dlasd8.c
Go to the documentation of this file.
00001 /* dlasd8.f -- translated by f2c (version 20061008).
00002    You must link the resulting object file with libf2c:
00003         on Microsoft Windows system, link with libf2c.lib;
00004         on Linux or Unix systems, link with .../path/to/libf2c.a -lm
00005         or, if you install libf2c.a in a standard place, with -lf2c -lm
00006         -- in that order, at the end of the command line, as in
00007                 cc *.o -lf2c -lm
00008         Source for libf2c is in /netlib/f2c/libf2c.zip, e.g.,
00009 
00010                 http://www.netlib.org/f2c/libf2c.zip
00011 */
00012 
00013 #include "f2c.h"
00014 #include "blaswrap.h"
00015 
00016 /* Table of constant values */
00017 
00018 static integer c__1 = 1;
00019 static integer c__0 = 0;
00020 static doublereal c_b8 = 1.;
00021 
00022 /* Subroutine */ int dlasd8_(integer *icompq, integer *k, doublereal *d__, 
00023         doublereal *z__, doublereal *vf, doublereal *vl, doublereal *difl, 
00024         doublereal *difr, integer *lddifr, doublereal *dsigma, doublereal *
00025         work, integer *info)
00026 {
00027     /* System generated locals */
00028     integer difr_dim1, difr_offset, i__1, i__2;
00029     doublereal d__1, d__2;
00030 
00031     /* Builtin functions */
00032     double sqrt(doublereal), d_sign(doublereal *, doublereal *);
00033 
00034     /* Local variables */
00035     integer i__, j;
00036     doublereal dj, rho;
00037     integer iwk1, iwk2, iwk3;
00038     extern doublereal ddot_(integer *, doublereal *, integer *, doublereal *, 
00039             integer *);
00040     doublereal temp;
00041     extern doublereal dnrm2_(integer *, doublereal *, integer *);
00042     integer iwk2i, iwk3i;
00043     doublereal diflj, difrj, dsigj;
00044     extern /* Subroutine */ int dcopy_(integer *, doublereal *, integer *, 
00045             doublereal *, integer *);
00046     extern doublereal dlamc3_(doublereal *, doublereal *);
00047     extern /* Subroutine */ int dlasd4_(integer *, integer *, doublereal *, 
00048             doublereal *, doublereal *, doublereal *, doublereal *, 
00049             doublereal *, integer *), dlascl_(char *, integer *, integer *, 
00050             doublereal *, doublereal *, integer *, integer *, doublereal *, 
00051             integer *, integer *), dlaset_(char *, integer *, integer 
00052             *, doublereal *, doublereal *, doublereal *, integer *), 
00053             xerbla_(char *, integer *);
00054     doublereal dsigjp;
00055 
00056 
00057 /*  -- LAPACK auxiliary routine (version 3.2) -- */
00058 /*     Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
00059 /*     October 2006 */
00060 
00061 /*     .. Scalar Arguments .. */
00062 /*     .. */
00063 /*     .. Array Arguments .. */
00064 /*     .. */
00065 
00066 /*  Purpose */
00067 /*  ======= */
00068 
00069 /*  DLASD8 finds the square roots of the roots of the secular equation, */
00070 /*  as defined by the values in DSIGMA and Z. It makes the appropriate */
00071 /*  calls to DLASD4, and stores, for each  element in D, the distance */
00072 /*  to its two nearest poles (elements in DSIGMA). It also updates */
00073 /*  the arrays VF and VL, the first and last components of all the */
00074 /*  right singular vectors of the original bidiagonal matrix. */
00075 
00076 /*  DLASD8 is called from DLASD6. */
00077 
00078 /*  Arguments */
00079 /*  ========= */
00080 
00081 /*  ICOMPQ  (input) INTEGER */
00082 /*          Specifies whether singular vectors are to be computed in */
00083 /*          factored form in the calling routine: */
00084 /*          = 0: Compute singular values only. */
00085 /*          = 1: Compute singular vectors in factored form as well. */
00086 
00087 /*  K       (input) INTEGER */
00088 /*          The number of terms in the rational function to be solved */
00089 /*          by DLASD4.  K >= 1. */
00090 
00091 /*  D       (output) DOUBLE PRECISION array, dimension ( K ) */
00092 /*          On output, D contains the updated singular values. */
00093 
00094 /*  Z       (input/output) DOUBLE PRECISION array, dimension ( K ) */
00095 /*          On entry, the first K elements of this array contain the */
00096 /*          components of the deflation-adjusted updating row vector. */
00097 /*          On exit, Z is updated. */
00098 
