14 typedef long long BLASLONG;
15 typedef unsigned long long BLASULONG;
17 typedef long BLASLONG;
18 typedef unsigned long BLASULONG;
22 typedef BLASLONG blasint;
24 #define blasabs(x) llabs(x)
26 #define blasabs(x) labs(x)
30 #define blasabs(x) abs(x)
33 typedef blasint integer;
35 typedef unsigned int uinteger;
36 typedef char *address;
37 typedef short int shortint;
39 typedef double doublereal;
40 typedef struct { real r, i; } complex;
41 typedef struct { doublereal r, i; } doublecomplex;
43 static inline _Fcomplex Cf(complex *z) {_Fcomplex zz={z->r , z->i}; return zz;}
44 static inline _Dcomplex Cd(doublecomplex *z) {_Dcomplex zz={z->r , z->i};return zz;}
45 static inline _Fcomplex * _pCf(complex *z) {return (_Fcomplex*)z;}
46 static inline _Dcomplex * _pCd(doublecomplex *z) {return (_Dcomplex*)z;}
48 static inline _Complex float Cf(complex *z) {return z->r + z->i*_Complex_I;}
49 static inline _Complex double Cd(doublecomplex *z) {return z->r + z->i*_Complex_I;}
50 static inline _Complex float * _pCf(complex *z) {return (_Complex float*)z;}
51 static inline _Complex double * _pCd(doublecomplex *z) {return (_Complex double*)z;}
53 #define pCf(z) (*_pCf(z))
54 #define pCd(z) (*_pCd(z))
56 typedef short int shortlogical;
57 typedef char logical1;
58 typedef char integer1;
63 /* Extern is for use with -E */
74 /*external read, write*/
83 /*internal read, write*/
113 /*rewind, backspace, endfile*/
125 ftnint *inex; /*parameters in standard's order*/
151 union Multitype { /* for multiple entry points */
162 typedef union Multitype Multitype;
164 struct Vardesc { /* for Namelist */
170 typedef struct Vardesc Vardesc;
177 typedef struct Namelist Namelist;
179 #define abs(x) ((x) >= 0 ? (x) : -(x))
180 #define dabs(x) (fabs(x))
181 #define f2cmin(a,b) ((a) <= (b) ? (a) : (b))
182 #define f2cmax(a,b) ((a) >= (b) ? (a) : (b))
183 #define dmin(a,b) (f2cmin(a,b))
184 #define dmax(a,b) (f2cmax(a,b))
185 #define bit_test(a,b) ((a) >> (b) & 1)
186 #define bit_clear(a,b) ((a) & ~((uinteger)1 << (b)))
187 #define bit_set(a,b) ((a) | ((uinteger)1 << (b)))
189 #define abort_() { sig_die("Fortran abort routine called", 1); }
190 #define c_abs(z) (cabsf(Cf(z)))
191 #define c_cos(R,Z) { pCf(R)=ccos(Cf(Z)); }
193 #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]);}
194 #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]);}
196 #define c_div(c, a, b) {pCf(c) = Cf(a)/Cf(b);}
197 #define z_div(c, a, b) {pCd(c) = Cd(a)/Cd(b);}
199 #define c_exp(R, Z) {pCf(R) = cexpf(Cf(Z));}
200 #define c_log(R, Z) {pCf(R) = clogf(Cf(Z));}
201 #define c_sin(R, Z) {pCf(R) = csinf(Cf(Z));}
202 //#define c_sqrt(R, Z) {*(R) = csqrtf(Cf(Z));}
203 #define c_sqrt(R, Z) {pCf(R) = csqrtf(Cf(Z));}
204 #define d_abs(x) (fabs(*(x)))
205 #define d_acos(x) (acos(*(x)))
206 #define d_asin(x) (asin(*(x)))
207 #define d_atan(x) (atan(*(x)))
208 #define d_atn2(x, y) (atan2(*(x),*(y)))
209 #define d_cnjg(R, Z) { pCd(R) = conj(Cd(Z)); }
210 #define r_cnjg(R, Z) { pCf(R) = conjf(Cf(Z)); }
211 #define d_cos(x) (cos(*(x)))
212 #define d_cosh(x) (cosh(*(x)))
213 #define d_dim(__a, __b) ( *(__a) > *(__b) ? *(__a) - *(__b) : 0.0 )
214 #define d_exp(x) (exp(*(x)))
215 #define d_imag(z) (cimag(Cd(z)))
216 #define r_imag(z) (cimagf(Cf(z)))
217 #define d_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
218 #define r_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
