14 typedef long long BLASLONG;
15 typedef unsigned long long BLASULONG;
17 typedef long BLASLONG;
18 typedef unsigned long BLASULONG;
22 typedef BLASLONG blasint;
24 #define blasabs(x) llabs(x)
26 #define blasabs(x) labs(x)
30 #define blasabs(x) abs(x)
33 typedef blasint integer;
35 typedef unsigned int uinteger;
36 typedef char *address;
37 typedef short int shortint;
39 typedef double doublereal;
40 typedef struct { real r, i; } complex;
41 typedef struct { doublereal r, i; } doublecomplex;
43 static inline _Fcomplex Cf(complex *z) {_Fcomplex zz={z->r , z->i}; return zz;}
44 static inline _Dcomplex Cd(doublecomplex *z) {_Dcomplex zz={z->r , z->i};return zz;}
45 static inline _Fcomplex * _pCf(complex *z) {return (_Fcomplex*)z;}
46 static inline _Dcomplex * _pCd(doublecomplex *z) {return (_Dcomplex*)z;}
48 static inline _Complex float Cf(complex *z) {return z->r + z->i*_Complex_I;}
49 static inline _Complex double Cd(doublecomplex *z) {return z->r + z->i*_Complex_I;}
50 static inline _Complex float * _pCf(complex *z) {return (_Complex float*)z;}
51 static inline _Complex double * _pCd(doublecomplex *z) {return (_Complex double*)z;}
53 #define pCf(z) (*_pCf(z))
54 #define pCd(z) (*_pCd(z))
56 typedef short int shortlogical;
57 typedef char logical1;
58 typedef char integer1;
63 /* Extern is for use with -E */
74 /*external read, write*/
83 /*internal read, write*/
113 /*rewind, backspace, endfile*/
125 ftnint *inex; /*parameters in standard's order*/
151 union Multitype { /* for multiple entry points */
162 typedef union Multitype Multitype;
164 struct Vardesc { /* for Namelist */
170 typedef struct Vardesc Vardesc;
177 typedef struct Namelist Namelist;
179 #define abs(x) ((x) >= 0 ? (x) : -(x))
180 #define dabs(x) (fabs(x))
181 #define f2cmin(a,b) ((a) <= (b) ? (a) : (b))
182 #define f2cmax(a,b) ((a) >= (b) ? (a) : (b))
183 #define dmin(a,b) (f2cmin(a,b))
184 #define dmax(a,b) (f2cmax(a,b))
185 #define bit_test(a,b) ((a) >> (b) & 1)
186 #define bit_clear(a,b) ((a) & ~((uinteger)1 << (b)))
187 #define bit_set(a,b) ((a) | ((uinteger)1 << (b)))
189 #define abort_() { sig_die("Fortran abort routine called", 1); }
190 #define c_abs(z) (cabsf(Cf(z)))
191 #define c_cos(R,Z) { pCf(R)=ccos(Cf(Z)); }
193 #define c_div(c, a, b) {Cf(c)._Val[0] = (Cf(a)._Val[0]/Cf(b)._Val[0]); Cf(c)._Val[1]=(Cf(a)._Val[1]/Cf(b)._Val[1]);}
194 #define z_div(c, a, b) {Cd(c)._Val[0] = (Cd(a)._Val[0]/Cd(b)._Val[0]); Cd(c)._Val[1]=(Cd(a)._Val[1]/df(b)._Val[1]);}
196 #define c_div(c, a, b) {pCf(c) = Cf(a)/Cf(b);}
197 #define z_div(c, a, b) {pCd(c) = Cd(a)/Cd(b);}
199 #define c_exp(R, Z) {pCf(R) = cexpf(Cf(Z));}
200 #define c_log(R, Z) {pCf(R) = clogf(Cf(Z));}
201 #define c_sin(R, Z) {pCf(R) = csinf(Cf(Z));}
202 //#define c_sqrt(R, Z) {*(R) = csqrtf(Cf(Z));}
203 #define c_sqrt(R, Z) {pCf(R) = csqrtf(Cf(Z));}
