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_n1 = -1;
516 static integer c__0 = 0;
517 static doublereal c_b63 = 0.;
518 static integer c__1 = 1;
519 static doublereal c_b84 = 1.;
521 /* > \brief \b DGESDD */
523 /* =========== DOCUMENTATION =========== */
525 /* Online html documentation available at */
526 /* http://www.netlib.org/lapack/explore-html/ */
529 /* > Download DGESDD + dependencies */
530 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dgesdd.
533 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dgesdd.
536 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dgesdd.
544 /* SUBROUTINE DGESDD( JOBZ, M, N, A, LDA, S, U, LDU, VT, LDVT, */
545 /* WORK, LWORK, IWORK, INFO ) */
548 /* INTEGER INFO, LDA, LDU, LDVT, LWORK, M, N */
549 /* INTEGER IWORK( * ) */
550 /* DOUBLE PRECISION A( LDA, * ), S( * ), U( LDU, * ), */
551 /* $ VT( LDVT, * ), WORK( * ) */
554 /* > \par Purpose: */
559 /* > DGESDD computes the singular value decomposition (SVD) of a real */
560 /* > M-by-N matrix A, optionally computing the left and right singular */
561 /* > vectors. If singular vectors are desired, it uses a */
562 /* > divide-and-conquer algorithm. */
564 /* > The SVD is written */
566 /* > A = U * SIGMA * transpose(V) */
568 /* > where SIGMA is an M-by-N matrix which is zero except for its */
569 /* > f2cmin(m,n) diagonal elements, U is an M-by-M orthogonal matrix, and */
570 /* > V is an N-by-N orthogonal matrix. The diagonal elements of SIGMA */
571 /* > are the singular values of A; they are real and non-negative, and */
572 /* > are returned in descending order. The first f2cmin(m,n) columns of */
573 /* > U and V are the left and right singular vectors of A. */
575 /* > Note that the routine returns VT = V**T, not V. */
577 /* > The divide and conquer algorithm makes very mild assumptions about */
578 /* > floating point arithmetic. It will work on machines with a guard */
579 /* > digit in add/subtract, or on those binary machines without guard */
580 /* > digits which subtract like the Cray X-MP, Cray Y-MP, Cray C-90, or */
581 /* > Cray-2. It could conceivably fail on hexadecimal or decimal machines */
582 /* > without guard digits, but we know of none. */
588 /* > \param[in] JOBZ */
590 /* > JOBZ is CHARACTER*1 */
591 /* > Specifies options for computing all or part of the matrix U: */
592 /* > = 'A': all M columns of U and all N rows of V**T are */
593 /* > returned in the arrays U and VT; */
594 /* > = 'S': the first f2cmin(M,N) columns of U and the first */
595 /* > f2cmin(M,N) rows of V**T are returned in the arrays U */
597 /* > = 'O': If M >= N, the first N columns of U are overwritten */
598 /* > on the array A and all rows of V**T are returned in */
599 /* > the array VT; */
600 /* > otherwise, all columns of U are returned in the */
601 /* > array U and the first M rows of V**T are overwritten */
602 /* > in the array A; */
603 /* > = 'N': no columns of U or rows of V**T are computed. */
609 /* > The number of rows of the input matrix A. M >= 0. */
615 /* > The number of columns of the input matrix A. N >= 0. */
618 /* > \param[in,out] A */
620 /* > A is DOUBLE PRECISION array, dimension (LDA,N) */
621 /* > On entry, the M-by-N matrix A. */
623 /* > if JOBZ = 'O', A is overwritten with the first N columns */
624 /* > of U (the left singular vectors, stored */
625 /* > columnwise) if M >= N; */
626 /* > A is overwritten with the first M rows */
627 /* > of V**T (the right singular vectors, stored */
628 /* > rowwise) otherwise. */
629 /* > if JOBZ .ne. 'O', the contents of A are destroyed. */
632 /* > \param[in] LDA */
634 /* > LDA is INTEGER */
635 /* > The leading dimension of the array A. LDA >= f2cmax(1,M). */
638 /* > \param[out] S */
640 /* > S is DOUBLE PRECISION array, dimension (f2cmin(M,N)) */
641 /* > The singular values of A, sorted so that S(i) >= S(i+1). */
644 /* > \param[out] U */
646 /* > U is DOUBLE PRECISION array, dimension (LDU,UCOL) */
647 /* > UCOL = M if JOBZ = 'A' or JOBZ = 'O' and M < N; */
648 /* > UCOL = f2cmin(M,N) if JOBZ = 'S'. */
649 /* > If JOBZ = 'A' or JOBZ = 'O' and M < N, U contains the M-by-M */
650 /* > orthogonal matrix U; */
651 /* > if JOBZ = 'S', U contains the first f2cmin(M,N) columns of U */
652 /* > (the left singular vectors, stored columnwise); */
653 /* > if JOBZ = 'O' and M >= N, or JOBZ = 'N', U is not referenced. */
656 /* > \param[in] LDU */
658 /* > LDU is INTEGER */
659 /* > The leading dimension of the array U. LDU >= 1; if */
660 /* > JOBZ = 'S' or 'A' or JOBZ = 'O' and M < N, LDU >= M. */
663 /* > \param[out] VT */
665 /* > VT is DOUBLE PRECISION array, dimension (LDVT,N) */
666 /* > If JOBZ = 'A' or JOBZ = 'O' and M >= N, VT contains the */
667 /* > N-by-N orthogonal matrix V**T; */
668 /* > if JOBZ = 'S', VT contains the first f2cmin(M,N) rows of */
669 /* > V**T (the right singular vectors, stored rowwise); */
670 /* > if JOBZ = 'O' and M < N, or JOBZ = 'N', VT is not referenced. */
673 /* > \param[in] LDVT */
675 /* > LDVT is INTEGER */
676 /* > The leading dimension of the array VT. LDVT >= 1; */
677 /* > if JOBZ = 'A' or JOBZ = 'O' and M >= N, LDVT >= N; */
678 /* > if JOBZ = 'S', LDVT >= f2cmin(M,N). */
681 /* > \param[out] WORK */
683 /* > WORK is DOUBLE PRECISION array, dimension (MAX(1,LWORK)) */
684 /* > On exit, if INFO = 0, WORK(1) returns the optimal LWORK; */
687 /* > \param[in] LWORK */
689 /* > LWORK is INTEGER */
690 /* > The dimension of the array WORK. LWORK >= 1. */
691 /* > If LWORK = -1, a workspace query is assumed. The optimal */
692 /* > size for the WORK array is calculated and stored in WORK(1), */
693 /* > and no other work except argument checking is performed. */
695 /* > Let mx = f2cmax(M,N) and mn = f2cmin(M,N). */
696 /* > If JOBZ = 'N', LWORK >= 3*mn + f2cmax( mx, 7*mn ). */
697 /* > If JOBZ = 'O', LWORK >= 3*mn + f2cmax( mx, 5*mn*mn + 4*mn ). */
698 /* > If JOBZ = 'S', LWORK >= 4*mn*mn + 7*mn. */
699 /* > If JOBZ = 'A', LWORK >= 4*mn*mn + 6*mn + mx. */
700 /* > These are not tight minimums in all cases; see comments inside code. */
701 /* > For good performance, LWORK should generally be larger; */
702 /* > a query is recommended. */
705 /* > \param[out] IWORK */
707 /* > IWORK is INTEGER array, dimension (8*f2cmin(M,N)) */
710 /* > \param[out] INFO */
712 /* > INFO is INTEGER */