00099 /*  VF      (input/output) DOUBLE PRECISION array, dimension ( K ) */
00100 /*          On entry, VF contains  information passed through DBEDE8. */
00101 /*          On exit, VF contains the first K components of the first */
00102 /*          components of all right singular vectors of the bidiagonal */
00103 /*          matrix. */
00104 
00105 /*  VL      (input/output) DOUBLE PRECISION array, dimension ( K ) */
00106 /*          On entry, VL contains  information passed through DBEDE8. */
00107 /*          On exit, VL contains the first K components of the last */
00108 /*          components of all right singular vectors of the bidiagonal */
00109 /*          matrix. */
00110 
00111 /*  DIFL    (output) DOUBLE PRECISION array, dimension ( K ) */
00112 /*          On exit, DIFL(I) = D(I) - DSIGMA(I). */
00113 
00114 /*  DIFR    (output) DOUBLE PRECISION array, */
00115 /*                   dimension ( LDDIFR, 2 ) if ICOMPQ = 1 and */
00116 /*                   dimension ( K ) if ICOMPQ = 0. */
00117 /*          On exit, DIFR(I,1) = D(I) - DSIGMA(I+1), DIFR(K,1) is not */
00118 /*          defined and will not be referenced. */
00119 
00120 /*          If ICOMPQ = 1, DIFR(1:K,2) is an array containing the */
00121 /*          normalizing factors for the right singular vector matrix. */
00122 
00123 /*  LDDIFR  (input) INTEGER */
00124 /*          The leading dimension of DIFR, must be at least K. */
00125 
00126 /*  DSIGMA  (input/output) DOUBLE PRECISION array, dimension ( K ) */
00127 /*          On entry, the first K elements of this array contain the old */
00128 /*          roots of the deflated updating problem.  These are the poles */
00129 /*          of the secular equation. */
00130 /*          On exit, the elements of DSIGMA may be very slightly altered */
00131 /*          in value. */
00132 
00133 /*  WORK    (workspace) DOUBLE PRECISION array, dimension at least 3 * K */
00134 
00135 /*  INFO    (output) INTEGER */
00136 /*          = 0:  successful exit. */
00137 /*          < 0:  if INFO = -i, the i-th argument had an illegal value. */
00138 /*          > 0:  if INFO = 1, an singular value did not converge */
00139 
00140 /*  Further Details */
00141 /*  =============== */
00142 
00143 /*  Based on contributions by */
00144 /*     Ming Gu and Huan Ren, Computer Science Division, University of */
00145 /*     California at Berkeley, USA */
00146 
00147 /*  ===================================================================== */
00148 
00149 /*     .. Parameters .. */
00150 /*     .. */
00151 /*     .. Local Scalars .. */
00152 /*     .. */
00153 /*     .. External Subroutines .. */
00154 /*     .. */
00155 /*     .. External Functions .. */
00156 /*     .. */
00157 /*     .. Intrinsic Functions .. */
00158 /*     .. */
00159 /*     .. Executable Statements .. */
00160 
00161 /*     Test the input parameters. */
00162 
00163     /* Parameter adjustments */
00164     --d__;
00165     --z__;
00166     --vf;
00167     --vl;
00168     --difl;
00169     difr_dim1 = *lddifr;
00170     difr_offset = 1 + difr_dim1;
00171     difr -= difr_offset;
00172     --dsigma;
00173     --work;
00174 
00175     /* Function Body */
00176     *info = 0;
00177 
00178     if (*icompq < 0 || *icompq > 1) {
00179         *info = -1;
00180     } else if (*k < 1) {
00181         *info = -2;
00182     } else if (*lddifr < *k) {
00183         *info = -9;
00184     }
00185     if (*info != 0) {
00186         i__1 = -(*info);
00187         xerbla_("DLASD8", &i__1);
00188         return 0;
00189     }
00190 
00191 /*     Quick return if possible */
00192 
00193     if (*k == 1) {
00194         d__[1] = abs(z__[1]);
00195         difl[1] = d__[1];
00196         if (*icompq == 1) {
00197             difl[2] = 1.;
00198             difr[(difr_dim1 << 1) + 1] = 1.;
00199         }
00200         return 0;
00201     }
00202 
00203 /*     Modify values DSIGMA(i) to make sure all DSIGMA(i)-DSIGMA(j) can */
00204 /*     be computed with high relative accuracy (barring over/underflow). */
00205 /*     This is a problem on machines without a guard digit in */
00206 /*     add/subtract (Cray XMP, Cray YMP, Cray C 90 and Cray 2). */
00207 /*     The following code replaces DSIGMA(I) by 2*DSIGMA(I)-DSIGMA(I), */
00208 /*     which on any of these machines zeros out the bottommost */