204 #define d_abs(x) (fabs(*(x)))
205 #define d_acos(x) (acos(*(x)))
206 #define d_asin(x) (asin(*(x)))
207 #define d_atan(x) (atan(*(x)))
208 #define d_atn2(x, y) (atan2(*(x),*(y)))
209 #define d_cnjg(R, Z) { pCd(R) = conj(Cd(Z)); }
210 #define r_cnjg(R, Z) { pCf(R) = conjf(Cf(Z)); }
211 #define d_cos(x) (cos(*(x)))
212 #define d_cosh(x) (cosh(*(x)))
213 #define d_dim(__a, __b) ( *(__a) > *(__b) ? *(__a) - *(__b) : 0.0 )
214 #define d_exp(x) (exp(*(x)))
215 #define d_imag(z) (cimag(Cd(z)))
216 #define r_imag(z) (cimagf(Cf(z)))
217 #define d_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
218 #define r_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
219 #define d_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
220 #define r_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
221 #define d_log(x) (log(*(x)))
222 #define d_mod(x, y) (fmod(*(x), *(y)))
223 #define u_nint(__x) ((__x)>=0 ? floor((__x) + .5) : -floor(.5 - (__x)))
224 #define d_nint(x) u_nint(*(x))
225 #define u_sign(__a,__b) ((__b) >= 0 ? ((__a) >= 0 ? (__a) : -(__a)) : -((__a) >= 0 ? (__a) : -(__a)))
226 #define d_sign(a,b) u_sign(*(a),*(b))
227 #define r_sign(a,b) u_sign(*(a),*(b))
228 #define d_sin(x) (sin(*(x)))
229 #define d_sinh(x) (sinh(*(x)))
230 #define d_sqrt(x) (sqrt(*(x)))
231 #define d_tan(x) (tan(*(x)))
232 #define d_tanh(x) (tanh(*(x)))
233 #define i_abs(x) abs(*(x))
234 #define i_dnnt(x) ((integer)u_nint(*(x)))
235 #define i_len(s, n) (n)
236 #define i_nint(x) ((integer)u_nint(*(x)))
237 #define i_sign(a,b) ((integer)u_sign((integer)*(a),(integer)*(b)))
238 #define pow_dd(ap, bp) ( pow(*(ap), *(bp)))
239 #define pow_si(B,E) spow_ui(*(B),*(E))
240 #define pow_ri(B,E) spow_ui(*(B),*(E))
241 #define pow_di(B,E) dpow_ui(*(B),*(E))
242 #define pow_zi(p, a, b) {pCd(p) = zpow_ui(Cd(a), *(b));}
243 #define pow_ci(p, a, b) {pCf(p) = cpow_ui(Cf(a), *(b));}
244 #define pow_zz(R,A,B) {pCd(R) = cpow(Cd(A),*(B));}
245 #define s_cat(lpp, rpp, rnp, np, llp) { ftnlen i, nc, ll; char *f__rp, *lp; ll = (llp); lp = (lpp); for(i=0; i < (int)*(np); ++i) { nc = ll; if((rnp)[i] < nc) nc = (rnp)[i]; ll -= nc; f__rp = (rpp)[i]; while(--nc >= 0) *lp++ = *(f__rp)++; } while(--ll >= 0) *lp++ = ' '; }
246 #define s_cmp(a,b,c,d) ((integer)strncmp((a),(b),f2cmin((c),(d))))
247 #define s_copy(A,B,C,D) { int __i,__m; for (__i=0, __m=f2cmin((C),(D)); __i<__m && (B)[__i] != 0; ++__i) (A)[__i] = (B)[__i]; }
248 #define sig_die(s, kill) { exit(1); }
249 #define s_stop(s, n) {exit(0);}
250 static char junk[] = "\n@(#)LIBF77 VERSION 19990503\n";
251 #define z_abs(z) (cabs(Cd(z)))
252 #define z_exp(R, Z) {pCd(R) = cexp(Cd(Z));}
253 #define z_sqrt(R, Z) {pCd(R) = csqrt(Cd(Z));}
254 #define myexit_() break;
255 #define mycycle() continue;
256 #define myceiling(w) {ceil(w)}
257 #define myhuge(w) {HUGE_VAL}
