floraachy
/
OpenBLAS
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
OpenBLAS/lapack-netlib/SRC/dlaed4.c

1504 lines
37 KiB
C
Raw Blame History

#include <math.h>
#include <stdlib.h>
#include <string.h>
#include <stdio.h>
#include <complex.h>
#ifdef complex
#undef complex
#endif
#ifdef I
#undef I
#endif
#if defined(_WIN64)
typedef long long BLASLONG;
typedef unsigned long long BLASULONG;
#else
typedef long BLASLONG;
typedef unsigned long BLASULONG;
#endif
#ifdef LAPACK_ILP64
typedef BLASLONG blasint;
#if defined(_WIN64)
#define blasabs(x) llabs(x)
#else
#define blasabs(x) labs(x)
#endif
#else
typedef int blasint;
#define blasabs(x) abs(x)
#endif
typedef blasint integer;
typedef unsigned int uinteger;
typedef char *address;
typedef short int shortint;
typedef float real;
typedef double doublereal;
typedef struct { real r, i; } complex;
typedef struct { doublereal r, i; } doublecomplex;
#ifdef _MSC_VER
static inline _Fcomplex Cf(complex *z) {_Fcomplex zz={z->r , z->i}; return zz;}
static inline _Dcomplex Cd(doublecomplex *z) {_Dcomplex zz={z->r , z->i};return zz;}
static inline _Fcomplex * _pCf(complex *z) {return (_Fcomplex*)z;}
static inline _Dcomplex * _pCd(doublecomplex *z) {return (_Dcomplex*)z;}
#else
static inline _Complex float Cf(complex *z) {return z->r + z->i*_Complex_I;}
static inline _Complex double Cd(doublecomplex *z) {return z->r + z->i*_Complex_I;}
static inline _Complex float * _pCf(complex *z) {return (_Complex float*)z;}
static inline _Complex double * _pCd(doublecomplex *z) {return (_Complex double*)z;}
#endif
#define pCf(z) (*_pCf(z))
#define pCd(z) (*_pCd(z))
typedef int logical;
typedef short int shortlogical;
typedef char logical1;
typedef char integer1;
#define TRUE_ (1)
#define FALSE_ (0)
/* Extern is for use with -E */
#ifndef Extern
#define Extern extern
#endif
/* I/O stuff */
typedef int flag;
typedef int ftnlen;
typedef int ftnint;
/*external read, write*/
typedef struct
{ flag cierr;
ftnint ciunit;
flag ciend;
char *cifmt;
ftnint cirec;
} cilist;
/*internal read, write*/
typedef struct
{ flag icierr;
char *iciunit;
flag iciend;
char *icifmt;
ftnint icirlen;
ftnint icirnum;
} icilist;
/*open*/
typedef struct
{ flag oerr;
ftnint ounit;
char *ofnm;
ftnlen ofnmlen;
char *osta;
char *oacc;
char *ofm;
ftnint orl;
char *oblnk;
} olist;
/*close*/
typedef struct
{ flag cerr;
ftnint cunit;
char *csta;
} cllist;
/*rewind, backspace, endfile*/
typedef struct
{ flag aerr;
ftnint aunit;
} alist;
/* inquire */
typedef struct
{ flag inerr;
ftnint inunit;
char *infile;
ftnlen infilen;
ftnint *inex; /*parameters in standard's order*/
ftnint *inopen;
ftnint *innum;
ftnint *innamed;
char *inname;
ftnlen innamlen;
char *inacc;
ftnlen inacclen;
char *inseq;
ftnlen inseqlen;
char *indir;
ftnlen indirlen;
char *infmt;
ftnlen infmtlen;
char *inform;
ftnint informlen;
char *inunf;
ftnlen inunflen;
ftnint *inrecl;
ftnint *innrec;
char *inblank;
ftnlen inblanklen;
} inlist;
#define VOID void
union Multitype { /* for multiple entry points */
integer1 g;
shortint h;
integer i;
/* longint j; */
real r;
doublereal d;
complex c;
doublecomplex z;
};
typedef union Multitype Multitype;
struct Vardesc { /* for Namelist */
char *name;
char *addr;
ftnlen *dims;
int type;
};
typedef struct Vardesc Vardesc;
struct Namelist {
char *name;
Vardesc **vars;
int nvars;
};
typedef struct Namelist Namelist;
#define abs(x) ((x) >= 0 ? (x) : -(x))
#define dabs(x) (fabs(x))
#define f2cmin(a,b) ((a) <= (b) ? (a) : (b))
#define f2cmax(a,b) ((a) >= (b) ? (a) : (b))
#define dmin(a,b) (f2cmin(a,b))
#define dmax(a,b) (f2cmax(a,b))
#define bit_test(a,b) ((a) >> (b) & 1)
#define bit_clear(a,b) ((a) & ~((uinteger)1 << (b)))
#define bit_set(a,b) ((a) | ((uinteger)1 << (b)))
#define abort_() { sig_die("Fortran abort routine called", 1); }
#define c_abs(z) (cabsf(Cf(z)))
#define c_cos(R,Z) { pCf(R)=ccos(Cf(Z)); }
#ifdef _MSC_VER
#define c_div(c, a, b) {Cf(c)._Val[0] = (Cf(a)._Val[0]/Cf(b)._Val[0]); Cf(c)._Val[1]=(Cf(a)._Val[1]/Cf(b)._Val[1]);}
#define z_div(c, a, b) {Cd(c)._Val[0] = (Cd(a)._Val[0]/Cd(b)._Val[0]); Cd(c)._Val[1]=(Cd(a)._Val[1]/df(b)._Val[1]);}
#else
#define c_div(c, a, b) {pCf(c) = Cf(a)/Cf(b);}
#define z_div(c, a, b) {pCd(c) = Cd(a)/Cd(b);}
#endif
#define c_exp(R, Z) {pCf(R) = cexpf(Cf(Z));}
#define c_log(R, Z) {pCf(R) = clogf(Cf(Z));}
#define c_sin(R, Z) {pCf(R) = csinf(Cf(Z));}
//#define c_sqrt(R, Z) {*(R) = csqrtf(Cf(Z));}
#define c_sqrt(R, Z) {pCf(R) = csqrtf(Cf(Z));}
