aboutsummaryrefslogtreecommitdiffstats
path: root/contrib/libs/clapack/dlaed9.c
blob: 25b3466e8fc9793cc54dbbb27d7556c7f1de0c6d (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
/* dlaed9.f -- translated by f2c (version 20061008).
   You must link the resulting object file with libf2c:
	on Microsoft Windows system, link with libf2c.lib;
	on Linux or Unix systems, link with .../path/to/libf2c.a -lm
	or, if you install libf2c.a in a standard place, with -lf2c -lm
	-- in that order, at the end of the command line, as in
		cc *.o -lf2c -lm
	Source for libf2c is in /netlib/f2c/libf2c.zip, e.g.,

		http://www.netlib.org/f2c/libf2c.zip
*/

#include "f2c.h"
#include "blaswrap.h"

/* Table of constant values */

static integer c__1 = 1;

/* Subroutine */ int dlaed9_(integer *k, integer *kstart, integer *kstop, 
	integer *n, doublereal *d__, doublereal *q, integer *ldq, doublereal *
	rho, doublereal *dlamda, doublereal *w, doublereal *s, integer *lds, 
	integer *info)
{
    /* System generated locals */
    integer q_dim1, q_offset, s_dim1, s_offset, i__1, i__2;
    doublereal d__1;

    /* Builtin functions */
    double sqrt(doublereal), d_sign(doublereal *, doublereal *);

    /* Local variables */
    integer i__, j;
    doublereal temp;
    extern doublereal dnrm2_(integer *, doublereal *, integer *);
    extern /* Subroutine */ int dcopy_(integer *, doublereal *, integer *, 
	    doublereal *, integer *), dlaed4_(integer *, integer *, 
	    doublereal *, doublereal *, doublereal *, doublereal *, 
	    doublereal *, integer *);
    extern doublereal dlamc3_(doublereal *, doublereal *);
    extern /* Subroutine */ int xerbla_(char *, integer *);


/*  -- LAPACK routine (version 3.2) -- */
/*     Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. */
/*     November 2006 */

/*     .. Scalar Arguments .. */
/*     .. */
/*     .. Array Arguments .. */
/*     .. */

/*  Purpose */
/*  ======= */

/*  DLAED9 finds the roots of the secular equation, as defined by the */
/*  values in D, Z, and RHO, between KSTART and KSTOP.  It makes the */
/*  appropriate calls to DLAED4 and then stores the new matrix of */
/*  eigenvectors for use in calculating the next level of Z vectors. */

/*  Arguments */
/*  ========= */

/*  K       (input) INTEGER */
/*          The number of terms in the rational function to be solved by */
/*          DLAED4.  K >= 0. */

/*  KSTART  (input) INTEGER */
/*  KSTOP   (input) INTEGER */
/*          The updated eigenvalues Lambda(I), KSTART <= I <= KSTOP */
/*          are to be computed.  1 <= KSTART <= KSTOP <= K. */

/*  N       (input) INTEGER */
/*          The number of rows and columns in the Q matrix. */
/*          N >= K (delation may result in N > K). */

/*  D       (output) DOUBLE PRECISION array, dimension (N) */
/*          D(I) contains the updated eigenvalues */
/*          for KSTART <= I <= KSTOP. */

/*  Q       (workspace) DOUBLE PRECISION array, dimension (LDQ,N) */

/*  LDQ     (input) INTEGER */
/*          The leading dimension of the array Q.  LDQ >= max( 1, N ). */

/*  RHO     (input) DOUBLE PRECISION */
/*          The value of the parameter in the rank one update equation. */
/*          RHO >= 0 required. */

/*  DLAMDA  (input) DOUBLE PRECISION array, dimension (K) */
/*          The first K elements of this array contain the old roots */
/*          of the deflated updating problem.  These are the poles */
/*          of the secular equation. */

/*  W       (input) DOUBLE PRECISION array, dimension (K) */
/*          The first K elements of this array contain the components */
/*          of the deflation-adjusted updating vector. */

/*  S       (output) DOUBLE PRECISION array, dimension (LDS, K) */
/*          Will contain the eigenvectors of the repaired matrix which */
/*          will be stored for subsequent Z vector calculation and */
/*          multiplied by the previously accumulated eigenvectors */
/*          to update the system. */

/*  LDS     (input) INTEGER */
/*          The leading dimension of S.  LDS >= max( 1, K ). */

/*  INFO    (output) INTEGER */
/*          = 0:  successful exit. */
/*          < 0:  if INFO = -i, the i-th argument had an illegal value. */
/*          > 0:  if INFO = 1, an eigenvalue did not converge */

/*  Further Details */
/*  =============== */

/*  Based on contributions by */
/*     Jeff Rutter, Computer Science Division, University of California */
/*     at Berkeley, USA */

/*  ===================================================================== */

/*     .. Local Scalars .. */
/*     .. */
/*     .. External Functions .. */
/*     .. */
/*     .. External Subroutines .. */
/*     .. */
/*     .. Intrinsic Functions .. */
/*     .. */
/*     .. Executable Statements .. */

