of */
/* a real or complex matrix, with applications to condition estimation", */
/* ACM Trans. Math. Soft., vol. 14, no. 4, pp. 381-396, December 1988. */
/* Last modified: April, 1999 */
/* This is a thread safe version of ZLACON, which uses the array ISAVE */
/* in place of a SAVE statement, as follows: */
/* ZLACON ZLACN2 */
/* JUMP ISAVE(1) */
/* J ISAVE(2) */
/* ITER ISAVE(3) */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Parameter adjustments */
--isave;
--x;
--v;
/* Function Body */
safmin = dlamch_("Safe minimum");
if (*kase == 0) {
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = i__;
d__1 = 1. / (doublereal) (*n);
z__1.r = d__1, z__1.i = 0.;
x[i__2].r = z__1.r, x[i__2].i = z__1.i;
/* L10: */
}
*kase = 1;
isave[1] = 1;
return 0;
}
switch (isave[1]) {
case 1: goto L20;
case 2: goto L40;
case 3: goto L70;
case 4: goto L90;
case 5: goto L120;
}
/* ................ ENTRY (ISAVE( 1 ) = 1) */
/* FIRST ITERATION. X HAS BEEN OVERWRITTEN BY A*X. */
L20:
if (*n == 1) {
v[1].r = x[1].r, v[1].i = x[1].i;
*est = z_abs(&v[1]);
/* ... QUIT */
goto L130;
}
*est = dzsum1_(n, &x[1], &c__1);
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
absxi = z_abs(&x[i__]);
if (absxi > safmin) {
i__2 = i__;
i__3 = i__;
d__1 = x[i__3].r / absxi;
d__2 = d_imag(&x[i__]) / absxi;
z__1.r = d__1, z__1.i = d__2;
x[i__2].r = z__1.r, x[i__2].i = z__1.i;
} else {
i__2 = i__;
x[i__2].r = 1., x[i__2].i = 0.;
}
/* L30: */
}
*kase = 2;
isave[1] = 2;
return 0;
/* ................ ENTRY (ISAVE( 1 ) = 2) */
/* FIRST ITERATION. X HAS BEEN OVERWRITTEN BY CTRANS(A)*X. */
L40:
isave[2] = izmax1_(n, &x[1], &c__1);
isave[3] = 2;
/* MAIN LOOP - ITERATIONS 2,3,...,ITMAX. */
L50:
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = i__;
x[i__2].r = 0., x[i__2].i = 0.;
/* L60: */
}
i__1 = isave[2];
x[i__1].r = 1., x[i__1].i = 0.;
*kase = 1;
isave[1] = 3;
return 0;
/* ................ ENTRY (ISAVE( 1 ) = 3) */
/* X HAS BEEN OVERWRITTEN BY A*X. */
L70:
zcopy_(n, &x[1], &c__1, &v[1], &c__1);
estold = *est;
*est = dzsum1_(n, &v[1], &c__1);
/* TEST FOR CYCLING. */
if (*est <= estold) {
goto L100;
}
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
absxi = z_abs(&x[i__]);
if (absxi > safmin) {
i__2 = i__;
i__3 = i__;
d__1 = x[i__3].r / absxi;
d__2 = d_imag(&x[i__]) / absxi;
z__1.r = d__1, z__1.i = d__2;
x[i__2].r = z__1.r, x[i__2].i = z__1.i;
} else {
i__2 = i__;
x[i__2].r = 1., x[i__2].i = 0.;
}
/* L80: */
}
*kase = 2;
isave[1] = 4;
return 0;
/* ................ ENTRY (ISAVE( 1 ) = 4) */
/* X HAS BEEN OVERWRITTEN BY CTRANS(A)*X. */
L90:
jlast = isave[2];
isave[2] = izmax1_(n, &x[1], &c__1);
if (z_abs(&x[jlast]) != z_abs(&x[isave[2]]) && isave[3] < 5) {
++isave[3];
goto L50;
}
/* ITERATION COMPLETE. FINAL STAGE. */
L100:
altsgn = 1.;
i__1 = *n;
for (i__ = 1; i__ <= i__1; ++i__) {
i__2 = i__;
d__1 = altsgn * ((doublereal) (i__ - 1) / (doublereal) (*n - 1) + 1.);
z__1.r = d__1, z__1.i = 0.;
x[i__2].r = z__1.r, x[i__2].i = z__1.i;
altsgn = -altsgn;
/* L110: */
}
*kase = 1;
isave[1] = 5;
return 0;
/* ................ ENTRY (ISAVE( 1 ) = 5) */
/* X HAS BEEN OVERWRITTEN BY A*X. */
L120:
temp = dzsum1_(n, &x[1], &c__1) / (doublereal) (*n * 3) * 2.;
if (temp > *est) {
zcopy_(n, &x[1], &c__1, &v[1], &c__1);
*est = temp;
}
L130:
*kase = 0;
return 0;
/* End of ZLACN2 */
} /* zlacn2_ */
/* Subroutine */ int zporfs_(char *uplo, integer *n, integer *nrhs,
doublecomplex *a, integer *lda, doublecomplex *af, integer *ldaf,
doublecomplex *b, integer *ldb, doublecomplex *x, integer *ldx,
doublereal *ferr, doublereal *berr, doublecomplex *work, doublereal *
rwork, integer *info)
{
/* System generated locals */
integer a_dim1, a_offset, af_dim1, af_offset, b_dim1, b_offset, x_dim1,
x_offset, i__1, i__2, i__3, i__4, i__5;
doublereal d__1, d__2, d__3, d__4;
doublecomplex z__1;
/* Local variables */
integer i__, j, k;
doublereal s, xk;
integer nz;
doublereal eps;
integer kase;
doublereal safe1, safe2;
integer isave[3], count;
logical upper;
doublereal safmin;
doublereal lstres;
/* -- LAPACK routine (version 3.2) -- */
/* November 2006 */
/* Modified to call ZLACN2 in place of ZLACON, 10 Feb 03, SJH. */
/* Purpose */
/* ======= */
/* ZPORFS improves the computed solution to a system of linear */
/* equations when the coefficient matrix is Hermitian positive definite, */
/* and provides error bounds and backward error estimates for the */
/* solution. */
/* Arguments */
/* ========= */
/* UPLO (input) CHARACTER*1 */
/* = 'U': Upper triangle of A is stored; */
/* = 'L': Lower triangle of A is stored. */
/* N (input) INTEGER */
/* The order of the matrix A. N >= 0. */
/* NRHS (input) INTEGER */
/* The number of right hand sides, i.e., the number of columns */
/* of the matrices B and X. NRHS >= 0. */
/* A (input) COMPLEX*16 array, dimension (LDA,N) */
/* The Hermitian matrix A. If UPLO = 'U', the leading N-by-N */
/* upper triangular part of A contains the upper triangular part */
/* of the matrix A, and the strictly lower triangular part of A */
/* is not referenced. If UPLO = 'L', the leading N-by-N lower */
/* triangular part of A contains the lower triangular part of */
/* the matrix A, and the strictly upper triangular part of A is */
/* not referenced. */
/* LDA (input) INTEGER */
/* The leading dimension of the array A. LDA >= max(1,N). */
/* AF (input) COMPLEX*16 array, dimension (LDAF,N) */
/* The triangular factor U or L from the Cholesky factorization */
/* A = U**H*U or A = L*L**H, as computed by ZPOTRF. */
/* LDAF (input) INTEGER */
/* The leading dimension of the array AF. LDAF >= max(1,N). */
/* B (input) COMPLEX*16 array, dimension (LDB,NRHS) */
/* The right hand side matrix B. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= max(1,N). */
/* X (input/output) COMPLEX*16 array, dimension (LDX,NRHS) */
/* On entry, the solution matrix X, as computed by ZPOTRS. */
/* On exit, the improved solution matrix X. */
/* LDX (input) INTEGER */
/* The leading dimension of the array X. LDX >= max(1,N). */
/* FERR (output) DOUBLE PRECISION array, dimension (NRHS) */
/* The estimated forward error bound for each solution vector */
/* X(j) (the j-th column of the solution matrix X). */
/* If XTRUE is the true solution corresponding to X(j), FERR(j) */
/* is an estimated upper bound for the magnitude of the largest */
/* element in (X(j) - XTRUE) divided by the magnitude of the */
/* largest element in X(j). The estimate is as reliable as */
/* the estimate for RCOND, and is almost always a slight */
/* overestimate of the true error. */
/* BERR (output) DOUBLE PRECISION array, dimension (NRHS) */
/* The componentwise relative backward error of each solution */
/* vector X(j) (i.e., the smallest relative change in */
/* any element of A or B that makes X(j) an exact solution). */
/* WORK (workspace) COMPLEX*16 array, dimension (2*N) */
/* RWORK (workspace) DOUBLE PRECISION array, dimension (N) */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* Internal Parameters */
/* =================== */
/* ITMAX is the maximum number of steps of iterative refinement. */
