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;
517 /* > \brief <b> DGBSVX computes the solution to system of linear equations A * X = B for GB matrices</b> */
519 /* =========== DOCUMENTATION =========== */
521 /* Online html documentation available at */
522 /* http://www.netlib.org/lapack/explore-html/ */
525 /* > Download DGBSVX + dependencies */
526 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dgbsvx.
529 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dgbsvx.
532 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dgbsvx.
540 /* SUBROUTINE DGBSVX( FACT, TRANS, N, KL, KU, NRHS, AB, LDAB, AFB, */
541 /* LDAFB, IPIV, EQUED, R, C, B, LDB, X, LDX, */
542 /* RCOND, FERR, BERR, WORK, IWORK, INFO ) */
544 /* CHARACTER EQUED, FACT, TRANS */
545 /* INTEGER INFO, KL, KU, LDAB, LDAFB, LDB, LDX, N, NRHS */
546 /* DOUBLE PRECISION RCOND */
547 /* INTEGER IPIV( * ), IWORK( * ) */
548 /* DOUBLE PRECISION AB( LDAB, * ), AFB( LDAFB, * ), B( LDB, * ), */
549 /* $ BERR( * ), C( * ), FERR( * ), R( * ), */
550 /* $ WORK( * ), X( LDX, * ) */
553 /* > \par Purpose: */
558 /* > DGBSVX uses the LU factorization to compute the solution to a real */
559 /* > system of linear equations A * X = B, A**T * X = B, or A**H * X = B, */
560 /* > where A is a band matrix of order N with KL subdiagonals and KU */
561 /* > superdiagonals, and X and B are N-by-NRHS matrices. */
563 /* > Error bounds on the solution and a condition estimate are also */
567 /* > \par Description: */
568 /* ================= */
572 /* > The following steps are performed by this subroutine: */
574 /* > 1. If FACT = 'E', real scaling factors are computed to equilibrate */
576 /* > TRANS = 'N': diag(R)*A*diag(C) *inv(diag(C))*X = diag(R)*B */
577 /* > TRANS = 'T': (diag(R)*A*diag(C))**T *inv(diag(R))*X = diag(C)*B */
578 /* > TRANS = 'C': (diag(R)*A*diag(C))**H *inv(diag(R))*X = diag(C)*B */
579 /* > Whether or not the system will be equilibrated depends on the */
580 /* > scaling of the matrix A, but if equilibration is used, A is */
581 /* > overwritten by diag(R)*A*diag(C) and B by diag(R)*B (if TRANS='N') */
582 /* > or diag(C)*B (if TRANS = 'T' or 'C'). */
584 /* > 2. If FACT = 'N' or 'E', the LU decomposition is used to factor the */
585 /* > matrix A (after equilibration if FACT = 'E') as */
587 /* > where L is a product of permutation and unit lower triangular */
588 /* > matrices with KL subdiagonals, and U is upper triangular with */
589 /* > KL+KU superdiagonals. */
591 /* > 3. If some U(i,i)=0, so that U is exactly singular, then the routine */
592 /* > returns with INFO = i. Otherwise, the factored form of A is used */
593 /* > to estimate the condition number of the matrix A. If the */
594 /* > reciprocal of the condition number is less than machine precision, */
595 /* > INFO = N+1 is returned as a warning, but the routine still goes on */
596 /* > to solve for X and compute error bounds as described below. */
598 /* > 4. The system of equations is solved for X using the factored form */
601 /* > 5. Iterative refinement is applied to improve the computed solution */
602 /* > matrix and calculate error bounds and backward error estimates */
605 /* > 6. If equilibration was used, the matrix X is premultiplied by */
606 /* > diag(C) (if TRANS = 'N') or diag(R) (if TRANS = 'T' or 'C') so */
607 /* > that it solves the original system before equilibration. */
613 /* > \param[in] FACT */
615 /* > FACT is CHARACTER*1 */
