/* Subroutine */ int cdrgev_(integer *nsizes, integer *nn, integer *ntypes, logical *dotype, integer *iseed, real *thresh, integer *nounit, complex *a, integer *lda, complex *b, complex *s, complex *t, complex *q, integer *ldq, complex *z__, complex *qe, integer *ldqe, complex * alpha, complex *beta, complex *alpha1, complex *beta1, complex *work, integer *lwork, real *rwork, real *result, integer *info) { /* Initialized data */ static integer kclass[26] = { 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2, 2,2,2,3 }; static integer kbmagn[26] = { 1,1,1,1,1,1,1,1,3,2,3,2,2,3,1,1,1,1,1,1,1,3, 2,3,2,1 }; static integer ktrian[26] = { 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1, 1,1,1,1 }; static logical lasign[26] = { FALSE_,FALSE_,FALSE_,FALSE_,FALSE_,FALSE_, TRUE_,FALSE_,TRUE_,TRUE_,FALSE_,FALSE_,TRUE_,TRUE_,TRUE_,FALSE_, TRUE_,FALSE_,FALSE_,FALSE_,TRUE_,TRUE_,TRUE_,TRUE_,TRUE_,FALSE_ }; static logical lbsign[26] = { FALSE_,FALSE_,FALSE_,FALSE_,FALSE_,FALSE_, FALSE_,TRUE_,FALSE_,FALSE_,TRUE_,TRUE_,FALSE_,FALSE_,TRUE_,FALSE_, TRUE_,FALSE_,FALSE_,FALSE_,FALSE_,FALSE_,FALSE_,FALSE_,FALSE_, FALSE_ }; static integer kz1[6] = { 0,1,2,1,3,3 }; static integer kz2[6] = { 0,0,1,2,1,1 }; static integer kadd[6] = { 0,0,0,0,3,2 }; static integer katype[26] = { 0,1,0,1,2,3,4,1,4,4,1,1,4,4,4,2,4,5,8,7,9,4, 4,4,4,0 }; static integer kbtype[26] = { 0,0,1,1,2,-3,1,4,1,1,4,4,1,1,-4,2,-4,8,8,8, 8,8,8,8,8,0 }; static integer kazero[26] = { 1,1,1,1,1,1,2,1,2,2,1,1,2,2,3,1,3,5,5,5,5,3, 3,3,3,1 }; static integer kbzero[26] = { 1,1,1,1,1,1,1,2,1,1,2,2,1,1,4,1,4,6,6,6,6,4, 4,4,4,1 }; static integer kamagn[26] = { 1,1,1,1,1,1,1,1,2,3,2,3,2,3,1,1,1,1,1,1,1,2, 3,3,2,1 }; /* Format strings */ static char fmt_9999[] = "(\002 CDRGEV: \002,a,\002 returned INFO=\002,i" "6,\002.\002,/3x,\002N=\002,i6,\002, JTYPE=\002,i6,\002, ISEED=" "(\002,3(i5,\002,\002),i5,\002)\002)"; static char fmt_9998[] = "(\002 CDRGEV: \002,a,\002 Eigenvectors from" " \002,a,\002 incorrectly \002,\002normalized.\002,/\002 Bits of " "error=\002,0p,g10.3,\002,\002,3x,\002N=\002,i4,\002, JTYPE=\002," "i3,\002, ISEED=(\002,3(i4,\002,\002),i5,\002)\002)"; static char fmt_9997[] = "(/1x,a3,\002 -- Complex Generalized eigenvalue" " problem \002,\002driver\002)"; static char fmt_9996[] = "(\002 Matrix types (see CDRGEV for details):" " \002)"; static char fmt_9995[] = "(\002 Special Matrices:\002,23x,\002(J'=transp" "osed Jordan block)\002,/\002 1=(0,0) 2=(I,0) 3=(0,I) 4=(I,I" ") 5=(J',J') \002,\0026=(diag(J',I), diag(I,J'))\002,/\002 Diag" "onal Matrices: ( \002,\002D=diag(0,1,2,...) )\002,/\002 7=(D," "I) 9=(large*D, small*I\002,\002) 11=(large*I, small*D) 13=(l" "arge*D, large*I)\002,/\002 8=(I,D) 10=(small*D, large*I) 12=" "(small*I, large*D) \002,\002 14=(small*D, small*I)\002,/\002 15" "=(D, reversed D)\002)"; static char fmt_9994[] = "(\002 Matrices Rotated by Random \002,a,\002 M" "atrices U, V:\002,/\002 16=Transposed Jordan Blocks " " 19=geometric \002,\002alpha, beta=0,1\002,/\002 17=arithm. alp" "ha&beta \002,\002 20=arithmetic alpha, beta=0," "1\002,/\002 18=clustered \002,\002alpha, beta=0,1 21" "=random alpha, beta=0,1\002,/\002 Large & Small Matrices:\002," "/\002 22=(large, small) \002,\00223=(small,large) 24=(smal" "l,small) 25=(large,large)\002,/\002 26=random O(1) matrices" ".\002)"; static char fmt_9993[] = "(/\002 Tests performed: \002,/\002 1 = max " "| ( b A - a B )'*l | / const.,\002,/\002 2 = | |VR(i)| - 1 | / u" "lp,\002,/\002 3 = max | ( b A - a B )*r | / const.\002,/\002 4 =" " | |VL(i)| - 1 | / ulp,\002,/\002 5 = 0 if W same no matter if r" " or l computed,\002,/\002 6 = 0 if l same no matter if l compute" "d,\002,/\002 7 = 0 if r same no matter if r computed,\002,/1x)"; static char fmt_9992[] = "(\002 Matrix order=\002,i5,\002, type=\002,i2" ",\002, seed=\002,4(i4,\002,\002),\002 result \002,i2,\002 is\002" ",0p,f8.2)"; static char fmt_9991[] = "(\002 Matrix order=\002,i5,\002, type=\002,i2" ",\002, seed=\002,4(i4,\002,\002),\002 result \002,i2,\002 is\002" ",1p,e10.3)"; /* System generated locals */ integer a_dim1, a_offset, b_dim1, b_offset, q_dim1, q_offset, qe_dim1, qe_offset, s_dim1, s_offset, t_dim1, t_offset, z_dim1, z_offset, i__1, i__2, i__3, i__4, i__5, i__6, i__7; real r__1, r__2; complex q__1, q__2, q__3; /* Builtin functions */ double r_sign(real *, real *), c_abs(complex *); void r_cnjg(complex *, complex *); integer s_wsfe(cilist *), do_fio(integer *, char *, ftnlen), e_wsfe(void); /* Local variables */ static integer iadd, ierr, nmax, i__, j, n; static logical badnn; extern /* Subroutine */ int cget52_(logical *, integer *, complex *, integer *, complex *, integer *, complex *, integer *, complex *, complex *, complex *, real *, real *), cggev_(char *, char *, integer *, complex *, integer *, complex *, integer *, complex *, complex *, complex *, integer *, complex *, integer *, complex *, integer *, real *, integer *); static real rmagn[4]; static complex ctemp; static integer nmats, jsize, nerrs, jtype, n1; extern /* Subroutine */ int clatm4_(integer *, integer *, integer *, integer *, logical *, real *, real *, real *, integer *, integer * , complex *, integer *), cunm2r_(char *, char *, integer *, integer *, integer *, complex *, integer *, complex *, complex *, integer *, complex *, integer *); static integer jc, nb, in; extern /* Subroutine */ int slabad_(real *, real *); static integer jr; extern /* Subroutine */ int clarfg_(integer *, complex *, complex *, integer *, complex *); extern /* Complex */ VOID clarnd_(complex *, integer *, integer *); extern doublereal slamch_(char *); extern /* Subroutine */ int clacpy_(char *, integer *, integer *, complex *, integer *, complex *, integer *), claset_(char *, integer *, integer *, complex *, complex *, complex *, integer *); static real safmin, safmax; static integer ioldsd[4]; extern integer ilaenv_(integer *, char *, char *, integer *, integer *, integer *, integer *, ftnlen, ftnlen); extern /* Subroutine */ int alasvm_(char *, integer *, integer *, integer *, integer *), xerbla_(char *, integer *); static integer minwrk, maxwrk; static real ulpinv; static integer