/* ==================================================================== */
/* Test the input parameters. */
/* Parameter adjustments */
a_dim1 = *lda;
a_offset = 1 + a_dim1;
a -= a_offset;
af_dim1 = *ldaf;
af_offset = 1 + af_dim1;
af -= af_offset;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
x_dim1 = *ldx;
x_offset = 1 + x_dim1;
x -= x_offset;
--ferr;
--berr;
--work;
--rwork;
/* Function Body */
*info = 0;
upper = lsame_(uplo, "U");
if (! upper && ! lsame_(uplo, "L")) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*nrhs < 0) {
*info = -3;
} else if (*lda < max(1,*n)) {
*info = -5;
} else if (*ldaf < max(1,*n)) {
*info = -7;
} else if (*ldb < max(1,*n)) {
*info = -9;
} else if (*ldx < max(1,*n)) {
*info = -11;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("ZPORFS", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0 || *nrhs == 0) {
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
ferr[j] = 0.;
berr[j] = 0.;
}
return 0;
}
/* NZ = maximum number of nonzero elements in each row of A, plus 1 */
nz = *n + 1;
eps = dlamch_("Epsilon");
safmin = dlamch_("Safe minimum");
safe1 = nz * safmin;
safe2 = safe1 / eps;
/* Do for each right hand side */
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
count = 1;
lstres = 3.;
L20:
/* Loop until stopping criterion is satisfied. */
/* Compute residual R = B - A * X */
zcopy_(n, &b[j * b_dim1 + 1], &c__1, &work[1], &c__1);
z__1.r = -1., z__1.i = -0.;
zhemv_(uplo, n, &z__1, &a[a_offset], lda, &x[j * x_dim1 + 1], &c__1, &
c_b1, &work[1], &c__1);
/* Compute componentwise relative backward error from formula */
/* max(i) ( abs(R(i)) / ( abs(A)*abs(X) + abs(B) )(i) ) */
/* where abs(Z) is the componentwise absolute value of the matrix */
/* or vector Z. If the i-th component of the denominator is less */
/* than SAFE2, then SAFE1 is added to the i-th components of the */
/* numerator and denominator before dividing. */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__ + j * b_dim1;
rwork[i__] = (d__1 = b[i__3].r, abs(d__1)) + (d__2 = d_imag(&b[
i__ + j * b_dim1]), abs(d__2));
}
/* Compute abs(A)*abs(X) + abs(B). */
if (upper) {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
s = 0.;
i__3 = k + j * x_dim1;
xk = (d__1 = x[i__3].r, abs(d__1)) + (d__2 = d_imag(&x[k + j *
x_dim1]), abs(d__2));
i__3 = k - 1;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = i__ + k * a_dim1;
rwork[i__] += ((d__1 = a[i__4].r, abs(d__1)) + (d__2 =
d_imag(&a[i__ + k * a_dim1]), abs(d__2))) * xk;
i__4 = i__ + k * a_dim1;
i__5 = i__ + j * x_dim1;
s += ((d__1 = a[i__4].r, abs(d__1)) + (d__2 = d_imag(&a[
i__ + k * a_dim1]), abs(d__2))) * ((d__3 = x[i__5]
.r, abs(d__3)) + (d__4 = d_imag(&x[i__ + j *
x_dim1]), abs(d__4)));
}
i__3 = k + k * a_dim1;
rwork[k] = rwork[k] + (d__1 = a[i__3].r, abs(d__1)) * xk + s;
}
} else {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
s = 0.;
i__3 = k + j * x_dim1;
xk = (d__1 = x[i__3].r, abs(d__1)) + (d__2 = d_imag(&x[k + j *
x_dim1]), abs(d__2));
i__3 = k + k * a_dim1;
rwork[k] += (d__1 = a[i__3].r, abs(d__1)) * xk;
i__3 = *n;
for (i__ = k + 1; i__ <= i__3; ++i__) {
i__4 = i__ + k * a_dim1;
rwork[i__] += ((d__1 = a[i__4].r, abs(d__1)) + (d__2 =
d_imag(&a[i__ + k * a_dim1]), abs(d__2))) * xk;
i__4 = i__ + k * a_dim1;
i__5 = i__ + j * x_dim1;
s += ((d__1 = a[i__4].r, abs(d__1)) + (d__2 = d_imag(&a[
i__ + k * a_dim1]), abs(d__2))) * ((d__3 = x[i__5]
.r, abs(d__3)) + (d__4 = d_imag(&x[i__ + j *
x_dim1]), abs(d__4)));
}
rwork[k] += s;
}
}
s = 0.;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
/* Computing MAX */
i__3 = i__;
d__3 = s, d__4 = ((d__1 = work[i__3].r, abs(d__1)) + (d__2 =
d_imag(&work[i__]), abs(d__2))) / rwork[i__];
s = max(d__3,d__4);
} else {
/* Computing MAX */
i__3 = i__;
d__3 = s, d__4 = ((d__1 = work[i__3].r, abs(d__1)) + (d__2 =
d_imag(&work[i__]), abs(d__2)) + safe1) / (rwork[i__]
+ safe1);
s = max(d__3,d__4);
}
}
berr[j] = s;
/* Test stopping criterion. Continue iterating if */
/* 1) The residual BERR(J) is larger than machine epsilon, and */
/* 2) BERR(J) decreased by at least a factor of 2 during the */
/* last iteration, and */
/* 3) At most ITMAX iterations tried. */
if (berr[j] > eps && berr[j] * 2. <= lstres && count <= 5) {
/* Update solution and try again. */
zpotrs_(uplo, n, &c__1, &af[af_offset], ldaf, &work[1], n, info);
zaxpy_(n, &c_b1, &work[1], &c__1, &x[j * x_dim1 + 1], &c__1);
lstres = berr[j];
++count;
goto L20;
}
/* Bound error from formula */
/* norm(X - XTRUE) / norm(X) .le. FERR = */
/* norm( abs(inv(A))* */
/* ( abs(R) + NZ*EPS*( abs(A)*abs(X)+abs(B) ))) / norm(X) */
/* where */
/* norm(Z) is the magnitude of the largest component of Z */
/* inv(A) is the inverse of A */
/* abs(Z) is the componentwise absolute value of the matrix or */
/* vector Z */
/* NZ is the maximum number of nonzeros in any row of A, plus 1 */
/* EPS is machine epsilon */
/* The i-th component of abs(R)+NZ*EPS*(abs(A)*abs(X)+abs(B)) */
/* is incremented by SAFE1 if the i-th component of */
/* abs(A)*abs(X) + abs(B) is less than SAFE2. */
/* Use ZLACN2 to estimate the infinity-norm of the matrix */
/* inv(A) * diag(W), */
/* where W = abs(R) + NZ*EPS*( abs(A)*abs(X)+abs(B) ))) */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
i__3 = i__;
rwork[i__] = (d__1 = work[i__3].r, abs(d__1)) + (d__2 =
d_imag(&work[i__]), abs(d__2)) + nz * eps * rwork[i__]
;
} else {
i__3 = i__;
rwork[i__] = (d__1 = work[i__3].r, abs(d__1)) + (d__2 =
d_imag(&work[i__]), abs(d__2)) + nz * eps * rwork[i__]
+ safe1;
}
}
kase = 0;
L100:
zlacn2_(n, &work[*n + 1], &work[1], &ferr[j], &kase, isave);
if (kase != 0) {
if (kase == 1) {
/* Multiply by diag(W)*inv(A'). */
zpotrs_(uplo, n, &c__1, &af[af_offset], ldaf, &work[1], n,
info);
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__;
i__4 = i__;
i__5 = i__;
z__1.r = rwork[i__4] * work[i__5].r, z__1.i = rwork[i__4]
* work[i__5].i;
work[i__3].r = z__1.r, work[i__3].i = z__1.i;
}
} else if (kase == 2) {
/* Multiply by inv(A)*diag(W). */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__;
i__4 = i__;
i__5 = i__;
z__1.r = rwork[i__4] * work[i__5].r, z__1.i = rwork[i__4]
* work[i__5].i;
work[i__3].r = z__1.r, work[i__3].i = z__1.i;
}
zpotrs_(uplo, n, &c__1, &af[af_offset], ldaf, &work[1], n,
info);
}
goto L100;
}
/* Normalize error. */
lstres = 0.;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
/* Computing MAX */
i__3 = i__ + j * x_dim1;
d__3 = lstres, d__4 = (d__1 = x[i__3].r, abs(d__1)) + (d__2 =
d_imag(&x[i__ + j * x_dim1]), abs(d__2));
lstres = max(d__3,d__4);
}
if (lstres != 0.) {
ferr[j] /= lstres;
}
}
return 0;
/* End of ZPORFS */
} /* zporfs_ */
/* Subroutine */ int zgbrfs_(char *trans, integer *n, integer *kl, integer *
ku, integer *nrhs, doublecomplex *ab, integer *ldab, doublecomplex *
afb, integer *ldafb, integer *ipiv, doublecomplex *b, integer *ldb,
doublecomplex *x, integer *ldx, doublereal *ferr, doublereal *berr,
doublecomplex *work, doublereal *rwork, integer *info)
{
/* System generated locals */