258 //#define mymaxloc_(w,s,e,n) {if (sizeof(*(w)) == sizeof(double)) dmaxloc_((w),*(s),*(e),n); else dmaxloc_((w),*(s),*(e),n);}
259 #define mymaxloc(w,s,e,n) {dmaxloc_(w,*(s),*(e),n)}
261 /* procedure parameter types for -A and -C++ */
263 #define F2C_proc_par_types 1
265 typedef logical (*L_fp)(...);
267 typedef logical (*L_fp)();
270 static float spow_ui(float x, integer n) {
271 float pow=1.0; unsigned long int u;
273 if(n < 0) n = -n, x = 1/x;
282 static double dpow_ui(double x, integer n) {
283 double pow=1.0; unsigned long int u;
285 if(n < 0) n = -n, x = 1/x;
295 static _Fcomplex cpow_ui(complex x, integer n) {
296 complex pow={1.0,0.0}; unsigned long int u;
298 if(n < 0) n = -n, x.r = 1/x.r, x.i=1/x.i;
300 if(u & 01) pow.r *= x.r, pow.i *= x.i;
301 if(u >>= 1) x.r *= x.r, x.i *= x.i;
305 _Fcomplex p={pow.r, pow.i};
309 static _Complex float cpow_ui(_Complex float x, integer n) {
310 _Complex float pow=1.0; unsigned long int u;
312 if(n < 0) n = -n, x = 1/x;
323 static _Dcomplex zpow_ui(_Dcomplex x, integer n) {
324 _Dcomplex pow={1.0,0.0}; unsigned long int u;
326 if(n < 0) n = -n, x._Val[0] = 1/x._Val[0], x._Val[1] =1/x._Val[1];
328 if(u & 01) pow._Val[0] *= x._Val[0], pow._Val[1] *= x._Val[1];
329 if(u >>= 1) x._Val[0] *= x._Val[0], x._Val[1] *= x._Val[1];
333 _Dcomplex p = {pow._Val[0], pow._Val[1]};
337 static _Complex double zpow_ui(_Complex double x, integer n) {
338 _Complex double pow=1.0; unsigned long int u;
340 if(n < 0) n = -n, x = 1/x;
350 static integer pow_ii(integer x, integer n) {
351 integer pow; unsigned long int u;
353 if (n == 0 || x == 1) pow = 1;
354 else if (x != -1) pow = x == 0 ? 1/x : 0;
357 if ((n > 0) || !(n == 0 || x == 1 || x != -1)) {
367 static integer dmaxloc_(double *w, integer s, integer e, integer *n)
369 double m; integer i, mi;
370 for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
371 if (w[i-1]>m) mi=i ,m=w[i-1];
374 static integer smaxloc_(float *w, integer s, integer e, integer *n)
376 float m; integer i, mi;
377 for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
378 if (w[i-1]>m) mi=i ,m=w[i-1];
381 static inline void cdotc_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
382 integer n = *n_, incx = *incx_, incy = *incy_, i;
384 _Fcomplex zdotc = {0.0, 0.0};
385 if (incx == 1 && incy == 1) {
386 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
387 zdotc._Val[0] += conjf(Cf(&x[i]))._Val[0] * Cf(&y[i])._Val[0];
388 zdotc._Val[1] += conjf(Cf(&x[i]))._Val[1] * Cf(&y[i])._Val[1];
391 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
392 zdotc._Val[0] += conjf(Cf(&x[i*incx]))._Val[0] * Cf(&y[i*incy])._Val[0];
393 zdotc._Val[1] += conjf(Cf(&x[i*incx]))._Val[1] * Cf(&y[i*incy])._Val[1];
399 _Complex float zdotc = 0.0;
400 if (incx == 1 && incy == 1) {
