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 CHPTRF */
519 /* =========== DOCUMENTATION =========== */
521 /* Online html documentation available at */
522 /* http://www.netlib.org/lapack/explore-html/ */
525 /* > Download CHPTRF + dependencies */
526 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/chptrf.
529 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/chptrf.
532 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/chptrf.
540 /* SUBROUTINE CHPTRF( UPLO, N, AP, IPIV, INFO ) */
543 /* INTEGER INFO, N */
544 /* INTEGER IPIV( * ) */
545 /* COMPLEX AP( * ) */
548 /* > \par Purpose: */
553 /* > CHPTRF computes the factorization of a complex Hermitian packed */
554 /* > matrix A using the Bunch-Kaufman diagonal pivoting method: */
556 /* > A = U*D*U**H or A = L*D*L**H */
558 /* > where U (or L) is a product of permutation and unit upper (lower) */
559 /* > triangular matrices, and D is Hermitian and block diagonal with */
560 /* > 1-by-1 and 2-by-2 diagonal blocks. */
566 /* > \param[in] UPLO */
568 /* > UPLO is CHARACTER*1 */
569 /* > = 'U': Upper triangle of A is stored; */
570 /* > = 'L': Lower triangle of A is stored. */
576 /* > The order of the matrix A. N >= 0. */
579 /* > \param[in,out] AP */
581 /* > AP is COMPLEX array, dimension (N*(N+1)/2) */
582 /* > On entry, the upper or lower triangle of the Hermitian matrix */
583 /* > A, packed columnwise in a linear array. The j-th column of A */
584 /* > is stored in the array AP as follows: */
585 /* > if UPLO = 'U', AP(i + (j-1)*j/2) = A(i,j) for 1<=i<=j; */
586 /* > if UPLO = 'L', AP(i + (j-1)*(2n-j)/2) = A(i,j) for j<=i<=n. */
588 /* > On exit, the block diagonal matrix D and the multipliers used */
589 /* > to obtain the factor U or L, stored as a packed triangular */
590 /* > matrix overwriting A (see below for further details). */
593 /* > \param[out] IPIV */
595 /* > IPIV is INTEGER array, dimension (N) */
596 /* > Details of the interchanges and the block structure of D. */
597 /* > If IPIV(k) > 0, then rows and columns k and IPIV(k) were */
598 /* > interchanged and D(k,k) is a 1-by-1 diagonal block. */
599 /* > If UPLO = 'U' and IPIV(k) = IPIV(k-1) < 0, then rows and */
600 /* > columns k-1 and -IPIV(k) were interchanged and D(k-1:k,k-1:k) */
601 /* > is a 2-by-2 diagonal block. If UPLO = 'L' and IPIV(k) = */
602 /* > IPIV(k+1) < 0, then rows and columns k+1 and -IPIV(k) were */
603 /* > interchanged and D(k:k+1,k:k+1) is a 2-by-2 diagonal block. */
606 /* > \param[out] INFO */
608 /* > INFO is INTEGER */
609 /* > = 0: successful exit */
610 /* > < 0: if INFO = -i, the i-th argument had an illegal value */
611 /* > > 0: if INFO = i, D(i,i) is exactly zero. The factorization */
612 /* > has been completed, but the block diagonal matrix D is */
613 /* > exactly singular, and division by zero will occur if it */
614 /* > is used to solve a system of equations. */
620 /* > \author Univ. of Tennessee */
621 /* > \author Univ. of California Berkeley */
622 /* > \author Univ. of Colorado Denver */