mtypes, ntestt; static real ulp; /* Fortran I/O blocks */ static cilist io___40 = { 0, 0, 0, fmt_9999, 0 }; static cilist io___42 = { 0, 0, 0, fmt_9999, 0 }; static cilist io___43 = { 0, 0, 0, fmt_9998, 0 }; static cilist io___44 = { 0, 0, 0, fmt_9998, 0 }; static cilist io___45 = { 0, 0, 0, fmt_9999, 0 }; static cilist io___46 = { 0, 0, 0, fmt_9999, 0 }; static cilist io___47 = { 0, 0, 0, fmt_9999, 0 }; static cilist io___48 = { 0, 0, 0, fmt_9997, 0 }; static cilist io___49 = { 0, 0, 0, fmt_9996, 0 }; static cilist io___50 = { 0, 0, 0, fmt_9995, 0 }; static cilist io___51 = { 0, 0, 0, fmt_9994, 0 }; static cilist io___52 = { 0, 0, 0, fmt_9993, 0 }; static cilist io___53 = { 0, 0, 0, fmt_9992, 0 }; static cilist io___54 = { 0, 0, 0, fmt_9991, 0 }; #define a_subscr(a_1,a_2) (a_2)*a_dim1 + a_1 #define a_ref(a_1,a_2) a[a_subscr(a_1,a_2)] #define b_subscr(a_1,a_2) (a_2)*b_dim1 + a_1 #define b_ref(a_1,a_2) b[b_subscr(a_1,a_2)] #define q_subscr(a_1,a_2) (a_2)*q_dim1 + a_1 #define q_ref(a_1,a_2) q[q_subscr(a_1,a_2)] #define z___subscr(a_1,a_2) (a_2)*z_dim1 + a_1 #define z___ref(a_1,a_2) z__[z___subscr(a_1,a_2)] #define qe_subscr(a_1,a_2) (a_2)*qe_dim1 + a_1 #define qe_ref(a_1,a_2) qe[qe_subscr(a_1,a_2)] /* -- LAPACK test routine (version 3.0) -- Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., Courant Institute, Argonne National Lab, and Rice University June 30, 1999 Purpose ======= CDRGEV checks the nonsymmetric generalized eigenvalue problem driver routine CGGEV. CGGEV computes for a pair of n-by-n nonsymmetric matrices (A,B) the generalized eigenvalues and, optionally, the left and right eigenvectors. A generalized eigenvalue for a pair of matrices (A,B) is a scalar w or a ratio alpha/beta = w, such that A - w*B is singular. It is usually represented as the pair (alpha,beta), as there is reasonalbe interpretation for beta=0, and even for both being zero. A right generalized eigenvector corresponding to a generalized eigenvalue w for a pair of matrices (A,B) is a vector r such that (A - wB) * r = 0. A left generalized eigenvector is a vector l such that l**H * (A - wB) = 0, where l**H is the conjugate-transpose of l. When CDRGEV is called, a number of matrix "sizes" ("n's") and a number of matrix "types" are specified. For each size ("n") and each type of matrix, a pair of matrices (A, B) will be generated and used for testing. For each matrix pair, the following tests will be performed and compared with the threshhold THRESH. Results from CGGEV: (1) max over all left eigenvalue/-vector pairs (alpha/beta,l) of | VL**H * (beta A - alpha B) |/( ulp max(|beta A|, |alpha B|) ) where VL**H is the conjugate-transpose of VL. (2) | |VL(i)| - 1 | / ulp and whether largest component real VL(i) denotes the i-th column of VL. (3) max over all left eigenvalue/-vector pairs (alpha/beta,r) of | (beta A - alpha B) * VR | / ( ulp max(|beta A|, |alpha B|) ) (4) | |VR(i)| - 1 | / ulp and whether largest component real VR(i) denotes the i-th column of VR. (5) W(full) = W(partial) W(full) denotes the eigenvalues computed when both l and r are also computed, and W(partial) denotes the eigenvalues computed when only W, only W and r, or only W and l are computed. (6) VL(full) = VL(partial) VL(full) denotes the left eigenvectors computed when both l and r are computed, and VL(partial) denotes the result when only l is computed. (7) VR(full) = VR(partial) VR(full) denotes the right eigenvectors computed when both l and r are also computed, and VR(partial) denotes the result when only l is computed. Test Matrices ---- -------- The sizes of the test matrices are specified by an array NN(1:NSIZES); the value of each element NN(j) specifies one size. The "types" are specified by a logical array DOTYPE( 1:NTYPES ); if DOTYPE(j) is .TRUE., then matrix type "j" will be generated. Currently, the list of possible types is: (1) ( 0, 0 ) (a pair of zero matrices) (2) ( I, 0 ) (an identity and a zero matrix) (3) ( 0, I ) (an identity and a zero matrix) (4) ( I, I ) (a pair of identity matrices) t t (5) ( J , J ) (a pair of transposed Jordan blocks) t ( I 0 ) (6) ( X, Y ) where X = ( J 0 ) and Y = ( t ) ( 0 I ) ( 0 J ) and I is a k x k identity and J a (k+1)x(k+1) Jordan block; k=(N-1)/2 (7) ( D, I ) where D is diag( 0, 1,..., N-1 ) (a diagonal matrix with those diagonal entries.) (8) ( I, D ) (9) ( big*D, small*I ) where "big" is near overflow and small=1/big (10) ( small*D, big*I ) (11) ( big*I, small*D ) (12) ( small*I, big*D ) (13) ( big*D, big*I ) (14) ( small*D, small*I ) (15) ( D1, D2 ) where D1 is diag( 0, 0, 1, ..., N-3, 0 ) and D2 is diag( 0, N-3, N-4,..., 1, 0, 0 ) t t (16) Q ( J , J ) Z where Q and Z are random orthogonal matrices. (17) Q ( T1, T2 ) Z where T1 and T2 are upper triangular matrices with random O(1) entries above the diagonal and diagonal entries diag(T1) = ( 0, 0, 1, ..., N-3, 0 ) and diag(T2) = ( 0, N-3, N-4,..., 1, 0, 0 ) (18) Q ( T1, T2 ) Z diag(T1) = ( 0, 0, 1, 1, s, ..., s, 0 ) diag(T2) = ( 0, 1, 0, 1,..., 1, 0 ) s = machine precision. (19) Q ( T1, T2 ) Z diag(T1)=( 0,0,1,1, 1-d, ..., 1-(N-5)*d=s, 0 ) diag(T2) = ( 0, 1, 0, 1, ..., 1, 0 ) N-5 (20) Q ( T1, T2 ) Z diag(T1)=( 0, 0, 1, 1, a, ..., a =s, 0 ) diag(T2) = ( 0, 1, 0, 1, ..., 1, 0, 0 ) (21) Q ( T1, T2 ) Z diag(T1)=( 0, 0, 1, r1, r2, ..., r(N-4), 0 ) diag(T2) = ( 0, 1, 0, 1, ..., 1, 0, 0 ) where r1,..., r(N-4) are random. (22) Q ( big*T1, small*T2 ) Z diag(T1) = ( 0, 0, 1, ..., N-3, 0 ) diag(T2) = ( 0, 1, ..., 1, 0, 0 ) (23) Q ( small*T1, big*T2 ) Z diag(T1) = ( 0, 0, 1, ..., N-3, 0 ) diag(T2) = ( 0, 1, ..., 1, 0, 0 ) (24) Q ( small*T1, small*T2 ) Z diag(T1) = ( 0, 0, 1, ..., N-3, 0 ) diag(T2) = ( 0, 1, ..., 1, 0, 0 ) (25) Q ( big*T1, big*T2 ) Z diag(T1) = ( 0, 0, 1, ..., N-3, 0 ) diag(T2) = ( 0, 1, ..., 1, 0, 0 ) (26) Q ( T1, T2 ) Z where T1 and T2 are random upper-triangular matrices. Arguments ========= NSIZES (input) INTEGER The number of sizes of matrices to use. If it is zero, CDRGES does nothing. NSIZES >= 0. NN (input) INTEGER array, dimension (NSIZES) An array containing the sizes to be used for the matrices. Zero values will be skipped. NN >= 0. NTYPES (input) INTEGER