713 /* > = 0: successful exit. */
714 /* > < 0: if INFO = -i, the i-th argument had an illegal value. */
715 /* > > 0: DBDSDC did not converge, updating process failed. */
721 /* > \author Univ. of Tennessee */
722 /* > \author Univ. of California Berkeley */
723 /* > \author Univ. of Colorado Denver */
724 /* > \author NAG Ltd. */
726 /* > \date June 2016 */
728 /* > \ingroup doubleGEsing */
730 /* > \par Contributors: */
731 /* ================== */
733 /* > Ming Gu and Huan Ren, Computer Science Division, University of */
734 /* > California at Berkeley, USA */
736 /* ===================================================================== */
737 /* Subroutine */ int dgesdd_(char *jobz, integer *m, integer *n, doublereal *
738 a, integer *lda, doublereal *s, doublereal *u, integer *ldu,
739 doublereal *vt, integer *ldvt, doublereal *work, integer *lwork,
740 integer *iwork, integer *info)
742 /* System generated locals */
743 integer a_dim1, a_offset, u_dim1, u_offset, vt_dim1, vt_offset, i__1,
746 /* Local variables */
747 integer lwork_dorglq_mn__, lwork_dorglq_nn__, lwork_dorgqr_mm__,
748 lwork_dorgqr_mn__, iscl;
750 integer idum[1], ierr, itau, lwork_dormbr_qln_mm__, lwork_dormbr_qln_mn__,
751 lwork_dormbr_qln_nn__, lwork_dormbr_prt_mm__,
752 lwork_dormbr_prt_mn__, lwork_dormbr_prt_nn__, i__;
753 extern /* Subroutine */ int dgemm_(char *, char *, integer *, integer *,
754 integer *, doublereal *, doublereal *, integer *, doublereal *,
755 integer *, doublereal *, doublereal *, integer *);
756 extern logical lsame_(char *, char *);
757 integer chunk, minmn, wrkbl, itaup, itauq, mnthr;
760 logical wntqn, wntqo, wntqs;
761 integer ie, lwork_dorgbr_p_mm__;
762 extern /* Subroutine */ int dbdsdc_(char *, char *, integer *, doublereal
763 *, doublereal *, doublereal *, integer *, doublereal *, integer *,
764 doublereal *, integer *, doublereal *, integer *, integer *);
765 integer il, lwork_dorgbr_q_nn__;
766 extern /* Subroutine */ int dgebrd_(integer *, integer *, doublereal *,
767 integer *, doublereal *, doublereal *, doublereal *, doublereal *,
768 doublereal *, integer *, integer *);
769 extern doublereal dlamch_(char *);
771 extern doublereal dlange_(char *, integer *, integer *, doublereal *,
772 integer *, doublereal *);
774 extern /* Subroutine */ int dgelqf_(integer *, integer *, doublereal *,
775 integer *, doublereal *, doublereal *, integer *, integer *),
776 dlascl_(char *, integer *, integer *, doublereal *, doublereal *,
777 integer *, integer *, doublereal *, integer *, integer *),
778 dgeqrf_(integer *, integer *, doublereal *, integer *,
779 doublereal *, doublereal *, integer *, integer *), dlacpy_(char *,
780 integer *, integer *, doublereal *, integer *, doublereal *,
781 integer *), dlaset_(char *, integer *, integer *,
782 doublereal *, doublereal *, doublereal *, integer *),
783 xerbla_(char *, integer *, ftnlen), dorgbr_(char *, integer *,
784 integer *, integer *, doublereal *, integer *, doublereal *,
785 doublereal *, integer *, integer *);
786 extern logical disnan_(doublereal *);
788 extern /* Subroutine */ int dormbr_(char *, char *, char *, integer *,
789 integer *, integer *, doublereal *, integer *, doublereal *,
790 doublereal *, integer *, doublereal *, integer *, integer *), dorglq_(integer *, integer *, integer *,
791 doublereal *, integer *, doublereal *, doublereal *, integer *,
792 integer *), dorgqr_(integer *, integer *, integer *, doublereal *,
793 integer *, doublereal *, doublereal *, integer *, integer *);
794 integer ldwrkl, ldwrkr, minwrk, ldwrku, maxwrk, ldwkvt;
796 logical wntqas, lquery;
798 doublereal dum[1], eps;
799 integer ivt, lwork_dgebrd_mm__, lwork_dgebrd_mn__, lwork_dgebrd_nn__,
800 lwork_dgelqf_mn__, lwork_dgeqrf_mn__;
803 /* -- LAPACK driver routine (version 3.7.0) -- */
804 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
805 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
809 /* ===================================================================== */
812 /* Test the input arguments */
814 /* Parameter adjustments */
816 a_offset = 1 + a_dim1 * 1;
820 u_offset = 1 + u_dim1 * 1;
823 vt_offset = 1 + vt_dim1 * 1;
830 minmn = f2cmin(*m,*n);
831 wntqa = lsame_(jobz, "A");
832 wntqs = lsame_(jobz, "S");
833 wntqas = wntqa || wntqs;
834 wntqo = lsame_(jobz, "O");
835 wntqn = lsame_(jobz, "N");
836 lquery = *lwork == -1;
838 if (! (wntqa || wntqs || wntqo || wntqn)) {
844 } else if (*lda < f2cmax(1,*m)) {
846 } else if (*ldu < 1 || wntqas && *ldu < *m || wntqo && *m < *n && *ldu < *
849 } else if (*ldvt < 1 || wntqa && *ldvt < *n || wntqs && *ldvt < minmn ||
850 wntqo && *m >= *n && *ldvt < *n) {
854 /* Compute workspace */
855 /* Note: Comments in the code beginning "Workspace:" describe the */
856 /* minimal amount of workspace allocated at that point in the code, */
857 /* as well as the preferred amount for good performance. */
858 /* NB refers to the optimal block size for the immediately */
859 /* following subroutine, as returned by ILAENV. */
865 mnthr = (integer) (minmn * 11. / 6.);
866 if (*m >= *n && minmn > 0) {
868 /* Compute space needed for DBDSDC */
871 /* dbdsdc needs only 4*N (or 6*N for uplo=L for LAPACK <= 3.6) */
872 /* keep 7*N for backwards compatibility. */
875 bdspac = *n * 3 * *n + (*n << 2);
878 /* Compute space preferred for each routine */
879 dgebrd_(m, n, dum, m, dum, dum, dum, dum, dum, &c_n1, &ierr);
880 lwork_dgebrd_mn__ = (integer) dum[0];
882 dgebrd_(n, n, dum, n, dum, dum, dum, dum, dum, &c_n1, &ierr);
883 lwork_dgebrd_nn__ = (integer) dum[0];
885 dgeqrf_(m, n, dum, m, dum, dum, &c_n1, &ierr);
886 lwork_dgeqrf_mn__ = (integer) dum[0];
888 dorgbr_("Q", n, n, n, dum, n, dum, dum, &c_n1, &ierr);
889 lwork_dorgbr_q_nn__ = (integer) dum[0];
891 dorgqr_(m, m, n, dum, m, dum, dum, &c_n1, &ierr);
892 lwork_dorgqr_mm__ = (integer) dum[0];
894 dorgqr_(m, n, n, dum, m, dum, dum, &c_n1, &ierr);
895 lwork_dorgqr_mn__ = (integer) dum[0];
897 dormbr_("P", "R", "T", n, n, n, dum, n, dum, dum, n, dum, &c_n1, &
899 lwork_dormbr_prt_nn__ = (integer) dum[0];
901 dormbr_("Q", "L", "N", n, n, n, dum, n, dum, dum, n, dum, &c_n1, &
903 lwork_dormbr_qln_nn__ = (integer) dum[0];
905 dormbr_("Q", "L", "N", m, n, n, dum, m, dum, dum, m, dum, &c_n1, &
907 lwork_dormbr_qln_mn__ = (integer) dum[0];
909 dormbr_("Q", "L", "N", m, m, n, dum, m, dum, dum, m, dum, &c_n1, &