219 #define d_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
220 #define r_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
221 #define d_log(x) (log(*(x)))
222 #define d_mod(x, y) (fmod(*(x), *(y)))
223 #define u_nint(__x) ((__x)>=0 ? floor((__x) + .5) : -floor(.5 - (__x)))
224 #define d_nint(x) u_nint(*(x))
225 #define u_sign(__a,__b) ((__b) >= 0 ? ((__a) >= 0 ? (__a) : -(__a)) : -((__a) >= 0 ? (__a) : -(__a)))
226 #define d_sign(a,b) u_sign(*(a),*(b))
227 #define r_sign(a,b) u_sign(*(a),*(b))
228 #define d_sin(x) (sin(*(x)))
229 #define d_sinh(x) (sinh(*(x)))
230 #define d_sqrt(x) (sqrt(*(x)))
231 #define d_tan(x) (tan(*(x)))
232 #define d_tanh(x) (tanh(*(x)))
233 #define i_abs(x) abs(*(x))
234 #define i_dnnt(x) ((integer)u_nint(*(x)))
235 #define i_len(s, n) (n)
236 #define i_nint(x) ((integer)u_nint(*(x)))
237 #define i_sign(a,b) ((integer)u_sign((integer)*(a),(integer)*(b)))
238 #define pow_dd(ap, bp) ( pow(*(ap), *(bp)))
239 #define pow_si(B,E) spow_ui(*(B),*(E))
240 #define pow_ri(B,E) spow_ui(*(B),*(E))
241 #define pow_di(B,E) dpow_ui(*(B),*(E))
242 #define pow_zi(p, a, b) {pCd(p) = zpow_ui(Cd(a), *(b));}
243 #define pow_ci(p, a, b) {pCf(p) = cpow_ui(Cf(a), *(b));}
244 #define pow_zz(R,A,B) {pCd(R) = cpow(Cd(A),*(B));}
245 #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++ = ' '; }
246 #define s_cmp(a,b,c,d) ((integer)strncmp((a),(b),f2cmin((c),(d))))
247 #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]; }
248 #define sig_die(s, kill) { exit(1); }
249 #define s_stop(s, n) {exit(0);}
250 static char junk[] = "\n@(#)LIBF77 VERSION 19990503\n";
251 #define z_abs(z) (cabs(Cd(z)))
252 #define z_exp(R, Z) {pCd(R) = cexp(Cd(Z));}
253 #define z_sqrt(R, Z) {pCd(R) = csqrt(Cd(Z));}
254 #define myexit_() break;
255 #define mycycle() continue;
256 #define myceiling(w) {ceil(w)}
257 #define myhuge(w) {HUGE_VAL}
258 //#define mymaxloc_(w,s,e,n) {if (sizeof(*(w)) == sizeof(double)) dmaxloc_((w),*(s),*(e),n); else dmaxloc_((w),*(s),*(e),n);}
259 #define mymaxloc(w,s,e,n) {dmaxloc_(w,*(s),*(e),n)}
261 /* procedure parameter types for -A and -C++ */
263 #define F2C_proc_par_types 1
265 typedef logical (*L_fp)(...);
267 typedef logical (*L_fp)();
270 static float spow_ui(float x, integer n) {
271 float pow=1.0; unsigned long int u;
273 if(n < 0) n = -n, x = 1/x;
282 static double dpow_ui(double x, integer n) {
283 double pow=1.0; unsigned long int u;
285 if(n < 0) n = -n, x = 1/x;
295 static _Fcomplex cpow_ui(complex x, integer n) {
296 complex pow={1.0,0.0}; unsigned long int u;
298 if(n < 0) n = -n, x.r = 1/x.r, x.i=1/x.i;
300 if(u & 01) pow.r *= x.r, pow.i *= x.i;
301 if(u >>= 1) x.r *= x.r, x.i *= x.i;
305 _Fcomplex p={pow.r, pow.i};
309 static _Complex float cpow_ui(_Complex float x, integer n) {
310 _Complex float pow=1.0; unsigned long int u;
312 if(n < 0) n = -n, x = 1/x;
323 static _Dcomplex zpow_ui(_Dcomplex x, integer n) {
324 _Dcomplex pow={1.0,0.0}; unsigned long int u;
326 if(n < 0) n = -n, x._Val[0] = 1/x._Val[0], x._Val[1] =1/x._Val[1];
328 if(u & 01) pow._Val[0] *= x._Val[0], pow._Val[1] *= x._Val[1];
329 if(u >>= 1) x._Val[0] *= x._Val[0], x._Val[1] *= x._Val[1];
333 _Dcomplex p = {pow._Val[0], pow._Val[1]};
337 static _Complex double zpow_ui(_Complex double x, integer n) {
338 _Complex double pow=1.0; unsigned long int u;
340 if(n < 0) n = -n, x = 1/x;
350 static integer pow_ii(integer x, integer n) {
351 integer pow; unsigned long int u;