#define d_abs(x) (fabs(*(x)))
#define d_acos(x) (acos(*(x)))
#define d_asin(x) (asin(*(x)))
#define d_atan(x) (atan(*(x)))
#define d_atn2(x, y) (atan2(*(x),*(y)))
#define d_cnjg(R, Z) { pCd(R) = conj(Cd(Z)); }
#define r_cnjg(R, Z) { pCf(R) = conjf(Cf(Z)); }
#define d_cos(x) (cos(*(x)))
#define d_cosh(x) (cosh(*(x)))
#define d_dim(__a, __b) ( *(__a) > *(__b) ? *(__a) - *(__b) : 0.0 )
#define d_exp(x) (exp(*(x)))
#define d_imag(z) (cimag(Cd(z)))
#define r_imag(z) (cimagf(Cf(z)))
#define d_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
#define r_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
#define d_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
#define r_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
#define d_log(x) (log(*(x)))
#define d_mod(x, y) (fmod(*(x), *(y)))
#define u_nint(__x) ((__x)>=0 ? floor((__x) + .5) : -floor(.5 - (__x)))
#define d_nint(x) u_nint(*(x))
#define u_sign(__a,__b) ((__b) >= 0 ? ((__a) >= 0 ? (__a) : -(__a)) : -((__a) >= 0 ? (__a) : -(__a)))
#define d_sign(a,b) u_sign(*(a),*(b))
#define r_sign(a,b) u_sign(*(a),*(b))
#define d_sin(x) (sin(*(x)))
#define d_sinh(x) (sinh(*(x)))
#define d_sqrt(x) (sqrt(*(x)))
#define d_tan(x) (tan(*(x)))
#define d_tanh(x) (tanh(*(x)))
#define i_abs(x) abs(*(x))
#define i_dnnt(x) ((integer)u_nint(*(x)))
#define i_len(s, n) (n)
#define i_nint(x) ((integer)u_nint(*(x)))
#define i_sign(a,b) ((integer)u_sign((integer)*(a),(integer)*(b)))
#define pow_dd(ap, bp) ( pow(*(ap), *(bp)))
#define pow_si(B,E) spow_ui(*(B),*(E))
#define pow_ri(B,E) spow_ui(*(B),*(E))
#define pow_di(B,E) dpow_ui(*(B),*(E))
#define pow_zi(p, a, b) {pCd(p) = zpow_ui(Cd(a), *(b));}
#define pow_ci(p, a, b) {pCf(p) = cpow_ui(Cf(a), *(b));}
#define pow_zz(R,A,B) {pCd(R) = cpow(Cd(A),*(B));}
#define s_cat(lpp, rpp, rnp, np, llp) { ftnlen i, nc, ll; char *f__rp, *lp; ll = (llp); lp = (lpp); for(i=0; i < (int)*(np); ++i) { nc = ll; if((rnp)[i] < nc) nc = (rnp)[i]; ll -= nc; f__rp = (rpp)[i]; while(--nc >= 0) *lp++ = *(f__rp)++; } while(--ll >= 0) *lp++ = ' '; }
#define s_cmp(a,b,c,d) ((integer)strncmp((a),(b),f2cmin((c),(d))))
#define s_copy(A,B,C,D) { int __i,__m; for (__i=0, __m=f2cmin((C),(D)); __i<__m && (B)[__i] != 0; ++__i) (A)[__i] = (B)[__i]; }
#define sig_die(s, kill) { exit(1); }
#define s_stop(s, n) {exit(0);}
static char junk[] = "\n@(#)LIBF77 VERSION 19990503\n";
#define z_abs(z) (cabs(Cd(z)))
#define z_exp(R, Z) {pCd(R) = cexp(Cd(Z));}
#define z_sqrt(R, Z) {pCd(R) = csqrt(Cd(Z));}
#define myexit_() break;
#define mycycle() continue;
#define myceiling(w) {ceil(w)}
#define myhuge(w) {HUGE_VAL}
//#define mymaxloc_(w,s,e,n) {if (sizeof(*(w)) == sizeof(double)) dmaxloc_((w),*(s),*(e),n); else dmaxloc_((w),*(s),*(e),n);}
#define mymaxloc(w,s,e,n) {dmaxloc_(w,*(s),*(e),n)}
/* procedure parameter types for -A and -C++ */
#define F2C_proc_par_types 1
#ifdef __cplusplus
typedef logical (*L_fp)(...);
#else
typedef logical (*L_fp)();
#endif
static float spow_ui(float x, integer n) {
float pow=1.0; unsigned long int u;
if(n != 0) {
if(n < 0) n = -n, x = 1/x;
for(u = n; ; ) {
if(u & 01) pow *= x;
if(u >>= 1) x *= x;
else break;
}
}
return pow;
}
static double dpow_ui(double x, integer n) {
double pow=1.0; unsigned long int u;
if(n != 0) {
if(n < 0) n = -n, x = 1/x;
for(u = n; ; ) {
if(u & 01) pow *= x;
if(u >>= 1) x *= x;
else break;
}
}
return pow;
}
#ifdef _MSC_VER
static _Fcomplex cpow_ui(complex x, integer n) {
complex pow={1.0,0.0}; unsigned long int u;
if(n != 0) {
if(n < 0) n = -n, x.r = 1/x.r, x.i=1/x.i;
for(u = n; ; ) {
if(u & 01) pow.r *= x.r, pow.i *= x.i;
if(u >>= 1) x.r *= x.r, x.i *= x.i;
else break;
}
}
_Fcomplex p={pow.r, pow.i};
return p;
}
#else
static _Complex float cpow_ui(_Complex float x, integer n) {
_Complex float pow=1.0; unsigned long int u;
if(n != 0) {
if(n < 0) n = -n, x = 1/x;
for(u = n; ; ) {
if(u & 01) pow *= x;
if(u >>= 1) x *= x;
else break;
}
}
return pow;
}
#endif
#ifdef _MSC_VER
static _Dcomplex zpow_ui(_Dcomplex x, integer n) {
_Dcomplex pow={1.0,0.0}; unsigned long int u;
if(n != 0) {
if(n < 0) n = -n, x._Val[0] = 1/x._Val[0], x._Val[1] =1/x._Val[1];
for(u = n; ; ) {
if(u & 01) pow._Val[0] *= x._Val[0], pow._Val[1] *= x._Val[1];
if(u >>= 1) x._Val[0] *= x._Val[0], x._Val[1] *= x._Val[1];
else break;
}
}
_Dcomplex p = {pow._Val[0], pow._Val[1]};
return p;
}
#else
static _Complex double zpow_ui(_Complex double x, integer n) {