/*     Test the input parameters. */

    /* Parameter adjustments */
    --d__;
    q_dim1 = *ldq;
    q_offset = 1 + q_dim1;
    q -= q_offset;
    --dlamda;
    --w;
    s_dim1 = *lds;
    s_offset = 1 + s_dim1;
    s -= s_offset;

    /* Function Body */
    *info = 0;

    if (*k < 0) {
	*info = -1;
    } else if (*kstart < 1 || *kstart > max(1,*k)) {
	*info = -2;
    } else if (max(1,*kstop) < *kstart || *kstop > max(1,*k)) {
	*info = -3;
    } else if (*n < *k) {
	*info = -4;
    } else if (*ldq < max(1,*k)) {
	*info = -7;
    } else if (*lds < max(1,*k)) {
	*info = -12;
    }
    if (*info != 0) {
	i__1 = -(*info);
	xerbla_("DLAED9", &i__1);
	return 0;
    }

/*     Quick return if possible */

    if (*k == 0) {
	return 0;
    }

/*     Modify values DLAMDA(i) to make sure all DLAMDA(i)-DLAMDA(j) can */
/*     be computed with high relative accuracy (barring over/underflow). */
/*     This is a problem on machines without a guard digit in */
/*     add/subtract (Cray XMP, Cray YMP, Cray C 90 and Cray 2). */
/*     The following code replaces DLAMDA(I) by 2*DLAMDA(I)-DLAMDA(I), */
/*     which on any of these machines zeros out the bottommost */
/*     bit of DLAMDA(I) if it is 1; this makes the subsequent */
/*     subtractions DLAMDA(I)-DLAMDA(J) unproblematic when cancellation */
/*     occurs. On binary machines with a guard digit (almost all */
/*     machines) it does not change DLAMDA(I) at all. On hexadecimal */
/*     and decimal machines with a guard digit, it slightly */
/*     changes the bottommost bits of DLAMDA(I). It does not account */
/*     for hexadecimal or decimal machines without guard digits */
/*     (we know of none). We use a subroutine call to compute */
/*     2*DLAMBDA(I) to prevent optimizing compilers from eliminating */
/*     this code. */

    i__1 = *n;
    for (i__ = 1; i__ <= i__1; ++i__) {
	dlamda[i__] = dlamc3_(&dlamda[i__], &dlamda[i__]) - dlamda[i__];
/* L10: */
    }

    i__1 = *kstop;
    for (j = *kstart; j <= i__1; ++j) {
	dlaed4_(k, &j, &dlamda[1], &w[1], &q[j * q_dim1 + 1], rho, &d__[j], 
		info);

/*        If the zero finder fails, the computation is terminated. */

	if (*info != 0) {
	    goto L120;
	}
/* L20: */
    }

    if (*k == 1 || *k == 2) {
	i__1 = *k;
	for (i__ = 1; i__ <= i__1; ++i__) {
	    i__2 = *k;
	    for (j = 1; j <= i__2; ++j) {
		s[j + i__ * s_dim1] = q[j + i__ * q_dim1];
/* L30: */
	    }
/* L40: */
	}
	goto L120;
    }

/*     Compute updated W. */

    dcopy_(k, &w[1], &c__1, &s[s_offset], &c__1);

/*     Initialize W(I) = Q(I,I) */

    i__1 = *ldq + 1;
    dcopy_(k, &q[q_offset], &i__1, &w[1], &c__1);
    i__1 = *k;
    for (j = 1; j <= i__1; ++j) {
	i__2 = j - 1;
	for (i__ = 1; i__ <= i__2; ++i__) {
	    w[i__] *= q[i__ + j * q_dim1] / (dlamda[i__] - dlamda[j]);
/* L50: */
	}
	i__2 = *k;
	for (i__ = j + 1; i__ <= i__2; ++i__) {
	    w[i__] *= q[i__ + j * q_dim1] / (dlamda[i__] - dlamda[j]);
/* L60: */
	}
/* L70: */
    }
    i__1 = *k;
    for (i__ = 1; i__ <= i__1; ++i__) {
	d__1 = sqrt(-w[i__]);
	w[i__] = d_sign(&d__1, &s[i__ + s_dim1]);
/* L80: */
    }

/*     Compute eigenvectors of the modified rank-1 modification. */

    i__1 = *k;
    for (j = 1; j <= i__1; ++j) {
	i__2 = *k;
	for (i__ = 1; i__ <= i__2; ++i__) {
	    q[i__ + j * q_dim1] = w[i__] / q[i__ + j * q_dim1];
/* L90: */
	}
	temp = dnrm2_(k, &q[j * q_dim1 + 1], &c__1);
	i__2 = *k;
	for (i__ = 1; i__ <= i__2; ++i__) {
	    s[i__ + j * s_dim1] = q[i__ + j * q_dim1] / temp;
/* L100: */
	}
/* L110: */
    }

L120:
    return 0;

/*     End of DLAED9 */

} /* dlaed9_ */