616 /* > Specifies whether or not the factored form of the matrix A is */
617 /* > supplied on entry, and if not, whether the matrix A should be */
618 /* > equilibrated before it is factored. */
619 /* > = 'F': On entry, AFB and IPIV contain the factored form of */
620 /* > A. If EQUED is not 'N', the matrix A has been */
621 /* > equilibrated with scaling factors given by R and C. */
622 /* > AB, AFB, and IPIV are not modified. */
623 /* > = 'N': The matrix A will be copied to AFB and factored. */
624 /* > = 'E': The matrix A will be equilibrated if necessary, then */
625 /* > copied to AFB and factored. */
628 /* > \param[in] TRANS */
630 /* > TRANS is CHARACTER*1 */
631 /* > Specifies the form of the system of equations. */
632 /* > = 'N': A * X = B (No transpose) */
633 /* > = 'T': A**T * X = B (Transpose) */
634 /* > = 'C': A**H * X = B (Transpose) */
640 /* > The number of linear equations, i.e., the order of the */
641 /* > matrix A. N >= 0. */
644 /* > \param[in] KL */
646 /* > KL is INTEGER */
647 /* > The number of subdiagonals within the band of A. KL >= 0. */
650 /* > \param[in] KU */
652 /* > KU is INTEGER */
653 /* > The number of superdiagonals within the band of A. KU >= 0. */
656 /* > \param[in] NRHS */
658 /* > NRHS is INTEGER */
659 /* > The number of right hand sides, i.e., the number of columns */
660 /* > of the matrices B and X. NRHS >= 0. */
663 /* > \param[in,out] AB */
665 /* > AB is DOUBLE PRECISION array, dimension (LDAB,N) */
666 /* > On entry, the matrix A in band storage, in rows 1 to KL+KU+1. */
667 /* > The j-th column of A is stored in the j-th column of the */
668 /* > array AB as follows: */
669 /* > AB(KU+1+i-j,j) = A(i,j) for f2cmax(1,j-KU)<=i<=f2cmin(N,j+kl) */
671 /* > If FACT = 'F' and EQUED is not 'N', then A must have been */
672 /* > equilibrated by the scaling factors in R and/or C. AB is not */
673 /* > modified if FACT = 'F' or 'N', or if FACT = 'E' and */
674 /* > EQUED = 'N' on exit. */
676 /* > On exit, if EQUED .ne. 'N', A is scaled as follows: */
677 /* > EQUED = 'R': A := diag(R) * A */
678 /* > EQUED = 'C': A := A * diag(C) */
679 /* > EQUED = 'B': A := diag(R) * A * diag(C). */
682 /* > \param[in] LDAB */
684 /* > LDAB is INTEGER */
685 /* > The leading dimension of the array AB. LDAB >= KL+KU+1. */
688 /* > \param[in,out] AFB */
690 /* > AFB is DOUBLE PRECISION array, dimension (LDAFB,N) */
691 /* > If FACT = 'F', then AFB is an input argument and on entry */
692 /* > contains details of the LU factorization of the band matrix */
693 /* > A, as computed by DGBTRF. U is stored as an upper triangular */
694 /* > band matrix with KL+KU superdiagonals in rows 1 to KL+KU+1, */
695 /* > and the multipliers used during the factorization are stored */
696 /* > in rows KL+KU+2 to 2*KL+KU+1. If EQUED .ne. 'N', then AFB is */
697 /* > the factored form of the equilibrated matrix A. */
699 /* > If FACT = 'N', then AFB is an output argument and on exit */
700 /* > returns details of the LU factorization of A. */
702 /* > If FACT = 'E', then AFB is an output argument and on exit */
703 /* > returns details of the LU factorization of the equilibrated */
704 /* > matrix A (see the description of AB for the form of the */
705 /* > equilibrated matrix). */
708 /* > \param[in] LDAFB */
710 /* > LDAFB is INTEGER */
711 /* > The leading dimension of the array AFB. LDAFB >= 2*KL+KU+1. */
714 /* > \param[in,out] IPIV */
716 /* > IPIV is INTEGER array, dimension (N) */