401 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
402 zdotc += conjf(Cf(&x[i])) * Cf(&y[i]);
405 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
406 zdotc += conjf(Cf(&x[i*incx])) * Cf(&y[i*incy]);
412 static inline void zdotc_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
413 integer n = *n_, incx = *incx_, incy = *incy_, i;
415 _Dcomplex zdotc = {0.0, 0.0};
416 if (incx == 1 && incy == 1) {
417 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
418 zdotc._Val[0] += conj(Cd(&x[i]))._Val[0] * Cd(&y[i])._Val[0];
419 zdotc._Val[1] += conj(Cd(&x[i]))._Val[1] * Cd(&y[i])._Val[1];
422 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
423 zdotc._Val[0] += conj(Cd(&x[i*incx]))._Val[0] * Cd(&y[i*incy])._Val[0];
424 zdotc._Val[1] += conj(Cd(&x[i*incx]))._Val[1] * Cd(&y[i*incy])._Val[1];
430 _Complex double zdotc = 0.0;
431 if (incx == 1 && incy == 1) {
432 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
433 zdotc += conj(Cd(&x[i])) * Cd(&y[i]);
436 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
437 zdotc += conj(Cd(&x[i*incx])) * Cd(&y[i*incy]);
443 static inline void cdotu_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
444 integer n = *n_, incx = *incx_, incy = *incy_, i;
446 _Fcomplex zdotc = {0.0, 0.0};
447 if (incx == 1 && incy == 1) {
448 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
449 zdotc._Val[0] += Cf(&x[i])._Val[0] * Cf(&y[i])._Val[0];
450 zdotc._Val[1] += Cf(&x[i])._Val[1] * Cf(&y[i])._Val[1];
453 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
454 zdotc._Val[0] += Cf(&x[i*incx])._Val[0] * Cf(&y[i*incy])._Val[0];
455 zdotc._Val[1] += Cf(&x[i*incx])._Val[1] * Cf(&y[i*incy])._Val[1];
461 _Complex float zdotc = 0.0;
462 if (incx == 1 && incy == 1) {
463 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
464 zdotc += Cf(&x[i]) * Cf(&y[i]);
467 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
468 zdotc += Cf(&x[i*incx]) * Cf(&y[i*incy]);
474 static inline void zdotu_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
475 integer n = *n_, incx = *incx_, incy = *incy_, i;
477 _Dcomplex zdotc = {0.0, 0.0};
478 if (incx == 1 && incy == 1) {
479 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
480 zdotc._Val[0] += Cd(&x[i])._Val[0] * Cd(&y[i])._Val[0];
481 zdotc._Val[1] += Cd(&x[i])._Val[1] * Cd(&y[i])._Val[1];
484 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
485 zdotc._Val[0] += Cd(&x[i*incx])._Val[0] * Cd(&y[i*incy])._Val[0];
486 zdotc._Val[1] += Cd(&x[i*incx])._Val[1] * Cd(&y[i*incy])._Val[1];
492 _Complex double zdotc = 0.0;
493 if (incx == 1 && incy == 1) {
494 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
495 zdotc += Cd(&x[i]) * Cd(&y[i]);
498 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
499 zdotc += Cd(&x[i*incx]) * Cd(&y[i*incy]);
505 /* -- translated by f2c (version 20000121).