The number of elements in DOTYPE. If it is zero, CDRGEV does nothing. It must be at least zero. If it is MAXTYP+1 and NSIZES is 1, then an additional type, MAXTYP+1 is defined, which is to use whatever matrix is in A. This is only useful if DOTYPE(1:MAXTYP) is .FALSE. and DOTYPE(MAXTYP+1) is .TRUE. . DOTYPE (input) LOGICAL array, dimension (NTYPES) If DOTYPE(j) is .TRUE., then for each size in NN a matrix of that size and of type j will be generated. If NTYPES is smaller than the maximum number of types defined (PARAMETER MAXTYP), then types NTYPES+1 through MAXTYP will not be generated. If NTYPES is larger than MAXTYP, DOTYPE(MAXTYP+1) through DOTYPE(NTYPES) will be ignored. ISEED (input/output) INTEGER array, dimension (4) On entry ISEED specifies the seed of the random number generator. The array elements should be between 0 and 4095; if not they will be reduced mod 4096. Also, ISEED(4) must be odd. The random number generator uses a linear congruential sequence limited to small integers, and so should produce machine independent random numbers. The values of ISEED are changed on exit, and can be used in the next call to CDRGES to continue the same random number sequence. THRESH (input) REAL A test will count as "failed" if the "error", computed as described above, exceeds THRESH. Note that the error is scaled to be O(1), so THRESH should be a reasonably small multiple of 1, e.g., 10 or 100. In particular, it should not depend on the precision (single vs. double) or the size of the matrix. It must be at least zero. NOUNIT (input) INTEGER The FORTRAN unit number for printing out error messages (e.g., if a routine returns IERR not equal to 0.) A (input/workspace) COMPLEX array, dimension(LDA, max(NN)) Used to hold the original A matrix. Used as input only if NTYPES=MAXTYP+1, DOTYPE(1:MAXTYP)=.FALSE., and DOTYPE(MAXTYP+1)=.TRUE. LDA (input) INTEGER The leading dimension of A, B, S, and T. It must be at least 1 and at least max( NN ). B (input/workspace) COMPLEX array, dimension(LDA, max(NN)) Used to hold the original B matrix. Used as input only if NTYPES=MAXTYP+1, DOTYPE(1:MAXTYP)=.FALSE., and DOTYPE(MAXTYP+1)=.TRUE. S (workspace) COMPLEX array, dimension (LDA, max(NN)) The Schur form matrix computed from A by CGGEV. On exit, S contains the Schur form matrix corresponding to the matrix in A. T (workspace) COMPLEX array, dimension (LDA, max(NN)) The upper triangular matrix computed from B by CGGEV. Q (workspace) COMPLEX array, dimension (LDQ, max(NN)) The (left) eigenvectors matrix computed by CGGEV. LDQ (input) INTEGER The leading dimension of Q and Z. It must be at least 1 and at least max( NN ). Z (workspace) COMPLEX array, dimension( LDQ, max(NN) ) The (right) orthogonal matrix computed by CGGEV. QE (workspace) COMPLEX array, dimension( LDQ, max(NN) ) QE holds the computed right or left eigenvectors. LDQE (input) INTEGER The leading dimension of QE. LDQE >= max(1,max(NN)). ALPHA (workspace) COMPLEX array, dimension (max(NN)) BETA (workspace) COMPLEX array, dimension (max(NN)) The generalized eigenvalues of (A,B) computed by CGGEV. ( ALPHAR(k)+ALPHAI(k)*i ) / BETA(k) is the k-th generalized eigenvalue of A and B. ALPHA1 (workspace) COMPLEX array, dimension (max(NN)) BETA1 (workspace) COMPLEX array, dimension (max(NN)) Like ALPHAR, ALPHAI, BETA, these arrays contain the eigenvalues of A and B, but those computed when CGGEV only computes a partial eigendecomposition, i.e. not the eigenvalues and left and right eigenvectors. WORK (workspace) COMPLEX array, dimension (LWORK) LWORK (input) INTEGER The number of entries in WORK. LWORK >= N*(N+1) RWORK (workspace) REAL array, dimension (8*N) Real workspace. RESULT (output) REAL array, dimension (2) The values computed by the tests described above. The values are currently limited to 1/ulp, to avoid overflow. INFO (output) INTEGER = 0: successful exit < 0: if INFO = -i, the i-th argument had an illegal value. > 0: A routine returned an error code. INFO is the absolute value of the INFO value returned. ===================================================================== Parameter adjustments */ --nn; --dotype; --iseed; t_dim1 = *lda; t_offset = 1 + t_dim1 * 1; t -= t_offset; s_dim1 = *lda; s_offset = 1 + s_dim1 * 1; s -= s_offset; b_dim1 = *lda; b_offset = 1 + b_dim1 * 1; b -= b_offset; a_dim1 = *lda; a_offset = 1 + a_dim1 * 1; a -= a_offset; z_dim1 = *ldq; z_offset = 1 + z_dim1 * 1; z__ -= z_offset; q_dim1 = *ldq; q_offset = 1 + q_dim1 * 1; q -= q_offset; qe_dim1 = *ldqe; qe_offset = 1 + qe_dim1 * 1; qe -= qe_offset; --alpha; --beta; --alpha1; --beta1; --work; --rwork; --result; /* Function Body Check for errors */ *info = 0; badnn = FALSE_; nmax = 1; i__1 = *nsizes; for (j = 1; j <= i__1; ++j) { /* Computing MAX */ i__2 = nmax, i__3 = nn[j]; nmax = max(i__2,i__3); if (nn[j] < 0) { badnn = TRUE_; } /* L10: */ } if (*nsizes < 0) { *info = -1; } else if (badnn) { *info = -2; } else if (*ntypes < 0) { *info = -3; } else if (*thresh < 0.f) { *info = -6; } else if (*lda <= 1 || *lda < nmax) { *info = -9; } else if (*ldq <= 1 || *ldq < nmax) { *info = -14; } else if (*ldqe <= 1 || *ldqe < nmax) { *info = -17; } /* Compute workspace (Note: Comments in the code beginning "Workspace:" describe the minimal amount of workspace needed at that point in the code, as well as the preferred amount for good performance. NB refers to the optimal block size for the immediately following subroutine, as returned by ILAENV. */ minwrk = 1; if (*info == 0 && *lwork >= 1) { minwrk = nmax * (nmax + 1); /* Computing MAX */ i__1 = 1, i__2 = ilaenv_(&c__1, "CGEQRF", " ", &nmax, &nmax, &c_n1, & c_n1, (ftnlen)6, (ftnlen)1), i__1 = max(i__1,i__2), i__2 = ilaenv_(&c__1, "CUNMQR", "LC", &nmax, &nmax, &nmax, &c_n1, ( ftnlen)6, (ftnlen)2), i__1 = max(i__1,i__2), i__2 = ilaenv_(& c__1, "CUNGQR", " ", &nmax, &nmax, &nmax, &c_n1, (ftnlen)6, ( ftnlen)1); nb = max(i__1,i__2); /* Computing MAX */ i__1 = nmax << 1, i__2 = nmax * (nb + 1), i__1 = max(i__1,i__2), i__2 = nmax * (nmax + 1); maxwrk = max(i__1,i__2); work[1].r = (real) maxwrk, work[1].i = 0.f; } if (*lwork < minwrk) { *info = -23; } if (*info != 