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 integer c_n1 = -1;
517 static real c_b18 = -1.f;
518 static real c_b20 = 1.f;
520 /* > \brief \b SSYTRF_AA */
522 /* =========== DOCUMENTATION =========== */
524 /* Online html documentation available at */
525 /* http://www.netlib.org/lapack/explore-html/ */
528 /* > Download SSYTRF_AA + dependencies */
529 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/ssytrf_
532 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/ssytrf_
535 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/ssytrf_
543 /* SUBROUTINE SSYTRF_AA( UPLO, N, A, LDA, IPIV, WORK, LWORK, INFO ) */
546 /* INTEGER N, LDA, LWORK, INFO */
547 /* INTEGER IPIV( * ) */
548 /* REAL A( LDA, * ), WORK( * ) */
550 /* > \par Purpose: */
555 /* > SSYTRF_AA computes the factorization of a real symmetric matrix A */
556 /* > using the Aasen's algorithm. The form of the factorization is */
558 /* > A = U**T*T*U or A = L*T*L**T */
560 /* > where U (or L) is a product of permutation and unit upper (lower) */
561 /* > triangular matrices, and T is a symmetric tridiagonal matrix. */
563 /* > This is the blocked version of the algorithm, calling Level 3 BLAS. */
569 /* > \param[in] UPLO */
571 /* > UPLO is CHARACTER*1 */
572 /* > = 'U': Upper triangle of A is stored; */
573 /* > = 'L': Lower triangle of A is stored. */
579 /* > The order of the matrix A. N >= 0. */
582 /* > \param[in,out] A */
584 /* > A is REAL array, dimension (LDA,N) */
585 /* > On entry, the symmetric matrix A. If UPLO = 'U', the leading */
586 /* > N-by-N upper triangular part of A contains the upper */
587 /* > triangular part of the matrix A, and the strictly lower */
588 /* > triangular part of A is not referenced. If UPLO = 'L', the */
589 /* > leading N-by-N lower triangular part of A contains the lower */
590 /* > triangular part of the matrix A, and the strictly upper */
591 /* > triangular part of A is not referenced. */
593 /* > On exit, the tridiagonal matrix is stored in the diagonals */
594 /* > and the subdiagonals of A just below (or above) the diagonals, */
595 /* > and L is stored below (or above) the subdiaonals, when UPLO */
596 /* > is 'L' (or 'U'). */
599 /* > \param[in] LDA */
601 /* > LDA is INTEGER */
602 /* > The leading dimension of the array A. LDA >= f2cmax(1,N). */
605 /* > \param[out] IPIV */
607 /* > IPIV is INTEGER array, dimension (N) */
608 /* > On exit, it contains the details of the interchanges, i.e., */
609 /* > the row and column k of A were interchanged with the */
610 /* > row and column IPIV(k). */
613 /* > \param[out] WORK */
615 /* > WORK is REAL array, dimension (MAX(1,LWORK)) */
616 /* > On exit, if INFO = 0, WORK(1) returns the optimal LWORK. */
619 /* > \param[in] LWORK */
621 /* > LWORK is INTEGER */
622 /* > The length of WORK. LWORK >= MAX(1,2*N). For optimum performance */
623 /* > LWORK >= N*(1+NB), where NB is the optimal blocksize. */
625 /* > If LWORK = -1, then a workspace query is assumed; the routine */
626 /* > only calculates the optimal size of the WORK array, returns */
627 /* > this value as the first entry of the WORK array, and no error */
628 /* > message related to LWORK is issued by XERBLA. */
631 /* > \param[out] INFO */
633 /* > INFO is INTEGER */
634 /* > = 0: successful exit */
635 /* > < 0: if INFO = -i, the i-th argument had an illegal value. */
641 /* > \author Univ. of Tennessee */
642 /* > \author Univ. of California Berkeley */
643 /* > \author Univ. of Colorado Denver */
644 /* > \author NAG Ltd. */
646 /* > \date November 2017 */
648 /* > \ingroup realSYcomputational */
650 /* ===================================================================== */
651 /* Subroutine */ int ssytrf_aa_(char *uplo, integer *n, real *a, integer *
652 lda, integer *ipiv, real *work, integer *lwork, integer *info)
654 /* System generated locals */