integer ab_dim1, ab_offset, afb_dim1, afb_offset, b_dim1, b_offset,
x_dim1, x_offset, i__1, i__2, i__3, i__4, i__5, i__6, i__7;
doublereal d__1, d__2, d__3, d__4;
doublecomplex z__1;
/* Builtin functions */
double d_imag(doublecomplex *);
/* Local variables */
integer i__, j, k;
doublereal s;
integer kk;
doublereal xk;
integer nz;
doublereal eps;
integer kase;
doublereal safe1, safe2;
extern logical lsame_(char *, char *);
integer isave[3];
extern /* Subroutine */ int zgbmv_(char *, integer *, integer *, integer *
, integer *, doublecomplex *, doublecomplex *, integer *,
doublecomplex *, integer *, doublecomplex *, doublecomplex *,
integer *);
integer count;
extern /* Subroutine */ int zcopy_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *), zaxpy_(integer *, doublecomplex *,
doublecomplex *, integer *, doublecomplex *, integer *), zlacn2_(
integer *, doublecomplex *, doublecomplex *, doublereal *,
integer *, integer *);
extern doublereal dlamch_(char *);
doublereal safmin;
extern /* Subroutine */ int xerbla_(char *, integer *);
logical notran;
char transn[1], transt[1];
doublereal lstres;
extern /* Subroutine */ int zgbtrs_(char *, integer *, integer *, integer
*, integer *, doublecomplex *, integer *, integer *,
doublecomplex *, integer *, integer *);
/* -- LAPACK routine (version 3.2) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* Modified to call ZLACN2 in place of ZLACON, 10 Feb 03, SJH. */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* ZGBRFS improves the computed solution to a system of linear */
/* equations when the coefficient matrix is banded, and provides */
/* error bounds and backward error estimates for the solution. */
/* Arguments */
/* ========= */
/* TRANS (input) CHARACTER*1 */
/* Specifies the form of the system of equations: */
/* = 'N': A * X = B (No transpose) */
/* = 'T': A**T * X = B (Transpose) */
/* = 'C': A**H * X = B (Conjugate transpose) */
/* N (input) INTEGER */
/* The order of the matrix A. N >= 0. */
/* KL (input) INTEGER */
/* The number of subdiagonals within the band of A. KL >= 0. */
/* KU (input) INTEGER */
/* The number of superdiagonals within the band of A. KU >= 0. */
/* NRHS (input) INTEGER */
/* The number of right hand sides, i.e., the number of columns */
/* of the matrices B and X. NRHS >= 0. */
/* AB (input) COMPLEX*16 array, dimension (LDAB,N) */
/* The original band matrix A, stored in rows 1 to KL+KU+1. */
/* The j-th column of A is stored in the j-th column of the */
/* array AB as follows: */
/* AB(ku+1+i-j,j) = A(i,j) for max(1,j-ku)<=i<=min(n,j+kl). */
/* LDAB (input) INTEGER */
/* The leading dimension of the array AB. LDAB >= KL+KU+1. */
/* AFB (input) COMPLEX*16 array, dimension (LDAFB,N) */
/* Details of the LU factorization of the band matrix A, as */
/* computed by ZGBTRF. U is stored as an upper triangular band */
/* matrix with KL+KU superdiagonals in rows 1 to KL+KU+1, and */
/* the multipliers used during the factorization are stored in */
/* rows KL+KU+2 to 2*KL+KU+1. */
/* LDAFB (input) INTEGER */
/* The leading dimension of the array AFB. LDAFB >= 2*KL*KU+1. */
/* IPIV (input) INTEGER array, dimension (N) */
/* The pivot indices from ZGBTRF; for 1<=i<=N, row i of the */
/* matrix was interchanged with row IPIV(i). */
/* B (input) COMPLEX*16 array, dimension (LDB,NRHS) */
/* The right hand side matrix B. */
/* LDB (input) INTEGER */
/* The leading dimension of the array B. LDB >= max(1,N). */
/* X (input/output) COMPLEX*16 array, dimension (LDX,NRHS) */
/* On entry, the solution matrix X, as computed by ZGBTRS. */
/* On exit, the improved solution matrix X. */
/* LDX (input) INTEGER */
/* The leading dimension of the array X. LDX >= max(1,N). */
/* FERR (output) DOUBLE PRECISION array, dimension (NRHS) */
/* The estimated forward error bound for each solution vector */
/* X(j) (the j-th column of the solution matrix X). */
/* If XTRUE is the true solution corresponding to X(j), FERR(j) */
/* is an estimated upper bound for the magnitude of the largest */
/* element in (X(j) - XTRUE) divided by the magnitude of the */
/* largest element in X(j). The estimate is as reliable as */
/* the estimate for RCOND, and is almost always a slight */
/* overestimate of the true error. */
/* BERR (output) DOUBLE PRECISION array, dimension (NRHS) */
/* The componentwise relative backward error of each solution */
/* vector X(j) (i.e., the smallest relative change in */
/* any element of A or B that makes X(j) an exact solution). */
/* WORK (workspace) COMPLEX*16 array, dimension (2*N) */
/* RWORK (workspace) DOUBLE PRECISION array, dimension (N) */
/* INFO (output) INTEGER */
/* = 0: successful exit */
/* < 0: if INFO = -i, the i-th argument had an illegal value */
/* Internal Parameters */
/* =================== */
/* ITMAX is the maximum number of steps of iterative refinement. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. Local Arrays .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. Statement Functions .. */
/* .. */
/* .. Statement Function definitions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
ab_dim1 = *ldab;
ab_offset = 1 + ab_dim1;
ab -= ab_offset;
afb_dim1 = *ldafb;
afb_offset = 1 + afb_dim1;
afb -= afb_offset;
--ipiv;
b_dim1 = *ldb;
b_offset = 1 + b_dim1;
b -= b_offset;
x_dim1 = *ldx;
x_offset = 1 + x_dim1;
x -= x_offset;
--ferr;
--berr;
--work;
--rwork;
/* Function Body */
*info = 0;
notran = lsame_(trans, "N");
if (! notran && ! lsame_(trans, "T") && ! lsame_(
trans, "C")) {
*info = -1;
} else if (*n < 0) {
*info = -2;
} else if (*kl < 0) {
*info = -3;
} else if (*ku < 0) {
*info = -4;
} else if (*nrhs < 0) {
*info = -5;
} else if (*ldab < *kl + *ku + 1) {
*info = -7;
} else if (*ldafb < (*kl << 1) + *ku + 1) {
*info = -9;
} else if (*ldb < max(1,*n)) {
*info = -12;
} else if (*ldx < max(1,*n)) {
*info = -14;
}
if (*info != 0) {
i__1 = -(*info);
xerbla_("ZGBRFS", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0 || *nrhs == 0) {
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
ferr[j] = 0.;
berr[j] = 0.;
/* L10: */
}
return 0;
}
if (notran) {
*(unsigned char *)transn = 'N';
*(unsigned char *)transt = 'C';
} else {
*(unsigned char *)transn = 'C';
*(unsigned char *)transt = 'N';
}
/* NZ = maximum number of nonzero elements in each row of A, plus 1 */
/* Computing MIN */
i__1 = *kl + *ku + 2, i__2 = *n + 1;
nz = min(i__1,i__2);
eps = dlamch_("Epsilon");
safmin = dlamch_("Safe minimum");
safe1 = nz * safmin;
safe2 = safe1 / eps;
/* Do for each right hand side */
i__1 = *nrhs;
for (j = 1; j <= i__1; ++j) {
count = 1;
lstres = 3.;
L20:
/* Loop until stopping criterion is satisfied. */
/* Compute residual R = B - op(A) * X, */
/* where op(A) = A, A**T, or A**H, depending on TRANS. */
zcopy_(n, &b[j * b_dim1 + 1], &c__1, &work[1], &c__1);
z__1.r = -1., z__1.i = -0.;
zgbmv_(trans, n, n, kl, ku, &z__1, &ab[ab_offset], ldab, &x[j *
x_dim1 + 1], &c__1, &c_b1, &work[1], &c__1);
/* Compute componentwise relative backward error from formula */
/* max(i) ( abs(R(i)) / ( abs(op(A))*abs(X) + abs(B) )(i) ) */