911 lwork_dormbr_qln_mm__ = (integer) dum[0];
916 /* Path 1 (M >> N, JOBZ='N') */
918 wrkbl = *n + lwork_dgeqrf_mn__;
920 i__1 = wrkbl, i__2 = *n * 3 + lwork_dgebrd_nn__;
921 wrkbl = f2cmax(i__1,i__2);
923 i__1 = wrkbl, i__2 = bdspac + *n;
924 maxwrk = f2cmax(i__1,i__2);
925 minwrk = bdspac + *n;
928 /* Path 2 (M >> N, JOBZ='O') */
930 wrkbl = *n + lwork_dgeqrf_mn__;
932 i__1 = wrkbl, i__2 = *n + lwork_dorgqr_mn__;
933 wrkbl = f2cmax(i__1,i__2);
935 i__1 = wrkbl, i__2 = *n * 3 + lwork_dgebrd_nn__;
936 wrkbl = f2cmax(i__1,i__2);
938 i__1 = wrkbl, i__2 = *n * 3 + lwork_dormbr_qln_nn__;
939 wrkbl = f2cmax(i__1,i__2);
941 i__1 = wrkbl, i__2 = *n * 3 + lwork_dormbr_prt_nn__;
942 wrkbl = f2cmax(i__1,i__2);
944 i__1 = wrkbl, i__2 = *n * 3 + bdspac;
945 wrkbl = f2cmax(i__1,i__2);
946 maxwrk = wrkbl + (*n << 1) * *n;
947 minwrk = bdspac + (*n << 1) * *n + *n * 3;
950 /* Path 3 (M >> N, JOBZ='S') */
952 wrkbl = *n + lwork_dgeqrf_mn__;
954 i__1 = wrkbl, i__2 = *n + lwork_dorgqr_mn__;
955 wrkbl = f2cmax(i__1,i__2);
957 i__1 = wrkbl, i__2 = *n * 3 + lwork_dgebrd_nn__;
958 wrkbl = f2cmax(i__1,i__2);
960 i__1 = wrkbl, i__2 = *n * 3 + lwork_dormbr_qln_nn__;
961 wrkbl = f2cmax(i__1,i__2);
963 i__1 = wrkbl, i__2 = *n * 3 + lwork_dormbr_prt_nn__;
964 wrkbl = f2cmax(i__1,i__2);
966 i__1 = wrkbl, i__2 = *n * 3 + bdspac;
967 wrkbl = f2cmax(i__1,i__2);
968 maxwrk = wrkbl + *n * *n;
969 minwrk = bdspac + *n * *n + *n * 3;
972 /* Path 4 (M >> N, JOBZ='A') */
974 wrkbl = *n + lwork_dgeqrf_mn__;
976 i__1 = wrkbl, i__2 = *n + lwork_dorgqr_mm__;
977 wrkbl = f2cmax(i__1,i__2);
979 i__1 = wrkbl, i__2 = *n * 3 + lwork_dgebrd_nn__;
980 wrkbl = f2cmax(i__1,i__2);
982 i__1 = wrkbl, i__2 = *n * 3 + lwork_dormbr_qln_nn__;
983 wrkbl = f2cmax(i__1,i__2);
985 i__1 = wrkbl, i__2 = *n * 3 + lwork_dormbr_prt_nn__;
986 wrkbl = f2cmax(i__1,i__2);
988 i__1 = wrkbl, i__2 = *n * 3 + bdspac;
989 wrkbl = f2cmax(i__1,i__2);
990 maxwrk = wrkbl + *n * *n;
992 i__1 = *n * 3 + bdspac, i__2 = *n + *m;
993 minwrk = *n * *n + f2cmax(i__1,i__2);
997 /* Path 5 (M >= N, but not much larger) */
999 wrkbl = *n * 3 + lwork_dgebrd_mn__;
1001 /* Path 5n (M >= N, jobz='N') */
1003 i__1 = wrkbl, i__2 = *n * 3 + bdspac;
1004 maxwrk = f2cmax(i__1,i__2);
1005 minwrk = *n * 3 + f2cmax(*m,bdspac);
1007 /* Path 5o (M >= N, jobz='O') */
1009 i__1 = wrkbl, i__2 = *n * 3 + lwork_dormbr_prt_nn__;
1010 wrkbl = f2cmax(i__1,i__2);
1012 i__1 = wrkbl, i__2 = *n * 3 + lwork_dormbr_qln_mn__;
1013 wrkbl = f2cmax(i__1,i__2);
1015 i__1 = wrkbl, i__2 = *n * 3 + bdspac;
1016 wrkbl = f2cmax(i__1,i__2);
1017 maxwrk = wrkbl + *m * *n;
1019 i__1 = *m, i__2 = *n * *n + bdspac;
1020 minwrk = *n * 3 + f2cmax(i__1,i__2);
1022 /* Path 5s (M >= N, jobz='S') */
1024 i__1 = wrkbl, i__2 = *n * 3 + lwork_dormbr_qln_mn__;
1025 wrkbl = f2cmax(i__1,i__2);
1027 i__1 = wrkbl, i__2 = *n * 3 + lwork_dormbr_prt_nn__;
1028 wrkbl = f2cmax(i__1,i__2);
1030 i__1 = wrkbl, i__2 = *n * 3 + bdspac;
1031 maxwrk = f2cmax(i__1,i__2);
1032 minwrk = *n * 3 + f2cmax(*m,bdspac);
1034 /* Path 5a (M >= N, jobz='A') */
1036 i__1 = wrkbl, i__2 = *n * 3 + lwork_dormbr_qln_mm__;
1037 wrkbl = f2cmax(i__1,i__2);
1039 i__1 = wrkbl, i__2 = *n * 3 + lwork_dormbr_prt_nn__;
1040 wrkbl = f2cmax(i__1,i__2);
1042 i__1 = wrkbl, i__2 = *n * 3 + bdspac;
1043 maxwrk = f2cmax(i__1,i__2);
1044 minwrk = *n * 3 + f2cmax(*m,bdspac);
1047 } else if (minmn > 0) {
1049 /* Compute space needed for DBDSDC */
1052 /* dbdsdc needs only 4*N (or 6*N for uplo=L for LAPACK <= 3.6) */
1053 /* keep 7*N for backwards compatibility. */
1056 bdspac = *m * 3 * *m + (*m << 2);
1059 /* Compute space preferred for each routine */
1060 dgebrd_(m, n, dum, m, dum, dum, dum, dum, dum, &c_n1, &ierr);
1061 lwork_dgebrd_mn__ = (integer) dum[0];
1063 dgebrd_(m, m, &a[a_offset], m, &s[1], dum, dum, dum, dum, &c_n1, &
1065 lwork_dgebrd_mm__ = (integer) dum[0];
1067 dgelqf_(m, n, &a[a_offset], m, dum, dum, &c_n1, &ierr);
1068 lwork_dgelqf_mn__ = (integer) dum[0];
1070 dorglq_(n, n, m, dum, n, dum, dum, &c_n1, &ierr);
1071 lwork_dorglq_nn__ = (integer) dum[0];
1073 dorglq_(m, n, m, &a[a_offset], m, dum, dum, &c_n1, &ierr);
1074 lwork_dorglq_mn__ = (integer) dum[0];
1076 dorgbr_("P", m, m, m, &a[a_offset], n, dum, dum, &c_n1, &ierr);
1077 lwork_dorgbr_p_mm__ = (integer) dum[0];
1079 dormbr_("P", "R", "T", m, m, m, dum, m, dum, dum, m, dum, &c_n1, &
1081 lwork_dormbr_prt_mm__ = (integer) dum[0];
1083 dormbr_("P", "R", "T", m, n, m, dum, m, dum, dum, m, dum, &c_n1, &
1085 lwork_dormbr_prt_mn__ = (integer) dum[0];
1087 dormbr_("P", "R", "T", n, n, m, dum, n, dum, dum, n, dum, &c_n1, &
1089 lwork_dormbr_prt_nn__ = (integer) dum[0];
1091 dormbr_("Q", "L", "N", m, m, m, dum, m, dum, dum, m, dum, &c_n1, &
1093 lwork_dormbr_qln_mm__ = (integer) dum[0];
1098 /* Path 1t (N >> M, JOBZ='N') */
1100 wrkbl = *m + lwork_dgelqf_mn__;
1102 i__1 = wrkbl, i__2 = *m * 3 + lwork_dgebrd_mm__;
1103 wrkbl = f2cmax(i__1,i__2);
1105 i__1 = wrkbl, i__2 = bdspac + *m;
1106 maxwrk = f2cmax(i__1,i__2);
1107 minwrk = bdspac + *m;
1110 /* Path 2t (N >> M, JOBZ='O') */
1112 wrkbl = *m + lwork_dgelqf_mn__;
1114 i__1 = wrkbl, i__2 = *m + lwork_dorglq_mn__;
1115 wrkbl = f2cmax(i__1,i__2);
1117 i__1 = wrkbl, i__2 = *m * 3 + lwork_dgebrd_mm__;
1118 wrkbl = f2cmax(i__1,i__2);
1120 i__1 = wrkbl, i__2 = *m * 3 + lwork_dormbr_qln_mm__;
1121 wrkbl = f2cmax(i__1,i__2);
1123 i__1 = wrkbl, i__2 = *m * 3 + lwork_dormbr_prt_mm__;
1124 wrkbl = f2cmax(i__1,i__2);
1126 i__1 = wrkbl, i__2 = *m * 3 + bdspac;
1127 wrkbl = f2cmax(i__1,i__2);
1128 maxwrk = wrkbl + (*m << 1) * *m;
1129 minwrk = bdspac + (*m << 1) * *m + *m * 3;
1132 /* Path 3t (N >> M, JOBZ='S') */
1134 wrkbl = *m + lwork_dgelqf_mn__;
1136 i__1 = wrkbl, i__2 = *m + lwork_dorglq_mn__;
1137 wrkbl = f2cmax(i__1,i__2);
1139 i__1 = wrkbl, i__2 = *m * 3 + lwork_dgebrd_mm__;
1140 wrkbl = f2cmax(i__1,i__2);
1142 i__1 = wrkbl, i__2 = *m * 3 + lwork_dormbr_qln_mm__;
1143 wrkbl = f2cmax(i__1,i__2);
1145 i__1 = wrkbl, i__2 = *m * 3 + lwork_dormbr_prt_mm__;
1146 wrkbl = f2cmax(i__1,i__2);
1148 i__1 = wrkbl, i__2 = *m * 3 + bdspac;
1149 wrkbl = f2cmax(i__1,i__2);
1150 maxwrk = wrkbl + *m * *m;
1151 minwrk = bdspac + *m * *m + *m * 3;
1154 /* Path 4t (N >> M, JOBZ='A') */
1156 wrkbl = *m + lwork_dgelqf_mn__;
1158 i__1 = wrkbl, i__2 = *m + lwork_dorglq_nn__;