506 You must link the resulting object file with the libraries:
507 -lf2c -lm (in that order)
513 /* Table of constant values */
515 static integer c__1 = 1;
516 static real c_b18 = -1.f;
517 static real c_b19 = 1.f;
519 /* > \brief \b SGTRFS */
521 /* =========== DOCUMENTATION =========== */
523 /* Online html documentation available at */
524 /* http://www.netlib.org/lapack/explore-html/ */
527 /* > Download SGTRFS + dependencies */
528 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/sgtrfs.
531 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/sgtrfs.
534 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/sgtrfs.
542 /* SUBROUTINE SGTRFS( TRANS, N, NRHS, DL, D, DU, DLF, DF, DUF, DU2, */
543 /* IPIV, B, LDB, X, LDX, FERR, BERR, WORK, IWORK, */
546 /* CHARACTER TRANS */
547 /* INTEGER INFO, LDB, LDX, N, NRHS */
548 /* INTEGER IPIV( * ), IWORK( * ) */
549 /* REAL B( LDB, * ), BERR( * ), D( * ), DF( * ), */
550 /* $ DL( * ), DLF( * ), DU( * ), DU2( * ), DUF( * ), */
551 /* $ FERR( * ), WORK( * ), X( LDX, * ) */
554 /* > \par Purpose: */
559 /* > SGTRFS improves the computed solution to a system of linear */
560 /* > equations when the coefficient matrix is tridiagonal, and provides */
561 /* > error bounds and backward error estimates for the solution. */
567 /* > \param[in] TRANS */
569 /* > TRANS is CHARACTER*1 */
570 /* > Specifies the form of the system of equations: */
571 /* > = 'N': A * X = B (No transpose) */
572 /* > = 'T': A**T * X = B (Transpose) */
573 /* > = 'C': A**H * X = B (Conjugate transpose = Transpose) */
579 /* > The order of the matrix A. N >= 0. */
582 /* > \param[in] NRHS */
584 /* > NRHS is INTEGER */
585 /* > The number of right hand sides, i.e., the number of columns */
586 /* > of the matrix B. NRHS >= 0. */
589 /* > \param[in] DL */
591 /* > DL is REAL array, dimension (N-1) */
592 /* > The (n-1) subdiagonal elements of A. */
597 /* > D is REAL array, dimension (N) */
598 /* > The diagonal elements of A. */
601 /* > \param[in] DU */
603 /* > DU is REAL array, dimension (N-1) */
604 /* > The (n-1) superdiagonal elements of A. */
607 /* > \param[in] DLF */
609 /* > DLF is REAL array, dimension (N-1) */
610 /* > The (n-1) multipliers that define the matrix L from the */
611 /* > LU factorization of A as computed by SGTTRF. */
614 /* > \param[in] DF */
616 /* > DF is REAL array, dimension (N) */
617 /* > The n diagonal elements of the upper triangular matrix U from */
618 /* > the LU factorization of A. */
621 /* > \param[in] DUF */
623 /* > DUF is REAL array, dimension (N-1) */
624 /* > The (n-1) elements of the first superdiagonal of U. */
627 /* > \param[in] DU2 */
629 /* > DU2 is REAL array, dimension (N-2) */
630 /* > The (n-2) elements of the second superdiagonal of U. */
633 /* > \param[in] IPIV */
635 /* > IPIV is INTEGER array, dimension (N) */
636 /* > The pivot indices; for 1 <= i <= n, row i of the matrix was */
637 /* > interchanged with row IPIV(i). IPIV(i) will always be either */
638 /* > i or i+1; IPIV(i) = i indicates a row interchange was not */
644 /* > B is REAL array, dimension (LDB,NRHS) */