_Complex double pow=1.0; unsigned long int u;
if(n != 0) {
if(n < 0) n = -n, x = 1/x;
for(u = n; ; ) {
if(u & 01) pow *= x;
if(u >>= 1) x *= x;
else break;
}
}
return pow;
}
#endif
static integer pow_ii(integer x, integer n) {
integer pow; unsigned long int u;
if (n <= 0) {
if (n == 0 || x == 1) pow = 1;
else if (x != -1) pow = x == 0 ? 1/x : 0;
else n = -n;
}
if ((n > 0) || !(n == 0 || x == 1 || x != -1)) {
u = n;
for(pow = 1; ; ) {
if(u & 01) pow *= x;
if(u >>= 1) x *= x;
else break;
}
}
return pow;
}
static integer dmaxloc_(double *w, integer s, integer e, integer *n)
{
double m; integer i, mi;
for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
if (w[i-1]>m) mi=i ,m=w[i-1];
return mi-s+1;
}
static integer smaxloc_(float *w, integer s, integer e, integer *n)
{
float m; integer i, mi;
for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
if (w[i-1]>m) mi=i ,m=w[i-1];
return mi-s+1;
}
static inline void cdotc_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
integer n = *n_, incx = *incx_, incy = *incy_, i;
#ifdef _MSC_VER
_Fcomplex zdotc = {0.0, 0.0};
if (incx == 1 && incy == 1) {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc._Val[0] += conjf(Cf(&x[i]))._Val[0] * Cf(&y[i])._Val[0];
zdotc._Val[1] += conjf(Cf(&x[i]))._Val[1] * Cf(&y[i])._Val[1];
}
} else {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc._Val[0] += conjf(Cf(&x[i*incx]))._Val[0] * Cf(&y[i*incy])._Val[0];
zdotc._Val[1] += conjf(Cf(&x[i*incx]))._Val[1] * Cf(&y[i*incy])._Val[1];
}
}
pCf(z) = zdotc;
}
#else
_Complex float zdotc = 0.0;
if (incx == 1 && incy == 1) {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc += conjf(Cf(&x[i])) * Cf(&y[i]);
}
} else {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc += conjf(Cf(&x[i*incx])) * Cf(&y[i*incy]);
}
}
pCf(z) = zdotc;
}
#endif
static inline void zdotc_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
integer n = *n_, incx = *incx_, incy = *incy_, i;
#ifdef _MSC_VER
_Dcomplex zdotc = {0.0, 0.0};
if (incx == 1 && incy == 1) {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc._Val[0] += conj(Cd(&x[i]))._Val[0] * Cd(&y[i])._Val[0];
zdotc._Val[1] += conj(Cd(&x[i]))._Val[1] * Cd(&y[i])._Val[1];
}
} else {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc._Val[0] += conj(Cd(&x[i*incx]))._Val[0] * Cd(&y[i*incy])._Val[0];
zdotc._Val[1] += conj(Cd(&x[i*incx]))._Val[1] * Cd(&y[i*incy])._Val[1];
}
}
pCd(z) = zdotc;
}
#else
_Complex double zdotc = 0.0;
if (incx == 1 && incy == 1) {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc += conj(Cd(&x[i])) * Cd(&y[i]);
}
} else {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc += conj(Cd(&x[i*incx])) * Cd(&y[i*incy]);
}
}
pCd(z) = zdotc;
}
#endif
static inline void cdotu_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
integer n = *n_, incx = *incx_, incy = *incy_, i;
#ifdef _MSC_VER
_Fcomplex zdotc = {0.0, 0.0};
if (incx == 1 && incy == 1) {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc._Val[0] += Cf(&x[i])._Val[0] * Cf(&y[i])._Val[0];
zdotc._Val[1] += Cf(&x[i])._Val[1] * Cf(&y[i])._Val[1];
}
} else {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc._Val[0] += Cf(&x[i*incx])._Val[0] * Cf(&y[i*incy])._Val[0];
zdotc._Val[1] += Cf(&x[i*incx])._Val[1] * Cf(&y[i*incy])._Val[1];
}
}
pCf(z) = zdotc;
}
#else
_Complex float zdotc = 0.0;
if (incx == 1 && incy == 1) {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc += Cf(&x[i]) * Cf(&y[i]);
}
} else {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc += Cf(&x[i*incx]) * Cf(&y[i*incy]);
}
}
pCf(z) = zdotc;
}
#endif
static inline void zdotu_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
integer n = *n_, incx = *incx_, incy = *incy_, i;
#ifdef _MSC_VER
_Dcomplex zdotc = {0.0, 0.0};
if (incx == 1 && incy == 1) {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc._Val[0] += Cd(&x[i])._Val[0] * Cd(&y[i])._Val[0];
zdotc._Val[1] += Cd(&x[i])._Val[1] * Cd(&y[i])._Val[1];
}
} else {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc._Val[0] += Cd(&x[i*incx])._Val[0] * Cd(&y[i*incy])._Val[0];
zdotc._Val[1] += Cd(&x[i*incx])._Val[1] * Cd(&y[i*incy])._Val[1];
}
}
pCd(z) = zdotc;
}
#else
_Complex double zdotc = 0.0;
if (incx == 1 && incy == 1) {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc += Cd(&x[i]) * Cd(&y[i]);
}
} else {
for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
zdotc += Cd(&x[i*incx]) * Cd(&y[i*incy]);
}
}
pCd(z) = zdotc;
}
#endif
/* -- translated by f2c (version 20000121).