353 if (n == 0 || x == 1) pow = 1;
354 else if (x != -1) pow = x == 0 ? 1/x : 0;
357 if ((n > 0) || !(n == 0 || x == 1 || x != -1)) {
367 static integer dmaxloc_(double *w, integer s, integer e, integer *n)
369 double m; integer i, mi;
370 for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
371 if (w[i-1]>m) mi=i ,m=w[i-1];
374 static integer smaxloc_(float *w, integer s, integer e, integer *n)
376 float m; integer i, mi;
377 for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
378 if (w[i-1]>m) mi=i ,m=w[i-1];
381 static inline void cdotc_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
382 integer n = *n_, incx = *incx_, incy = *incy_, i;
384 _Fcomplex zdotc = {0.0, 0.0};
385 if (incx == 1 && incy == 1) {
386 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
387 zdotc._Val[0] += conjf(Cf(&x[i]))._Val[0] * Cf(&y[i])._Val[0];
388 zdotc._Val[1] += conjf(Cf(&x[i]))._Val[1] * Cf(&y[i])._Val[1];
391 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
392 zdotc._Val[0] += conjf(Cf(&x[i*incx]))._Val[0] * Cf(&y[i*incy])._Val[0];
393 zdotc._Val[1] += conjf(Cf(&x[i*incx]))._Val[1] * Cf(&y[i*incy])._Val[1];
399 _Complex float zdotc = 0.0;
400 if (incx == 1 && incy == 1) {
401 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
402 zdotc += conjf(Cf(&x[i])) * Cf(&y[i]);
405 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
406 zdotc += conjf(Cf(&x[i*incx])) * Cf(&y[i*incy]);
412 static inline void zdotc_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
413 integer n = *n_, incx = *incx_, incy = *incy_, i;
415 _Dcomplex zdotc = {0.0, 0.0};
416 if (incx == 1 && incy == 1) {
417 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
418 zdotc._Val[0] += conj(Cd(&x[i]))._Val[0] * Cd(&y[i])._Val[0];
419 zdotc._Val[1] += conj(Cd(&x[i]))._Val[1] * Cd(&y[i])._Val[1];
422 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
423 zdotc._Val[0] += conj(Cd(&x[i*incx]))._Val[0] * Cd(&y[i*incy])._Val[0];
424 zdotc._Val[1] += conj(Cd(&x[i*incx]))._Val[1] * Cd(&y[i*incy])._Val[1];
430 _Complex double zdotc = 0.0;
431 if (incx == 1 && incy == 1) {
432 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
433 zdotc += conj(Cd(&x[i])) * Cd(&y[i]);
436 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
437 zdotc += conj(Cd(&x[i*incx])) * Cd(&y[i*incy]);
443 static inline void cdotu_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
444 integer n = *n_, incx = *incx_, incy = *incy_, i;
446 _Fcomplex zdotc = {0.0, 0.0};
447 if (incx == 1 && incy == 1) {
448 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
449 zdotc._Val[0] += Cf(&x[i])._Val[0] * Cf(&y[i])._Val[0];
450 zdotc._Val[1] += Cf(&x[i])._Val[1] * Cf(&y[i])._Val[1];
453 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
454 zdotc._Val[0] += Cf(&x[i*incx])._Val[0] * Cf(&y[i*incy])._Val[0];
455 zdotc._Val[1] += Cf(&x[i*incx])._Val[1] * Cf(&y[i*incy])._Val[1];
461 _Complex float zdotc = 0.0;
462 if (incx == 1 && incy == 1) {
463 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
464 zdotc += Cf(&x[i]) * Cf(&y[i]);
467 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
468 zdotc += Cf(&x[i*incx]) * Cf(&y[i*incy]);
474 static inline void zdotu_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
475 integer n = *n_, incx = *incx_, incy = *incy_, i;
477 _Dcomplex zdotc = {0.0, 0.0};
478 if (incx == 1 && incy == 1) {
479 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