623 /* > \author NAG Ltd. */
625 /* > \date December 2016 */
627 /* > \ingroup complexOTHERcomputational */
629 /* > \par Further Details: */
630 /* ===================== */
634 /* > If UPLO = 'U', then A = U*D*U**H, where */
635 /* > U = P(n)*U(n)* ... *P(k)U(k)* ..., */
636 /* > i.e., U is a product of terms P(k)*U(k), where k decreases from n to */
637 /* > 1 in steps of 1 or 2, and D is a block diagonal matrix with 1-by-1 */
638 /* > and 2-by-2 diagonal blocks D(k). P(k) is a permutation matrix as */
639 /* > defined by IPIV(k), and U(k) is a unit upper triangular matrix, such */
640 /* > that if the diagonal block D(k) is of order s (s = 1 or 2), then */
642 /* > ( I v 0 ) k-s */
643 /* > U(k) = ( 0 I 0 ) s */
644 /* > ( 0 0 I ) n-k */
647 /* > If s = 1, D(k) overwrites A(k,k), and v overwrites A(1:k-1,k). */
648 /* > If s = 2, the upper triangle of D(k) overwrites A(k-1,k-1), A(k-1,k), */
649 /* > and A(k,k), and v overwrites A(1:k-2,k-1:k). */
651 /* > If UPLO = 'L', then A = L*D*L**H, where */
652 /* > L = P(1)*L(1)* ... *P(k)*L(k)* ..., */
653 /* > i.e., L is a product of terms P(k)*L(k), where k increases from 1 to */
654 /* > n in steps of 1 or 2, and D is a block diagonal matrix with 1-by-1 */
655 /* > and 2-by-2 diagonal blocks D(k). P(k) is a permutation matrix as */
656 /* > defined by IPIV(k), and L(k) is a unit lower triangular matrix, such */
657 /* > that if the diagonal block D(k) is of order s (s = 1 or 2), then */
659 /* > ( I 0 0 ) k-1 */
660 /* > L(k) = ( 0 I 0 ) s */
661 /* > ( 0 v I ) n-k-s+1 */
662 /* > k-1 s n-k-s+1 */
664 /* > If s = 1, D(k) overwrites A(k,k), and v overwrites A(k+1:n,k). */
665 /* > If s = 2, the lower triangle of D(k) overwrites A(k,k), A(k+1,k), */
666 /* > and A(k+1,k+1), and v overwrites A(k+2:n,k:k+1). */
669 /* > \par Contributors: */
670 /* ================== */
672 /* > J. Lewis, Boeing Computer Services Company */
674 /* ===================================================================== */
675 /* Subroutine */ int chptrf_(char *uplo, integer *n, complex *ap, integer *
678 /* System generated locals */
679 integer i__1, i__2, i__3, i__4, i__5, i__6;
680 real r__1, r__2, r__3, r__4;
681 complex q__1, q__2, q__3, q__4, q__5, q__6;
683 /* Local variables */
684 extern /* Subroutine */ int chpr_(char *, integer *, real *, complex *,
685 integer *, complex *);
691 extern logical lsame_(char *, char *);
692 extern /* Subroutine */ int cswap_(integer *, complex *, integer *,
693 complex *, integer *);
700 extern real slapy2_(real *, real *);
705 extern integer icamax_(integer *, complex *, integer *);
707 extern /* Subroutine */ int csscal_(integer *, real *, complex *, integer
708 *), xerbla_(char *, integer *, ftnlen);
710 integer knc, kpc, npp;
714 /* -- LAPACK computational routine (version 3.7.0) -- */
715 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
716 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
720 /* ===================================================================== */