1159 wrkbl = f2cmax(i__1,i__2);
1161 i__1 = wrkbl, i__2 = *m * 3 + lwork_dgebrd_mm__;
1162 wrkbl = f2cmax(i__1,i__2);
1164 i__1 = wrkbl, i__2 = *m * 3 + lwork_dormbr_qln_mm__;
1165 wrkbl = f2cmax(i__1,i__2);
1167 i__1 = wrkbl, i__2 = *m * 3 + lwork_dormbr_prt_mm__;
1168 wrkbl = f2cmax(i__1,i__2);
1170 i__1 = wrkbl, i__2 = *m * 3 + bdspac;
1171 wrkbl = f2cmax(i__1,i__2);
1172 maxwrk = wrkbl + *m * *m;
1174 i__1 = *m * 3 + bdspac, i__2 = *m + *n;
1175 minwrk = *m * *m + f2cmax(i__1,i__2);
1179 /* Path 5t (N > M, but not much larger) */
1181 wrkbl = *m * 3 + lwork_dgebrd_mn__;
1183 /* Path 5tn (N > M, jobz='N') */
1185 i__1 = wrkbl, i__2 = *m * 3 + bdspac;
1186 maxwrk = f2cmax(i__1,i__2);
1187 minwrk = *m * 3 + f2cmax(*n,bdspac);
1189 /* Path 5to (N > M, jobz='O') */
1191 i__1 = wrkbl, i__2 = *m * 3 + lwork_dormbr_qln_mm__;
1192 wrkbl = f2cmax(i__1,i__2);
1194 i__1 = wrkbl, i__2 = *m * 3 + lwork_dormbr_prt_mn__;
1195 wrkbl = f2cmax(i__1,i__2);
1197 i__1 = wrkbl, i__2 = *m * 3 + bdspac;
1198 wrkbl = f2cmax(i__1,i__2);
1199 maxwrk = wrkbl + *m * *n;
1201 i__1 = *n, i__2 = *m * *m + bdspac;
1202 minwrk = *m * 3 + f2cmax(i__1,i__2);
1204 /* Path 5ts (N > M, jobz='S') */
1206 i__1 = wrkbl, i__2 = *m * 3 + lwork_dormbr_qln_mm__;
1207 wrkbl = f2cmax(i__1,i__2);
1209 i__1 = wrkbl, i__2 = *m * 3 + lwork_dormbr_prt_mn__;
1210 wrkbl = f2cmax(i__1,i__2);
1212 i__1 = wrkbl, i__2 = *m * 3 + bdspac;
1213 maxwrk = f2cmax(i__1,i__2);
1214 minwrk = *m * 3 + f2cmax(*n,bdspac);
1216 /* Path 5ta (N > M, jobz='A') */
1218 i__1 = wrkbl, i__2 = *m * 3 + lwork_dormbr_qln_mm__;
1219 wrkbl = f2cmax(i__1,i__2);
1221 i__1 = wrkbl, i__2 = *m * 3 + lwork_dormbr_prt_nn__;
1222 wrkbl = f2cmax(i__1,i__2);
1224 i__1 = wrkbl, i__2 = *m * 3 + bdspac;
1225 maxwrk = f2cmax(i__1,i__2);
1226 minwrk = *m * 3 + f2cmax(*n,bdspac);
1230 maxwrk = f2cmax(maxwrk,minwrk);
1231 work[1] = (doublereal) maxwrk;
1233 if (*lwork < minwrk && ! lquery) {
1240 xerbla_("DGESDD", &i__1, (ftnlen)6);
1242 } else if (lquery) {
1246 /* Quick return if possible */
1248 if (*m == 0 || *n == 0) {
1252 /* Get machine constants */
1255 smlnum = sqrt(dlamch_("S")) / eps;
1256 bignum = 1. / smlnum;
1258 /* Scale A if f2cmax element outside range [SMLNUM,BIGNUM] */
1260 anrm = dlange_("M", m, n, &a[a_offset], lda, dum);
1261 if (disnan_(&anrm)) {
1266 if (anrm > 0. && anrm < smlnum) {
1268 dlascl_("G", &c__0, &c__0, &anrm, &smlnum, m, n, &a[a_offset], lda, &
1270 } else if (anrm > bignum) {
1272 dlascl_("G", &c__0, &c__0, &anrm, &bignum, m, n, &a[a_offset], lda, &
1278 /* A has at least as many rows as columns. If A has sufficiently */
1279 /* more rows than columns, first reduce using the QR */
1280 /* decomposition (if sufficient workspace available) */
1286 /* Path 1 (M >> N, JOBZ='N') */
1287 /* No singular vectors to be computed */
1293 /* Workspace: need N [tau] + N [work] */
1294 /* Workspace: prefer N [tau] + N*NB [work] */
1296 i__1 = *lwork - nwork + 1;
1297 dgeqrf_(m, n, &a[a_offset], lda, &work[itau], &work[nwork], &
1300 /* Zero out below R */
1304 dlaset_("L", &i__1, &i__2, &c_b63, &c_b63, &a[a_dim1 + 2],
1311 /* Bidiagonalize R in A */
1312 /* Workspace: need 3*N [e, tauq, taup] + N [work] */
1313 /* Workspace: prefer 3*N [e, tauq, taup] + 2*N*NB [work] */
1315 i__1 = *lwork - nwork + 1;
1316 dgebrd_(n, n, &a[a_offset], lda, &s[1], &work[ie], &work[
1317 itauq], &work[itaup], &work[nwork], &i__1, &ierr);
1320 /* Perform bidiagonal SVD, computing singular values only */
1321 /* Workspace: need N [e] + BDSPAC */
1323 dbdsdc_("U", "N", n, &s[1], &work[ie], dum, &c__1, dum, &c__1,
1324 dum, idum, &work[nwork], &iwork[1], info);
1328 /* Path 2 (M >> N, JOBZ = 'O') */
1329 /* N left singular vectors to be overwritten on A and */
1330 /* N right singular vectors to be computed in VT */
1334 /* WORK(IR) is LDWRKR by N */
1336 if (*lwork >= *lda * *n + *n * *n + *n * 3 + bdspac) {
1339 ldwrkr = (*lwork - *n * *n - *n * 3 - bdspac) / *n;
1341 itau = ir + ldwrkr * *n;
1345 /* Workspace: need N*N [R] + N [tau] + N [work] */
1346 /* Workspace: prefer N*N [R] + N [tau] + N*NB [work] */
1348 i__1 = *lwork - nwork + 1;
1349 dgeqrf_(m, n, &a[a_offset], lda, &work[itau], &work[nwork], &
1352 /* Copy R to WORK(IR), zeroing out below it */
1354 dlacpy_("U", n, n, &a[a_offset], lda, &work[ir], &ldwrkr);
1357 dlaset_("L", &i__1, &i__2, &c_b63, &c_b63, &work[ir + 1], &
1360 /* Generate Q in A */
1361 /* Workspace: need N*N [R] + N [tau] + N [work] */
1362 /* Workspace: prefer N*N [R] + N [tau] + N*NB [work] */
1364 i__1 = *lwork - nwork + 1;
1365 dorgqr_(m, n, n, &a[a_offset], lda, &work[itau], &work[nwork],
1372 /* Bidiagonalize R in WORK(IR) */
1373 /* Workspace: need N*N [R] + 3*N [e, tauq, taup] + N [work] */
1374 /* Workspace: prefer N*N [R] + 3*N [e, tauq, taup] + 2*N*NB [work] */
1376 i__1 = *lwork - nwork + 1;
1377 dgebrd_(n, n, &work[ir], &ldwrkr, &s[1], &work[ie], &work[
1378 itauq], &work[itaup], &work[nwork], &i__1, &ierr);
1380 /* WORK(IU) is N by N */
1383 nwork = iu + *n * *n;
1385 /* Perform bidiagonal SVD, computing left singular vectors */
1386 /* of bidiagonal matrix in WORK(IU) and computing right */
1387 /* singular vectors of bidiagonal matrix in VT */
1388 /* Workspace: need N*N [R] + 3*N [e, tauq, taup] + N*N [U] + BDSPAC */
1390 dbdsdc_("U", "I", n, &s[1], &work[ie], &work[iu], n, &vt[
1391 vt_offset], ldvt, dum, idum, &work[nwork], &iwork[1],
1394 /* Overwrite WORK(IU) by left singular vectors of R */
1395 /* and VT by right singular vectors of R */
1396 /* Workspace: need N*N [R] + 3*N [e, tauq, taup] + N*N [U] + N [work] */
1397 /* Workspace: prefer N*N [R] + 3*N [e, tauq, taup] + N*N [U] + N*NB [work] */
1399 i__1 = *lwork - nwork + 1;
1400 dormbr_("Q", "L", "N", n, n, n, &work[ir], &ldwrkr, &work[
1401 itauq], &work[iu], n, &work[nwork], &i__1, &ierr);
1402 i__1 = *lwork - nwork + 1;
1403 dormbr_("P", "R", "T", n, n, n, &work[ir], &ldwrkr, &work[
1404 itaup], &vt[vt_offset], ldvt, &work[nwork], &i__1, &
1407 /* Multiply Q in A by left singular vectors of R in */
1408 /* WORK(IU), storing result in WORK(IR) and copying to A */
1409 /* Workspace: need N*N [R] + 3*N [e, tauq, taup] + N*N [U] */
1410 /* Workspace: prefer M*N [R] + 3*N [e, tauq, taup] + N*N [U] */
1414 for (i__ = 1; i__2 < 0 ? i__ >= i__1 : i__ <= i__1; i__ +=
1417 i__3 = *m - i__ + 1;