/* where abs(Z) is the componentwise absolute value of the matrix */
/* or vector Z. If the i-th component of the denominator is less */
/* than SAFE2, then SAFE1 is added to the i-th components of the */
/* numerator and denominator before dividing. */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__3 = i__ + j * b_dim1;
rwork[i__] = (d__1 = b[i__3].r, abs(d__1)) + (d__2 = d_imag(&b[
i__ + j * b_dim1]), abs(d__2));
/* L30: */
}
/* Compute abs(op(A))*abs(X) + abs(B). */
if (notran) {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
kk = *ku + 1 - k;
i__3 = k + j * x_dim1;
xk = (d__1 = x[i__3].r, abs(d__1)) + (d__2 = d_imag(&x[k + j *
x_dim1]), abs(d__2));
/* Computing MAX */
i__3 = 1, i__4 = k - *ku;
/* Computing MIN */
i__6 = *n, i__7 = k + *kl;
i__5 = min(i__6,i__7);
for (i__ = max(i__3,i__4); i__ <= i__5; ++i__) {
i__3 = kk + i__ + k * ab_dim1;
rwork[i__] += ((d__1 = ab[i__3].r, abs(d__1)) + (d__2 =
d_imag(&ab[kk + i__ + k * ab_dim1]), abs(d__2))) *
xk;
/* L40: */
}
/* L50: */
}
} else {
i__2 = *n;
for (k = 1; k <= i__2; ++k) {
s = 0.;
kk = *ku + 1 - k;
/* Computing MAX */
i__5 = 1, i__3 = k - *ku;
/* Computing MIN */
i__6 = *n, i__7 = k + *kl;
i__4 = min(i__6,i__7);
for (i__ = max(i__5,i__3); i__ <= i__4; ++i__) {
i__5 = kk + i__ + k * ab_dim1;
i__3 = i__ + j * x_dim1;
s += ((d__1 = ab[i__5].r, abs(d__1)) + (d__2 = d_imag(&ab[
kk + i__ + k * ab_dim1]), abs(d__2))) * ((d__3 =
x[i__3].r, abs(d__3)) + (d__4 = d_imag(&x[i__ + j
* x_dim1]), abs(d__4)));
/* L60: */
}
rwork[k] += s;
/* L70: */
}
}
s = 0.;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
/* Computing MAX */
i__4 = i__;
d__3 = s, d__4 = ((d__1 = work[i__4].r, abs(d__1)) + (d__2 =
d_imag(&work[i__]), abs(d__2))) / rwork[i__];
s = max(d__3,d__4);
} else {
/* Computing MAX */
i__4 = i__;
d__3 = s, d__4 = ((d__1 = work[i__4].r, abs(d__1)) + (d__2 =
d_imag(&work[i__]), abs(d__2)) + safe1) / (rwork[i__]
+ safe1);
s = max(d__3,d__4);
}
/* L80: */
}
berr[j] = s;
/* Test stopping criterion. Continue iterating if */
/* 1) The residual BERR(J) is larger than machine epsilon, and */
/* 2) BERR(J) decreased by at least a factor of 2 during the */
/* last iteration, and */
/* 3) At most ITMAX iterations tried. */
if (berr[j] > eps && berr[j] * 2. <= lstres && count <= 5) {
/* Update solution and try again. */
zgbtrs_(trans, n, kl, ku, &c__1, &afb[afb_offset], ldafb, &ipiv[1]
, &work[1], n, info);
zaxpy_(n, &c_b1, &work[1], &c__1, &x[j * x_dim1 + 1], &c__1);
lstres = berr[j];
++count;
goto L20;
}
/* Bound error from formula */
/* norm(X - XTRUE) / norm(X) .le. FERR = */
/* norm( abs(inv(op(A)))* */
/* ( abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) / norm(X) */
/* where */
/* norm(Z) is the magnitude of the largest component of Z */
/* inv(op(A)) is the inverse of op(A) */
/* abs(Z) is the componentwise absolute value of the matrix or */
/* vector Z */
/* NZ is the maximum number of nonzeros in any row of A, plus 1 */
/* EPS is machine epsilon */
/* The i-th component of abs(R)+NZ*EPS*(abs(op(A))*abs(X)+abs(B)) */
/* is incremented by SAFE1 if the i-th component of */
/* abs(op(A))*abs(X) + abs(B) is less than SAFE2. */
/* Use ZLACN2 to estimate the infinity-norm of the matrix */
/* inv(op(A)) * diag(W), */
/* where W = abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
if (rwork[i__] > safe2) {
i__4 = i__;
rwork[i__] = (d__1 = work[i__4].r, abs(d__1)) + (d__2 =
d_imag(&work[i__]), abs(d__2)) + nz * eps * rwork[i__]
;
} else {
i__4 = i__;
rwork[i__] = (d__1 = work[i__4].r, abs(d__1)) + (d__2 =
d_imag(&work[i__]), abs(d__2)) + nz * eps * rwork[i__]
+ safe1;
}
/* L90: */
}
kase = 0;
L100:
zlacn2_(n, &work[*n + 1], &work[1], &ferr[j], &kase, isave);
if (kase != 0) {
if (kase == 1) {
/* Multiply by diag(W)*inv(op(A)**H). */
zgbtrs_(transt, n, kl, ku, &c__1, &afb[afb_offset], ldafb, &
ipiv[1], &work[1], n, info);
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__4 = i__;
i__5 = i__;
i__3 = i__;
z__1.r = rwork[i__5] * work[i__3].r, z__1.i = rwork[i__5]
* work[i__3].i;
work[i__4].r = z__1.r, work[i__4].i = z__1.i;
/* L110: */
}
} else {
/* Multiply by inv(op(A))*diag(W). */
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
i__4 = i__;
i__5 = i__;
i__3 = i__;
z__1.r = rwork[i__5] * work[i__3].r, z__1.i = rwork[i__5]
* work[i__3].i;
work[i__4].r = z__1.r, work[i__4].i = z__1.i;
/* L120: */
}
zgbtrs_(transn, n, kl, ku, &c__1, &afb[afb_offset], ldafb, &
ipiv[1], &work[1], n, info);
}
goto L100;
}
/* Normalize error. */
lstres = 0.;
i__2 = *n;
for (i__ = 1; i__ <= i__2; ++i__) {
/* Computing MAX */
i__4 = i__ + j * x_dim1;
d__3 = lstres, d__4 = (d__1 = x[i__4].r, abs(d__1)) + (d__2 =
d_imag(&x[i__ + j * x_dim1]), abs(d__2));
lstres = max(d__3,d__4);
/* L130: */
}
if (lstres != 0.) {
ferr[j] /= lstres;
}
/* L140: */
}
return 0;
/* End of ZGBRFS */
} /* zgbrfs_ */
/* Subroutine */
int zhptri_(char *uplo, integer *n, doublecomplex *ap, integer *ipiv, doublecomplex *work, integer *info)
{
/* System generated locals */
integer i__1, i__2, i__3;
doublereal d__1;
doublecomplex z__1, z__2;
/* Builtin functions */
double z_abs(doublecomplex *);
void d_cnjg(doublecomplex *, doublecomplex *);
/* Local variables */
doublereal d__;
integer j, k;
doublereal t, ak;
integer kc, kp, kx, kpc, npp;
doublereal akp1;
doublecomplex temp, akkp1;
extern logical lsame_(char *, char *);
extern /* Double Complex */
VOID zdotc_f2c_(doublecomplex *, integer *, doublecomplex *, integer *, doublecomplex *, integer *);
integer kstep;
logical upper;
extern /* Subroutine */
int zcopy_(integer *, doublecomplex *, integer *, doublecomplex *, integer *), zhpmv_(char *, integer *, doublecomplex *, doublecomplex *, doublecomplex *, integer *, doublecomplex *, doublecomplex *, integer *), zswap_( integer *, doublecomplex *, integer *, doublecomplex *, integer *) , xerbla_(char *, integer *);
integer kcnext;
/* -- LAPACK computational routine (version 3.4.0) -- */
/* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
/* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
/* November 2011 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* ===================================================================== */
/* .. Parameters .. */
/* .. */
/* .. Local Scalars .. */
/* .. */
/* .. External Functions .. */
/* .. */
/* .. External Subroutines .. */
/* .. */
/* .. Intrinsic Functions .. */
/* .. */
/* .. Executable Statements .. */
/* Test the input parameters. */
/* Parameter adjustments */
--work;
--ipiv;
--ap;
/* Function Body */
*info = 0;
upper = lsame_(uplo, "U");
if (! upper && ! lsame_(uplo, "L"))
{
*info = -1;
}
else if (*n < 0)
{
*info = -2;
}
if (*info != 0)
{
i__1 = -(*info);
xerbla_("ZHPTRI", &i__1);
return 0;
}
/* Quick return if possible */
if (*n == 0)
{
return 0;
}
/* Check that the diagonal matrix D is nonsingular. */
if (upper)
{
/* Upper triangular storage: examine D from bottom to top */
kp = *n * (*n + 1) / 2;
for (*info = *n;
*info >= 1;
--(*info))
{
i__1 = kp;
if (ipiv[*info] > 0 && (ap[i__1].r == 0. && ap[i__1].i == 0.))