480 zdotc._Val[0] += Cd(&x[i])._Val[0] * Cd(&y[i])._Val[0];
481 zdotc._Val[1] += Cd(&x[i])._Val[1] * Cd(&y[i])._Val[1];
484 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
485 zdotc._Val[0] += Cd(&x[i*incx])._Val[0] * Cd(&y[i*incy])._Val[0];
486 zdotc._Val[1] += Cd(&x[i*incx])._Val[1] * Cd(&y[i*incy])._Val[1];
492 _Complex double zdotc = 0.0;
493 if (incx == 1 && incy == 1) {
494 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
495 zdotc += Cd(&x[i]) * Cd(&y[i]);
498 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
499 zdotc += Cd(&x[i*incx]) * Cd(&y[i*incy]);
505 /* -- translated by f2c (version 20000121).
506 You must link the resulting object file with the libraries:
507 -lf2c -lm (in that order)
513 /* Table of constant values */
515 static integer c__1 = 1;
516 static real c_b30 = 0.f;
518 /* > \brief \b SLASD2 merges the two sets of singular values together into a single sorted set. Used by sbdsdc
521 /* =========== DOCUMENTATION =========== */
523 /* Online html documentation available at */
524 /* http://www.netlib.org/lapack/explore-html/ */
527 /* > Download SLASD2 + dependencies */
528 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/slasd2.
531 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/slasd2.
534 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/slasd2.
542 /* SUBROUTINE SLASD2( NL, NR, SQRE, K, D, Z, ALPHA, BETA, U, LDU, VT, */
543 /* LDVT, DSIGMA, U2, LDU2, VT2, LDVT2, IDXP, IDX, */
544 /* IDXC, IDXQ, COLTYP, INFO ) */
546 /* INTEGER INFO, K, LDU, LDU2, LDVT, LDVT2, NL, NR, SQRE */
547 /* REAL ALPHA, BETA */
548 /* INTEGER COLTYP( * ), IDX( * ), IDXC( * ), IDXP( * ), */
550 /* REAL D( * ), DSIGMA( * ), U( LDU, * ), */
551 /* $ U2( LDU2, * ), VT( LDVT, * ), VT2( LDVT2, * ), */
555 /* > \par Purpose: */
560 /* > SLASD2 merges the two sets of singular values together into a single */
561 /* > sorted set. Then it tries to deflate the size of the problem. */
562 /* > There are two ways in which deflation can occur: when two or more */
563 /* > singular values are close together or if there is a tiny entry in the */
564 /* > Z vector. For each such occurrence the order of the related secular */
565 /* > equation problem is reduced by one. */
567 /* > SLASD2 is called from SLASD1. */
573 /* > \param[in] NL */
575 /* > NL is INTEGER */
576 /* > The row dimension of the upper block. NL >= 1. */
579 /* > \param[in] NR */
581 /* > NR is INTEGER */
582 /* > The row dimension of the lower block. NR >= 1. */
585 /* > \param[in] SQRE */
587 /* > SQRE is INTEGER */
588 /* > = 0: the lower block is an NR-by-NR square matrix. */
589 /* > = 1: the lower block is an NR-by-(NR+1) rectangular matrix. */
591 /* > The bidiagonal matrix has N = NL + NR + 1 rows and */
592 /* > M = N + SQRE >= N columns. */
595 /* > \param[out] K */
598 /* > Contains the dimension of the non-deflated matrix, */
599 /* > This is the order of the related secular equation. 1 <= K <=N. */
602 /* > \param[in,out] D */
604 /* > D is REAL array, dimension (N) */
605 /* > On entry D contains the singular values of the two submatrices */
606 /* > to be combined. On exit D contains the trailing (N-K) updated */
607 /* > singular values (those which were deflated) sorted into */
608 /* > increasing order. */
611 /* > \param[out] Z */
613 /* > Z is REAL array, dimension (N) */
614 /* > On exit Z contains the updating row vector in the secular */