0) { i__1 = -(*info); xerbla_("CDRGEV", &i__1); return 0; } /* Quick return if possible */ if (*nsizes == 0 || *ntypes == 0) { return 0; } ulp = slamch_("Precision"); safmin = slamch_("Safe minimum"); safmin /= ulp; safmax = 1.f / safmin; slabad_(&safmin, &safmax); ulpinv = 1.f / ulp; /* The values RMAGN(2:3) depend on N, see below. */ rmagn[0] = 0.f; rmagn[1] = 1.f; /* Loop over sizes, types */ ntestt = 0; nerrs = 0; nmats = 0; i__1 = *nsizes; for (jsize = 1; jsize <= i__1; ++jsize) { n = nn[jsize]; n1 = max(1,n); rmagn[2] = safmax * ulp / (real) n1; rmagn[3] = safmin * ulpinv * n1; if (*nsizes != 1) { mtypes = min(26,*ntypes); } else { mtypes = min(27,*ntypes); } i__2 = mtypes; for (jtype = 1; jtype <= i__2; ++jtype) { if (! dotype[jtype]) { goto L210; } ++nmats; /* Save ISEED in case of an error. */ for (j = 1; j <= 4; ++j) { ioldsd[j - 1] = iseed[j]; /* L20: */ } /* Generate test matrices A and B Description of control parameters: KCLASS: =1 means w/o rotation, =2 means w/ rotation, =3 means random. KATYPE: the "type" to be passed to CLATM4 for computing A. KAZERO: the pattern of zeros on the diagonal for A: =1: ( xxx ), =2: (0, xxx ) =3: ( 0, 0, xxx, 0 ), =4: ( 0, xxx, 0, 0 ), =5: ( 0, 0, 1, xxx, 0 ), =6: ( 0, 1, 0, xxx, 0 ). (xxx means a string of non-zero entries.) KAMAGN: the magnitude of the matrix: =0: zero, =1: O(1), =2: large, =3: small. LASIGN: .TRUE. if the diagonal elements of A are to be multiplied by a random magnitude 1 number. KBTYPE, KBZERO, KBMAGN, LBSIGN: the same, but for B. KTRIAN: =0: don't fill in the upper triangle, =1: do. KZ1, KZ2, KADD: used to implement KAZERO and KBZERO. RMAGN: used to implement KAMAGN and KBMAGN. */ if (mtypes > 26) { goto L100; } ierr = 0; if (kclass[jtype - 1] < 3) { /* Generate A (w/o rotation) */ if ((i__3 = katype[jtype - 1], abs(i__3)) == 3) { in = ((n - 1) / 2 << 1) + 1; if (in != n) { claset_("Full", &n, &n, &c_b1, &c_b1, &a[a_offset], lda); } } else { in = n; } clatm4_(&katype[jtype - 1], &in, &kz1[kazero[jtype - 1] - 1], &kz2[kazero[jtype - 1] - 1], &lasign[jtype - 1], & rmagn[kamagn[jtype - 1]], &ulp, &rmagn[ktrian[jtype - 1] * kamagn[jtype - 1]], &c__2, &iseed[1], &a[ a_offset], lda); iadd = kadd[kazero[jtype - 1] - 1]; if (iadd > 0 && iadd <= n) { i__3 = a_subscr(iadd, iadd); i__4 = kamagn[jtype - 1]; a[i__3].r = rmagn[i__4], a[i__3].i = 0.f; } /* Generate B (w/o rotation) */ if ((i__3 = kbtype[jtype - 1], abs(i__3)) == 3) { in = ((n - 1) / 2 << 1) + 1; if (in != n) { claset_("Full", &n, &n, &c_b1, &c_b1, &b[b_offset], lda); } } else { in = n; } clatm4_(&kbtype[jtype - 1], &in, &kz1[kbzero[jtype - 1] - 1], &kz2[kbzero[jtype - 1] - 1], &lbsign[jtype - 1], & rmagn[kbmagn[jtype - 1]], &c_b28, &rmagn[ktrian[jtype - 1] * kbmagn[jtype - 1]], &c__2, &iseed[1], &b[ b_offset], lda); iadd = kadd[kbzero[jtype - 1] - 1]; if (iadd != 0 && iadd <= n) { i__3 = b_subscr(iadd, iadd); i__4 = kbmagn[jtype - 1]; b[i__3].r = rmagn[i__4], b[i__3].i = 0.f; } if (kclass[jtype - 1] == 2 && n > 0) { /* Include rotations Generate Q, Z as Householder transformations times a diagonal matrix. */ i__3 = n - 1; for (jc = 1; jc <= i__3; ++jc) { i__4 = n; for (jr = jc; jr <= i__4; ++jr) { i__5 = q_subscr(jr, jc); clarnd_(&q__1, &c__3, &iseed[1]); q[i__5].r = q__1.r, q[i__5].i = q__1.i; i__5 = z___subscr(jr, jc); clarnd_(&q__1, &c__3, &iseed[1]); z__[i__5].r = q__1.r, z__[i__5].i = q__1.i; /* L30: */ } i__4 = n + 1 - jc; clarfg_(&i__4, &q_ref(jc, jc), &q_ref(jc + 1, jc), & c__1, &work[jc]); i__4 = (n << 1) + jc; i__5 = q_subscr(jc, jc); r__2 = q[i__5].r; r__1 = r_sign(&c_b28, &r__2); work[i__4].r = r__1, work[i__4].i = 0.f; i__4 = q_subscr(jc, jc); q[i__4].r = 1.f, q[i__4].i = 0.f; i__4 = n + 1 - jc; clarfg_(&i__4, &z___ref(jc, jc), &z___ref(jc + 1, jc), &c__1, &work[n + jc]); i__4 = n * 3 + jc; i__5 = z___subscr(jc, jc); r__2 = z__[i__5].r; r__1 = r_sign(&c_b28, &r__2); work[i__4].r = r__1, work[i__4].i = 0.f; i__4 = z___subscr(jc, jc); z__[i__4].r = 1.f, z__[i__4].i = 0.f; /* L40: */ } clarnd_(&q__1, &c__3, &iseed[1]); ctemp.r = q__1.r, ctemp.i = q__1.i; i__3 = q_subscr(n, n); q[i__3].r = 1.f, q[i__3].i = 0.f; i__3 = n; work[i__3].r = 0.f, work[i__3].i = 0.f; i__3 = n * 3; r__1 = c_abs(&ctemp); q__1.r = ctemp.r / r__1, q__1.i = ctemp.i / r__1; work[i__3].r = q__1.r, work[i__3].i = q__1.i; clarnd_(&q__1, &c__3, &iseed[1]); ctemp.r = q__1.r, ctemp.i = q__1.i; i__3 = z___subscr(n, n); z__[i__3].r = 1.f, z__[i__3].i = 0.f; i__3 = n << 1; work[i__3].r = 0.f, work[i__3].i = 0.f; i__3 = n << 2; r__1 = c_abs(&ctemp); q__1.r = ctemp.r / r__1, q__1.i = ctemp.i / r__1; work[i__3].r = q__1.r, work[i__3].i = q__1.i; /* Apply the diagonal matrices */ i__3 = n; for (jc = 1; jc <= i__3; ++jc) { i__4 = n; for (jr = 1; jr <= i__4; ++jr) { i__5 = a_subscr(jr, jc); i__6 = (n << 1) + jr; r_cnjg(&q__3, &work[n * 3 + jc]); q__2.r = work[i__6].r * q__3.r - work[i__6].i * q__3.i, q__2.i = work[i__6].r * q__3.i + work[i__6].i * q__3.r; i__7 = a_subscr(jr, jc); q__1.r = q__2.r * a[i__7].r - q__2.i * a[i__7].i, q__1.i = q__2.r * a[i__7].i + q__2.i * a[ i__7].r; a[i__5].r = q__1.r, a[i__5].i = q__1.i; i__5 = b_subscr(jr, jc); i__6 = (n << 1) + jr; r_cnjg(&q__3, &work[n * 3 + jc]); q__2.r = work[i__6].r * q__3.r - work[i__6].i * q__3.i, q__2.i = work[i__6].r * q__3.i + work[i__6].i * q__3.r; i__7 = b_subscr(jr, jc); q__1.r = q__2.r * b[i__7].r - q__2.i * b[i__7].i, q__1.i = q__2.r * b[i__7].i + q__2.i * b[ i__7].r; b[i__5].r = q__1.r, b[i__5].i = q__1.i; /* L50: */ } /* L60: */ } i__3 = n - 1; cunm2r_("L", "N", &n, &n, &i__3, &q[q_offset], ldq, &work[ 1], &a[a_offset], lda, &work[(n << 1) + 1], &ierr); if (ierr != 0) { goto L90; } i__3 = n - 1; cunm2r_("R", "C", &n, &n, &i__3, &z__[z_offset], ldq, & work[n + 1], &a[a_offset], lda, &work[(n << 1) + 1], &ierr); if (ierr != 0) { goto L90; } i__3 = n - 1; cunm2r_("L", "N", &n, &n, &i__3, &q[q_offset], ldq, &work[ 1], &b[b_offset], lda, &work[(n << 1) + 1], &ierr); if (ierr != 0) { goto L90; } i__3 = n - 1; cunm2r_("R", "C", &n, &n, &i__3, &z__[z_offset], ldq, & work[n + 1], &b[b_offset], lda, &work[(n << 1) + 1], &ierr); if (ierr != 0) { goto L90; } } } else { /* Random