1418 chunk = f2cmin(i__3,ldwrkr);
1419 dgemm_("N", "N", &chunk, n, n, &c_b84, &a[i__ + a_dim1],
1420 lda, &work[iu], n, &c_b63, &work[ir], &ldwrkr);
1421 dlacpy_("F", &chunk, n, &work[ir], &ldwrkr, &a[i__ +
1428 /* Path 3 (M >> N, JOBZ='S') */
1429 /* N left singular vectors to be computed in U and */
1430 /* N right singular vectors to be computed in VT */
1434 /* WORK(IR) is N by N */
1437 itau = ir + ldwrkr * *n;
1441 /* Workspace: need N*N [R] + N [tau] + N [work] */
1442 /* Workspace: prefer N*N [R] + N [tau] + N*NB [work] */
1444 i__2 = *lwork - nwork + 1;
1445 dgeqrf_(m, n, &a[a_offset], lda, &work[itau], &work[nwork], &
1448 /* Copy R to WORK(IR), zeroing out below it */
1450 dlacpy_("U", n, n, &a[a_offset], lda, &work[ir], &ldwrkr);
1453 dlaset_("L", &i__2, &i__1, &c_b63, &c_b63, &work[ir + 1], &
1456 /* Generate Q in A */
1457 /* Workspace: need N*N [R] + N [tau] + N [work] */
1458 /* Workspace: prefer N*N [R] + N [tau] + N*NB [work] */
1460 i__2 = *lwork - nwork + 1;
1461 dorgqr_(m, n, n, &a[a_offset], lda, &work[itau], &work[nwork],
1468 /* Bidiagonalize R in WORK(IR) */
1469 /* Workspace: need N*N [R] + 3*N [e, tauq, taup] + N [work] */
1470 /* Workspace: prefer N*N [R] + 3*N [e, tauq, taup] + 2*N*NB [work] */
1472 i__2 = *lwork - nwork + 1;
1473 dgebrd_(n, n, &work[ir], &ldwrkr, &s[1], &work[ie], &work[
1474 itauq], &work[itaup], &work[nwork], &i__2, &ierr);
1476 /* Perform bidiagonal SVD, computing left singular vectors */
1477 /* of bidiagoal matrix in U and computing right singular */
1478 /* vectors of bidiagonal matrix in VT */
1479 /* Workspace: need N*N [R] + 3*N [e, tauq, taup] + BDSPAC */
1481 dbdsdc_("U", "I", n, &s[1], &work[ie], &u[u_offset], ldu, &vt[
1482 vt_offset], ldvt, dum, idum, &work[nwork], &iwork[1],
1485 /* Overwrite U by left singular vectors of R and VT */
1486 /* by right singular vectors of R */
1487 /* Workspace: need N*N [R] + 3*N [e, tauq, taup] + N [work] */
1488 /* Workspace: prefer N*N [R] + 3*N [e, tauq, taup] + N*NB [work] */
1490 i__2 = *lwork - nwork + 1;
1491 dormbr_("Q", "L", "N", n, n, n, &work[ir], &ldwrkr, &work[
1492 itauq], &u[u_offset], ldu, &work[nwork], &i__2, &ierr);
1494 i__2 = *lwork - nwork + 1;
1495 dormbr_("P", "R", "T", n, n, n, &work[ir], &ldwrkr, &work[
1496 itaup], &vt[vt_offset], ldvt, &work[nwork], &i__2, &
1499 /* Multiply Q in A by left singular vectors of R in */
1500 /* WORK(IR), storing result in U */
1501 /* Workspace: need N*N [R] */
1503 dlacpy_("F", n, n, &u[u_offset], ldu, &work[ir], &ldwrkr);
1504 dgemm_("N", "N", m, n, n, &c_b84, &a[a_offset], lda, &work[ir]
1505 , &ldwrkr, &c_b63, &u[u_offset], ldu);
1509 /* Path 4 (M >> N, JOBZ='A') */
1510 /* M left singular vectors to be computed in U and */
1511 /* N right singular vectors to be computed in VT */
1515 /* WORK(IU) is N by N */
1518 itau = iu + ldwrku * *n;
1521 /* Compute A=Q*R, copying result to U */
1522 /* Workspace: need N*N [U] + N [tau] + N [work] */
1523 /* Workspace: prefer N*N [U] + N [tau] + N*NB [work] */
1525 i__2 = *lwork - nwork + 1;
1526 dgeqrf_(m, n, &a[a_offset], lda, &work[itau], &work[nwork], &
1528 dlacpy_("L", m, n, &a[a_offset], lda, &u[u_offset], ldu);
1530 /* Generate Q in U */
1531 /* Workspace: need N*N [U] + N [tau] + M [work] */
1532 /* Workspace: prefer N*N [U] + N [tau] + M*NB [work] */
1533 i__2 = *lwork - nwork + 1;
1534 dorgqr_(m, m, n, &u[u_offset], ldu, &work[itau], &work[nwork],
1537 /* Produce R in A, zeroing out other entries */
1541 dlaset_("L", &i__2, &i__1, &c_b63, &c_b63, &a[a_dim1 + 2],
1548 /* Bidiagonalize R in A */
1549 /* Workspace: need N*N [U] + 3*N [e, tauq, taup] + N [work] */
1550 /* Workspace: prefer N*N [U] + 3*N [e, tauq, taup] + 2*N*NB [work] */
1552 i__2 = *lwork - nwork + 1;
1553 dgebrd_(n, n, &a[a_offset], lda, &s[1], &work[ie], &work[
1554 itauq], &work[itaup], &work[nwork], &i__2, &ierr);
1556 /* Perform bidiagonal SVD, computing left singular vectors */
1557 /* of bidiagonal matrix in WORK(IU) and computing right */
1558 /* singular vectors of bidiagonal matrix in VT */
1559 /* Workspace: need N*N [U] + 3*N [e, tauq, taup] + BDSPAC */
1561 dbdsdc_("U", "I", n, &s[1], &work[ie], &work[iu], n, &vt[
1562 vt_offset], ldvt, dum, idum, &work[nwork], &iwork[1],
1565 /* Overwrite WORK(IU) by left singular vectors of R and VT */
1566 /* by right singular vectors of R */
1567 /* Workspace: need N*N [U] + 3*N [e, tauq, taup] + N [work] */
1568 /* Workspace: prefer N*N [U] + 3*N [e, tauq, taup] + N*NB [work] */
1570 i__2 = *lwork - nwork + 1;
1571 dormbr_("Q", "L", "N", n, n, n, &a[a_offset], lda, &work[
1572 itauq], &work[iu], &ldwrku, &work[nwork], &i__2, &
1574 i__2 = *lwork - nwork + 1;
1575 dormbr_("P", "R", "T", n, n, n, &a[a_offset], lda, &work[
1576 itaup], &vt[vt_offset], ldvt, &work[nwork], &i__2, &
1579 /* Multiply Q in U by left singular vectors of R in */
1580 /* WORK(IU), storing result in A */
1581 /* Workspace: need N*N [U] */
1583 dgemm_("N", "N", m, n, n, &c_b84, &u[u_offset], ldu, &work[iu]
1584 , &ldwrku, &c_b63, &a[a_offset], lda);
1586 /* Copy left singular vectors of A from A to U */
1588 dlacpy_("F", m, n, &a[a_offset], lda, &u[u_offset], ldu);
1596 /* Path 5 (M >= N, but not much larger) */
1597 /* Reduce to bidiagonal form without QR decomposition */
1604 /* Bidiagonalize A */
1605 /* Workspace: need 3*N [e, tauq, taup] + M [work] */
1606 /* Workspace: prefer 3*N [e, tauq, taup] + (M+N)*NB [work] */
1608 i__2 = *lwork - nwork + 1;
1609 dgebrd_(m, n, &a[a_offset], lda, &s[1], &work[ie], &work[itauq], &
1610 work[itaup], &work[nwork], &i__2, &ierr);
1613 /* Path 5n (M >= N, JOBZ='N') */
1614 /* Perform bidiagonal SVD, only computing singular values */
1615 /* Workspace: need 3*N [e, tauq, taup] + BDSPAC */
1617 dbdsdc_("U", "N", n, &s[1], &work[ie], dum, &c__1, dum, &c__1,
1618 dum, idum, &work[nwork], &iwork[1], info);
1620 /* Path 5o (M >= N, JOBZ='O') */
1622 if (*lwork >= *m * *n + *n * 3 + bdspac) {
1624 /* WORK( IU ) is M by N */
1627 nwork = iu + ldwrku * *n;
1628 dlaset_("F", m, n, &c_b63, &c_b63, &work[iu], &ldwrku);
1629 /* IR is unused; silence compile warnings */
1633 /* WORK( IU ) is N by N */
1636 nwork = iu + ldwrku * *n;
1638 /* WORK(IR) is LDWRKR by N */
1641 ldwrkr = (*lwork - *n * *n - *n * 3) / *n;
1643 nwork = iu + ldwrku * *n;
1645 /* Perform bidiagonal SVD, computing left singular vectors */