618 /* > \param[in] ALPHA */
620 /* > ALPHA is REAL */
621 /* > Contains the diagonal element associated with the added row. */
624 /* > \param[in] BETA */
627 /* > Contains the off-diagonal element associated with the added */
631 /* > \param[in,out] U */
633 /* > U is REAL array, dimension (LDU,N) */
634 /* > On entry U contains the left singular vectors of two */
635 /* > submatrices in the two square blocks with corners at (1,1), */
636 /* > (NL, NL), and (NL+2, NL+2), (N,N). */
637 /* > On exit U contains the trailing (N-K) updated left singular */
638 /* > vectors (those which were deflated) in its last N-K columns. */
641 /* > \param[in] LDU */
643 /* > LDU is INTEGER */
644 /* > The leading dimension of the array U. LDU >= N. */
647 /* > \param[in,out] VT */
649 /* > VT is REAL array, dimension (LDVT,M) */
650 /* > On entry VT**T contains the right singular vectors of two */
651 /* > submatrices in the two square blocks with corners at (1,1), */
652 /* > (NL+1, NL+1), and (NL+2, NL+2), (M,M). */
653 /* > On exit VT**T contains the trailing (N-K) updated right singular */
654 /* > vectors (those which were deflated) in its last N-K columns. */
655 /* > In case SQRE =1, the last row of VT spans the right null */
659 /* > \param[in] LDVT */
661 /* > LDVT is INTEGER */
662 /* > The leading dimension of the array VT. LDVT >= M. */
665 /* > \param[out] DSIGMA */
667 /* > DSIGMA is REAL array, dimension (N) */
668 /* > Contains a copy of the diagonal elements (K-1 singular values */
669 /* > and one zero) in the secular equation. */
672 /* > \param[out] U2 */
674 /* > U2 is REAL array, dimension (LDU2,N) */
675 /* > Contains a copy of the first K-1 left singular vectors which */
676 /* > will be used by SLASD3 in a matrix multiply (SGEMM) to solve */
677 /* > for the new left singular vectors. U2 is arranged into four */
678 /* > blocks. The first block contains a column with 1 at NL+1 and */
679 /* > zero everywhere else; the second block contains non-zero */
680 /* > entries only at and above NL; the third contains non-zero */
681 /* > entries only below NL+1; and the fourth is dense. */
684 /* > \param[in] LDU2 */
686 /* > LDU2 is INTEGER */
687 /* > The leading dimension of the array U2. LDU2 >= N. */
690 /* > \param[out] VT2 */
692 /* > VT2 is REAL array, dimension (LDVT2,N) */
693 /* > VT2**T contains a copy of the first K right singular vectors */
694 /* > which will be used by SLASD3 in a matrix multiply (SGEMM) to */
695 /* > solve for the new right singular vectors. VT2 is arranged into */
696 /* > three blocks. The first block contains a row that corresponds */
697 /* > to the special 0 diagonal element in SIGMA; the second block */
698 /* > contains non-zeros only at and before NL +1; the third block */
699 /* > contains non-zeros only at and after NL +2. */
702 /* > \param[in] LDVT2 */
704 /* > LDVT2 is INTEGER */
705 /* > The leading dimension of the array VT2. LDVT2 >= M. */
708 /* > \param[out] IDXP */
710 /* > IDXP is INTEGER array, dimension (N) */
711 /* > This will contain the permutation used to place deflated */
712 /* > values of D at the end of the array. On output IDXP(2:K) */
713 /* > points to the nondeflated D-values and IDXP(K+1:N) */
714 /* > points to the deflated singular values. */
717 /* > \param[out] IDX */
719 /* > IDX is INTEGER array, dimension (N) */