matrices */ i__3 = n; for (jc = 1; jc <= i__3; ++jc) { i__4 = n; for (jr = 1; jr <= i__4; ++jr) { i__5 = a_subscr(jr, jc); i__6 = kamagn[jtype - 1]; clarnd_(&q__2, &c__4, &iseed[1]); q__1.r = rmagn[i__6] * q__2.r, q__1.i = rmagn[i__6] * q__2.i; a[i__5].r = q__1.r, a[i__5].i = q__1.i; i__5 = b_subscr(jr, jc); i__6 = kbmagn[jtype - 1]; clarnd_(&q__2, &c__4, &iseed[1]); q__1.r = rmagn[i__6] * q__2.r, q__1.i = rmagn[i__6] * q__2.i; b[i__5].r = q__1.r, b[i__5].i = q__1.i; /* L70: */ } /* L80: */ } } L90: if (ierr != 0) { io___40.ciunit = *nounit; s_wsfe(&io___40); do_fio(&c__1, "Generator", (ftnlen)9); do_fio(&c__1, (char *)&ierr, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)); do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof(integer)); e_wsfe(); *info = abs(ierr); return 0; } L100: for (i__ = 1; i__ <= 7; ++i__) { result[i__] = -1.f; /* L110: */ } /* Call CGGEV to compute eigenvalues and eigenvectors. */ clacpy_(" ", &n, &n, &a[a_offset], lda, &s[s_offset], lda); clacpy_(" ", &n, &n, &b[b_offset], lda, &t[t_offset], lda); cggev_("V", "V", &n, &s[s_offset], lda, &t[t_offset], lda, &alpha[ 1], &beta[1], &q[q_offset], ldq, &z__[z_offset], ldq, & work[1], lwork, &rwork[1], &ierr); if (ierr != 0 && ierr != n + 1) { result[1] = ulpinv; io___42.ciunit = *nounit; s_wsfe(&io___42); do_fio(&c__1, "CGGEV1", (ftnlen)6); do_fio(&c__1, (char *)&ierr, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)); do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof(integer)); e_wsfe(); *info = abs(ierr); goto L190; } /* Do the tests (1) and (2) */ cget52_(&c_true, &n, &a[a_offset], lda, &b[b_offset], lda, &q[ q_offset], ldq, &alpha[1], &beta[1], &work[1], &rwork[1], &result[1]); if (result[2] > *thresh) { io___43.ciunit = *nounit; s_wsfe(&io___43); do_fio(&c__1, "Left", (ftnlen)4); do_fio(&c__1, "CGGEV1", (ftnlen)6); do_fio(&c__1, (char *)&result[2], (ftnlen)sizeof(real)); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)); do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof(integer)); e_wsfe(); } /* Do the tests (3) and (4) */ cget52_(&c_false, &n, &a[a_offset], lda, &b[b_offset], lda, &z__[ z_offset], ldq, &alpha[1], &beta[1], &work[1], &rwork[1], &result[3]); if (result[4] > *thresh) { io___44.ciunit = *nounit; s_wsfe(&io___44); do_fio(&c__1, "Right", (ftnlen)5); do_fio(&c__1, "CGGEV1", (ftnlen)6); do_fio(&c__1, (char *)&result[4], (ftnlen)sizeof(real)); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)); do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof(integer)); e_wsfe(); } /* Do test (5) */ clacpy_(" ", &n, &n, &a[a_offset], lda, &s[s_offset], lda); clacpy_(" ", &n, &n, &b[b_offset], lda, &t[t_offset], lda); cggev_("N", "N", &n, &s[s_offset], lda, &t[t_offset], lda, & alpha1[1], &beta1[1], &q[q_offset], ldq, &z__[z_offset], ldq, &work[1], lwork, &rwork[1], &ierr); if (ierr != 0 && ierr != n + 1) { result[1] = ulpinv; io___45.ciunit = *nounit; s_wsfe(&io___45); do_fio(&c__1, "CGGEV2", (ftnlen)6); do_fio(&c__1, (char *)&ierr, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)); do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof(integer)); e_wsfe(); *info = abs(ierr); goto L190; } i__3 = n; for (j = 1; j <= i__3; ++j) { i__4 = j; i__5 = j; i__6 = j; i__7 = j; if (alpha[i__4].r != alpha1[i__5].r || alpha[i__4].i != alpha1[i__5].i || (beta[i__6].r != beta1[i__7].r || beta[i__6].i != beta1[i__7].i)) { result[5] = ulpinv; } /* L120: */ } /* Do test (6): Compute eigenvalues and left eigenvectors, and test them */ clacpy_(" ", &n, &n, &a[a_offset], lda, &s[s_offset], lda); clacpy_(" ", &n, &n, &b[b_offset], lda, &t[t_offset], lda); cggev_("V", "N", &n, &s[s_offset], lda, &t[t_offset], lda, & alpha1[1], &beta1[1], &qe[qe_offset], ldqe, &z__[z_offset] , ldq, &work[1], lwork, &rwork[1], &ierr); if (ierr != 0 && ierr != n + 1) { result[1] = ulpinv; io___46.ciunit = *nounit; s_wsfe(&io___46); do_fio(&c__1, "CGGEV3", (ftnlen)6); do_fio(&c__1, (char *)&ierr, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)); do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof(integer)); e_wsfe(); *info = abs(ierr); goto L190; } i__3 = n; for (j = 1; j <= i__3; ++j) { i__4 = j; i__5 = j; i__6 = j; i__7 = j; if (alpha[i__4].r != alpha1[i__5].r || alpha[i__4].i != alpha1[i__5].i || (beta[i__6].r != beta1[i__7].r || beta[i__6].i != beta1[i__7].i)) { result[6] = ulpinv; } /* L130: */ } i__3 = n; for (j = 1; j <= i__3; ++j) { i__4 = n; for (jc = 1; jc <= i__4; ++jc) { i__5 = q_subscr(j, jc); i__6 = qe_subscr(j, jc); if (q[i__5].r != qe[i__6].r || q[i__5].i != qe[i__6].i) { result[6] = ulpinv; } /* L140: */ } /* L150: */ } /* Do test (7): Compute eigenvalues and right eigenvectors, and test them */ clacpy_(" ", &n, &n, &a[a_offset], lda, &s[s_offset], lda); clacpy_(" ", &n, &n, &b[b_offset], lda, &t[t_offset], lda); cggev_("N", "V", &n, &s[s_offset], lda, &t[t_offset], lda, & alpha1[1], &beta1[1], &q[q_offset], ldq, &qe[qe_offset], ldqe, &work[1], lwork, &rwork[1], &ierr); if (ierr != 0 && ierr != n + 1) { result[1] = ulpinv; io___47.ciunit = *nounit; s_wsfe(&io___47); do_fio(&c__1, "CGGEV4", (ftnlen)6); do_fio(&c__1, (char *)&ierr, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)); do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof(integer)); e_wsfe(); *info = abs(ierr); goto L190; } i__3 = n; for (j = 1; j <= i__3; ++j) { i__4 = j; i__5 = j; i__6 = j; i__7 = j; if (alpha[i__4].r != alpha1[i__5].r || alpha[i__4].i != alpha1[i__5].i || (beta[i__6].r != beta1[i__7].r || beta[i__6].i != beta1[i__7].i)) { result[7] = ulpinv; } /* L160: */ } i__3 = n; for (j = 1; j <= i__3; ++j) { i__4 = n; for (jc = 1; jc <= i__4; ++jc) { i__5 = z___subscr(j, jc); i__6 = qe_subscr(j, jc); if (z__[i__5].r != qe[i__6].r || z__[i__5].i != qe[i__6] .i) { result[7] = ulpinv; } /* L170: */ } /* L180: */ } /* End of Loop -- Check for RESULT(j) > THRESH */ L190: ntestt += 7; /* Print out tests which fail. */ for (jr = 1; jr <= 9; ++jr) { if (result[jr] >= *thresh) { /* If this is the first test to fail, print a header to the data file. */ if (nerrs == 0) { io___48.ciunit = *nounit; s_wsfe(&io___48); do_fio(&c__1, "CGV", (ftnlen)3); e_wsfe(); /* Matrix