645 /* > The right hand side matrix B. */
648 /* > \param[in] LDB */
650 /* > LDB is INTEGER */
651 /* > The leading dimension of the array B. LDB >= f2cmax(1,N). */
654 /* > \param[in,out] X */
656 /* > X is REAL array, dimension (LDX,NRHS) */
657 /* > On entry, the solution matrix X, as computed by SGTTRS. */
658 /* > On exit, the improved solution matrix X. */
661 /* > \param[in] LDX */
663 /* > LDX is INTEGER */
664 /* > The leading dimension of the array X. LDX >= f2cmax(1,N). */
667 /* > \param[out] FERR */
669 /* > FERR is REAL array, dimension (NRHS) */
670 /* > The estimated forward error bound for each solution vector */
671 /* > X(j) (the j-th column of the solution matrix X). */
672 /* > If XTRUE is the true solution corresponding to X(j), FERR(j) */
673 /* > is an estimated upper bound for the magnitude of the largest */
674 /* > element in (X(j) - XTRUE) divided by the magnitude of the */
675 /* > largest element in X(j). The estimate is as reliable as */
676 /* > the estimate for RCOND, and is almost always a slight */
677 /* > overestimate of the true error. */
680 /* > \param[out] BERR */
682 /* > BERR is REAL array, dimension (NRHS) */
683 /* > The componentwise relative backward error of each solution */
684 /* > vector X(j) (i.e., the smallest relative change in */
685 /* > any element of A or B that makes X(j) an exact solution). */
688 /* > \param[out] WORK */
690 /* > WORK is REAL array, dimension (3*N) */
693 /* > \param[out] IWORK */
695 /* > IWORK is INTEGER array, dimension (N) */
698 /* > \param[out] INFO */
700 /* > INFO is INTEGER */
701 /* > = 0: successful exit */
702 /* > < 0: if INFO = -i, the i-th argument had an illegal value */
705 /* > \par Internal Parameters: */
706 /* ========================= */
709 /* > ITMAX is the maximum number of steps of iterative refinement. */
715 /* > \author Univ. of Tennessee */
716 /* > \author Univ. of California Berkeley */
717 /* > \author Univ. of Colorado Denver */
718 /* > \author NAG Ltd. */
720 /* > \date December 2016 */
722 /* > \ingroup realGTcomputational */
724 /* ===================================================================== */
725 /* Subroutine */ int sgtrfs_(char *trans, integer *n, integer *nrhs, real *dl,
726 real *d__, real *du, real *dlf, real *df, real *duf, real *du2,
727 integer *ipiv, real *b, integer *ldb, real *x, integer *ldx, real *
728 ferr, real *berr, real *work, integer *iwork, integer *info)
730 /* System generated locals */
731 integer b_dim1, b_offset, x_dim1, x_offset, i__1, i__2;
732 real r__1, r__2, r__3, r__4;
734 /* Local variables */
739 extern logical lsame_(char *, char *);
740 integer isave[3], count;
741 extern /* Subroutine */ int scopy_(integer *, real *, integer *, real *,
742 integer *), saxpy_(integer *, real *, real *, integer *, real *,
743 integer *), slacn2_(integer *, real *, real *, integer *, real *,
744 integer *, integer *);
745 extern real slamch_(char *);
748 extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen), slagtm_(
749 char *, integer *, integer *, real *, real *, real *, real *,