720 /* > This will contain the permutation used to sort the contents of */
721 /* > D into ascending order. */
724 /* > \param[out] IDXC */
726 /* > IDXC is INTEGER array, dimension (N) */
727 /* > This will contain the permutation used to arrange the columns */
728 /* > of the deflated U matrix into three groups: the first group */
729 /* > contains non-zero entries only at and above NL, the second */
730 /* > contains non-zero entries only below NL+2, and the third is */
734 /* > \param[in,out] IDXQ */
736 /* > IDXQ is INTEGER array, dimension (N) */
737 /* > This contains the permutation which separately sorts the two */
738 /* > sub-problems in D into ascending order. Note that entries in */
739 /* > the first hlaf of this permutation must first be moved one */
740 /* > position backward; and entries in the second half */
741 /* > must first have NL+1 added to their values. */
744 /* > \param[out] COLTYP */
746 /* > COLTYP is INTEGER array, dimension (N) */
747 /* > As workspace, this will contain a label which will indicate */
748 /* > which of the following types a column in the U2 matrix or a */
749 /* > row in the VT2 matrix is: */
750 /* > 1 : non-zero in the upper half only */
751 /* > 2 : non-zero in the lower half only */
755 /* > On exit, it is an array of dimension 4, with COLTYP(I) being */
756 /* > the dimension of the I-th type columns. */
759 /* > \param[out] INFO */
761 /* > INFO is INTEGER */
762 /* > = 0: successful exit. */
763 /* > < 0: if INFO = -i, the i-th argument had an illegal value. */
769 /* > \author Univ. of Tennessee */
770 /* > \author Univ. of California Berkeley */
771 /* > \author Univ. of Colorado Denver */
772 /* > \author NAG Ltd. */
774 /* > \date December 2016 */
776 /* > \ingroup OTHERauxiliary */
778 /* > \par Contributors: */
779 /* ================== */
781 /* > Ming Gu and Huan Ren, Computer Science Division, University of */
782 /* > California at Berkeley, USA */
784 /* ===================================================================== */
785 /* Subroutine */ int slasd2_(integer *nl, integer *nr, integer *sqre, integer
786 *k, real *d__, real *z__, real *alpha, real *beta, real *u, integer *
787 ldu, real *vt, integer *ldvt, real *dsigma, real *u2, integer *ldu2,
788 real *vt2, integer *ldvt2, integer *idxp, integer *idx, integer *idxc,
789 integer *idxq, integer *coltyp, integer *info)
791 /* System generated locals */
792 integer u_dim1, u_offset, u2_dim1, u2_offset, vt_dim1, vt_offset,
793 vt2_dim1, vt2_offset, i__1;
796 /* Local variables */
797 integer idxi, idxj, ctot[4];
798 extern /* Subroutine */ int srot_(integer *, real *, integer *, real *,
799 integer *, real *, real *);
801 integer i__, j, m, n;
803 integer idxjp, jprev, k2;
804 extern /* Subroutine */ int scopy_(integer *, real *, integer *, real *,
807 extern real slapy2_(real *, real *);
809 extern real slamch_(char *);
810 extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen), slamrg_(
811 integer *, integer *, real *, integer *, integer *, integer *);
813 extern /* Subroutine */ int slacpy_(char *, integer *, integer *, real *,
814 integer *, real *, integer *), slaset_(char *, integer *,
815 integer *, real *, real *, real *, integer *);
817 integer psm[4], nlp1, nlp2;
820 /* -- LAPACK auxiliary routine (version 3.7.0) -- */
821 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
822 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