types */ io___49.ciunit = *nounit; s_wsfe(&io___49); e_wsfe(); io___50.ciunit = *nounit; s_wsfe(&io___50); e_wsfe(); io___51.ciunit = *nounit; s_wsfe(&io___51); do_fio(&c__1, "Orthogonal", (ftnlen)10); e_wsfe(); /* Tests performed */ io___52.ciunit = *nounit; s_wsfe(&io___52); e_wsfe(); } ++nerrs; if (result[jr] < 1e4f) { io___53.ciunit = *nounit; s_wsfe(&io___53); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)) ; do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof( integer)); do_fio(&c__1, (char *)&jr, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&result[jr], (ftnlen)sizeof( real)); e_wsfe(); } else { io___54.ciunit = *nounit; s_wsfe(&io___54); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)) ; do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof( integer)); do_fio(&c__1, (char *)&jr, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&result[jr], (ftnlen)sizeof( real)); e_wsfe(); } } /* L200: */ } L210: ; } /* L220: */ } /* Summary */ alasvm_("CGV", nounit, &nerrs, &ntestt, &c__0); work[1].r = (real) maxwrk, work[1].i = 0.f; return 0; /* End of CDRGEV */ } /* cdrgev_ */
/* Subroutine */ int chbev_(char *jobz, char *uplo, integer *n, integer *kd, complex *ab, integer *ldab, real *w, complex *z__, integer *ldz, complex *work, real *rwork, integer *info) { /* -- LAPACK driver routine (version 3.0) -- Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., Courant Institute, Argonne National Lab, and Rice University June 30, 1999 Purpose ======= CHBEV computes all the eigenvalues and, optionally, eigenvectors of a complex Hermitian band matrix A. Arguments ========= JOBZ (input) CHARACTER*1 = 'N': Compute eigenvalues only; = 'V': Compute eigenvalues and eigenvectors. UPLO (input) CHARACTER*1 = 'U': Upper triangle of A is stored; = 'L': Lower triangle of A is stored. N (input) INTEGER The order of the matrix A. N >= 0. KD (input) INTEGER The number of superdiagonals of the matrix A if UPLO = 'U', or the number of subdiagonals if UPLO = 'L'. KD >= 0. AB (input/output) COMPLEX array, dimension (LDAB, N) On entry, the upper or lower triangle of the Hermitian band matrix A, stored in the first KD+1 rows of the array. The j-th column of A is stored in the j-th column of the array AB as follows: if UPLO = 'U', AB(kd+1+i-j,j) = A(i,j) for max(1,j-kd)<=i<=j; if UPLO = 'L', AB(1+i-j,j) = A(i,j) for j<=i<=min(n,j+kd). On exit, AB is overwritten by values generated during the reduction to tridiagonal form. If UPLO = 'U', the first superdiagonal and the diagonal of the tridiagonal matrix T are returned in rows KD and KD+1 of AB, and if UPLO = 'L', the diagonal and first subdiagonal of T are returned in the first two rows of AB. LDAB (input) INTEGER The leading dimension of the array AB. LDAB >= KD + 1. W (output) REAL array, dimension (N) If INFO = 0, the eigenvalues in ascending order. Z (output) COMPLEX array, dimension (LDZ, N) If JOBZ = 'V', then if INFO = 0, Z contains the orthonormal eigenvectors of the matrix A, with the i-th column of Z holding the eigenvector associated with W(i). If JOBZ = 'N', then Z is not referenced. LDZ (input) INTEGER The leading dimension of the array Z. LDZ >= 1, and if JOBZ = 'V', LDZ >= max(1,N). WORK (workspace) COMPLEX array, dimension (N) RWORK (workspace) REAL array, dimension (max(1,3*N-2)) INFO (output) INTEGER = 0: successful exit. < 0: if INFO = -i, the i-th argument had an illegal value. > 0: if INFO = i, the algorithm failed to converge; i off-diagonal elements of an intermediate tridiagonal form did not converge to zero. ===================================================================== Test the input parameters. Parameter adjustments */ /* Table of constant values */ static real c_b11 = 1.f; static integer c__1 = 1; /* System generated locals */ integer ab_dim1, ab_offset, z_dim1, z_offset, i__1; real r__1; /* Builtin functions */ double sqrt(doublereal); /* Local variables */ static integer inde; static real anrm; static integer imax; static real rmin, rmax, sigma; extern logical lsame_(char *, char *); static integer iinfo; extern /* Subroutine */ int sscal_(integer *, real *, real *, integer *); static logical lower, wantz; extern doublereal clanhb_(char *, char *, integer *, integer *, complex *, integer *, real *); static integer iscale; extern /* Subroutine */ int clascl_(char *, integer *, integer *, real *, real *, integer *, integer *, complex *, integer *, integer *), chbtrd_(char *, char *, integer *, integer *, complex *, integer *, real *, real *, complex *, integer *, complex *, integer *); extern doublereal slamch_(char *); static real safmin; extern /* Subroutine */ int xerbla_(char *, integer *); static real bignum; static integer indrwk; extern /* Subroutine */ int csteqr_(char *, integer *, real *, real *, complex *, integer *, real *, integer *), ssterf_(integer *, real *, real *, integer *); static real smlnum, eps; #define z___subscr(a_1,a_2) (a_2)*z_dim1 + a_1 #define z___ref(a_1,a_2) z__[z___subscr(a_1,a_2)] #define ab_subscr(a_1,a_2) (a_2)*ab_dim1 + a_1 #define ab_ref(a_1,a_2) ab[ab_subscr(a_1,a_2)] ab_dim1 = *ldab; ab_offset = 1 + ab_dim1 * 1; ab -= ab_offset; --w; z_dim1 = *ldz; z_offset = 1 + z_dim1 * 1; z__ -= z_offset; --work; --rwork; /* Function Body */ wantz = lsame_(jobz, "V"); lower = lsame_(uplo, "L"); *info = 0; if (! (wantz || lsame_(jobz, "N"))) { *info = -1; } else if (! (lower || lsame_(uplo, "U"))) { *info = -2; } else if (*n < 0) { *info = -3; } else if (*kd < 0) { *info = -4; } else if (*ldab < *kd + 1) { *info = -6; } else if (*ldz < 1 || wantz && *ldz < *n) { *info = -9; } if (*info != 0) { i__1 = -(*info); xerbla_("CHBEV ", &i__1); return 0; } /* Quick return if possible */ if (*n == 0) { return 0; } if (*n == 1) { if (lower) { i__1 = ab_subscr(1, 1); w[1] = ab[i__1].r; } else { i__1 = ab_subscr(*kd + 1, 1); w[1] = ab[i__1].r; } if (wantz) { i__1 = z___subscr(1, 1); z__[i__1].r = 1.f, z__[i__1].i = 0.f; } return 0; } /* Get machine constants. */ safmin = slamch_("Safe minimum"); eps = slamch_("Precision"); smlnum = safmin / eps; bignum = 1.f / smlnum; rmin = sqrt(smlnum); rmax = sqrt(bignum); /* Scale matrix to allowable range, if necessary. */ anrm = clanhb_("M", uplo, n, kd, &ab[ab_offset], ldab, &rwork[1]); iscale = 0; if (anrm > 0.f && anrm < rmin) { iscale = 1; sigma = rmin / anrm; } else if (anrm > rmax) { iscale = 1; sigma = rmax / anrm; } if (iscale == 1) { if (lower) { clascl_("B", kd, kd, &c_b11, &sigma, n, n, &ab[ab_offset], ldab, info); } else { clascl_("Q", kd, kd, &c_b11, &sigma, n, n, &ab[ab_offset], ldab, info); } } /* Call CHBTRD to reduce Hermitian band matrix to tridiagonal form. */ inde = 1; chbtrd_(jobz, uplo, n, kd, &ab[ab_offset], ldab, &w[1], &rwork[inde], & z__[z_offset], ldz, &work[1], &iinfo); /* For eigenvalues only, call SSTERF. For eigenvectors, call CSTEQR. */ if (! wantz) { ssterf_(n, &w[1], &rwork[inde], info); } else { indrwk = inde + *n; csteqr_(jobz, n, &w[1], &rwork[inde], &z__[z_offset], ldz, &rwork[ indrwk], info); } /* If matrix was scaled, then rescale eigenvalues appropriately. */ if (iscale == 1) { if (*info == 0) { imax = *n; } else { imax = *info - 1; } r__1 = 1.f / sigma; sscal_(&imax, &r__1, &w[1], &c__1); } return 0; /* End of CHBEV */ } /* chbev_ */
/* Subroutine */ int zhpev_(char *jobz, char *uplo, integer *n, doublecomplex *ap, doublereal *w, doublecomplex *z__, integer *ldz, doublecomplex * work, doublereal *rwork, integer *info) { /* -- LAPACK driver routine (version 3.0) -- Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., Courant Institute, Argonne National Lab, and Rice University March 31, 1993 Purpose ======= ZHPEV computes all the eigenvalues and, optionally, eigenvectors of a complex Hermitian matrix in packed storage. Arguments ========= JOBZ (input) CHARACTER*1 = 'N': Compute eigenvalues only; = 'V': Compute eigenvalues and eigenvectors. UPLO (input) CHARACTER*1 = 'U': Upper triangle of A is stored; = 'L': Lower triangle of A is stored. N (input) INTEGER The order of the matrix A. N >= 0. AP (input/output) COMPLEX*16 array, dimension (N*(N+1)/2) On entry, the upper or lower triangle of the Hermitian matrix A, packed columnwise in a linear array. The j-th column of A is stored in the array AP as follows: if UPLO = 'U', AP(i + (j-1)*j/2) = A(i,j) for 1<=i<=j; if UPLO = 'L', AP(i + (j-1)*(2*n-j)/2) = A(i,j) for j<=i<=n. On exit, AP is overwritten by values generated during the reduction to tridiagonal form. If UPLO = 'U', the diagonal and first superdiagonal of the tridiagonal matrix T overwrite the corresponding elements of A, and if UPLO = 'L', the diagonal and first subdiagonal of T overwrite the corresponding elements of A. W (output) DOUBLE PRECISION array, dimension (N) If INFO = 0, the eigenvalues in ascending order. Z (output) COMPLEX*16 array, dimension (LDZ, N) If JOBZ = 'V', then if INFO = 0, Z contains the orthonormal eigenvectors of the matrix A, with the i-th column of Z holding the eigenvector associated with W(i). If JOBZ = 'N', then Z is not referenced. LDZ (input) INTEGER The leading dimension of the array Z. LDZ >= 1, and if JOBZ = 'V', LDZ >= max(1,N). WORK (workspace) COMPLEX*16 array, dimension (max(1, 2*N-1)) RWORK (workspace) DOUBLE PRECISION array, dimension (max(1, 3*N-2)) INFO (output) INTEGER = 0: successful exit. < 0: if INFO = -i, the i-th argument had an illegal value. > 0: if INFO = i, the algorithm failed to converge; i off-diagonal elements of an intermediate tridiagonal form did not converge to zero. ===================================================================== Test the input parameters. Parameter adjustments */ /* Table of constant values */ static integer c__1 = 1; /* System generated locals */ integer z_dim1, z_offset, i__1; doublereal d__1; /* Builtin functions */ double sqrt(doublereal); /* Local variables */ static integer inde; static doublereal anrm; static integer imax; static doublereal rmin, rmax; extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *, integer *); static doublereal sigma; extern logical lsame_(char *, char *); static integer iinfo; static logical wantz; extern doublereal dlamch_(char *); static integer iscale; static doublereal safmin; extern /* Subroutine */ int xerbla_(char *, integer *), zdscal_( integer *, doublereal *, doublecomplex *, integer *); static doublereal bignum; static integer indtau; extern /* Subroutine */ int dsterf_(integer *, doublereal *, doublereal *, integer *); extern doublereal zlanhp_(char *, char *, integer *, doublecomplex *, doublereal *); static integer indrwk, indwrk; static doublereal smlnum; extern /* Subroutine */ int zhptrd_(char *, integer *, doublecomplex *, doublereal *, doublereal *, doublecomplex *, integer *), zsteqr_(char *, integer *, doublereal *, doublereal *, doublecomplex *, integer *, doublereal *, integer *), zupgtr_(char *, integer *, doublecomplex *, doublecomplex *, doublecomplex *, integer *, doublecomplex *, integer *); static doublereal eps; #define z___subscr(a_1,a_2) (a_2)*z_dim1 + a_1 #define z___ref(a_1,a_2) z__[z___subscr(a_1,a_2)] --ap; --w; z_dim1 = *ldz; z_offset = 1 + z_dim1 * 1; z__ -= z_offset; --work; --rwork; /* Function Body */ wantz = lsame_(jobz, "V"); *info = 0; if (! (wantz || lsame_(jobz, "N"))) { *info = -1; } else if (! (lsame_(uplo, "L") || lsame_(uplo, "U"))) { *info = -2; } else if (*n < 0) { *info = -3; } else if (*ldz < 1 || wantz && *ldz < *n) { *info = -7; } if (*info != 0) { i__1 = -(*info); xerbla_("ZHPEV ", &i__1); return 0; } /* Quick return if possible */ if (*n == 0) { return 0; } if (*n == 1) { w[1] = ap[1].r; rwork[1] = 1.; if (wantz) { i__1 = z___subscr(1, 1); z__[i__1].r = 1., z__[i__1].i = 0.; } return 0; } /* Get machine constants. */ safmin = dlamch_("Safe minimum"); eps = dlamch_("Precision"); smlnum = safmin / eps; bignum = 1. / smlnum; rmin = sqrt(smlnum); rmax = sqrt(bignum); /* Scale matrix to allowable range, if necessary. */ anrm = zlanhp_("M", uplo, n, &ap[1], &rwork[1]); iscale = 0; if (anrm > 0. && anrm < rmin) { iscale = 1; sigma = rmin / anrm; } else if (anrm > rmax) { iscale = 1; sigma = rmax / anrm; } if (iscale == 1) { i__1 = *n * (*n + 1) / 2; zdscal_(&i__1, &sigma, &ap[1], &c__1); } /* Call ZHPTRD to reduce Hermitian packed matrix to tridiagonal form. */ inde = 1; indtau = 1; zhptrd_(uplo, n, &ap[1], &w[1], &rwork[inde], &work[indtau], &iinfo); /* For eigenvalues only, call DSTERF. For eigenvectors, first call ZUPGTR to generate the orthogonal matrix, then call ZSTEQR. */ if (! wantz) { dsterf_(n, &w[1], &rwork[inde], info); } else { indwrk = indtau + *n; zupgtr_(uplo, n, &ap[1], &work[indtau], &z__[z_offset], ldz, &work[ indwrk], &iinfo); indrwk = inde + *n; zsteqr_(jobz, n, &w[1], &rwork[inde], &z__[z_offset], ldz, &rwork[ indrwk], info); } /* If matrix was scaled, then rescale eigenvalues appropriately. */ if (iscale == 1) { if (*info == 0) { imax = *n; } else { imax = *info - 1; } d__1 = 1. / sigma; dscal_(&imax, &d__1, &w[1], &c__1); } return 0; /* End of ZHPEV */ } /* zhpev_ */