750 real *, integer *, real *, real *, integer *);
752 char transn[1], transt[1];
754 extern /* Subroutine */ int sgttrs_(char *, integer *, integer *, real *,
755 real *, real *, real *, integer *, real *, integer *, integer *);
759 /* -- LAPACK computational routine (version 3.7.0) -- */
760 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
761 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
765 /* ===================================================================== */
768 /* Test the input parameters. */
770 /* Parameter adjustments */
780 b_offset = 1 + b_dim1 * 1;
783 x_offset = 1 + x_dim1 * 1;
792 notran = lsame_(trans, "N");
793 if (! notran && ! lsame_(trans, "T") && ! lsame_(
798 } else if (*nrhs < 0) {
800 } else if (*ldb < f2cmax(1,*n)) {
802 } else if (*ldx < f2cmax(1,*n)) {
807 xerbla_("SGTRFS", &i__1, (ftnlen)6);
811 /* Quick return if possible */
813 if (*n == 0 || *nrhs == 0) {
815 for (j = 1; j <= i__1; ++j) {
824 *(unsigned char *)transn = 'N';
825 *(unsigned char *)transt = 'T';
827 *(unsigned char *)transn = 'T';
828 *(unsigned char *)transt = 'N';
831 /* NZ = maximum number of nonzero elements in each row of A, plus 1 */
834 eps = slamch_("Epsilon");
835 safmin = slamch_("Safe minimum");
839 /* Do for each right hand side */
842 for (j = 1; j <= i__1; ++j) {
848 /* Loop until stopping criterion is satisfied. */
850 /* Compute residual R = B - op(A) * X, */
851 /* where op(A) = A, A**T, or A**H, depending on TRANS. */
853 scopy_(n, &b[j * b_dim1 + 1], &c__1, &work[*n + 1], &c__1);
854 slagtm_(trans, n, &c__1, &c_b18, &dl[1], &d__[1], &du[1], &x[j *
855 x_dim1 + 1], ldx, &c_b19, &work[*n + 1], n);
857 /* Compute abs(op(A))*abs(x) + abs(b) for use in the backward */
862 work[1] = (r__1 = b[j * b_dim1 + 1], abs(r__1)) + (r__2 = d__[
863 1] * x[j * x_dim1 + 1], abs(r__2));
865 work[1] = (r__1 = b[j * b_dim1 + 1], abs(r__1)) + (r__2 = d__[
866 1] * x[j * x_dim1 + 1], abs(r__2)) + (r__3 = du[1] *
867 x[j * x_dim1 + 2], abs(r__3));
869 for (i__ = 2; i__ <= i__2; ++i__) {
870 work[i__] = (r__1 = b[i__ + j * b_dim1], abs(r__1)) + (
871 r__2 = dl[i__ - 1] * x[i__ - 1 + j * x_dim1], abs(
872 r__2)) + (r__3 = d__[i__] * x[i__ + j * x_dim1],
873 abs(r__3)) + (r__4 = du[i__] * x[i__ + 1 + j *
877 work[*n] = (r__1 = b[*n + j * b_dim1], abs(r__1)) + (r__2 =
878 dl[*n - 1] * x[*n - 1 + j * x_dim1], abs(r__2)) + (
879 r__3 = d__[*n] * x[*n + j * x_dim1], abs(r__3));
883 work[1] = (r__1 = b[j * b_dim1 + 1], abs(r__1)) + (r__2 = d__[
884 1] * x[j * x_dim1 + 1], abs(r__2));
886 work[1] = (r__1 = b[j * b_dim1 + 1], abs(r__1)) + (r__2 = d__[
887 1] * x[j * x_dim1 + 1], abs(r__2)) + (r__3 = dl[1] *
888 x[j * x_dim1 + 2], abs(r__3));
890 for (i__ = 2; i__ <= i__2; ++i__) {
891 work[i__] = (r__1 = b[i__ + j * b_dim1], abs(r__1)) + (
892 r__2 = du[i__ - 1] * x[i__ - 1 + j * x_dim1], abs(
893 r__2)) + (r__3 = d__[i__] * x[i__ + j * x_dim1],
894 abs(r__3)) + (r__4 = dl[i__] * x[i__ + 1 + j *
898 work[*n] = (r__1 = b[*n + j * b_dim1], abs(r__1)) + (r__2 =
899 du[*n - 1] * x[*n - 1 + j * x_dim1], abs(r__2)) + (