717 /* > If FACT = 'F', then IPIV is an input argument and on entry */
718 /* > contains the pivot indices from the factorization A = L*U */
719 /* > as computed by DGBTRF; row i of the matrix was interchanged */
720 /* > with row IPIV(i). */
722 /* > If FACT = 'N', then IPIV is an output argument and on exit */
723 /* > contains the pivot indices from the factorization A = L*U */
724 /* > of the original matrix A. */
726 /* > If FACT = 'E', then IPIV is an output argument and on exit */
727 /* > contains the pivot indices from the factorization A = L*U */
728 /* > of the equilibrated matrix A. */
731 /* > \param[in,out] EQUED */
733 /* > EQUED is CHARACTER*1 */
734 /* > Specifies the form of equilibration that was done. */
735 /* > = 'N': No equilibration (always true if FACT = 'N'). */
736 /* > = 'R': Row equilibration, i.e., A has been premultiplied by */
738 /* > = 'C': Column equilibration, i.e., A has been postmultiplied */
740 /* > = 'B': Both row and column equilibration, i.e., A has been */
741 /* > replaced by diag(R) * A * diag(C). */
742 /* > EQUED is an input argument if FACT = 'F'; otherwise, it is an */
743 /* > output argument. */
746 /* > \param[in,out] R */
748 /* > R is DOUBLE PRECISION array, dimension (N) */
749 /* > The row scale factors for A. If EQUED = 'R' or 'B', A is */
750 /* > multiplied on the left by diag(R); if EQUED = 'N' or 'C', R */
751 /* > is not accessed. R is an input argument if FACT = 'F'; */
752 /* > otherwise, R is an output argument. If FACT = 'F' and */
753 /* > EQUED = 'R' or 'B', each element of R must be positive. */
756 /* > \param[in,out] C */
758 /* > C is DOUBLE PRECISION array, dimension (N) */
759 /* > The column scale factors for A. If EQUED = 'C' or 'B', A is */
760 /* > multiplied on the right by diag(C); if EQUED = 'N' or 'R', C */
761 /* > is not accessed. C is an input argument if FACT = 'F'; */
762 /* > otherwise, C is an output argument. If FACT = 'F' and */
763 /* > EQUED = 'C' or 'B', each element of C must be positive. */
766 /* > \param[in,out] B */
768 /* > B is DOUBLE PRECISION array, dimension (LDB,NRHS) */
769 /* > On entry, the right hand side matrix B. */
771 /* > if EQUED = 'N', B is not modified; */
772 /* > if TRANS = 'N' and EQUED = 'R' or 'B', B is overwritten by */
774 /* > if TRANS = 'T' or 'C' and EQUED = 'C' or 'B', B is */
775 /* > overwritten by diag(C)*B. */
778 /* > \param[in] LDB */
780 /* > LDB is INTEGER */
781 /* > The leading dimension of the array B. LDB >= f2cmax(1,N). */
784 /* > \param[out] X */
786 /* > X is DOUBLE PRECISION array, dimension (LDX,NRHS) */
787 /* > If INFO = 0 or INFO = N+1, the N-by-NRHS solution matrix X */
788 /* > to the original system of equations. Note that A and B are */
789 /* > modified on exit if EQUED .ne. 'N', and the solution to the */
790 /* > equilibrated system is inv(diag(C))*X if TRANS = 'N' and */
791 /* > EQUED = 'C' or 'B', or inv(diag(R))*X if TRANS = 'T' or 'C' */
792 /* > and EQUED = 'R' or 'B'. */
795 /* > \param[in] LDX */
797 /* > LDX is INTEGER */
798 /* > The leading dimension of the array X. LDX >= f2cmax(1,N). */
801 /* > \param[out] RCOND */
803 /* > RCOND is DOUBLE PRECISION */
804 /* > The estimate of the reciprocal condition number of the matrix */
805 /* > A after equilibration (if done). If RCOND is less than the */
806 /* > machine precision (in particular, if RCOND = 0), the matrix */
807 /* > is singular to working precision. This condition is */
808 /* > indicated by a return code of INFO > 0. */