655 integer a_dim1, a_offset, i__1, i__2, i__3, i__4;
657 /* Local variables */
660 extern logical lsame_(char *, char *);
661 extern /* Subroutine */ int sscal_(integer *, real *, real *, integer *),
662 sgemm_(char *, char *, integer *, integer *, integer *, real *,
663 real *, integer *, real *, integer *, real *, real *, integer *), slasyf_aa_(char *, integer *, integer *,
664 integer *, real *, integer *, integer *, real *, integer *, real *
665 ), sgemv_(char *, integer *, integer *, real *, real *,
666 integer *, real *, integer *, real *, real *, integer *);
668 integer k1, k2, j1, j2, j3;
669 extern /* Subroutine */ int scopy_(integer *, real *, integer *, real *,
670 integer *), sswap_(integer *, real *, integer *, real *, integer *
672 integer jb, nb, mj, nj;
673 extern /* Subroutine */ int xerbla_(char *, integer *, ftnlen);
674 extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
675 integer *, integer *, ftnlen, ftnlen);
680 /* -- LAPACK computational routine (version 3.8.0) -- */
681 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
682 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
687 /* ===================================================================== */
690 /* Determine the block size */
692 /* Parameter adjustments */
694 a_offset = 1 + a_dim1 * 1;
700 nb = ilaenv_(&c__1, "SSYTRF_AA", uplo, n, &c_n1, &c_n1, &c_n1, (ftnlen)9,
703 /* Test the input parameters. */
706 upper = lsame_(uplo, "U");
707 lquery = *lwork == -1;
708 if (! upper && ! lsame_(uplo, "L")) {
712 } else if (*lda < f2cmax(1,*n)) {
714 } else /* if(complicated condition) */ {
716 i__1 = 1, i__2 = *n << 1;
717 if (*lwork < f2cmax(i__1,i__2) && ! lquery) {
723 lwkopt = (nb + 1) * *n;
724 work[1] = (real) lwkopt;
729 xerbla_("SSYTRF_AA", &i__1, (ftnlen)9);
745 /* Adjust block size based on the workspace size */
747 if (*lwork < (nb + 1) * *n) {
748 nb = (*lwork - *n) / *n;
753 /* ..................................................... */
754 /* Factorize A as U**T*D*U using the upper triangle of A */
755 /* ..................................................... */
757 /* Copy first row A(1, 1:N) into H(1:n) (stored in WORK(1:N)) */
759 scopy_(n, &a[a_dim1 + 1], lda, &work[1], &c__1);
761 /* J is the main loop index, increasing from 1 to N in steps of */
762 /* JB, where JB is the number of columns factorized by SLASYF; */
763 /* JB is either NB, or N-J+1 for the last block */
771 /* each step of the main loop */
772 /* J is the last column of the previous panel */
773 /* J1 is the first column of the current panel */
774 /* K1 identifies if the previous column of the panel has been */
775 /* explicitly stored, e.g., K1=1 for the first panel, and */
776 /* K1=0 for the rest */
781 jb = f2cmin(i__1,nb);
782 k1 = f2cmax(1,j) - j;
784 /* Panel factorization */
788 slasyf_aa_(uplo, &i__1, &i__2, &jb, &a[f2cmax(1,j) + (j + 1) * a_dim1],
789 lda, &ipiv[j + 1], &work[1], n, &work[*n * nb + 1])
792 /* Adjust IPIV and apply it back (J-th step picks (J+1)-th pivot) */
795 i__2 = *n, i__3 = j + jb + 1;
796 i__1 = f2cmin(i__2,i__3);
797 for (j2 = j + 2; j2 <= i__1; ++j2) {
799 if (j2 != ipiv[j2] && j1 - k1 > 2) {
801 sswap_(&i__2, &a[j2 * a_dim1 + 1], &c__1, &a[ipiv[j2] *
807 /* Trailing submatrix update, where */
808 /* the row A(J1-1, J2-1:N) stores U(J1, J2+1:N) and */
809 /* WORK stores the current block of the auxiriarly matrix H */
813 /* If first panel and JB=1 (NB=1), then nothing to do */
815 if (j1 > 1 || jb > 1) {
817 /* Merge rank-1 update with BLAS-3 update */
819 alpha = a[j + (j + 1) * a_dim1];
820 a[j + (j + 1) * a_dim1] = 1.f;
822 scopy_(&i__1, &a[j - 1 + (j + 1) * a_dim1], lda, &work[j + 1
823 - j1 + 1 + jb * *n], &c__1);
825 sscal_(&i__1, &alpha, &work[j + 1 - j1 + 1 + jb * *n], &c__1);
827 /* K1 identifies if the previous column of the panel has been */