00209 /*     bit of DSIGMA(I) if it is 1; this makes the subsequent */
00210 /*     subtractions DSIGMA(I)-DSIGMA(J) unproblematic when cancellation */
00211 /*     occurs. On binary machines with a guard digit (almost all */
00212 /*     machines) it does not change DSIGMA(I) at all. On hexadecimal */
00213 /*     and decimal machines with a guard digit, it slightly */
00214 /*     changes the bottommost bits of DSIGMA(I). It does not account */
00215 /*     for hexadecimal or decimal machines without guard digits */
00216 /*     (we know of none). We use a subroutine call to compute */
00217 /*     2*DLAMBDA(I) to prevent optimizing compilers from eliminating */
00218 /*     this code. */
00219 
00220     i__1 = *k;
00221     for (i__ = 1; i__ <= i__1; ++i__) {
00222         dsigma[i__] = dlamc3_(&dsigma[i__], &dsigma[i__]) - dsigma[i__];
00223 /* L10: */
00224     }
00225 
00226 /*     Book keeping. */
00227 
00228     iwk1 = 1;
00229     iwk2 = iwk1 + *k;
00230     iwk3 = iwk2 + *k;
00231     iwk2i = iwk2 - 1;
00232     iwk3i = iwk3 - 1;
00233 
00234 /*     Normalize Z. */
00235 
00236     rho = dnrm2_(k, &z__[1], &c__1);
00237     dlascl_("G", &c__0, &c__0, &rho, &c_b8, k, &c__1, &z__[1], k, info);
00238     rho *= rho;
00239 
00240 /*     Initialize WORK(IWK3). */
00241 
00242     dlaset_("A", k, &c__1, &c_b8, &c_b8, &work[iwk3], k);
00243 
00244 /*     Compute the updated singular values, the arrays DIFL, DIFR, */
00245 /*     and the updated Z. */
00246 
00247     i__1 = *k;
00248     for (j = 1; j <= i__1; ++j) {
00249         dlasd4_(k, &j, &dsigma[1], &z__[1], &work[iwk1], &rho, &d__[j], &work[
00250                 iwk2], info);
00251 
00252 /*        If the root finder fails, the computation is terminated. */
00253 
00254         if (*info != 0) {
00255             return 0;
00256         }
00257         work[iwk3i + j] = work[iwk3i + j] * work[j] * work[iwk2i + j];
00258         difl[j] = -work[j];
00259         difr[j + difr_dim1] = -work[j + 1];
00260         i__2 = j - 1;
00261         for (i__ = 1; i__ <= i__2; ++i__) {
00262             work[iwk3i + i__] = work[iwk3i + i__] * work[i__] * work[iwk2i + 
00263                     i__] / (dsigma[i__] - dsigma[j]) / (dsigma[i__] + dsigma[
00264                     j]);
00265 /* L20: */
00266         }
00267         i__2 = *k;
00268         for (i__ = j + 1; i__ <= i__2; ++i__) {
00269             work[iwk3i + i__] = work[iwk3i + i__] * work[i__] * work[iwk2i + 
00270                     i__] / (dsigma[i__] - dsigma[j]) / (dsigma[i__] + dsigma[
00271                     j]);
00272 /* L30: */
00273         }
00274 /* L40: */
00275     }
00276 
00277 /*     Compute updated Z. */
00278 
00279     i__1 = *k;
00280     for (i__ = 1; i__ <= i__1; ++i__) {
00281         d__2 = sqrt((d__1 = work[iwk3i + i__], abs(d__1)));
00282         z__[i__] = d_sign(&d__2, &z__[i__]);
00283 /* L50: */
00284     }
00285 
00286 /*     Update VF and VL. */
00287 
00288     i__1 = *k;
00289     for (j = 1; j <= i__1; ++j) {
00290         diflj = difl[j];
00291         dj = d__[j];
00292         dsigj = -dsigma[j];
00293         if (j < *k) {
00294             difrj = -difr[j + difr_dim1];
00295             dsigjp = -dsigma[j + 1];
00296         }
00297         work[j] = -z__[j] / diflj / (dsigma[j] + dj);
00298         i__2 = j - 1;
00299         for (i__ = 1; i__ <= i__2; ++i__) {
00300             work[i__] = z__[i__] / (dlamc3_(&dsigma[i__], &dsigj) - diflj) / (
00301                     dsigma[i__] + dj);
00302 /* L60: */
00303         }
00304         i__2 = *k;
00305         for (i__ = j + 1; i__ <= i__2; ++i__) {
00306             work[i__] = z__[i__] / (dlamc3_(&dsigma[i__], &dsigjp) + difrj) / 
00307                     (dsigma[i__] + dj);
00308 /* L70: */
00309         }
00310         temp = dnrm2_(k, &work[1], &c__1);
00311         work[iwk2i + j] = ddot_(k, &work[1], &c__1, &vf[1], &c__1) / temp;
00312         work[iwk3i + j] = ddot_(k, &work[1], &c__1, &vl[1], &c__1) / temp;
00313         if (*icompq == 1) {
00314             difr[j + (difr_dim1 << 1)] = temp;
00315         }
00316 /* L80: */
00317     }
00318 
00319     dcopy_(k, &work[iwk2], &c__1, &vf[1], &c__1);
00320     dcopy_(k, &work[iwk3], &c__1, &vl[1], &c__1);
00321 
00322     return 0;
00323 
00324 /*     End of DLASD8 */
00325 
00326 } /* dlasd8_ */


swiftnav
Author(s):
autogenerated on Sat Jun 8 2019 18:55:46