1646 /* of bidiagonal matrix in WORK(IU) and computing right */
1647 /* singular vectors of bidiagonal matrix in VT */
1648 /* Workspace: need 3*N [e, tauq, taup] + N*N [U] + BDSPAC */
1650 dbdsdc_("U", "I", n, &s[1], &work[ie], &work[iu], &ldwrku, &
1651 vt[vt_offset], ldvt, dum, idum, &work[nwork], &iwork[
1654 /* Overwrite VT by right singular vectors of A */
1655 /* Workspace: need 3*N [e, tauq, taup] + N*N [U] + N [work] */
1656 /* Workspace: prefer 3*N [e, tauq, taup] + N*N [U] + N*NB [work] */
1658 i__2 = *lwork - nwork + 1;
1659 dormbr_("P", "R", "T", n, n, n, &a[a_offset], lda, &work[
1660 itaup], &vt[vt_offset], ldvt, &work[nwork], &i__2, &
1663 if (*lwork >= *m * *n + *n * 3 + bdspac) {
1666 /* Overwrite WORK(IU) by left singular vectors of A */
1667 /* Workspace: need 3*N [e, tauq, taup] + M*N [U] + N [work] */
1668 /* Workspace: prefer 3*N [e, tauq, taup] + M*N [U] + N*NB [work] */
1670 i__2 = *lwork - nwork + 1;
1671 dormbr_("Q", "L", "N", m, n, n, &a[a_offset], lda, &work[
1672 itauq], &work[iu], &ldwrku, &work[nwork], &i__2, &
1675 /* Copy left singular vectors of A from WORK(IU) to A */
1677 dlacpy_("F", m, n, &work[iu], &ldwrku, &a[a_offset], lda);
1681 /* Generate Q in A */
1682 /* Workspace: need 3*N [e, tauq, taup] + N*N [U] + N [work] */
1683 /* Workspace: prefer 3*N [e, tauq, taup] + N*N [U] + N*NB [work] */
1685 i__2 = *lwork - nwork + 1;
1686 dorgbr_("Q", m, n, n, &a[a_offset], lda, &work[itauq], &
1687 work[nwork], &i__2, &ierr);
1689 /* Multiply Q in A by left singular vectors of */
1690 /* bidiagonal matrix in WORK(IU), storing result in */
1691 /* WORK(IR) and copying to A */
1692 /* Workspace: need 3*N [e, tauq, taup] + N*N [U] + NB*N [R] */
1693 /* Workspace: prefer 3*N [e, tauq, taup] + N*N [U] + M*N [R] */
1697 for (i__ = 1; i__1 < 0 ? i__ >= i__2 : i__ <= i__2; i__ +=
1700 i__3 = *m - i__ + 1;
1701 chunk = f2cmin(i__3,ldwrkr);
1702 dgemm_("N", "N", &chunk, n, n, &c_b84, &a[i__ +
1703 a_dim1], lda, &work[iu], &ldwrku, &c_b63, &
1705 dlacpy_("F", &chunk, n, &work[ir], &ldwrkr, &a[i__ +
1713 /* Path 5s (M >= N, JOBZ='S') */
1714 /* Perform bidiagonal SVD, computing left singular vectors */
1715 /* of bidiagonal matrix in U and computing right singular */
1716 /* vectors of bidiagonal matrix in VT */
1717 /* Workspace: need 3*N [e, tauq, taup] + BDSPAC */
1719 dlaset_("F", m, n, &c_b63, &c_b63, &u[u_offset], ldu);
1720 dbdsdc_("U", "I", n, &s[1], &work[ie], &u[u_offset], ldu, &vt[
1721 vt_offset], ldvt, dum, idum, &work[nwork], &iwork[1],
1724 /* Overwrite U by left singular vectors of A and VT */
1725 /* by right singular vectors of A */
1726 /* Workspace: need 3*N [e, tauq, taup] + N [work] */
1727 /* Workspace: prefer 3*N [e, tauq, taup] + N*NB [work] */
1729 i__1 = *lwork - nwork + 1;
1730 dormbr_("Q", "L", "N", m, n, n, &a[a_offset], lda, &work[
1731 itauq], &u[u_offset], ldu, &work[nwork], &i__1, &ierr);
1732 i__1 = *lwork - nwork + 1;
1733 dormbr_("P", "R", "T", n, n, n, &a[a_offset], lda, &work[
1734 itaup], &vt[vt_offset], ldvt, &work[nwork], &i__1, &
1738 /* Path 5a (M >= N, JOBZ='A') */
1739 /* Perform bidiagonal SVD, computing left singular vectors */
1740 /* of bidiagonal matrix in U and computing right singular */
1741 /* vectors of bidiagonal matrix in VT */
1742 /* Workspace: need 3*N [e, tauq, taup] + BDSPAC */
1744 dlaset_("F", m, m, &c_b63, &c_b63, &u[u_offset], ldu);
1745 dbdsdc_("U", "I", n, &s[1], &work[ie], &u[u_offset], ldu, &vt[
1746 vt_offset], ldvt, dum, idum, &work[nwork], &iwork[1],
1749 /* Set the right corner of U to identity matrix */
1754 dlaset_("F", &i__1, &i__2, &c_b63, &c_b84, &u[*n + 1 + (*
1755 n + 1) * u_dim1], ldu);
1758 /* Overwrite U by left singular vectors of A and VT */
1759 /* by right singular vectors of A */
1760 /* Workspace: need 3*N [e, tauq, taup] + M [work] */
1761 /* Workspace: prefer 3*N [e, tauq, taup] + M*NB [work] */
1763 i__1 = *lwork - nwork + 1;
1764 dormbr_("Q", "L", "N", m, m, n, &a[a_offset], lda, &work[
1765 itauq], &u[u_offset], ldu, &work[nwork], &i__1, &ierr);
1766 i__1 = *lwork - nwork + 1;
1767 dormbr_("P", "R", "T", n, n, m, &a[a_offset], lda, &work[
1768 itaup], &vt[vt_offset], ldvt, &work[nwork], &i__1, &
1776 /* A has more columns than rows. If A has sufficiently more */
1777 /* columns than rows, first reduce using the LQ decomposition (if */
1778 /* sufficient workspace available) */
1784 /* Path 1t (N >> M, JOBZ='N') */
1785 /* No singular vectors to be computed */
1791 /* Workspace: need M [tau] + M [work] */
1792 /* Workspace: prefer M [tau] + M*NB [work] */
1794 i__1 = *lwork - nwork + 1;
1795 dgelqf_(m, n, &a[a_offset], lda, &work[itau], &work[nwork], &
1798 /* Zero out above L */
1802 dlaset_("U", &i__1, &i__2, &c_b63, &c_b63, &a[(a_dim1 << 1) +
1809 /* Bidiagonalize L in A */
1810 /* Workspace: need 3*M [e, tauq, taup] + M [work] */
1811 /* Workspace: prefer 3*M [e, tauq, taup] + 2*M*NB [work] */
1813 i__1 = *lwork - nwork + 1;
1814 dgebrd_(m, m, &a[a_offset], lda, &s[1], &work[ie], &work[
1815 itauq], &work[itaup], &work[nwork], &i__1, &ierr);
1818 /* Perform bidiagonal SVD, computing singular values only */
1819 /* Workspace: need M [e] + BDSPAC */
1821 dbdsdc_("U", "N", m, &s[1], &work[ie], dum, &c__1, dum, &c__1,
1822 dum, idum, &work[nwork], &iwork[1], info);
1826 /* Path 2t (N >> M, JOBZ='O') */
1827 /* M right singular vectors to be overwritten on A and */
1828 /* M left singular vectors to be computed in U */
1832 /* WORK(IVT) is M by M */
1833 /* WORK(IL) is M by M; it is later resized to M by chunk for gemm */
1836 if (*lwork >= *m * *n + *m * *m + *m * 3 + bdspac) {
1841 chunk = (*lwork - *m * *m) / *m;
1843 itau = il + ldwrkl * *m;
1847 /* Workspace: need M*M [VT] + M*M [L] + M [tau] + M [work] */
1848 /* Workspace: prefer M*M [VT] + M*M [L] + M [tau] + M*NB [work] */
1850 i__1 = *lwork - nwork + 1;
1851 dgelqf_(m, n, &a[a_offset], lda, &work[itau], &work[nwork], &
1854 /* Copy L to WORK(IL), zeroing about above it */
1856 dlacpy_("L", m, m, &a[a_offset], lda, &work[il], &ldwrkl);
1859 dlaset_("U", &i__1, &i__2, &c_b63, &c_b63, &work[il + ldwrkl],
1862 /* Generate Q in A */
1863 /* Workspace: need M*M [VT] + M*M [L] + M [tau] + M [work] */
1864 /* Workspace: prefer M*M [VT] + M*M [L] + M [tau] + M*NB [work] */
1866 i__1 = *lwork - nwork + 1;