828 /* explicitly stored, e.g., K1=1 and K2= 0 for the first panel, */
829 /* while K1=0 and K2=1 for the rest */
833 /* Not first panel */
842 /* First update skips the first column */
849 for (j2 = j + 1; i__2 < 0 ? j2 >= i__1 : j2 <= i__1; j2 +=
852 i__3 = nb, i__4 = *n - j2 + 1;
853 nj = f2cmin(i__3,i__4);
855 /* Update (J2, J2) diagonal block with SGEMV */
858 for (mj = nj - 1; mj >= 1; --mj) {
860 sgemv_("No transpose", &mj, &i__3, &c_b18, &work[j3 -
861 j1 + 1 + k1 * *n], n, &a[j1 - k2 + j3 *
862 a_dim1], &c__1, &c_b20, &a[j3 + j3 * a_dim1],
867 /* Update off-diagonal block of J2-th block row with SGEMM */
871 sgemm_("Transpose", "Transpose", &nj, &i__3, &i__4, &
872 c_b18, &a[j1 - k2 + j2 * a_dim1], lda, &work[j3 -
873 j1 + 1 + k1 * *n], n, &c_b20, &a[j2 + j3 * a_dim1]
877 /* Recover T( J, J+1 ) */
879 a[j + (j + 1) * a_dim1] = alpha;
882 /* WORK(J+1, 1) stores H(J+1, 1) */
885 scopy_(&i__2, &a[j + 1 + (j + 1) * a_dim1], lda, &work[1], &c__1);
890 /* ..................................................... */
891 /* Factorize A as L*D*L**T using the lower triangle of A */
892 /* ..................................................... */
894 /* copy first column A(1:N, 1) into H(1:N, 1) */
895 /* (stored in WORK(1:N)) */
897 scopy_(n, &a[a_dim1 + 1], &c__1, &work[1], &c__1);
899 /* J is the main loop index, increasing from 1 to N in steps of */
900 /* JB, where JB is the number of columns factorized by SLASYF; */
901 /* JB is either NB, or N-J+1 for the last block */
909 /* each step of the main loop */
910 /* J is the last column of the previous panel */
911 /* J1 is the first column of the current panel */
912 /* K1 identifies if the previous column of the panel has been */
913 /* explicitly stored, e.g., K1=1 for the first panel, and */
914 /* K1=0 for the rest */
919 jb = f2cmin(i__2,nb);
920 k1 = f2cmax(1,j) - j;
922 /* Panel factorization */
926 slasyf_aa_(uplo, &i__2, &i__1, &jb, &a[j + 1 + f2cmax(1,j) * a_dim1],
927 lda, &ipiv[j + 1], &work[1], n, &work[*n * nb + 1])
930 /* Adjust IPIV and apply it back (J-th step picks (J+1)-th pivot) */
933 i__1 = *n, i__3 = j + jb + 1;
934 i__2 = f2cmin(i__1,i__3);
935 for (j2 = j + 2; j2 <= i__2; ++j2) {
937 if (j2 != ipiv[j2] && j1 - k1 > 2) {
939 sswap_(&i__1, &a[j2 + a_dim1], lda, &a[ipiv[j2] + a_dim1],
945 /* Trailing submatrix update, where */
946 /* A(J2+1, J1-1) stores L(J2+1, J1) and */
947 /* WORK(J2+1, 1) stores H(J2+1, 1) */
951 /* if first panel and JB=1 (NB=1), then nothing to do */
953 if (j1 > 1 || jb > 1) {
955 /* Merge rank-1 update with BLAS-3 update */
957 alpha = a[j + 1 + j * a_dim1];
958 a[j + 1 + j * a_dim1] = 1.f;
960 scopy_(&i__2, &a[j + 1 + (j - 1) * a_dim1], &c__1, &work[j +
961 1 - j1 + 1 + jb * *n], &c__1);
963 sscal_(&i__2, &alpha, &work[j + 1 - j1 + 1 + jb * *n], &c__1);
965 /* K1 identifies if the previous column of the panel has been */
966 /* explicitly stored, e.g., K1=1 and K2= 0 for the first panel, */
967 /* while K1=0 and K2=1 for the rest */
971 /* Not first panel */
980 /* First update skips the first column */
987 for (j2 = j + 1; i__1 < 0 ? j2 >= i__2 : j2 <= i__2; j2 +=
990 i__3 = nb, i__4 = *n - j2 + 1;
991 nj = f2cmin(i__3,i__4);
993 /* Update (J2, J2) diagonal block with SGEMV */
996 for (mj = nj - 1; mj >= 1; --mj) {
998 sgemv_("No transpose", &mj, &i__3, &c_b18, &work[j3 -
999 j1 + 1 + k1 * *n], n, &a[j3 + (j1 - k2) *
1000 a_dim1], lda, &c_b20, &a[j3 + j3 * a_dim1], &
1005 /* Update off-diagonal block in J2-th block column with SGEMM */
1009 sgemm_("No transpose", "Transpose", &i__3, &nj, &i__4, &
1010 c_b18, &work[j3 - j1 + 1 + k1 * *n], n, &a[j2 + (
1011 j1 - k2) * a_dim1], lda, &c_b20, &a[j3 + j2 *
1015 /* Recover T( J+1, J ) */
1017 a[j + 1 + j * a_dim1] = alpha;
1020 /* WORK(J+1, 1) stores H(J+1, 1) */
1023 scopy_(&i__1, &a[j + 1 + (j + 1) * a_dim1], &c__1, &work[1], &
1032 /* End of SSYTRF_AA */