723 /* Test the input parameters. */
725 /* Parameter adjustments */
731 upper = lsame_(uplo, "U");
732 if (! upper && ! lsame_(uplo, "L")) {
739 xerbla_("CHPTRF", &i__1, (ftnlen)6);
743 /* Initialize ALPHA for use in choosing pivot block size. */
745 alpha = (sqrt(17.f) + 1.f) / 8.f;
749 /* Factorize A as U*D*U**H using the upper triangle of A */
751 /* K is the main loop index, decreasing from N to 1 in steps of */
755 kc = (*n - 1) * *n / 2 + 1;
759 /* If K < 1, exit from loop */
766 /* Determine rows and columns to be interchanged and whether */
767 /* a 1-by-1 or 2-by-2 pivot block will be used */
770 absakk = (r__1 = ap[i__1].r, abs(r__1));
772 /* IMAX is the row-index of the largest off-diagonal element in */
773 /* column K, and COLMAX is its absolute value */
777 imax = icamax_(&i__1, &ap[kc], &c__1);
778 i__1 = kc + imax - 1;
779 colmax = (r__1 = ap[i__1].r, abs(r__1)) + (r__2 = r_imag(&ap[kc +
780 imax - 1]), abs(r__2));
785 if (f2cmax(absakk,colmax) == 0.f) {
787 /* Column K is zero: set INFO and continue */
796 ap[i__1].r = r__1, ap[i__1].i = 0.f;
798 if (absakk >= alpha * colmax) {
800 /* no interchange, use 1-by-1 pivot block */
805 /* JMAX is the column-index of the largest off-diagonal */
806 /* element in row IMAX, and ROWMAX is its absolute value */
810 kx = imax * (imax + 1) / 2 + imax;
812 for (j = imax + 1; j <= i__1; ++j) {
814 if ((r__1 = ap[i__2].r, abs(r__1)) + (r__2 = r_imag(&ap[
815 kx]), abs(r__2)) > rowmax) {
817 rowmax = (r__1 = ap[i__2].r, abs(r__1)) + (r__2 =
818 r_imag(&ap[kx]), abs(r__2));
824 kpc = (imax - 1) * imax / 2 + 1;
827 jmax = icamax_(&i__1, &ap[kpc], &c__1);
829 i__1 = kpc + jmax - 1;
830 r__3 = rowmax, r__4 = (r__1 = ap[i__1].r, abs(r__1)) + (
831 r__2 = r_imag(&ap[kpc + jmax - 1]), abs(r__2));
832 rowmax = f2cmax(r__3,r__4);
835 if (absakk >= alpha * colmax * (colmax / rowmax)) {
837 /* no interchange, use 1-by-1 pivot block */
840 } else /* if(complicated condition) */ {
841 i__1 = kpc + imax - 1;
842 if ((r__1 = ap[i__1].r, abs(r__1)) >= alpha * rowmax) {
844 /* interchange rows and columns K and IMAX, use 1-by-1 */
850 /* interchange rows and columns K-1 and IMAX, use 2-by-2 */
865 /* Interchange rows and columns KK and KP in the leading */
866 /* submatrix A(1:k,1:k) */
869 cswap_(&i__1, &ap[knc], &c__1, &ap[kpc], &c__1);
872 for (j = kp + 1; j <= i__1; ++j) {
874 r_cnjg(&q__1, &ap[knc + j - 1]);
875 t.r = q__1.r, t.i = q__1.i;
877 r_cnjg(&q__1, &ap[kx]);
878 ap[i__2].r = q__1.r, ap[i__2].i = q__1.i;
880 ap[i__2].r = t.r, ap[i__2].i = t.i;
884 r_cnjg(&q__1, &ap[kx + kk - 1]);
885 ap[i__1].r = q__1.r, ap[i__1].i = q__1.i;
891 ap[i__1].r = r__1, ap[i__1].i = 0.f;
893 ap[i__1].r = r1, ap[i__1].i = 0.f;
898 ap[i__1].r = r__1, ap[i__1].i = 0.f;
900 t.r = ap[i__1].r, t.i = ap[i__1].i;
903 ap[i__1].r = ap[i__2].r, ap[i__1].i = ap[i__2].i;
905 ap[i__1].r = t.r, ap[i__1].i = t.i;
911 ap[i__1].r = r__1, ap[i__1].i = 0.f;
916 ap[i__1].r = r__1, ap[i__1].i = 0.f;
920 /* Update the leading submatrix */
924 /* 1-by-1 pivot block D(k): column k now holds */