You must link the resulting object file with the libraries:
-lf2c -lm (in that order)
*/
/* > \brief \b DLAED4 used by sstedc. Finds a single root of the secular equation. */
/* =========== DOCUMENTATION =========== */
/* Online html documentation available at */
/* http://www.netlib.org/lapack/explore-html/ */
/* > \htmlonly */
/* > Download DLAED4 + dependencies */
/* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dlaed4.
f"> */
/* > [TGZ]</a> */
/* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dlaed4.
f"> */
/* > [ZIP]</a> */
/* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dlaed4.
f"> */
/* > [TXT]</a> */
/* > \endhtmlonly */
/* Definition: */
/* =========== */
/* SUBROUTINE DLAED4( N, I, D, Z, DELTA, RHO, DLAM, INFO ) */
/* INTEGER I, INFO, N */
/* DOUBLE PRECISION DLAM, RHO */
/* DOUBLE PRECISION D( * ), DELTA( * ), Z( * ) */
/* > \par Purpose: */
/* ============= */
/* > */
/* > \verbatim */
/* > */
/* > This subroutine computes the I-th updated eigenvalue of a symmetric */
/* > rank-one modification to a diagonal matrix whose elements are */
/* > given in the array d, and that */
/* > */
/* > D(i) < D(j) for i < j */
/* > */
/* > and that RHO > 0. This is arranged by the calling routine, and is */
/* > no loss in generality. The rank-one modified system is thus */
/* > */
/* > diag( D ) + RHO * Z * Z_transpose. */
/* > */
/* > where we assume the Euclidean norm of Z is 1. */
/* > */
/* > The method consists of approximating the rational functions in the */
/* > secular equation by simpler interpolating rational functions. */
/* > \endverbatim */
/* Arguments: */
/* ========== */
/* > \param[in] N */
/* > \verbatim */
/* > N is INTEGER */
/* > The length of all arrays. */
/* > \endverbatim */
/* > */
/* > \param[in] I */
/* > \verbatim */
/* > I is INTEGER */
/* > The index of the eigenvalue to be computed. 1 <= I <= N. */
/* > \endverbatim */
/* > */
/* > \param[in] D */
/* > \verbatim */
/* > D is DOUBLE PRECISION array, dimension (N) */
/* > The original eigenvalues. It is assumed that they are in */
/* > order, D(I) < D(J) for I < J. */
/* > \endverbatim */
/* > */
/* > \param[in] Z */
/* > \verbatim */
/* > Z is DOUBLE PRECISION array, dimension (N) */
/* > The components of the updating vector. */
/* > \endverbatim */
/* > */
/* > \param[out] DELTA */
/* > \verbatim */
/* > DELTA is DOUBLE PRECISION array, dimension (N) */
/* > If N > 2, DELTA contains (D(j) - lambda_I) in its j-th */
/* > component. If N = 1, then DELTA(1) = 1. If N = 2, see DLAED5 */
/* > for detail. The vector DELTA contains the information necessary */
/* > to construct the eigenvectors by DLAED3 and DLAED9. */
/* > \endverbatim */
/* > */
/* > \param[in] RHO */
/* > \verbatim */
/* > RHO is DOUBLE PRECISION */
/* > The scalar in the symmetric updating formula. */
/* > \endverbatim */
/* > */
/* > \param[out] DLAM */
/* > \verbatim */
/* > DLAM is DOUBLE PRECISION */
/* > The computed lambda_I, the I-th updated eigenvalue. */
/* > \endverbatim */
/* > */
/* > \param[out] INFO */
/* > \verbatim */
/* > INFO is INTEGER */
/* > = 0: successful exit */
/* > > 0: if INFO = 1, the updating process failed. */
/* > \endverbatim */
/* > \par Internal Parameters: */
/* ========================= */
/* > */
/* > \verbatim */
/* > Logical variable ORGATI (origin-at-i?) is used for distinguishing */
/* > whether D(i) or D(i+1) is treated as the origin. */
/* > */
/* > ORGATI = .true. origin at i */
/* > ORGATI = .false. origin at i+1 */
/* > */
/* > Logical variable SWTCH3 (switch-for-3-poles?) is for noting */
/* > if we are working with THREE poles! */
/* > */
/* > MAXIT is the maximum number of iterations allowed for each */
/* > eigenvalue. */
/* > \endverbatim */
/* Authors: */
/* ======== */
/* > \author Univ. of Tennessee */
/* > \author Univ. of California Berkeley */
/* > \author Univ. of Colorado Denver */
/* > \author NAG Ltd. */
/* > \date December 2016 */
/* > \ingroup auxOTHERcomputational */
/* > \par Contributors: */
/* ================== */
/* > */
/* > Ren-Cang Li, Computer Science Division, University of California */