1867 dorglq_(m, n, m, &a[a_offset], lda, &work[itau], &work[nwork],
1874 /* Bidiagonalize L in WORK(IL) */
1875 /* Workspace: need M*M [VT] + M*M [L] + 3*M [e, tauq, taup] + M [work] */
1876 /* Workspace: prefer M*M [VT] + M*M [L] + 3*M [e, tauq, taup] + 2*M*NB [work] */
1878 i__1 = *lwork - nwork + 1;
1879 dgebrd_(m, m, &work[il], &ldwrkl, &s[1], &work[ie], &work[
1880 itauq], &work[itaup], &work[nwork], &i__1, &ierr);
1882 /* Perform bidiagonal SVD, computing left singular vectors */
1883 /* of bidiagonal matrix in U, and computing right singular */
1884 /* vectors of bidiagonal matrix in WORK(IVT) */
1885 /* Workspace: need M*M [VT] + M*M [L] + 3*M [e, tauq, taup] + BDSPAC */
1887 dbdsdc_("U", "I", m, &s[1], &work[ie], &u[u_offset], ldu, &
1888 work[ivt], m, dum, idum, &work[nwork], &iwork[1],
1891 /* Overwrite U by left singular vectors of L and WORK(IVT) */
1892 /* by right singular vectors of L */
1893 /* Workspace: need M*M [VT] + M*M [L] + 3*M [e, tauq, taup] + M [work] */
1894 /* Workspace: prefer M*M [VT] + M*M [L] + 3*M [e, tauq, taup] + M*NB [work] */
1896 i__1 = *lwork - nwork + 1;
1897 dormbr_("Q", "L", "N", m, m, m, &work[il], &ldwrkl, &work[
1898 itauq], &u[u_offset], ldu, &work[nwork], &i__1, &ierr);
1899 i__1 = *lwork - nwork + 1;
1900 dormbr_("P", "R", "T", m, m, m, &work[il], &ldwrkl, &work[
1901 itaup], &work[ivt], m, &work[nwork], &i__1, &ierr);
1903 /* Multiply right singular vectors of L in WORK(IVT) by Q */
1904 /* in A, storing result in WORK(IL) and copying to A */
1905 /* Workspace: need M*M [VT] + M*M [L] */
1906 /* Workspace: prefer M*M [VT] + M*N [L] */
1907 /* At this point, L is resized as M by chunk. */
1911 for (i__ = 1; i__2 < 0 ? i__ >= i__1 : i__ <= i__1; i__ +=
1914 i__3 = *n - i__ + 1;
1915 blk = f2cmin(i__3,chunk);
1916 dgemm_("N", "N", m, &blk, m, &c_b84, &work[ivt], m, &a[
1917 i__ * a_dim1 + 1], lda, &c_b63, &work[il], &
1919 dlacpy_("F", m, &blk, &work[il], &ldwrkl, &a[i__ * a_dim1
1926 /* Path 3t (N >> M, JOBZ='S') */
1927 /* M right singular vectors to be computed in VT and */
1928 /* M left singular vectors to be computed in U */
1932 /* WORK(IL) is M by M */
1935 itau = il + ldwrkl * *m;
1939 /* Workspace: need M*M [L] + M [tau] + M [work] */
1940 /* Workspace: prefer M*M [L] + M [tau] + M*NB [work] */
1942 i__2 = *lwork - nwork + 1;
1943 dgelqf_(m, n, &a[a_offset], lda, &work[itau], &work[nwork], &
1946 /* Copy L to WORK(IL), zeroing out above it */
1948 dlacpy_("L", m, m, &a[a_offset], lda, &work[il], &ldwrkl);
1951 dlaset_("U", &i__2, &i__1, &c_b63, &c_b63, &work[il + ldwrkl],
1954 /* Generate Q in A */
1955 /* Workspace: need M*M [L] + M [tau] + M [work] */
1956 /* Workspace: prefer M*M [L] + M [tau] + M*NB [work] */
1958 i__2 = *lwork - nwork + 1;
1959 dorglq_(m, n, m, &a[a_offset], lda, &work[itau], &work[nwork],
1966 /* Bidiagonalize L in WORK(IU). */
1967 /* Workspace: need M*M [L] + 3*M [e, tauq, taup] + M [work] */
1968 /* Workspace: prefer M*M [L] + 3*M [e, tauq, taup] + 2*M*NB [work] */
1970 i__2 = *lwork - nwork + 1;
1971 dgebrd_(m, m, &work[il], &ldwrkl, &s[1], &work[ie], &work[
1972 itauq], &work[itaup], &work[nwork], &i__2, &ierr);
1974 /* Perform bidiagonal SVD, computing left singular vectors */
1975 /* of bidiagonal matrix in U and computing right singular */
1976 /* vectors of bidiagonal matrix in VT */
1977 /* Workspace: need M*M [L] + 3*M [e, tauq, taup] + BDSPAC */
1979 dbdsdc_("U", "I", m, &s[1], &work[ie], &u[u_offset], ldu, &vt[
1980 vt_offset], ldvt, dum, idum, &work[nwork], &iwork[1],
1983 /* Overwrite U by left singular vectors of L and VT */
1984 /* by right singular vectors of L */
1985 /* Workspace: need M*M [L] + 3*M [e, tauq, taup] + M [work] */
1986 /* Workspace: prefer M*M [L] + 3*M [e, tauq, taup] + M*NB [work] */
1988 i__2 = *lwork - nwork + 1;
1989 dormbr_("Q", "L", "N", m, m, m, &work[il], &ldwrkl, &work[
1990 itauq], &u[u_offset], ldu, &work[nwork], &i__2, &ierr);
1991 i__2 = *lwork - nwork + 1;
1992 dormbr_("P", "R", "T", m, m, m, &work[il], &ldwrkl, &work[
1993 itaup], &vt[vt_offset], ldvt, &work[nwork], &i__2, &
1996 /* Multiply right singular vectors of L in WORK(IL) by */
1997 /* Q in A, storing result in VT */
1998 /* Workspace: need M*M [L] */
2000 dlacpy_("F", m, m, &vt[vt_offset], ldvt, &work[il], &ldwrkl);
2001 dgemm_("N", "N", m, n, m, &c_b84, &work[il], &ldwrkl, &a[
2002 a_offset], lda, &c_b63, &vt[vt_offset], ldvt);
2006 /* Path 4t (N >> M, JOBZ='A') */
2007 /* N right singular vectors to be computed in VT and */
2008 /* M left singular vectors to be computed in U */
2012 /* WORK(IVT) is M by M */
2015 itau = ivt + ldwkvt * *m;
2018 /* Compute A=L*Q, copying result to VT */
2019 /* Workspace: need M*M [VT] + M [tau] + M [work] */
2020 /* Workspace: prefer M*M [VT] + M [tau] + M*NB [work] */
2022 i__2 = *lwork - nwork + 1;
2023 dgelqf_(m, n, &a[a_offset], lda, &work[itau], &work[nwork], &
2025 dlacpy_("U", m, n, &a[a_offset], lda, &vt[vt_offset], ldvt);
2027 /* Generate Q in VT */
2028 /* Workspace: need M*M [VT] + M [tau] + N [work] */
2029 /* Workspace: prefer M*M [VT] + M [tau] + N*NB [work] */
2031 i__2 = *lwork - nwork + 1;
2032 dorglq_(n, n, m, &vt[vt_offset], ldvt, &work[itau], &work[
2033 nwork], &i__2, &ierr);
2035 /* Produce L in A, zeroing out other entries */
2039 dlaset_("U", &i__2, &i__1, &c_b63, &c_b63, &a[(a_dim1 << 1) +
2046 /* Bidiagonalize L in A */
2047 /* Workspace: need M*M [VT] + 3*M [e, tauq, taup] + M [work] */
2048 /* Workspace: prefer M*M [VT] + 3*M [e, tauq, taup] + 2*M*NB [work] */
2050 i__2 = *lwork - nwork + 1;
2051 dgebrd_(m, m, &a[a_offset], lda, &s[1], &work[ie], &work[
2052 itauq], &work[itaup], &work[nwork], &i__2, &ierr);
2054 /* Perform bidiagonal SVD, computing left singular vectors */
2055 /* of bidiagonal matrix in U and computing right singular */
2056 /* vectors of bidiagonal matrix in WORK(IVT) */
2057 /* Workspace: need M*M [VT] + 3*M [e, tauq, taup] + BDSPAC */
2059 dbdsdc_("U", "I", m, &s[1], &work[ie], &u[u_offset], ldu, &
2060 work[ivt], &ldwkvt, dum, idum, &work[nwork], &iwork[1]
2063 /* Overwrite U by left singular vectors of L and WORK(IVT) */
2064 /* by right singular vectors of L */
2065 /* Workspace: need M*M [VT] + 3*M [e, tauq, taup]+ M [work] */
2066 /* Workspace: prefer M*M [VT] + 3*M [e, tauq, taup]+ M*NB [work] */
2068 i__2 = *lwork - nwork + 1;