926 /* W(k) = U(k)*D(k) */
928 /* where U(k) is the k-th column of U */
930 /* Perform a rank-1 update of A(1:k-1,1:k-1) as */
932 /* A := A - U(k)*D(k)*U(k)**H = A - W(k)*1/D(k)*W(k)**H */
935 r1 = 1.f / ap[i__1].r;
938 chpr_(uplo, &i__1, &r__1, &ap[kc], &c__1, &ap[1]);
940 /* Store U(k) in column k */
943 csscal_(&i__1, &r1, &ap[kc], &c__1);
946 /* 2-by-2 pivot block D(k): columns k and k-1 now hold */
948 /* ( W(k-1) W(k) ) = ( U(k-1) U(k) )*D(k) */
950 /* where U(k) and U(k-1) are the k-th and (k-1)-th columns */
953 /* Perform a rank-2 update of A(1:k-2,1:k-2) as */
955 /* A := A - ( U(k-1) U(k) )*D(k)*( U(k-1) U(k) )**H */
956 /* = A - ( W(k-1) W(k) )*inv(D(k))*( W(k-1) W(k) )**H */
960 i__1 = k - 1 + (k - 1) * k / 2;
962 r__2 = r_imag(&ap[k - 1 + (k - 1) * k / 2]);
963 d__ = slapy2_(&r__1, &r__2);
964 i__1 = k - 1 + (k - 2) * (k - 1) / 2;
965 d22 = ap[i__1].r / d__;
966 i__1 = k + (k - 1) * k / 2;
967 d11 = ap[i__1].r / d__;
968 tt = 1.f / (d11 * d22 - 1.f);
969 i__1 = k - 1 + (k - 1) * k / 2;
970 q__1.r = ap[i__1].r / d__, q__1.i = ap[i__1].i / d__;
971 d12.r = q__1.r, d12.i = q__1.i;
974 for (j = k - 2; j >= 1; --j) {
975 i__1 = j + (k - 2) * (k - 1) / 2;
976 q__3.r = d11 * ap[i__1].r, q__3.i = d11 * ap[i__1].i;
978 i__2 = j + (k - 1) * k / 2;
979 q__4.r = q__5.r * ap[i__2].r - q__5.i * ap[i__2].i,
980 q__4.i = q__5.r * ap[i__2].i + q__5.i * ap[
982 q__2.r = q__3.r - q__4.r, q__2.i = q__3.i - q__4.i;
983 q__1.r = d__ * q__2.r, q__1.i = d__ * q__2.i;
984 wkm1.r = q__1.r, wkm1.i = q__1.i;
985 i__1 = j + (k - 1) * k / 2;
986 q__3.r = d22 * ap[i__1].r, q__3.i = d22 * ap[i__1].i;
987 i__2 = j + (k - 2) * (k - 1) / 2;
988 q__4.r = d12.r * ap[i__2].r - d12.i * ap[i__2].i,
989 q__4.i = d12.r * ap[i__2].i + d12.i * ap[i__2]
991 q__2.r = q__3.r - q__4.r, q__2.i = q__3.i - q__4.i;
992 q__1.r = d__ * q__2.r, q__1.i = d__ * q__2.i;
993 wk.r = q__1.r, wk.i = q__1.i;
994 for (i__ = j; i__ >= 1; --i__) {
995 i__1 = i__ + (j - 1) * j / 2;
996 i__2 = i__ + (j - 1) * j / 2;
997 i__3 = i__ + (k - 1) * k / 2;
999 q__3.r = ap[i__3].r * q__4.r - ap[i__3].i *
1000 q__4.i, q__3.i = ap[i__3].r * q__4.i + ap[
1002 q__2.r = ap[i__2].r - q__3.r, q__2.i = ap[i__2].i
1004 i__4 = i__ + (k - 2) * (k - 1) / 2;
1005 r_cnjg(&q__6, &wkm1);
1006 q__5.r = ap[i__4].r * q__6.r - ap[i__4].i *
1007 q__6.i, q__5.i = ap[i__4].r * q__6.i + ap[
1009 q__1.r = q__2.r - q__5.r, q__1.i = q__2.i -
1011 ap[i__1].r = q__1.r, ap[i__1].i = q__1.i;
1014 i__1 = j + (k - 1) * k / 2;
1015 ap[i__1].r = wk.r, ap[i__1].i = wk.i;
1016 i__1 = j + (k - 2) * (k - 1) / 2;
1017 ap[i__1].r = wkm1.r, ap[i__1].i = wkm1.i;
1018 i__1 = j + (j - 1) * j / 2;
1019 i__2 = j + (j - 1) * j / 2;
1021 q__1.r = r__1, q__1.i = 0.f;
1022 ap[i__1].r = q__1.r, ap[i__1].i = q__1.i;
1031 /* Store details of the interchanges in IPIV */
1040 /* Decrease K and return to the start of the main loop */
1048 /* Factorize A as L*D*L**H using the lower triangle of A */
1050 /* K is the main loop index, increasing from 1 to N in steps of */