/* > at Berkeley, USA */
/* > */
/* ===================================================================== */
/* Subroutine */ void dlaed4_(integer *n, integer *i__, doublereal *d__,
doublereal *z__, doublereal *delta, doublereal *rho, doublereal *dlam,
integer *info)
{
/* System generated locals */
integer i__1;
doublereal d__1;
/* Local variables */
doublereal dphi, dpsi;
integer iter;
doublereal temp, prew, temp1, a, b, c__;
integer j;
doublereal w, dltlb, dltub, midpt;
integer niter;
logical swtch;
extern /* Subroutine */ void dlaed5_(integer *, doublereal *, doublereal *,
doublereal *, doublereal *, doublereal *), dlaed6_(integer *,
logical *, doublereal *, doublereal *, doublereal *, doublereal *,
doublereal *, integer *);
logical swtch3;
integer ii;
extern doublereal dlamch_(char *);
doublereal dw, zz[3];
logical orgati;
doublereal erretm, rhoinv;
integer ip1;
doublereal del, eta, phi, eps, tau, psi;
integer iim1, iip1;
/* -- LAPACK computational routine (version 3.7.0) -- */
/* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
/* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
/* December 2016 */
/* ===================================================================== */
/* Since this routine is called in an inner loop, we do no argument */
/* checking. */
/* Quick return for N=1 and 2. */
/* Parameter adjustments */
--delta;
--z__;
--d__;
/* Function Body */
*info = 0;
if (*n == 1) {
/* Presumably, I=1 upon entry */
*dlam = d__[1] + *rho * z__[1] * z__[1];
delta[1] = 1.;
return;
}
if (*n == 2) {
dlaed5_(i__, &d__[1], &z__[1], &delta[1], rho, dlam);
return;
}
/* Compute machine epsilon */
eps = dlamch_("Epsilon");
rhoinv = 1. / *rho;
/* The case I = N */
if (*i__ == *n) {
/* Initialize some basic variables */
ii = *n - 1;
niter = 1;
/* Calculate initial guess */
midpt = *rho / 2.;
/* If ||Z||_2 is not one, then TEMP should be set to */
/* RHO * ||Z||_2^2 / TWO */
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
delta[j] = d__[j] - d__[*i__] - midpt;
/* L10: */
}
psi = 0.;
i__1 = *n - 2;
for (j = 1; j <= i__1; ++j) {
psi += z__[j] * z__[j] / delta[j];
/* L20: */
}
c__ = rhoinv + psi;
w = c__ + z__[ii] * z__[ii] / delta[ii] + z__[*n] * z__[*n] / delta[*
n];
if (w <= 0.) {
temp = z__[*n - 1] * z__[*n - 1] / (d__[*n] - d__[*n - 1] + *rho)
+ z__[*n] * z__[*n] / *rho;
if (c__ <= temp) {
tau = *rho;
} else {
del = d__[*n] - d__[*n - 1];
a = -c__ * del + z__[*n - 1] * z__[*n - 1] + z__[*n] * z__[*n]
;
b = z__[*n] * z__[*n] * del;
if (a < 0.) {
tau = b * 2. / (sqrt(a * a + b * 4. * c__) - a);
} else {
tau = (a + sqrt(a * a + b * 4. * c__)) / (c__ * 2.);
}
}
/* It can be proved that */
/* D(N)+RHO/2 <= LAMBDA(N) < D(N)+TAU <= D(N)+RHO */
dltlb = midpt;
dltub = *rho;
} else {
del = d__[*n] - d__[*n - 1];
a = -c__ * del + z__[*n - 1] * z__[*n - 1] + z__[*n] * z__[*n];
b = z__[*n] * z__[*n] * del;
if (a < 0.) {
tau = b * 2. / (sqrt(a * a + b * 4. * c__) - a);
} else {
tau = (a + sqrt(a * a + b * 4. * c__)) / (c__ * 2.);
}
/* It can be proved that */
/* D(N) < D(N)+TAU < LAMBDA(N) < D(N)+RHO/2 */
dltlb = 0.;
dltub = midpt;
}
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
delta[j] = d__[j] - d__[*i__] - tau;
/* L30: */
}
/* Evaluate PSI and the derivative DPSI */
dpsi = 0.;
psi = 0.;
erretm = 0.;
i__1 = ii;
for (j = 1; j <= i__1; ++j) {
temp = z__[j] / delta[j];
psi += z__[j] * temp;
dpsi += temp * temp;
erretm += psi;
/* L40: */
}
erretm = abs(erretm);
/* Evaluate PHI and the derivative DPHI */
temp = z__[*n] / delta[*n];
phi = z__[*n] * temp;
dphi = temp * temp;
erretm = (-phi - psi) * 8. + erretm - phi + rhoinv + abs(tau) * (dpsi
+ dphi);
w = rhoinv + phi + psi;
/* Test for convergence */
if (abs(w) <= eps * erretm) {
*dlam = d__[*i__] + tau;
goto L250;
}
if (w <= 0.) {
dltlb = f2cmax(dltlb,tau);
} else {
dltub = f2cmin(dltub,tau);
}
/* Calculate the new step */
++niter;
c__ = w - delta[*n - 1] * dpsi - delta[*n] * dphi;
a = (delta[*n - 1] + delta[*n]) * w - delta[*n - 1] * delta[*n] * (
dpsi + dphi);
b = delta[*n - 1] * delta[*n] * w;
if (c__ < 0.) {
c__ = abs(c__);
}
if (c__ == 0.) {
/* ETA = B/A */