{
return 0;
}
kp -= *info;
/* L10: */
}
}
else
{
/* Lower triangular storage: examine D from top to bottom. */
kp = 1;
i__1 = *n;
for (*info = 1;
*info <= i__1;
++(*info))
{
i__2 = kp;
if (ipiv[*info] > 0 && (ap[i__2].r == 0. && ap[i__2].i == 0.))
{
return 0;
}
kp = kp + *n - *info + 1;
/* L20: */
}
}
*info = 0;
if (upper)
{
/* Compute inv(A) from the factorization A = U*D*U**H. */
/* K is the main loop index, increasing from 1 to N in steps of */
/* 1 or 2, depending on the size of the diagonal blocks. */
k = 1;
kc = 1;
L30: /* If K > N, exit from loop. */
if (k > *n)
{
goto L50;
}
kcnext = kc + k;
if (ipiv[k] > 0)
{
/* 1 x 1 diagonal block */
/* Invert the diagonal block. */
i__1 = kc + k - 1;
i__2 = kc + k - 1;
d__1 = 1. / ap[i__2].r;
ap[i__1].r = d__1;
ap[i__1].i = 0.; // , expr subst
/* Compute column K of the inverse. */
if (k > 1)
{
i__1 = k - 1;
zcopy_(&i__1, &ap[kc], &c__1, &work[1], &c__1);
i__1 = k - 1;
z__1.r = -1.;
z__1.i = -0.; // , expr subst
zhpmv_(uplo, &i__1, &z__1, &ap[1], &work[1], &c__1, &c_b2, & ap[kc], &c__1);
i__1 = kc + k - 1;
i__2 = kc + k - 1;
i__3 = k - 1;
zdotc_f2c_(&z__2, &i__3, &work[1], &c__1, &ap[kc], &c__1);
d__1 = z__2.r;
z__1.r = ap[i__2].r - d__1;
z__1.i = ap[i__2].i; // , expr subst
ap[i__1].r = z__1.r;
ap[i__1].i = z__1.i; // , expr subst
}
kstep = 1;
}
else
{
/* 2 x 2 diagonal block */
/* Invert the diagonal block. */
t = z_abs(&ap[kcnext + k - 1]);
i__1 = kc + k - 1;
ak = ap[i__1].r / t;
i__1 = kcnext + k;
akp1 = ap[i__1].r / t;
i__1 = kcnext + k - 1;
z__1.r = ap[i__1].r / t;
z__1.i = ap[i__1].i / t; // , expr subst
akkp1.r = z__1.r;
akkp1.i = z__1.i; // , expr subst
d__ = t * (ak * akp1 - 1.);
i__1 = kc + k - 1;
d__1 = akp1 / d__;
ap[i__1].r = d__1;
ap[i__1].i = 0.; // , expr subst
i__1 = kcnext + k;
d__1 = ak / d__;
ap[i__1].r = d__1;
ap[i__1].i = 0.; // , expr subst
i__1 = kcnext + k - 1;
z__2.r = -akkp1.r;
z__2.i = -akkp1.i; // , expr subst
z__1.r = z__2.r / d__;
z__1.i = z__2.i / d__; // , expr subst
ap[i__1].r = z__1.r;
ap[i__1].i = z__1.i; // , expr subst
/* Compute columns K and K+1 of the inverse. */
if (k > 1)
{
i__1 = k - 1;
zcopy_(&i__1, &ap[kc], &c__1, &work[1], &c__1);
i__1 = k - 1;
z__1.r = -1.;
z__1.i = -0.; // , expr subst
zhpmv_(uplo, &i__1, &z__1, &ap[1], &work[1], &c__1, &c_b2, & ap[kc], &c__1);
i__1 = kc + k - 1;
i__2 = kc + k - 1;
i__3 = k - 1;
zdotc_f2c_(&z__2, &i__3, &work[1], &c__1, &ap[kc], &c__1);
d__1 = z__2.r;
z__1.r = ap[i__2].r - d__1;
z__1.i = ap[i__2].i; // , expr subst
ap[i__1].r = z__1.r;
ap[i__1].i = z__1.i; // , expr subst
i__1 = kcnext + k - 1;
i__2 = kcnext + k - 1;
i__3 = k - 1;
zdotc_f2c_(&z__2, &i__3, &ap[kc], &c__1, &ap[kcnext], &c__1);
z__1.r = ap[i__2].r - z__2.r;
z__1.i = ap[i__2].i - z__2.i; // , expr subst
ap[i__1].r = z__1.r;
ap[i__1].i = z__1.i; // , expr subst
i__1 = k - 1;
zcopy_(&i__1, &ap[kcnext], &c__1, &work[1], &c__1);
i__1 = k - 1;
z__1.r = -1.;
z__1.i = -0.; // , expr subst
zhpmv_(uplo, &i__1, &z__1, &ap[1], &work[1], &c__1, &c_b2, & ap[kcnext], &c__1);
i__1 = kcnext + k;
i__2 = kcnext + k;
i__3 = k - 1;
zdotc_f2c_(&z__2, &i__3, &work[1], &c__1, &ap[kcnext], &c__1);
d__1 = z__2.r;
z__1.r = ap[i__2].r - d__1;
z__1.i = ap[i__2].i; // , expr subst
ap[i__1].r = z__1.r;
ap[i__1].i = z__1.i; // , expr subst
}
kstep = 2;
kcnext = kcnext + k + 1;
}
kp = (i__1 = ipiv[k], f2c_abs(i__1));
if (kp != k)
{
/* Interchange rows and columns K and KP in the leading */
/* submatrix A(1:k+1,1:k+1) */
kpc = (kp - 1) * kp / 2 + 1;
i__1 = kp - 1;
zswap_(&i__1, &ap[kc], &c__1, &ap[kpc], &c__1);
kx = kpc + kp - 1;
i__1 = k - 1;
for (j = kp + 1;
j <= i__1;
++j)
{
kx = kx + j - 1;
d_cnjg(&z__1, &ap[kc + j - 1]);
temp.r = z__1.r;
temp.i = z__1.i; // , expr subst
i__2 = kc + j - 1;
d_cnjg(&z__1, &ap[kx]);
ap[i__2].r = z__1.r;
ap[i__2].i = z__1.i; // , expr subst
i__2 = kx;
ap[i__2].r = temp.r;
ap[i__2].i = temp.i; // , expr subst
/* L40: */
}
i__1 = kc + kp - 1;
d_cnjg(&z__1, &ap[kc + kp - 1]);
ap[i__1].r = z__1.r;
ap[i__1].i = z__1.i; // , expr subst
i__1 = kc + k - 1;
temp.r = ap[i__1].r;
temp.i = ap[i__1].i; // , expr subst
i__1 = kc + k - 1;
i__2 = kpc + kp - 1;
ap[i__1].r = ap[i__2].r;
ap[i__1].i = ap[i__2].i; // , expr subst
i__1 = kpc + kp - 1;
ap[i__1].r = temp.r;
ap[i__1].i = temp.i; // , expr subst
if (kstep == 2)
{
i__1 = kc + k + k - 1;
temp.r = ap[i__1].r;
temp.i = ap[i__1].i; // , expr subst
i__1 = kc + k + k - 1;
i__2 = kc + k + kp - 1;
ap[i__1].r = ap[i__2].r;
ap[i__1].i = ap[i__2].i; // , expr subst
i__1 = kc + k + kp - 1;
ap[i__1].r = temp.r;
ap[i__1].i = temp.i; // , expr subst
}
}
k += kstep;
kc = kcnext;
goto L30;
L50:
;
}
else
{
/* Compute inv(A) from the factorization A = L*D*L**H. */
/* K is the main loop index, increasing from 1 to N in steps of */
/* 1 or 2, depending on the size of the diagonal blocks. */
npp = *n * (*n + 1) / 2;
k = *n;
kc = npp;
L60: /* If K < 1, exit from loop. */
if (k < 1)
{
goto L80;
}
kcnext = kc - (*n - k + 2);
if (ipiv[k] > 0)
{
/* 1 x 1 diagonal block */
/* Invert the diagonal block. */
i__1 = kc;
i__2 = kc;
d__1 = 1. / ap[i__2].r;
ap[i__1].r = d__1;
ap[i__1].i = 0.; // , expr subst
/* Compute column K of the inverse. */
if (k < *n)
{
i__1 = *n - k;
zcopy_(&i__1, &ap[kc + 1], &c__1, &work[1], &c__1);
i__1 = *n - k;
z__1.r = -1.;
z__1.i = -0.; // , expr subst
zhpmv_(uplo, &i__1, &z__1, &ap[kc + *n - k + 1], &work[1], & c__1, &c_b2, &ap[kc + 1], &c__1);
i__1 = kc;
i__2 = kc;
i__3 = *n - k;
zdotc_f2c_(&z__2, &i__3, &work[1], &c__1, &ap[kc + 1], &c__1);
d__1 = z__2.r;
z__1.r = ap[i__2].r - d__1;
z__1.i = ap[i__2].i; // , expr subst
ap[i__1].r = z__1.r;
ap[i__1].i = z__1.i; // , expr subst
}
kstep = 1;
}
else
{
/* 2 x 2 diagonal block */
/* Invert the diagonal block. */
t = z_abs(&ap[kcnext + 1]);
i__1 = kcnext;
ak = ap[i__1].r / t;