1055 npp = *n * (*n + 1) / 2;
1059 /* If K > N, exit from loop */
1066 /* Determine rows and columns to be interchanged and whether */
1067 /* a 1-by-1 or 2-by-2 pivot block will be used */
1070 absakk = (r__1 = ap[i__1].r, abs(r__1));
1072 /* IMAX is the row-index of the largest off-diagonal element in */
1073 /* column K, and COLMAX is its absolute value */
1077 imax = k + icamax_(&i__1, &ap[kc + 1], &c__1);
1078 i__1 = kc + imax - k;
1079 colmax = (r__1 = ap[i__1].r, abs(r__1)) + (r__2 = r_imag(&ap[kc +
1080 imax - k]), abs(r__2));
1085 if (f2cmax(absakk,colmax) == 0.f) {
1087 /* Column K is zero: set INFO and continue */
1096 ap[i__1].r = r__1, ap[i__1].i = 0.f;
1098 if (absakk >= alpha * colmax) {
1100 /* no interchange, use 1-by-1 pivot block */
1105 /* JMAX is the column-index of the largest off-diagonal */
1106 /* element in row IMAX, and ROWMAX is its absolute value */
1111 for (j = k; j <= i__1; ++j) {
1113 if ((r__1 = ap[i__2].r, abs(r__1)) + (r__2 = r_imag(&ap[
1114 kx]), abs(r__2)) > rowmax) {
1116 rowmax = (r__1 = ap[i__2].r, abs(r__1)) + (r__2 =
1117 r_imag(&ap[kx]), abs(r__2));
1123 kpc = npp - (*n - imax + 1) * (*n - imax + 2) / 2 + 1;
1126 jmax = imax + icamax_(&i__1, &ap[kpc + 1], &c__1);
1128 i__1 = kpc + jmax - imax;
1129 r__3 = rowmax, r__4 = (r__1 = ap[i__1].r, abs(r__1)) + (
1130 r__2 = r_imag(&ap[kpc + jmax - imax]), abs(r__2));
1131 rowmax = f2cmax(r__3,r__4);
1134 if (absakk >= alpha * colmax * (colmax / rowmax)) {
1136 /* no interchange, use 1-by-1 pivot block */
1139 } else /* if(complicated condition) */ {
1141 if ((r__1 = ap[i__1].r, abs(r__1)) >= alpha * rowmax) {
1143 /* interchange rows and columns K and IMAX, use 1-by-1 */
1149 /* interchange rows and columns K+1 and IMAX, use 2-by-2 */
1160 knc = knc + *n - k + 1;
1164 /* Interchange rows and columns KK and KP in the trailing */
1165 /* submatrix A(k:n,k:n) */
1169 cswap_(&i__1, &ap[knc + kp - kk + 1], &c__1, &ap[kpc + 1],
1174 for (j = kk + 1; j <= i__1; ++j) {
1175 kx = kx + *n - j + 1;
1176 r_cnjg(&q__1, &ap[knc + j - kk]);
1177 t.r = q__1.r, t.i = q__1.i;
1178 i__2 = knc + j - kk;
1179 r_cnjg(&q__1, &ap[kx]);
1180 ap[i__2].r = q__1.r, ap[i__2].i = q__1.i;
1182 ap[i__2].r = t.r, ap[i__2].i = t.i;
1185 i__1 = knc + kp - kk;
1186 r_cnjg(&q__1, &ap[knc + kp - kk]);
1187 ap[i__1].r = q__1.r, ap[i__1].i = q__1.i;
1193 ap[i__1].r = r__1, ap[i__1].i = 0.f;
1195 ap[i__1].r = r1, ap[i__1].i = 0.f;
1200 ap[i__1].r = r__1, ap[i__1].i = 0.f;
1202 t.r = ap[i__1].r, t.i = ap[i__1].i;
1205 ap[i__1].r = ap[i__2].r, ap[i__1].i = ap[i__2].i;
1207 ap[i__1].r = t.r, ap[i__1].i = t.i;
1213 ap[i__1].r = r__1, ap[i__1].i = 0.f;
1218 ap[i__1].r = r__1, ap[i__1].i = 0.f;
1222 /* Update the trailing submatrix */
1226 /* 1-by-1 pivot block D(k): column k now holds */
1228 /* W(k) = L(k)*D(k) */
1230 /* where L(k) is the k-th column of L */
1234 /* Perform a rank-1 update of A(k+1:n,k+1:n) as */
1236 /* A := A - L(k)*D(k)*L(k)**H = A - W(k)*(1/D(k))*W(k)**H */
1239 r1 = 1.f / ap[i__1].r;