/* ETA = RHO - TAU */
eta = dltub - tau;
} else if (a >= 0.) {
eta = (a + sqrt((d__1 = a * a - b * 4. * c__, abs(d__1)))) / (c__
* 2.);
} else {
eta = b * 2. / (a - sqrt((d__1 = a * a - b * 4. * c__, abs(d__1)))
);
}
/* Note, eta should be positive if w is negative, and */
/* eta should be negative otherwise. However, */
/* if for some reason caused by roundoff, eta*w > 0, */
/* we simply use one Newton step instead. This way */
/* will guarantee eta*w < 0. */
if (w * eta > 0.) {
eta = -w / (dpsi + dphi);
}
temp = tau + eta;
if (temp > dltub || temp < dltlb) {
if (w < 0.) {
eta = (dltub - tau) / 2.;
} else {
eta = (dltlb - tau) / 2.;
}
}
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
delta[j] -= eta;
/* L50: */
}
tau += eta;
/* Evaluate PSI and the derivative DPSI */
dpsi = 0.;
psi = 0.;
erretm = 0.;
i__1 = ii;
for (j = 1; j <= i__1; ++j) {
temp = z__[j] / delta[j];
psi += z__[j] * temp;
dpsi += temp * temp;
erretm += psi;
/* L60: */
}
erretm = abs(erretm);
/* Evaluate PHI and the derivative DPHI */
temp = z__[*n] / delta[*n];
phi = z__[*n] * temp;
dphi = temp * temp;
erretm = (-phi - psi) * 8. + erretm - phi + rhoinv + abs(tau) * (dpsi
+ dphi);
w = rhoinv + phi + psi;
/* Main loop to update the values of the array DELTA */
iter = niter + 1;
for (niter = iter; niter <= 30; ++niter) {
/* Test for convergence */
if (abs(w) <= eps * erretm) {
*dlam = d__[*i__] + tau;
goto L250;
}
if (w <= 0.) {
dltlb = f2cmax(dltlb,tau);
} else {
dltub = f2cmin(dltub,tau);
}
/* Calculate the new step */
c__ = w - delta[*n - 1] * dpsi - delta[*n] * dphi;
a = (delta[*n - 1] + delta[*n]) * w - delta[*n - 1] * delta[*n] *
(dpsi + dphi);
b = delta[*n - 1] * delta[*n] * w;
if (a >= 0.) {
eta = (a + sqrt((d__1 = a * a - b * 4. * c__, abs(d__1)))) / (
c__ * 2.);
} else {
eta = b * 2. / (a - sqrt((d__1 = a * a - b * 4. * c__, abs(
d__1))));
}
/* Note, eta should be positive if w is negative, and */
/* eta should be negative otherwise. However, */
/* if for some reason caused by roundoff, eta*w > 0, */
/* we simply use one Newton step instead. This way */
/* will guarantee eta*w < 0. */
if (w * eta > 0.) {
eta = -w / (dpsi + dphi);
}
temp = tau + eta;
if (temp > dltub || temp < dltlb) {
if (w < 0.) {
eta = (dltub - tau) / 2.;
} else {
eta = (dltlb - tau) / 2.;
}
}
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
delta[j] -= eta;
/* L70: */
}
tau += eta;
/* Evaluate PSI and the derivative DPSI */
dpsi = 0.;
psi = 0.;
erretm = 0.;
i__1 = ii;
for (j = 1; j <= i__1; ++j) {
temp = z__[j] / delta[j];
psi += z__[j] * temp;
dpsi += temp * temp;
erretm += psi;
/* L80: */
}
erretm = abs(erretm);
/* Evaluate PHI and the derivative DPHI */
temp = z__[*n] / delta[*n];
phi = z__[*n] * temp;
dphi = temp * temp;
erretm = (-phi - psi) * 8. + erretm - phi + rhoinv + abs(tau) * (
dpsi + dphi);
w = rhoinv + phi + psi;
/* L90: */
}
/* Return with INFO = 1, NITER = MAXIT and not converged */
*info = 1;
*dlam = d__[*i__] + tau;
goto L250;
/* End for the case I = N */
} else {
/* The case for I < N */
niter = 1;
ip1 = *i__ + 1;
/* Calculate initial guess */
del = d__[ip1] - d__[*i__];
midpt = del / 2.;
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
delta[j] = d__[j] - d__[*i__] - midpt;
/* L100: */
}
psi = 0.;
i__1 = *i__ - 1;
for (j = 1; j <= i__1; ++j) {
psi += z__[j] * z__[j] / delta[j];
/* L110: */
}
phi = 0.;
i__1 = *i__ + 2;
for (j = *n; j >= i__1; --j) {
phi += z__[j] * z__[j] / delta[j];
/* L120: */
}
c__ = rhoinv + psi + phi;
w = c__ + z__[*i__] * z__[*i__] / delta[*i__] + z__[ip1] * z__[ip1] /
delta[ip1];
if (w > 0.) {
/* d(i)< the ith eigenvalue < (d(i)+d(i+1))/2 */
/* We choose d(i) as origin. */
orgati = TRUE_;
a = c__ * del + z__[*i__] * z__[*i__] + z__[ip1] * z__[ip1];
b = z__[*i__] * z__[*i__] * del;
if (a > 0.) {
tau = b * 2. / (a + sqrt((d__1 = a * a - b * 4. * c__, abs(
d__1))));
} else {
tau = (a - sqrt((d__1 = a * a - b * 4. * c__, abs(d__1)))) / (
c__ * 2.);
}
dltlb = 0.;
dltub = midpt;
} else {
/* (d(i)+d(i+1))/2 <= the ith eigenvalue < d(i+1) */
/* We choose d(i+1) as origin. */
orgati = FALSE_;
a = c__ * del - z__[*i__] * z__[*i__] - z__[ip1] * z__[ip1];