i__1 = kc;
akp1 = ap[i__1].r / t;
i__1 = kcnext + 1;
z__1.r = ap[i__1].r / t;
z__1.i = ap[i__1].i / t; // , expr subst
akkp1.r = z__1.r;
akkp1.i = z__1.i; // , expr subst
d__ = t * (ak * akp1 - 1.);
i__1 = kcnext;
d__1 = akp1 / d__;
ap[i__1].r = d__1;
ap[i__1].i = 0.; // , expr subst
i__1 = kc;
d__1 = ak / d__;
ap[i__1].r = d__1;
ap[i__1].i = 0.; // , expr subst
i__1 = kcnext + 1;
z__2.r = -akkp1.r;
z__2.i = -akkp1.i; // , expr subst
z__1.r = z__2.r / d__;
z__1.i = z__2.i / d__; // , expr subst
ap[i__1].r = z__1.r;
ap[i__1].i = z__1.i; // , expr subst
/* Compute columns K-1 and K of the inverse. */
if (k < *n)
{
i__1 = *n - k;
zcopy_(&i__1, &ap[kc + 1], &c__1, &work[1], &c__1);
i__1 = *n - k;
z__1.r = -1.;
z__1.i = -0.; // , expr subst
zhpmv_(uplo, &i__1, &z__1, &ap[kc + (*n - k + 1)], &work[1], & c__1, &c_b2, &ap[kc + 1], &c__1);
i__1 = kc;
i__2 = kc;
i__3 = *n - k;
zdotc_f2c_(&z__2, &i__3, &work[1], &c__1, &ap[kc + 1], &c__1);
d__1 = z__2.r;
z__1.r = ap[i__2].r - d__1;
z__1.i = ap[i__2].i; // , expr subst
ap[i__1].r = z__1.r;
ap[i__1].i = z__1.i; // , expr subst
i__1 = kcnext + 1;
i__2 = kcnext + 1;
i__3 = *n - k;
zdotc_f2c_(&z__2, &i__3, &ap[kc + 1], &c__1, &ap[kcnext + 2], & c__1);
z__1.r = ap[i__2].r - z__2.r;
z__1.i = ap[i__2].i - z__2.i; // , expr subst
ap[i__1].r = z__1.r;
ap[i__1].i = z__1.i; // , expr subst
i__1 = *n - k;
zcopy_(&i__1, &ap[kcnext + 2], &c__1, &work[1], &c__1);
i__1 = *n - k;
z__1.r = -1.;
z__1.i = -0.; // , expr subst
zhpmv_(uplo, &i__1, &z__1, &ap[kc + (*n - k + 1)], &work[1], & c__1, &c_b2, &ap[kcnext + 2], &c__1);
i__1 = kcnext;
i__2 = kcnext;
i__3 = *n - k;
zdotc_f2c_(&z__2, &i__3, &work[1], &c__1, &ap[kcnext + 2], &c__1);
d__1 = z__2.r;
z__1.r = ap[i__2].r - d__1;
z__1.i = ap[i__2].i; // , expr subst
ap[i__1].r = z__1.r;
ap[i__1].i = z__1.i; // , expr subst
}
kstep = 2;
kcnext -= *n - k + 3;
}
kp = (i__1 = ipiv[k], f2c_abs(i__1));
if (kp != k)
{
/* Interchange rows and columns K and KP in the trailing */
/* submatrix A(k-1:n,k-1:n) */
kpc = npp - (*n - kp + 1) * (*n - kp + 2) / 2 + 1;
if (kp < *n)
{
i__1 = *n - kp;
zswap_(&i__1, &ap[kc + kp - k + 1], &c__1, &ap[kpc + 1], & c__1);
}
kx = kc + kp - k;
i__1 = kp - 1;
for (j = k + 1;
j <= i__1;
++j)
{
kx = kx + *n - j + 1;
d_cnjg(&z__1, &ap[kc + j - k]);
temp.r = z__1.r;
temp.i = z__1.i; // , expr subst
i__2 = kc + j - k;
d_cnjg(&z__1, &ap[kx]);
ap[i__2].r = z__1.r;
ap[i__2].i = z__1.i; // , expr subst
i__2 = kx;
ap[i__2].r = temp.r;
ap[i__2].i = temp.i; // , expr subst
/* L70: */
}
i__1 = kc + kp - k;
d_cnjg(&z__1, &ap[kc + kp - k]);
ap[i__1].r = z__1.r;
ap[i__1].i = z__1.i; // , expr subst
i__1 = kc;
temp.r = ap[i__1].r;
temp.i = ap[i__1].i; // , expr subst
i__1 = kc;
i__2 = kpc;
ap[i__1].r = ap[i__2].r;
ap[i__1].i = ap[i__2].i; // , expr subst
i__1 = kpc;
ap[i__1].r = temp.r;
ap[i__1].i = temp.i; // , expr subst
if (kstep == 2)
{
i__1 = kc - *n + k - 1;
temp.r = ap[i__1].r;
temp.i = ap[i__1].i; // , expr subst
i__1 = kc - *n + k - 1;
i__2 = kc - *n + kp - 1;
ap[i__1].r = ap[i__2].r;
ap[i__1].i = ap[i__2].i; // , expr subst
i__1 = kc - *n + kp - 1;
ap[i__1].r = temp.r;
ap[i__1].i = temp.i; // , expr subst
}
}
k -= kstep;
kc = kcnext;
goto L60;
L80:
;
}
return 0;
/* End of ZHPTRI */
}
/* Subroutine */ int zchkhp_(logical *dotype, integer *nn, integer *nval,
integer *nns, integer *nsval, doublereal *thresh, logical *tsterr,
integer *nmax, doublecomplex *a, doublecomplex *afac, doublecomplex *
ainv, doublecomplex *b, doublecomplex *x, doublecomplex *xact,
doublecomplex *work, doublereal *rwork, integer *iwork, integer *nout)
{
/* Initialized data */
static integer iseedy[4] = { 1988,1989,1990,1991 };
static char uplos[1*2] = "U" "L";
/* Format strings */
static char fmt_9999[] = "(\002 UPLO = '\002,a1,\002', N =\002,i5,\002, "
"type \002,i2,\002, test \002,i2,\002, ratio =\002,g12.5)";
static char fmt_9998[] = "(\002 UPLO = '\002,a1,\002', N =\002,i5,\002, "
"NRHS=\002,i3,\002, type \002,i2,\002, test(\002,i2,\002) =\002,g"
"12.5)";
/* System generated locals */
integer i__1, i__2, i__3, i__4, i__5;
/* Builtin functions */
/* Subroutine */ int s_copy(char *, char *, ftnlen, ftnlen);
integer s_wsfe(cilist *), do_fio(integer *, char *, ftnlen), e_wsfe(void);
/* Local variables */
integer i__, j, k, n, i1, i2, in, kl, ku, nt, lda, npp, ioff, mode, imat,
info;
char path[3], dist[1];
integer irhs, nrhs;
char uplo[1], type__[1];
integer nrun;
extern /* Subroutine */ int alahd_(integer *, char *);
integer nfail, iseed[4];
extern doublereal dget06_(doublereal *, doublereal *);
extern logical lsame_(char *, char *);
doublereal rcond;
integer nimat;
doublereal anorm;
extern /* Subroutine */ int zget04_(integer *, integer *, doublecomplex *,
integer *, doublecomplex *, integer *, doublereal *, doublereal *
), zhpt01_(char *, integer *, doublecomplex *, doublecomplex *,
integer *, doublecomplex *, integer *, doublereal *, doublereal *);
integer iuplo, izero, nerrs;
extern /* Subroutine */ int zppt02_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublereal *, doublereal *), zppt03_(char *,
integer *, doublecomplex *, doublecomplex *, doublecomplex *,
integer *, doublereal *, doublereal *, doublereal *);
logical zerot;
extern /* Subroutine */ int zcopy_(integer *, doublecomplex *, integer *,
doublecomplex *, integer *), zppt05_(char *, integer *, integer *,
doublecomplex *, doublecomplex *, integer *, doublecomplex *,
integer *, doublecomplex *, integer *, doublereal *, doublereal *,
doublereal *);
char xtype[1];
extern /* Subroutine */ int zlatb4_(char *, integer *, integer *, integer
*, char *, integer *, integer *, doublereal *, integer *,
doublereal *, char *), alaerh_(char *,
char *, integer *, integer *, char *, integer *, integer *,
integer *, integer *, integer *, integer *, integer *, integer *,