826 /* ===================================================================== */
829 /* Test the input parameters. */
831 /* Parameter adjustments */
835 u_offset = 1 + u_dim1 * 1;
838 vt_offset = 1 + vt_dim1 * 1;
842 u2_offset = 1 + u2_dim1 * 1;
845 vt2_offset = 1 + vt2_dim1 * 1;
858 } else if (*nr < 1) {
860 } else if (*sqre != 1 && *sqre != 0) {
869 } else if (*ldvt < m) {
871 } else if (*ldu2 < n) {
873 } else if (*ldvt2 < m) {
878 xerbla_("SLASD2", &i__1, (ftnlen)6);
885 /* Generate the first part of the vector Z; and move the singular */
886 /* values in the first part of D one position backward. */
888 z1 = *alpha * vt[nlp1 + nlp1 * vt_dim1];
890 for (i__ = *nl; i__ >= 1; --i__) {
891 z__[i__ + 1] = *alpha * vt[i__ + nlp1 * vt_dim1];
892 d__[i__ + 1] = d__[i__];
893 idxq[i__ + 1] = idxq[i__] + 1;
897 /* Generate the second part of the vector Z. */
900 for (i__ = nlp2; i__ <= i__1; ++i__) {
901 z__[i__] = *beta * vt[i__ + nlp2 * vt_dim1];
905 /* Initialize some reference arrays. */
908 for (i__ = 2; i__ <= i__1; ++i__) {
913 for (i__ = nlp2; i__ <= i__1; ++i__) {
918 /* Sort the singular values into increasing order */
921 for (i__ = nlp2; i__ <= i__1; ++i__) {
926 /* DSIGMA, IDXC, IDXC, and the first column of U2 */
927 /* are used as storage space. */
930 for (i__ = 2; i__ <= i__1; ++i__) {
931 dsigma[i__] = d__[idxq[i__]];
932 u2[i__ + u2_dim1] = z__[idxq[i__]];
933 idxc[i__] = coltyp[idxq[i__]];
937 slamrg_(nl, nr, &dsigma[2], &c__1, &c__1, &idx[2]);
940 for (i__ = 2; i__ <= i__1; ++i__) {
942 d__[i__] = dsigma[idxi];
943 z__[i__] = u2[idxi + u2_dim1];
944 coltyp[i__] = idxc[idxi];
948 /* Calculate the allowable deflation tolerance */
950 eps = slamch_("Epsilon");
952 r__1 = abs(*alpha), r__2 = abs(*beta);
953 tol = f2cmax(r__1,r__2);
955 r__2 = (r__1 = d__[n], abs(r__1));
956 tol = eps * 8.f * f2cmax(r__2,tol);
958 /* There are 2 kinds of deflation -- first a value in the z-vector */
959 /* is small, second two (or more) singular values are very close */
960 /* together (their difference is small). */
962 /* If the value in the z-vector is small, we simply permute the */
963 /* array so that the corresponding singular value is moved to the */
966 /* If two values in the D-vector are close, we perform a two-sided */
967 /* rotation designed to make one of the corresponding z-vector */
968 /* entries zero, and then permute the array so that the deflated */
969 /* singular value is moved to the end. */
971 /* If there are multiple singular values then the problem deflates. */
972 /* Here the number of equal singular values are found. As each equal */
973 /* singular value is found, an elementary reflector is computed to */
974 /* rotate the corresponding singular subspace so that the */
975 /* corresponding components of Z are zero in this new basis. */
980 for (j = 2; j <= i__1; ++j) {
981 if ((r__1 = z__[j], abs(r__1)) <= tol) {
983 /* Deflate due to small z component. */
1004 if ((r__1 = z__[j], abs(r__1)) <= tol) {
1006 /* Deflate due to small z component. */
1013 /* Check if singular values are close enough to allow deflation. */
1015 if ((r__1 = d__[j] - d__[jprev], abs(r__1)) <= tol) {
1017 /* Deflation is possible. */
1022 /* Find sqrt(a**2+b**2) without overflow or */
1023 /* destructive underflow. */
1025 tau = slapy2_(&c__, &s);
1031 /* Apply back the Givens rotation to the left and right */
1032 /* singular vector matrices. */