b = z__[ip1] * z__[ip1] * del;
if (a < 0.) {
tau = b * 2. / (a - sqrt((d__1 = a * a + b * 4. * c__, abs(
d__1))));
} else {
tau = -(a + sqrt((d__1 = a * a + b * 4. * c__, abs(d__1)))) /
(c__ * 2.);
}
dltlb = -midpt;
dltub = 0.;
}
if (orgati) {
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
delta[j] = d__[j] - d__[*i__] - tau;
/* L130: */
}
} else {
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
delta[j] = d__[j] - d__[ip1] - tau;
/* L140: */
}
}
if (orgati) {
ii = *i__;
} else {
ii = *i__ + 1;
}
iim1 = ii - 1;
iip1 = ii + 1;
/* Evaluate PSI and the derivative DPSI */
dpsi = 0.;
psi = 0.;
erretm = 0.;
i__1 = iim1;
for (j = 1; j <= i__1; ++j) {
temp = z__[j] / delta[j];
psi += z__[j] * temp;
dpsi += temp * temp;
erretm += psi;
/* L150: */
}
erretm = abs(erretm);
/* Evaluate PHI and the derivative DPHI */
dphi = 0.;
phi = 0.;
i__1 = iip1;
for (j = *n; j >= i__1; --j) {
temp = z__[j] / delta[j];
phi += z__[j] * temp;
dphi += temp * temp;
erretm += phi;
/* L160: */
}
w = rhoinv + phi + psi;
/* W is the value of the secular function with */
/* its ii-th element removed. */
swtch3 = FALSE_;
if (orgati) {
if (w < 0.) {
swtch3 = TRUE_;
}
} else {
if (w > 0.) {
swtch3 = TRUE_;
}
}
if (ii == 1 || ii == *n) {
swtch3 = FALSE_;
}
temp = z__[ii] / delta[ii];
dw = dpsi + dphi + temp * temp;
temp = z__[ii] * temp;
w += temp;
erretm = (phi - psi) * 8. + erretm + rhoinv * 2. + abs(temp) * 3. +
abs(tau) * dw;
/* Test for convergence */
if (abs(w) <= eps * erretm) {
if (orgati) {
*dlam = d__[*i__] + tau;
} else {
*dlam = d__[ip1] + tau;
}
goto L250;
}
if (w <= 0.) {
dltlb = f2cmax(dltlb,tau);
} else {
dltub = f2cmin(dltub,tau);
}
/* Calculate the new step */
++niter;
if (! swtch3) {
if (orgati) {
/* Computing 2nd power */
d__1 = z__[*i__] / delta[*i__];
c__ = w - delta[ip1] * dw - (d__[*i__] - d__[ip1]) * (d__1 *
d__1);
} else {
/* Computing 2nd power */
d__1 = z__[ip1] / delta[ip1];
c__ = w - delta[*i__] * dw - (d__[ip1] - d__[*i__]) * (d__1 *
d__1);
}
a = (delta[*i__] + delta[ip1]) * w - delta[*i__] * delta[ip1] *
dw;
b = delta[*i__] * delta[ip1] * w;
if (c__ == 0.) {
if (a == 0.) {
if (orgati) {
a = z__[*i__] * z__[*i__] + delta[ip1] * delta[ip1] *
(dpsi + dphi);
} else {
a = z__[ip1] * z__[ip1] + delta[*i__] * delta[*i__] *
(dpsi + dphi);
}
}
eta = b / a;
} else if (a <= 0.) {
eta = (a - sqrt((d__1 = a * a - b * 4. * c__, abs(d__1)))) / (
c__ * 2.);
} else {
eta = b * 2. / (a + sqrt((d__1 = a * a - b * 4. * c__, abs(
d__1))));
}
} else {
/* Interpolation using THREE most relevant poles */
temp = rhoinv + psi + phi;
if (orgati) {
temp1 = z__[iim1] / delta[iim1];
temp1 *= temp1;
c__ = temp - delta[iip1] * (dpsi + dphi) - (d__[iim1] - d__[
iip1]) * temp1;
zz[0] = z__[iim1] * z__[iim1];
zz[2] = delta[iip1] * delta[iip1] * (dpsi - temp1 + dphi);
} else {
temp1 = z__[iip1] / delta[iip1];
temp1 *= temp1;
c__ = temp - delta[iim1] * (dpsi + dphi) - (d__[iip1] - d__[
iim1]) * temp1;
zz[0] = delta[iim1] * delta[iim1] * (dpsi + (dphi - temp1));
zz[2] = z__[iip1] * z__[iip1];
}
zz[1] = z__[ii] * z__[ii];
dlaed6_(&niter, &orgati, &c__, &delta[iim1], zz, &w, &eta, info);
if (*info != 0) {
goto L250;
}
}
/* Note, eta should be positive if w is negative, and */
/* eta should be negative otherwise. However, */
/* if for some reason caused by roundoff, eta*w > 0, */
/* we simply use one Newton step instead. This way */
/* will guarantee eta*w < 0. */
if (w * eta >= 0.) {
eta = -w / dw;
}
temp = tau + eta;
if (temp > dltub || temp < dltlb) {
if (w < 0.) {
eta = (dltub - tau) / 2.;
} else {
eta = (dltlb - tau) / 2.;
}
}
prew = w;
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
delta[j] -= eta;
/* L180: */
}
/* Evaluate PSI and the derivative DPSI */
dpsi = 0.;
psi = 0.;
erretm = 0.;
i__1 = iim1;
for (j = 1; j <= i__1; ++j) {
temp = z__[j] / delta[j];
psi += z__[j] * temp;
dpsi += temp * temp;
erretm += psi;
/* L190: */
}
erretm = abs(erretm);
/* Evaluate PHI and the derivative DPHI */
dphi = 0.;
phi = 0.;
i__1 = iip1;
for (j = *n; j >= i__1; --j) {
temp = z__[j] / delta[j];
phi += z__[j] * temp;
dphi += temp * temp;