integer *);
doublereal rcondc;
char packit[1];
extern /* Subroutine */ int alasum_(char *, integer *, integer *, integer
*, integer *);
doublereal cndnum;
extern /* Subroutine */ int zlaipd_(integer *, doublecomplex *, integer *,
integer *);
logical trfcon;
extern doublereal zlanhp_(char *, char *, integer *, doublecomplex *,
doublereal *);
extern /* Subroutine */ int zhpcon_(char *, integer *, doublecomplex *,
integer *, doublereal *, doublereal *, doublecomplex *, integer *), zlacpy_(char *, integer *, integer *, doublecomplex *,
integer *, doublecomplex *, integer *), zlarhs_(char *,
char *, char *, char *, integer *, integer *, integer *, integer *
, integer *, doublecomplex *, integer *, doublecomplex *, integer
*, doublecomplex *, integer *, integer *, integer *), zlatms_(integer *, integer *, char *,
integer *, char *, doublereal *, integer *, doublereal *,
doublereal *, integer *, integer *, char *, doublecomplex *,
integer *, doublecomplex *, integer *),
zhprfs_(char *, integer *, integer *, doublecomplex *,
doublecomplex *, integer *, doublecomplex *, integer *,
doublecomplex *, integer *, doublereal *, doublereal *,
doublecomplex *, doublereal *, integer *), zhptrf_(char *,
integer *, doublecomplex *, integer *, integer *);
doublereal result[8];
extern /* Subroutine */ int zhptri_(char *, integer *, doublecomplex *,
integer *, doublecomplex *, integer *), zhptrs_(char *,
integer *, integer *, doublecomplex *, integer *, doublecomplex *,
integer *, integer *), zerrsy_(char *, integer *)
;
/* Fortran I/O blocks */
static cilist io___38 = { 0, 0, 0, fmt_9999, 0 };
static cilist io___41 = { 0, 0, 0, fmt_9998, 0 };
static cilist io___43 = { 0, 0, 0, fmt_9999, 0 };
/* -- LAPACK test routine (version 3.1) -- */
/* Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/* November 2006 */
/* .. Scalar Arguments .. */
/* .. */
/* .. Array Arguments .. */
/* .. */
/* Purpose */
/* ======= */
/* ZCHKHP tests ZHPTRF, -TRI, -TRS, -RFS, and -CON */
/* 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. */
/* 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 dimension N. */
/* NNS (input) INTEGER */
/* The number of values of NRHS contained in the vector NSVAL. */
/* NSVAL (input) INTEGER array, dimension (NNS) */
/* The values of the number of right hand sides NRHS. */
/* 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. */
/* NMAX (input) INTEGER */
/* The maximum value permitted for N, used in dimensioning the */
/* work arrays. */
/* A (workspace) COMPLEX*16 array, dimension */
/* (NMAX*(NMAX+1)/2) */
/* AFAC (workspace) COMPLEX*16 array, dimension */
/* (NMAX*(NMAX+1)/2) */
/* AINV (workspace) COMPLEX*16 array, dimension */
/* (NMAX*(NMAX+1)/2) */
/* B (workspace) COMPLEX*16 array, dimension (NMAX*NSMAX) */
/* where NSMAX is the largest entry in NSVAL. */
/* X (workspace) COMPLEX*16 array, dimension (NMAX*NSMAX) */
/* XACT (workspace) COMPLEX*16 array, dimension (NMAX*NSMAX) */
/* WORK (workspace) COMPLEX*16 array, dimension */
/* (NMAX*max(2,NSMAX)) */
/* RWORK (workspace) DOUBLE PRECISION array, */
/* dimension (NMAX+2*NSMAX) */
/* 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;
--rwork;
--work;
--xact;
--x;
--b;
--ainv;
--afac;
--a;
--nsval;
--nval;
--dotype;
/* Function Body */
/* .. */
/* .. Executable Statements .. */
/* Initialize constants and the random number seed. */
s_copy(path, "Zomplex precision", (ftnlen)1, (ftnlen)17);
s_copy(path + 1, "HP", (ftnlen)2, (ftnlen)2);
nrun = 0;
nfail = 0;
nerrs = 0;
for (i__ = 1; i__ <= 4; ++i__) {
iseed[i__ - 1] = iseedy[i__ - 1];
/* L10: */
}
/* Test the error exits */
if (*tsterr) {
zerrsy_(path, nout);
}
infoc_1.infot = 0;
/* Do for each value of N in NVAL */
i__1 = *nn;
for (in = 1; in <= i__1; ++in) {
n = nval[in];
lda = max(n,1);
*(unsigned char *)xtype = 'N';
nimat = 10;
if (n <= 0) {
nimat = 1;
}
izero = 0;
i__2 = nimat;
for (imat = 1; imat <= i__2; ++imat) {
/* Do the tests only if DOTYPE( IMAT ) is true. */
if (! dotype[imat]) {
goto L160;
}
/* Skip types 3, 4, 5, or 6 if the matrix size is too small. */
zerot = imat >= 3 && imat <= 6;
if (zerot && n < imat - 2) {
goto L160;
}
/* Do first for UPLO = 'U', then for UPLO = 'L' */
for (iuplo = 1; iuplo <= 2; ++iuplo) {
*(unsigned char *)uplo = *(unsigned char *)&uplos[iuplo - 1];
if (lsame_(uplo, "U")) {
*(unsigned char *)packit = 'C';
} else {
*(unsigned char *)packit = 'R';
}
/* Set up parameters with ZLATB4 and generate a test matrix */
/* with ZLATMS. */
zlatb4_(path, &imat, &n, &n, type__, &kl, &ku, &anorm, &mode,
&cndnum, dist);
s_copy(srnamc_1.srnamt, "ZLATMS", (ftnlen)6, (ftnlen)6);
zlatms_(&n, &n, dist, iseed, type__, &rwork[1], &mode, &
cndnum, &anorm, &kl, &ku, packit, &a[1], &lda, &work[
1], &info);
/* Check error code from ZLATMS. */
if (info != 0) {
alaerh_(path, "ZLATMS", &info, &c__0, uplo, &n, &n, &c_n1,
&c_n1, &c_n1, &imat, &nfail, &nerrs, nout);
goto L150;
}
/* For types 3-6, zero one or more rows and columns of */
/* the matrix to test that INFO is returned correctly. */
if (zerot) {
if (imat == 3) {
izero = 1;
} else if (imat == 4) {
izero = n;
} else {
izero = n / 2 + 1;
}
if (imat < 6) {
/* Set row and column IZERO to zero. */
if (iuplo == 1) {
ioff = (izero - 1) * izero / 2;
i__3 = izero - 1;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = ioff + i__;
a[i__4].r = 0., a[i__4].i = 0.;
/* L20: */
}
ioff += izero;
i__3 = n;
for (i__ = izero; i__ <= i__3; ++i__) {
i__4 = ioff;
a[i__4].r = 0., a[i__4].i = 0.;
ioff += i__;
/* L30: */
}
} else {
ioff = izero;
i__3 = izero - 1;
for (i__ = 1; i__ <= i__3; ++i__) {
i__4 = ioff;
a[i__4].r = 0., a[i__4].i = 0.;
ioff = ioff + n - i__;
/* L40: */
}
ioff -= izero;
i__3 = n;
for (i__ = izero; i__ <= i__3; ++i__) {
i__4 = ioff + i__;
a[i__4].r = 0., a[i__4].i = 0.;