2069 dormbr_("Q", "L", "N", m, m, m, &a[a_offset], lda, &work[
2070 itauq], &u[u_offset], ldu, &work[nwork], &i__2, &ierr);
2071 i__2 = *lwork - nwork + 1;
2072 dormbr_("P", "R", "T", m, m, m, &a[a_offset], lda, &work[
2073 itaup], &work[ivt], &ldwkvt, &work[nwork], &i__2, &
2076 /* Multiply right singular vectors of L in WORK(IVT) by */
2077 /* Q in VT, storing result in A */
2078 /* Workspace: need M*M [VT] */
2080 dgemm_("N", "N", m, n, m, &c_b84, &work[ivt], &ldwkvt, &vt[
2081 vt_offset], ldvt, &c_b63, &a[a_offset], lda);
2083 /* Copy right singular vectors of A from A to VT */
2085 dlacpy_("F", m, n, &a[a_offset], lda, &vt[vt_offset], ldvt);
2093 /* Path 5t (N > M, but not much larger) */
2094 /* Reduce to bidiagonal form without LQ decomposition */
2101 /* Bidiagonalize A */
2102 /* Workspace: need 3*M [e, tauq, taup] + N [work] */
2103 /* Workspace: prefer 3*M [e, tauq, taup] + (M+N)*NB [work] */
2105 i__2 = *lwork - nwork + 1;
2106 dgebrd_(m, n, &a[a_offset], lda, &s[1], &work[ie], &work[itauq], &
2107 work[itaup], &work[nwork], &i__2, &ierr);
2110 /* Path 5tn (N > M, JOBZ='N') */
2111 /* Perform bidiagonal SVD, only computing singular values */
2112 /* Workspace: need 3*M [e, tauq, taup] + BDSPAC */
2114 dbdsdc_("L", "N", m, &s[1], &work[ie], dum, &c__1, dum, &c__1,
2115 dum, idum, &work[nwork], &iwork[1], info);
2117 /* Path 5to (N > M, JOBZ='O') */
2120 if (*lwork >= *m * *n + *m * 3 + bdspac) {
2122 /* WORK( IVT ) is M by N */
2124 dlaset_("F", m, n, &c_b63, &c_b63, &work[ivt], &ldwkvt);
2125 nwork = ivt + ldwkvt * *n;
2126 /* IL is unused; silence compile warnings */
2130 /* WORK( IVT ) is M by M */
2132 nwork = ivt + ldwkvt * *m;
2135 /* WORK(IL) is M by CHUNK */
2137 chunk = (*lwork - *m * *m - *m * 3) / *m;
2140 /* Perform bidiagonal SVD, computing left singular vectors */
2141 /* of bidiagonal matrix in U and computing right singular */
2142 /* vectors of bidiagonal matrix in WORK(IVT) */
2143 /* Workspace: need 3*M [e, tauq, taup] + M*M [VT] + BDSPAC */
2145 dbdsdc_("L", "I", m, &s[1], &work[ie], &u[u_offset], ldu, &
2146 work[ivt], &ldwkvt, dum, idum, &work[nwork], &iwork[1]
2149 /* Overwrite U by left singular vectors of A */
2150 /* Workspace: need 3*M [e, tauq, taup] + M*M [VT] + M [work] */
2151 /* Workspace: prefer 3*M [e, tauq, taup] + M*M [VT] + M*NB [work] */
2153 i__2 = *lwork - nwork + 1;
2154 dormbr_("Q", "L", "N", m, m, n, &a[a_offset], lda, &work[
2155 itauq], &u[u_offset], ldu, &work[nwork], &i__2, &ierr);
2157 if (*lwork >= *m * *n + *m * 3 + bdspac) {
2160 /* Overwrite WORK(IVT) by left singular vectors of A */
2161 /* Workspace: need 3*M [e, tauq, taup] + M*N [VT] + M [work] */
2162 /* Workspace: prefer 3*M [e, tauq, taup] + M*N [VT] + M*NB [work] */
2164 i__2 = *lwork - nwork + 1;
2165 dormbr_("P", "R", "T", m, n, m, &a[a_offset], lda, &work[
2166 itaup], &work[ivt], &ldwkvt, &work[nwork], &i__2,
2169 /* Copy right singular vectors of A from WORK(IVT) to A */
2171 dlacpy_("F", m, n, &work[ivt], &ldwkvt, &a[a_offset], lda);
2175 /* Generate P**T in A */
2176 /* Workspace: need 3*M [e, tauq, taup] + M*M [VT] + M [work] */
2177 /* Workspace: prefer 3*M [e, tauq, taup] + M*M [VT] + M*NB [work] */
2179 i__2 = *lwork - nwork + 1;
2180 dorgbr_("P", m, n, m, &a[a_offset], lda, &work[itaup], &
2181 work[nwork], &i__2, &ierr);
2183 /* Multiply Q in A by right singular vectors of */
2184 /* bidiagonal matrix in WORK(IVT), storing result in */
2185 /* WORK(IL) and copying to A */
2186 /* Workspace: need 3*M [e, tauq, taup] + M*M [VT] + M*NB [L] */
2187 /* Workspace: prefer 3*M [e, tauq, taup] + M*M [VT] + M*N [L] */
2191 for (i__ = 1; i__1 < 0 ? i__ >= i__2 : i__ <= i__2; i__ +=
2194 i__3 = *n - i__ + 1;
2195 blk = f2cmin(i__3,chunk);
2196 dgemm_("N", "N", m, &blk, m, &c_b84, &work[ivt], &
2197 ldwkvt, &a[i__ * a_dim1 + 1], lda, &c_b63, &
2199 dlacpy_("F", m, &blk, &work[il], m, &a[i__ * a_dim1 +
2206 /* Path 5ts (N > M, JOBZ='S') */
2207 /* Perform bidiagonal SVD, computing left singular vectors */
2208 /* of bidiagonal matrix in U and computing right singular */
2209 /* vectors of bidiagonal matrix in VT */
2210 /* Workspace: need 3*M [e, tauq, taup] + BDSPAC */
2212 dlaset_("F", m, n, &c_b63, &c_b63, &vt[vt_offset], ldvt);
2213 dbdsdc_("L", "I", m, &s[1], &work[ie], &u[u_offset], ldu, &vt[
2214 vt_offset], ldvt, dum, idum, &work[nwork], &iwork[1],
2217 /* Overwrite U by left singular vectors of A and VT */
2218 /* by right singular vectors of A */
2219 /* Workspace: need 3*M [e, tauq, taup] + M [work] */
2220 /* Workspace: prefer 3*M [e, tauq, taup] + M*NB [work] */
2222 i__1 = *lwork - nwork + 1;
2223 dormbr_("Q", "L", "N", m, m, n, &a[a_offset], lda, &work[
2224 itauq], &u[u_offset], ldu, &work[nwork], &i__1, &ierr);
2225 i__1 = *lwork - nwork + 1;
2226 dormbr_("P", "R", "T", m, n, m, &a[a_offset], lda, &work[
2227 itaup], &vt[vt_offset], ldvt, &work[nwork], &i__1, &
2231 /* Path 5ta (N > M, JOBZ='A') */
2232 /* Perform bidiagonal SVD, computing left singular vectors */
2233 /* of bidiagonal matrix in U and computing right singular */
2234 /* vectors of bidiagonal matrix in VT */
2235 /* Workspace: need 3*M [e, tauq, taup] + BDSPAC */
2237 dlaset_("F", n, n, &c_b63, &c_b63, &vt[vt_offset], ldvt);
2238 dbdsdc_("L", "I", m, &s[1], &work[ie], &u[u_offset], ldu, &vt[
2239 vt_offset], ldvt, dum, idum, &work[nwork], &iwork[1],
2242 /* Set the right corner of VT to identity matrix */
2247 dlaset_("F", &i__1, &i__2, &c_b63, &c_b84, &vt[*m + 1 + (*
2248 m + 1) * vt_dim1], ldvt);
2251 /* Overwrite U by left singular vectors of A and VT */
2252 /* by right singular vectors of A */
2253 /* Workspace: need 3*M [e, tauq, taup] + N [work] */
2254 /* Workspace: prefer 3*M [e, tauq, taup] + N*NB [work] */
2256 i__1 = *lwork - nwork + 1;
2257 dormbr_("Q", "L", "N", m, m, n, &a[a_offset], lda, &work[
2258 itauq], &u[u_offset], ldu, &work[nwork], &i__1, &ierr);
2259 i__1 = *lwork - nwork + 1;
2260 dormbr_("P", "R", "T", n, n, m, &a[a_offset], lda, &work[
2261 itaup], &vt[vt_offset], ldvt, &work[nwork], &i__1, &
2269 /* Undo scaling if necessary */
2272 if (anrm > bignum) {
2273 dlascl_("G", &c__0, &c__0, &bignum, &anrm, &minmn, &c__1, &s[1], &
2276 if (anrm < smlnum) {
2277 dlascl_("G", &c__0, &c__0, &smlnum, &anrm, &minmn, &c__1, &s[1], &
2282 /* Return optimal workspace in WORK(1) */
2284 work[1] = (doublereal) maxwrk;