811 /* > \param[out] FERR */
813 /* > FERR is DOUBLE PRECISION array, dimension (NRHS) */
814 /* > The estimated forward error bound for each solution vector */
815 /* > X(j) (the j-th column of the solution matrix X). */
816 /* > If XTRUE is the true solution corresponding to X(j), FERR(j) */
817 /* > is an estimated upper bound for the magnitude of the largest */
818 /* > element in (X(j) - XTRUE) divided by the magnitude of the */
819 /* > largest element in X(j). The estimate is as reliable as */
820 /* > the estimate for RCOND, and is almost always a slight */
821 /* > overestimate of the true error. */
824 /* > \param[out] BERR */
826 /* > BERR is DOUBLE PRECISION array, dimension (NRHS) */
827 /* > The componentwise relative backward error of each solution */
828 /* > vector X(j) (i.e., the smallest relative change in */
829 /* > any element of A or B that makes X(j) an exact solution). */
832 /* > \param[out] WORK */
834 /* > WORK is DOUBLE PRECISION array, dimension (3*N) */
835 /* > On exit, WORK(1) contains the reciprocal pivot growth */
836 /* > factor norm(A)/norm(U). The "f2cmax absolute element" norm is */
837 /* > used. If WORK(1) is much less than 1, then the stability */
838 /* > of the LU factorization of the (equilibrated) matrix A */
839 /* > could be poor. This also means that the solution X, condition */
840 /* > estimator RCOND, and forward error bound FERR could be */
841 /* > unreliable. If factorization fails with 0<INFO<=N, then */
842 /* > WORK(1) contains the reciprocal pivot growth factor for the */
843 /* > leading INFO columns of A. */
846 /* > \param[out] IWORK */
848 /* > IWORK is INTEGER array, dimension (N) */
851 /* > \param[out] INFO */
853 /* > INFO is INTEGER */
854 /* > = 0: successful exit */
855 /* > < 0: if INFO = -i, the i-th argument had an illegal value */
856 /* > > 0: if INFO = i, and i is */
857 /* > <= N: U(i,i) is exactly zero. The factorization */
858 /* > has been completed, but the factor U is exactly */
859 /* > singular, so the solution and error bounds */
860 /* > could not be computed. RCOND = 0 is returned. */
861 /* > = N+1: U is nonsingular, but RCOND is less than machine */
862 /* > precision, meaning that the matrix is singular */
863 /* > to working precision. Nevertheless, the */
864 /* > solution and error bounds are computed because */
865 /* > there are a number of situations where the */
866 /* > computed solution can be more accurate than the */
867 /* > value of RCOND would suggest. */
873 /* > \author Univ. of Tennessee */
874 /* > \author Univ. of California Berkeley */
875 /* > \author Univ. of Colorado Denver */
876 /* > \author NAG Ltd. */
878 /* > \date April 2012 */
880 /* > \ingroup doubleGBsolve */
882 /* ===================================================================== */
883 /* Subroutine */ int dgbsvx_(char *fact, char *trans, integer *n, integer *kl,
884 integer *ku, integer *nrhs, doublereal *ab, integer *ldab,
885 doublereal *afb, integer *ldafb, integer *ipiv, char *equed,
886 doublereal *r__, doublereal *c__, doublereal *b, integer *ldb,
887 doublereal *x, integer *ldx, doublereal *rcond, doublereal *ferr,
888 doublereal *berr, doublereal *work, integer *iwork, integer *info)
890 /* System generated locals */
891 integer ab_dim1, ab_offset, afb_dim1, afb_offset, b_dim1, b_offset,
892 x_dim1, x_offset, i__1, i__2, i__3, i__4, i__5;
893 doublereal d__1, d__2, d__3;
895 /* Local variables */