/* L50: */
}
}
} else {
ioff = 0;
if (iuplo == 1) {
/* Set the first IZERO rows and columns to zero. */
i__3 = n;
for (j = 1; j <= i__3; ++j) {
i2 = min(j,izero);
i__4 = i2;
for (i__ = 1; i__ <= i__4; ++i__) {
i__5 = ioff + i__;
a[i__5].r = 0., a[i__5].i = 0.;
/* L60: */
}
ioff += j;
/* L70: */
}
} else {
/* Set the last IZERO rows and columns to zero. */
i__3 = n;
for (j = 1; j <= i__3; ++j) {
i1 = max(j,izero);
i__4 = n;
for (i__ = i1; i__ <= i__4; ++i__) {
i__5 = ioff + i__;
a[i__5].r = 0., a[i__5].i = 0.;
/* L80: */
}
ioff = ioff + n - j;
/* L90: */
}
}
}
} else {
izero = 0;
}
/* Set the imaginary part of the diagonals. */
if (iuplo == 1) {
zlaipd_(&n, &a[1], &c__2, &c__1);
} else {
zlaipd_(&n, &a[1], &n, &c_n1);
}
/* Compute the L*D*L' or U*D*U' factorization of the matrix. */
npp = n * (n + 1) / 2;
zcopy_(&npp, &a[1], &c__1, &afac[1], &c__1);
s_copy(srnamc_1.srnamt, "ZHPTRF", (ftnlen)6, (ftnlen)6);
zhptrf_(uplo, &n, &afac[1], &iwork[1], &info);
/* Adjust the expected value of INFO to account for */
/* pivoting. */
k = izero;
if (k > 0) {
L100:
if (iwork[k] < 0) {
if (iwork[k] != -k) {
k = -iwork[k];
goto L100;
}
} else if (iwork[k] != k) {
k = iwork[k];
goto L100;
}
}
/* Check error code from ZHPTRF. */
if (info != k) {
alaerh_(path, "ZHPTRF", &info, &k, uplo, &n, &n, &c_n1, &
c_n1, &c_n1, &imat, &nfail, &nerrs, nout);
}
if (info != 0) {
trfcon = TRUE_;
} else {
trfcon = FALSE_;
}
/* + TEST 1 */
/* Reconstruct matrix from factors and compute residual. */
zhpt01_(uplo, &n, &a[1], &afac[1], &iwork[1], &ainv[1], &lda,
&rwork[1], result);
nt = 1;
/* + TEST 2 */
/* Form the inverse and compute the residual. */
if (! trfcon) {
zcopy_(&npp, &afac[1], &c__1, &ainv[1], &c__1);
s_copy(srnamc_1.srnamt, "ZHPTRI", (ftnlen)6, (ftnlen)6);
zhptri_(uplo, &n, &ainv[1], &iwork[1], &work[1], &info);
/* Check error code from ZHPTRI. */
if (info != 0) {
alaerh_(path, "ZHPTRI", &info, &c__0, uplo, &n, &n, &
c_n1, &c_n1, &c_n1, &imat, &nfail, &nerrs,
nout);
}
zppt03_(uplo, &n, &a[1], &ainv[1], &work[1], &lda, &rwork[
1], &rcondc, &result[1]);
nt = 2;
}
/* Print information about the tests that did not pass */
/* the threshold. */
i__3 = nt;
for (k = 1; k <= i__3; ++k) {
if (result[k - 1] >= *thresh) {
if (nfail == 0 && nerrs == 0) {
alahd_(nout, path);
}
io___38.ciunit = *nout;
s_wsfe(&io___38);
do_fio(&c__1, uplo, (ftnlen)1);
do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&imat, (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;
}
/* L110: */
}
nrun += nt;
/* Do only the condition estimate if INFO is not 0. */
if (trfcon) {
rcondc = 0.;
goto L140;
}
i__3 = *nns;
for (irhs = 1; irhs <= i__3; ++irhs) {
nrhs = nsval[irhs];
/* + TEST 3 */
/* Solve and compute residual for A * X = B. */
s_copy(srnamc_1.srnamt, "ZLARHS", (ftnlen)6, (ftnlen)6);
zlarhs_(path, xtype, uplo, " ", &n, &n, &kl, &ku, &nrhs, &
a[1], &lda, &xact[1], &lda, &b[1], &lda, iseed, &
info);
*(unsigned char *)xtype = 'C';
zlacpy_("Full", &n, &nrhs, &b[1], &lda, &x[1], &lda);
s_copy(srnamc_1.srnamt, "ZHPTRS", (ftnlen)6, (ftnlen)6);
zhptrs_(uplo, &n, &nrhs, &afac[1], &iwork[1], &x[1], &lda,
&info);
/* Check error code from ZHPTRS. */
if (info != 0) {
alaerh_(path, "ZHPTRS", &info, &c__0, uplo, &n, &n, &
c_n1, &c_n1, &nrhs, &imat, &nfail, &nerrs,
nout);
}
zlacpy_("Full", &n, &nrhs, &b[1], &lda, &work[1], &lda);
zppt02_(uplo, &n, &nrhs, &a[1], &x[1], &lda, &work[1], &
lda, &rwork[1], &result[2]);
/* + TEST 4 */
/* Check solution from generated exact solution. */
zget04_(&n, &nrhs, &x[1], &lda, &xact[1], &lda, &rcondc, &
result[3]);
/* + TESTS 5, 6, and 7 */
/* Use iterative refinement to improve the solution. */
s_copy(srnamc_1.srnamt, "ZHPRFS", (ftnlen)6, (ftnlen)6);
zhprfs_(uplo, &n, &nrhs, &a[1], &afac[1], &iwork[1], &b[1]
, &lda, &x[1], &lda, &rwork[1], &rwork[nrhs + 1],
&work[1], &rwork[(nrhs << 1) + 1], &info);
/* Check error code from ZHPRFS. */
if (info != 0) {
alaerh_(path, "ZHPRFS", &info, &c__0, uplo, &n, &n, &
c_n1, &c_n1, &nrhs, &imat, &nfail, &nerrs,
nout);
}
zget04_(&n, &nrhs, &x[1], &lda, &xact[1], &lda, &rcondc, &
result[4]);
zppt05_(uplo, &n, &nrhs, &a[1], &b[1], &lda, &x[1], &lda,
&xact[1], &lda, &rwork[1], &rwork[nrhs + 1], &
result[5]);
/* Print information about the tests that did not pass */
/* the threshold. */
for (k = 3; k <= 7; ++k) {
if (result[k - 1] >= *thresh) {
if (nfail == 0 && nerrs == 0) {
alahd_(nout, path);
}
io___41.ciunit = *nout;
s_wsfe(&io___41);
do_fio(&c__1, uplo, (ftnlen)1);
do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer))
;
do_fio(&c__1, (char *)&nrhs, (ftnlen)sizeof(
integer));
do_fio(&c__1, (char *)&imat, (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;
}
/* L120: */
}
nrun += 5;
/* L130: */
}
/* + TEST 8 */
/* Get an estimate of RCOND = 1/CNDNUM. */
L140:
anorm = zlanhp_("1", uplo, &n, &a[1], &rwork[1]);
s_copy(srnamc_1.srnamt, "ZHPCON", (ftnlen)6, (ftnlen)6);
zhpcon_(uplo, &n, &afac[1], &iwork[1], &anorm, &rcond, &work[
1], &info);
/* Check error code from ZHPCON. */
if (info != 0) {
alaerh_(path, "ZHPCON", &info, &c__0, uplo, &n, &n, &c_n1,
&c_n1, &c_n1, &imat, &nfail, &nerrs, nout);
}
result[7] = dget06_(&rcond, &rcondc);
/* Print the test ratio if it is .GE. THRESH. */
if (result[7] >= *thresh) {
if (nfail == 0 && nerrs == 0) {
alahd_(nout, path);
}
io___43.ciunit = *nout;
s_wsfe(&io___43);
do_fio(&c__1, uplo, (ftnlen)1);
do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&imat, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&c__8, (ftnlen)sizeof(integer));
do_fio(&c__1, (char *)&result[7], (ftnlen)sizeof(
doublereal));
e_wsfe();
++nfail;
}
++nrun;
L150:
;
}
L160:
;
}
/* L170: */
}
/* Print a summary of the results. */
alasum_(path, nout, &nfail, &nrun, &nerrs);
return 0;
/* End of ZCHKHP */
} /* zchkhp_ */