erretm += phi;
/* L200: */
}
temp = z__[ii] / delta[ii];
dw = dpsi + dphi + temp * temp;
temp = z__[ii] * temp;
w = rhoinv + phi + psi + temp;
erretm = (phi - psi) * 8. + erretm + rhoinv * 2. + abs(temp) * 3. + (
d__1 = tau + eta, abs(d__1)) * dw;
swtch = FALSE_;
if (orgati) {
if (-w > abs(prew) / 10.) {
swtch = TRUE_;
}
} else {
if (w > abs(prew) / 10.) {
swtch = TRUE_;
}
}
tau += eta;
/* Main loop to update the values of the array DELTA */
iter = niter + 1;
for (niter = iter; niter <= 30; ++niter) {
/* Test for convergence */
if (abs(w) <= eps * erretm) {
if (orgati) {
*dlam = d__[*i__] + tau;
} else {
*dlam = d__[ip1] + tau;
}
goto L250;
}
if (w <= 0.) {
dltlb = f2cmax(dltlb,tau);
} else {
dltub = f2cmin(dltub,tau);
}
/* Calculate the new step */
if (! swtch3) {
if (! swtch) {
if (orgati) {
/* Computing 2nd power */
d__1 = z__[*i__] / delta[*i__];
c__ = w - delta[ip1] * dw - (d__[*i__] - d__[ip1]) * (
d__1 * d__1);
} else {
/* Computing 2nd power */
d__1 = z__[ip1] / delta[ip1];
c__ = w - delta[*i__] * dw - (d__[ip1] - d__[*i__]) *
(d__1 * d__1);
}
} else {
temp = z__[ii] / delta[ii];
if (orgati) {
dpsi += temp * temp;
} else {
dphi += temp * temp;
}
c__ = w - delta[*i__] * dpsi - delta[ip1] * dphi;
}
a = (delta[*i__] + delta[ip1]) * w - delta[*i__] * delta[ip1]
* dw;
b = delta[*i__] * delta[ip1] * w;
if (c__ == 0.) {
if (a == 0.) {
if (! swtch) {
if (orgati) {
a = z__[*i__] * z__[*i__] + delta[ip1] *
delta[ip1] * (dpsi + dphi);
} else {
a = z__[ip1] * z__[ip1] + delta[*i__] * delta[
*i__] * (dpsi + dphi);
}
} else {
a = delta[*i__] * delta[*i__] * dpsi + delta[ip1]
* delta[ip1] * dphi;
}
}
eta = b / a;
} else if (a <= 0.) {
eta = (a - sqrt((d__1 = a * a - b * 4. * c__, abs(d__1))))
/ (c__ * 2.);
} else {
eta = b * 2. / (a + sqrt((d__1 = a * a - b * 4. * c__,
abs(d__1))));
}
} else {
/* Interpolation using THREE most relevant poles */
temp = rhoinv + psi + phi;
if (swtch) {
c__ = temp - delta[iim1] * dpsi - delta[iip1] * dphi;
zz[0] = delta[iim1] * delta[iim1] * dpsi;
zz[2] = delta[iip1] * delta[iip1] * dphi;
} else {
if (orgati) {
temp1 = z__[iim1] / delta[iim1];
temp1 *= temp1;
c__ = temp - delta[iip1] * (dpsi + dphi) - (d__[iim1]
- d__[iip1]) * temp1;
zz[0] = z__[iim1] * z__[iim1];
zz[2] = delta[iip1] * delta[iip1] * (dpsi - temp1 +
dphi);
} else {
temp1 = z__[iip1] / delta[iip1];
temp1 *= temp1;
c__ = temp - delta[iim1] * (dpsi + dphi) - (d__[iip1]
- d__[iim1]) * temp1;
zz[0] = delta[iim1] * delta[iim1] * (dpsi + (dphi -
temp1));
zz[2] = z__[iip1] * z__[iip1];
}
}
dlaed6_(&niter, &orgati, &c__, &delta[iim1], zz, &w, &eta,
info);
if (*info != 0) {
goto L250;
}
}
/* Note, eta should be positive if w is negative, and */
/* eta should be negative otherwise. However, */
/* if for some reason caused by roundoff, eta*w > 0, */
/* we simply use one Newton step instead. This way */
/* will guarantee eta*w < 0. */
if (w * eta >= 0.) {
eta = -w / dw;
}
temp = tau + eta;
if (temp > dltub || temp < dltlb) {
if (w < 0.) {
eta = (dltub - tau) / 2.;
} else {
eta = (dltlb - tau) / 2.;
}
}
i__1 = *n;
for (j = 1; j <= i__1; ++j) {
delta[j] -= eta;
/* L210: */
}
tau += eta;
prew = w;
/* Evaluate PSI and the derivative DPSI */
dpsi = 0.;
psi = 0.;
erretm = 0.;
i__1 = iim1;
for (j = 1; j <= i__1; ++j) {
temp = z__[j] / delta[j];
psi += z__[j] * temp;
dpsi += temp * temp;
erretm += psi;
/* L220: */
}
erretm = abs(erretm);
/* Evaluate PHI and the derivative DPHI */
dphi = 0.;
phi = 0.;
i__1 = iip1;
for (j = *n; j >= i__1; --j) {
temp = z__[j] / delta[j];
phi += z__[j] * temp;
dphi += temp * temp;
erretm += phi;
/* L230: */
}
temp = z__[ii] / delta[ii];
dw = dpsi + dphi + temp * temp;
temp = z__[ii] * temp;
w = rhoinv + phi + psi + temp;
erretm = (phi - psi) * 8. + erretm + rhoinv * 2. + abs(temp) * 3.
+ abs(tau) * dw;
if (w * prew > 0. && abs(w) > abs(prew) / 10.) {
swtch = ! swtch;
}
/* L240: */
}
/* Return with INFO = 1, NITER = MAXIT and not converged */
*info = 1;
if (orgati) {
*dlam = d__[*i__] + tau;
} else {
*dlam = d__[ip1] + tau;
}
}
L250:
return;
/* End of DLAED4 */
} /* dlaed4_ */
Reference in New Issue View Git Blame Copy Permalink