1034 idxjp = idxq[idx[jprev] + 1];
1035 idxj = idxq[idx[j] + 1];
1036 if (idxjp <= nlp1) {
1042 srot_(&n, &u[idxjp * u_dim1 + 1], &c__1, &u[idxj * u_dim1 + 1], &
1044 srot_(&m, &vt[idxjp + vt_dim1], ldvt, &vt[idxj + vt_dim1], ldvt, &
1046 if (coltyp[j] != coltyp[jprev]) {
1055 u2[*k + u2_dim1] = z__[jprev];
1056 dsigma[*k] = d__[jprev];
1064 /* Record the last singular value. */
1067 u2[*k + u2_dim1] = z__[jprev];
1068 dsigma[*k] = d__[jprev];
1073 /* Count up the total number of the various types of columns, then */
1074 /* form a permutation which positions the four column types into */
1075 /* four groups of uniform structure (although one or more of these */
1076 /* groups may be empty). */
1078 for (j = 1; j <= 4; ++j) {
1083 for (j = 2; j <= i__1; ++j) {
1089 /* PSM(*) = Position in SubMatrix (of types 1 through 4) */
1092 psm[1] = ctot[0] + 2;
1093 psm[2] = psm[1] + ctot[1];
1094 psm[3] = psm[2] + ctot[2];
1096 /* Fill out the IDXC array so that the permutation which it induces */
1097 /* will place all type-1 columns first, all type-2 columns next, */
1098 /* then all type-3's, and finally all type-4's, starting from the */
1099 /* second column. This applies similarly to the rows of VT. */
1102 for (j = 2; j <= i__1; ++j) {
1105 idxc[psm[ct - 1]] = j;
1110 /* Sort the singular values and corresponding singular vectors into */
1111 /* DSIGMA, U2, and VT2 respectively. The singular values/vectors */
1112 /* which were not deflated go into the first K slots of DSIGMA, U2, */
1113 /* and VT2 respectively, while those which were deflated go into the */
1114 /* last N - K slots, except that the first column/row will be treated */
1118 for (j = 2; j <= i__1; ++j) {
1120 dsigma[j] = d__[jp];
1121 idxj = idxq[idx[idxp[idxc[j]]] + 1];
1125 scopy_(&n, &u[idxj * u_dim1 + 1], &c__1, &u2[j * u2_dim1 + 1], &c__1);
1126 scopy_(&m, &vt[idxj + vt_dim1], ldvt, &vt2[j + vt2_dim1], ldvt2);
1130 /* Determine DSIGMA(1), DSIGMA(2) and Z(1) */
1134 if (abs(dsigma[2]) <= hlftol) {
1138 z__[1] = slapy2_(&z1, &z__[m]);
1139 if (z__[1] <= tol) {
1145 s = z__[m] / z__[1];
1148 if (abs(z1) <= tol) {
1155 /* Move the rest of the updating row to Z. */
1158 scopy_(&i__1, &u2[u2_dim1 + 2], &c__1, &z__[2], &c__1);
1160 /* Determine the first column of U2, the first row of VT2 and the */
1161 /* last row of VT. */
1163 slaset_("A", &n, &c__1, &c_b30, &c_b30, &u2[u2_offset], ldu2);
1164 u2[nlp1 + u2_dim1] = 1.f;
1167 for (i__ = 1; i__ <= i__1; ++i__) {
1168 vt[m + i__ * vt_dim1] = -s * vt[nlp1 + i__ * vt_dim1];
1169 vt2[i__ * vt2_dim1 + 1] = c__ * vt[nlp1 + i__ * vt_dim1];
1173 for (i__ = nlp2; i__ <= i__1; ++i__) {
1174 vt2[i__ * vt2_dim1 + 1] = s * vt[m + i__ * vt_dim1];
1175 vt[m + i__ * vt_dim1] = c__ * vt[m + i__ * vt_dim1];
1179 scopy_(&m, &vt[nlp1 + vt_dim1], ldvt, &vt2[vt2_dim1 + 1], ldvt2);
1182 scopy_(&m, &vt[m + vt_dim1], ldvt, &vt2[m + vt2_dim1], ldvt2);
1185 /* The deflated singular values and their corresponding vectors go */
1186 /* into the back of D, U, and V respectively. */
1190 scopy_(&i__1, &dsigma[*k + 1], &c__1, &d__[*k + 1], &c__1);
1192 slacpy_("A", &n, &i__1, &u2[(*k + 1) * u2_dim1 + 1], ldu2, &u[(*k + 1)
1193 * u_dim1 + 1], ldu);
1195 slacpy_("A", &i__1, &m, &vt2[*k + 1 + vt2_dim1], ldvt2, &vt[*k + 1 +
1199 /* Copy CTOT into COLTYP for referencing in SLASD3. */
1201 for (j = 1; j <= 4; ++j) {
1202 coltyp[j] = ctot[j - 1];