899 extern logical lsame_(char *, char *);
900 doublereal rcmin, rcmax, anorm;
901 extern /* Subroutine */ int dcopy_(integer *, doublereal *, integer *,
902 doublereal *, integer *);
905 extern doublereal dlamch_(char *), dlangb_(char *, integer *,
906 integer *, integer *, doublereal *, integer *, doublereal *);
907 extern /* Subroutine */ int dlaqgb_(integer *, integer *, integer *,
908 integer *, doublereal *, integer *, doublereal *, doublereal *,
909 doublereal *, doublereal *, doublereal *, char *),
910 dgbcon_(char *, integer *, integer *, integer *, doublereal *,
911 integer *, integer *, doublereal *, doublereal *, doublereal *,
912 integer *, integer *);
914 extern doublereal dlantb_(char *, char *, char *, integer *, integer *,
915 doublereal *, integer *, doublereal *);
916 extern /* Subroutine */ int dgbequ_(integer *, integer *, integer *,
917 integer *, doublereal *, integer *, doublereal *, doublereal *,
918 doublereal *, doublereal *, doublereal *, integer *), dgbrfs_(
919 char *, integer *, integer *, integer *, integer *, doublereal *,
920 integer *, doublereal *, integer *, integer *, doublereal *,
921 integer *, doublereal *, integer *, doublereal *, doublereal *,
922 doublereal *, integer *, integer *), dgbtrf_(integer *,
923 integer *, integer *, integer *, doublereal *, integer *, integer
926 extern /* Subroutine */ int dlacpy_(char *, integer *, integer *,
927 doublereal *, integer *, doublereal *, integer *),
928 xerbla_(char *, integer *, ftnlen);
930 extern /* Subroutine */ int dgbtrs_(char *, integer *, integer *, integer
931 *, integer *, doublereal *, integer *, integer *, doublereal *,
932 integer *, integer *);
942 /* -- LAPACK driver routine (version 3.7.0) -- */
943 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
944 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
948 /* ===================================================================== */
951 /* Parameter adjustments */
953 ab_offset = 1 + ab_dim1 * 1;
956 afb_offset = 1 + afb_dim1 * 1;
962 b_offset = 1 + b_dim1 * 1;
965 x_offset = 1 + x_dim1 * 1;
974 nofact = lsame_(fact, "N");
975 equil = lsame_(fact, "E");
976 notran = lsame_(trans, "N");
977 if (nofact || equil) {
978 *(unsigned char *)equed = 'N';
982 rowequ = lsame_(equed, "R") || lsame_(equed,
984 colequ = lsame_(equed, "C") || lsame_(equed,
986 smlnum = dlamch_("Safe minimum");
987 bignum = 1. / smlnum;
990 /* Test the input parameters. */
992 if (! nofact && ! equil && ! lsame_(fact, "F")) {
994 } else if (! notran && ! lsame_(trans, "T") && !
995 lsame_(trans, "C")) {
999 } else if (*kl < 0) {
1001 } else if (*ku < 0) {
1003 } else if (*nrhs < 0) {
1005 } else if (*ldab < *kl + *ku + 1) {
1007 } else if (*ldafb < (*kl << 1) + *ku + 1) {
1009 } else if (lsame_(fact, "F") && ! (rowequ || colequ
1010 || lsame_(equed, "N"))) {
1017 for (j = 1; j <= i__1; ++j) {
1019 d__1 = rcmin, d__2 = r__[j];
1020 rcmin = f2cmin(d__1,d__2);
1022 d__1 = rcmax, d__2 = r__[j];
1023 rcmax = f2cmax(d__1,d__2);
1028 } else if (*n > 0) {
1029 rowcnd = f2cmax(rcmin,smlnum) / f2cmin(rcmax,bignum);
1034 if (colequ && *info == 0) {
1038 for (j = 1; j <= i__1; ++j) {
1040 d__1 = rcmin, d__2 = c__[j];
1041 rcmin = f2cmin(d__1,d__2);
1043 d__1 = rcmax, d__2 = c__[j];
1044 rcmax = f2cmax(d__1,d__2);
1049 } else if (*n > 0) {
1050 colcnd = f2cmax(rcmin,smlnum) / f2cmin(rcmax,bignum);
1056 if (*ldb < f2cmax(1,*n)) {
1058 } else if (*ldx < f2cmax(1,*n)) {
1066 xerbla_("DGBSVX", &i__1, (ftnlen)6);