900 r__3 = d__[*n] * x[*n + j * x_dim1], abs(r__3));
904 /* Compute componentwise relative backward error from formula */
906 /* f2cmax(i) ( abs(R(i)) / ( abs(op(A))*abs(X) + abs(B) )(i) ) */
908 /* where abs(Z) is the componentwise absolute value of the matrix */
909 /* or vector Z. If the i-th component of the denominator is less */
910 /* than SAFE2, then SAFE1 is added to the i-th components of the */
911 /* numerator and denominator before dividing. */
915 for (i__ = 1; i__ <= i__2; ++i__) {
916 if (work[i__] > safe2) {
918 r__2 = s, r__3 = (r__1 = work[*n + i__], abs(r__1)) / work[
920 s = f2cmax(r__2,r__3);
923 r__2 = s, r__3 = ((r__1 = work[*n + i__], abs(r__1)) + safe1)
924 / (work[i__] + safe1);
925 s = f2cmax(r__2,r__3);
931 /* Test stopping criterion. Continue iterating if */
932 /* 1) The residual BERR(J) is larger than machine epsilon, and */
933 /* 2) BERR(J) decreased by at least a factor of 2 during the */
934 /* last iteration, and */
935 /* 3) At most ITMAX iterations tried. */
937 if (berr[j] > eps && berr[j] * 2.f <= lstres && count <= 5) {
939 /* Update solution and try again. */
941 sgttrs_(trans, n, &c__1, &dlf[1], &df[1], &duf[1], &du2[1], &ipiv[
942 1], &work[*n + 1], n, info);
943 saxpy_(n, &c_b19, &work[*n + 1], &c__1, &x[j * x_dim1 + 1], &c__1)
950 /* Bound error from formula */
952 /* norm(X - XTRUE) / norm(X) .le. FERR = */
953 /* norm( abs(inv(op(A)))* */
954 /* ( abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) / norm(X) */
957 /* norm(Z) is the magnitude of the largest component of Z */
958 /* inv(op(A)) is the inverse of op(A) */
959 /* abs(Z) is the componentwise absolute value of the matrix or */
961 /* NZ is the maximum number of nonzeros in any row of A, plus 1 */
962 /* EPS is machine epsilon */
964 /* The i-th component of abs(R)+NZ*EPS*(abs(op(A))*abs(X)+abs(B)) */
965 /* is incremented by SAFE1 if the i-th component of */
966 /* abs(op(A))*abs(X) + abs(B) is less than SAFE2. */
968 /* Use SLACN2 to estimate the infinity-norm of the matrix */
969 /* inv(op(A)) * diag(W), */
970 /* where W = abs(R) + NZ*EPS*( abs(op(A))*abs(X)+abs(B) ))) */
973 for (i__ = 1; i__ <= i__2; ++i__) {
974 if (work[i__] > safe2) {
975 work[i__] = (r__1 = work[*n + i__], abs(r__1)) + nz * eps *
978 work[i__] = (r__1 = work[*n + i__], abs(r__1)) + nz * eps *
986 slacn2_(n, &work[(*n << 1) + 1], &work[*n + 1], &iwork[1], &ferr[j], &
991 /* Multiply by diag(W)*inv(op(A)**T). */
993 sgttrs_(transt, n, &c__1, &dlf[1], &df[1], &duf[1], &du2[1], &
994 ipiv[1], &work[*n + 1], n, info);
996 for (i__ = 1; i__ <= i__2; ++i__) {
997 work[*n + i__] = work[i__] * work[*n + i__];
1002 /* Multiply by inv(op(A))*diag(W). */
1005 for (i__ = 1; i__ <= i__2; ++i__) {
1006 work[*n + i__] = work[i__] * work[*n + i__];
1009 sgttrs_(transn, n, &c__1, &dlf[1], &df[1], &duf[1], &du2[1], &
1010 ipiv[1], &work[*n + 1], n, info);
1015 /* Normalize error. */
1019 for (i__ = 1; i__ <= i__2; ++i__) {
1021 r__2 = lstres, r__3 = (r__1 = x[i__ + j * x_dim1], abs(r__1));
1022 lstres = f2cmax(r__2,r__3);
1025 if (lstres != 0.f) {