1242 chpr_(uplo, &i__1, &r__1, &ap[kc + 1], &c__1, &ap[kc + *n
1245 /* Store L(k) in column K */
1248 csscal_(&i__1, &r1, &ap[kc + 1], &c__1);
1252 /* 2-by-2 pivot block D(k): columns K and K+1 now hold */
1254 /* ( W(k) W(k+1) ) = ( L(k) L(k+1) )*D(k) */
1256 /* where L(k) and L(k+1) are the k-th and (k+1)-th columns */
1261 /* Perform a rank-2 update of A(k+2:n,k+2:n) as */
1263 /* A := A - ( L(k) L(k+1) )*D(k)*( L(k) L(k+1) )**H */
1264 /* = A - ( W(k) W(k+1) )*inv(D(k))*( W(k) W(k+1) )**H */
1266 /* where L(k) and L(k+1) are the k-th and (k+1)-th */
1269 i__1 = k + 1 + (k - 1) * ((*n << 1) - k) / 2;
1271 r__2 = r_imag(&ap[k + 1 + (k - 1) * ((*n << 1) - k) / 2]);
1272 d__ = slapy2_(&r__1, &r__2);
1273 i__1 = k + 1 + k * ((*n << 1) - k - 1) / 2;
1274 d11 = ap[i__1].r / d__;
1275 i__1 = k + (k - 1) * ((*n << 1) - k) / 2;
1276 d22 = ap[i__1].r / d__;
1277 tt = 1.f / (d11 * d22 - 1.f);
1278 i__1 = k + 1 + (k - 1) * ((*n << 1) - k) / 2;
1279 q__1.r = ap[i__1].r / d__, q__1.i = ap[i__1].i / d__;
1280 d21.r = q__1.r, d21.i = q__1.i;
1284 for (j = k + 2; j <= i__1; ++j) {
1285 i__2 = j + (k - 1) * ((*n << 1) - k) / 2;
1286 q__3.r = d11 * ap[i__2].r, q__3.i = d11 * ap[i__2].i;
1287 i__3 = j + k * ((*n << 1) - k - 1) / 2;
1288 q__4.r = d21.r * ap[i__3].r - d21.i * ap[i__3].i,
1289 q__4.i = d21.r * ap[i__3].i + d21.i * ap[i__3]
1291 q__2.r = q__3.r - q__4.r, q__2.i = q__3.i - q__4.i;
1292 q__1.r = d__ * q__2.r, q__1.i = d__ * q__2.i;
1293 wk.r = q__1.r, wk.i = q__1.i;
1294 i__2 = j + k * ((*n << 1) - k - 1) / 2;
1295 q__3.r = d22 * ap[i__2].r, q__3.i = d22 * ap[i__2].i;
1296 r_cnjg(&q__5, &d21);
1297 i__3 = j + (k - 1) * ((*n << 1) - k) / 2;
1298 q__4.r = q__5.r * ap[i__3].r - q__5.i * ap[i__3].i,
1299 q__4.i = q__5.r * ap[i__3].i + q__5.i * ap[
1301 q__2.r = q__3.r - q__4.r, q__2.i = q__3.i - q__4.i;
1302 q__1.r = d__ * q__2.r, q__1.i = d__ * q__2.i;
1303 wkp1.r = q__1.r, wkp1.i = q__1.i;
1305 for (i__ = j; i__ <= i__2; ++i__) {
1306 i__3 = i__ + (j - 1) * ((*n << 1) - j) / 2;
1307 i__4 = i__ + (j - 1) * ((*n << 1) - j) / 2;
1308 i__5 = i__ + (k - 1) * ((*n << 1) - k) / 2;
1310 q__3.r = ap[i__5].r * q__4.r - ap[i__5].i *
1311 q__4.i, q__3.i = ap[i__5].r * q__4.i + ap[
1313 q__2.r = ap[i__4].r - q__3.r, q__2.i = ap[i__4].i
1315 i__6 = i__ + k * ((*n << 1) - k - 1) / 2;
1316 r_cnjg(&q__6, &wkp1);
1317 q__5.r = ap[i__6].r * q__6.r - ap[i__6].i *
1318 q__6.i, q__5.i = ap[i__6].r * q__6.i + ap[
1320 q__1.r = q__2.r - q__5.r, q__1.i = q__2.i -
1322 ap[i__3].r = q__1.r, ap[i__3].i = q__1.i;
1325 i__2 = j + (k - 1) * ((*n << 1) - k) / 2;
1326 ap[i__2].r = wk.r, ap[i__2].i = wk.i;
1327 i__2 = j + k * ((*n << 1) - k - 1) / 2;
1328 ap[i__2].r = wkp1.r, ap[i__2].i = wkp1.i;
1329 i__2 = j + (j - 1) * ((*n << 1) - j) / 2;
1330 i__3 = j + (j - 1) * ((*n << 1) - j) / 2;
1332 q__1.r = r__1, q__1.i = 0.f;
1333 ap[i__2].r = q__1.r, ap[i__2].i = q__1.i;
1340 /* Store details of the interchanges in IPIV */
1349 /* Increase K and return to the start of the main loop */
1352 kc = knc + *n - k + 2;