1072 /* Compute row and column scalings to equilibrate the matrix A. */
1074 dgbequ_(n, n, kl, ku, &ab[ab_offset], ldab, &r__[1], &c__[1], &rowcnd,
1075 &colcnd, &amax, &infequ);
1078 /* Equilibrate the matrix. */
1080 dlaqgb_(n, n, kl, ku, &ab[ab_offset], ldab, &r__[1], &c__[1], &
1081 rowcnd, &colcnd, &amax, equed);
1082 rowequ = lsame_(equed, "R") || lsame_(equed,
1084 colequ = lsame_(equed, "C") || lsame_(equed,
1089 /* Scale the right hand side. */
1094 for (j = 1; j <= i__1; ++j) {
1096 for (i__ = 1; i__ <= i__2; ++i__) {
1097 b[i__ + j * b_dim1] = r__[i__] * b[i__ + j * b_dim1];
1103 } else if (colequ) {
1105 for (j = 1; j <= i__1; ++j) {
1107 for (i__ = 1; i__ <= i__2; ++i__) {
1108 b[i__ + j * b_dim1] = c__[i__] * b[i__ + j * b_dim1];
1115 if (nofact || equil) {
1117 /* Compute the LU factorization of the band matrix A. */
1120 for (j = 1; j <= i__1; ++j) {
1123 j1 = f2cmax(i__2,1);
1126 j2 = f2cmin(i__2,*n);
1128 dcopy_(&i__2, &ab[*ku + 1 - j + j1 + j * ab_dim1], &c__1, &afb[*
1129 kl + *ku + 1 - j + j1 + j * afb_dim1], &c__1);
1133 dgbtrf_(n, n, kl, ku, &afb[afb_offset], ldafb, &ipiv[1], info);
1135 /* Return if INFO is non-zero. */
1139 /* Compute the reciprocal pivot growth factor of the */
1140 /* leading rank-deficient INFO columns of A. */
1144 for (j = 1; j <= i__1; ++j) {
1148 i__4 = *n + *ku + 1 - j, i__5 = *kl + *ku + 1;
1149 i__3 = f2cmin(i__4,i__5);
1150 for (i__ = f2cmax(i__2,1); i__ <= i__3; ++i__) {
1152 d__2 = anorm, d__3 = (d__1 = ab[i__ + j * ab_dim1], abs(
1154 anorm = f2cmax(d__2,d__3);
1160 i__3 = *info - 1, i__2 = *kl + *ku;
1161 i__1 = f2cmin(i__3,i__2);
1163 i__4 = 1, i__5 = *kl + *ku + 2 - *info;
1164 rpvgrw = dlantb_("M", "U", "N", info, &i__1, &afb[f2cmax(i__4,i__5)
1165 + afb_dim1], ldafb, &work[1]);
1169 rpvgrw = anorm / rpvgrw;
1177 /* Compute the norm of the matrix A and the */
1178 /* reciprocal pivot growth factor RPVGRW. */
1181 *(unsigned char *)norm = '1';
1183 *(unsigned char *)norm = 'I';
1185 anorm = dlangb_(norm, n, kl, ku, &ab[ab_offset], ldab, &work[1]);
1187 rpvgrw = dlantb_("M", "U", "N", n, &i__1, &afb[afb_offset], ldafb, &work[
1192 rpvgrw = dlangb_("M", n, kl, ku, &ab[ab_offset], ldab, &work[1]) / rpvgrw;
1195 /* Compute the reciprocal of the condition number of A. */
1197 dgbcon_(norm, n, kl, ku, &afb[afb_offset], ldafb, &ipiv[1], &anorm, rcond,
1198 &work[1], &iwork[1], info);
1200 /* Compute the solution matrix X. */
1202 dlacpy_("Full", n, nrhs, &b[b_offset], ldb, &x[x_offset], ldx);
1203 dgbtrs_(trans, n, kl, ku, nrhs, &afb[afb_offset], ldafb, &ipiv[1], &x[
1204 x_offset], ldx, info);
1206 /* Use iterative refinement to improve the computed solution and */
1207 /* compute error bounds and backward error estimates for it. */
1209 dgbrfs_(trans, n, kl, ku, nrhs, &ab[ab_offset], ldab, &afb[afb_offset],
1210 ldafb, &ipiv[1], &b[b_offset], ldb, &x[x_offset], ldx, &ferr[1], &
1211 berr[1], &work[1], &iwork[1], info);
1213 /* Transform the solution matrix X to a solution of the original */
1219 for (j = 1; j <= i__1; ++j) {
1221 for (i__ = 1; i__ <= i__3; ++i__) {
1222 x[i__ + j * x_dim1] = c__[i__] * x[i__ + j * x_dim1];
1228 for (j = 1; j <= i__1; ++j) {
1233 } else if (rowequ) {
1235 for (j = 1; j <= i__1; ++j) {
1237 for (i__ = 1; i__ <= i__3; ++i__) {
1238 x[i__ + j * x_dim1] = r__[i__] * x[i__ + j * x_dim1];
1244 for (j = 1; j <= i__1; ++j) {
1250 /* Set INFO = N+1 if the matrix is singular to working precision. */
1252 if (*rcond < dlamch_("Epsilon")) {