/* Subroutine */ int zhgeqz_(char *job, char *compq, char *compz, integer *n, integer *ilo, integer *ihi, doublecomplex *a, integer *lda, doublecomplex *b, integer *ldb, doublecomplex *alpha, doublecomplex * beta, doublecomplex *q, integer *ldq, doublecomplex *z__, integer * ldz, doublecomplex *work, integer *lwork, doublereal *rwork, integer * info) { /* System generated locals */ integer a_dim1, a_offset, b_dim1, b_offset, q_dim1, q_offset, z_dim1, z_offset, i__1, i__2, i__3, i__4, i__5, i__6; doublereal d__1, d__2, d__3, d__4, d__5, d__6; doublecomplex z__1, z__2, z__3, z__4, z__5, z__6; /* Builtin functions */ double z_abs(doublecomplex *); void d_cnjg(doublecomplex *, doublecomplex *); double d_imag(doublecomplex *); void z_div(doublecomplex *, doublecomplex *, doublecomplex *), pow_zi( doublecomplex *, doublecomplex *, integer *), z_sqrt( doublecomplex *, doublecomplex *); /* Local variables */ static doublereal absb, atol, btol, temp, opst; extern /* Subroutine */ int zrot_(integer *, doublecomplex *, integer *, doublecomplex *, integer *, doublereal *, doublecomplex *); static doublereal temp2, c__; static integer j; static doublecomplex s, t; extern logical lsame_(char *, char *); static doublecomplex ctemp; static integer iiter, ilast, jiter; static doublereal anorm; static integer maxit; static doublereal bnorm; static doublecomplex shift; extern /* Subroutine */ int zscal_(integer *, doublecomplex *, doublecomplex *, integer *); static doublereal tempr; static doublecomplex ctemp2, ctemp3; static logical ilazr2; static integer jc, in; static doublereal ascale, bscale; static doublecomplex u12; extern doublereal dlamch_(char *); static integer jr, nq; static doublecomplex signbc; static integer nz; static doublereal safmin; extern /* Subroutine */ int xerbla_(char *, integer *); static doublecomplex eshift; static logical ilschr; static integer icompq, ilastm; static doublecomplex rtdisc; static integer ischur; extern doublereal zlanhs_(char *, integer *, doublecomplex *, integer *, doublereal *); static logical ilazro; static integer icompz, ifirst; extern /* Subroutine */ int zlartg_(doublecomplex *, doublecomplex *, doublereal *, doublecomplex *, doublecomplex *); static integer ifrstm; extern /* Subroutine */ int zlaset_(char *, integer *, integer *, doublecomplex *, doublecomplex *, doublecomplex *, integer *); static integer istart; static logical lquery; static doublecomplex ad11, ad12, ad21, ad22; static integer jch; static logical ilq, ilz; static doublereal ulp; static doublecomplex abi22; #define a_subscr(a_1,a_2) (a_2)*a_dim1 + a_1 #define a_ref(a_1,a_2) a[a_subscr(a_1,a_2)] #define b_subscr(a_1,a_2) (a_2)*b_dim1 + a_1 #define b_ref(a_1,a_2) b[b_subscr(a_1,a_2)] #define q_subscr(a_1,a_2) (a_2)*q_dim1 + a_1 #define q_ref(a_1,a_2) q[q_subscr(a_1,a_2)] #define z___subscr(a_1,a_2) (a_2)*z_dim1 + a_1 #define z___ref(a_1,a_2) z__[z___subscr(a_1,a_2)] /* -- LAPACK routine (instrumented to count operations, version 3.0) -- Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., Courant Institute, Argonne National Lab, and Rice University June 30, 1999 ----------------------- Begin Timing Code ------------------------ Common block to return operation count and iteration count ITCNT is initialized to 0, OPS is only incremented OPST is used to accumulate small contributions to OPS to avoid roundoff error ------------------------ End Timing Code ------------------------- Purpose ======= ZHGEQZ implements a single-shift version of the QZ method for finding the generalized eigenvalues w(i)=ALPHA(i)/BETA(i) of the equation det( A - w(i) B ) = 0 If JOB='S', then the pair (A,B) is simultaneously reduced to Schur form (i.e., A and B are both upper triangular) by applying one unitary tranformation (usually called Q) on the left and another (usually called Z) on the right. The diagonal elements of A are then ALPHA(1),...,ALPHA(N), and of B are BETA(1),...,BETA(N). If JOB='S' and COMPQ and COMPZ are 'V' or 'I', then the unitary transformations used to reduce (A,B) are accumulated into the arrays Q and Z s.t.: Q(in) A(in) Z(in)* = Q(out) A(out) Z(out)* Q(in) B(in) Z(in)* = Q(out) B(out) Z(out)* Ref: C.B. Moler & G.W. Stewart, "An Algorithm for Generalized Matrix Eigenvalue Problems", SIAM J. Numer. Anal., 10(1973), pp. 241--256. Arguments ========= JOB (input) CHARACTER*1 = 'E': compute only ALPHA and BETA. A and B will not necessarily be put into generalized Schur form. = 'S': put A and B into generalized Schur form, as well as computing ALPHA and BETA. COMPQ (input) CHARACTER*1 = 'N': do not modify Q. = 'V': multiply the array Q on the right by the conjugate transpose of the unitary tranformation that is applied to the left side of A and B to reduce them to Schur form. = 'I': like COMPQ='V', except that Q will be initialized to the identity first. COMPZ (input) CHARACTER*1 = 'N': do not modify Z. = 'V': multiply the array Z on the right by the unitary tranformation that is applied to the right side of A and B to reduce them to Schur form. = 'I': like COMPZ='V', except that Z will be initialized to the identity first. N (input) INTEGER The order of the matrices A, B, Q, and Z. N >= 0. ILO (input) INTEGER IHI (input) INTEGER It is assumed that A is already upper triangular in rows and columns 1:ILO-1 and IHI+1:N. 1 <= ILO <= IHI <= N, if N > 0; ILO=1 and IHI=0, if N=0. A (input/output) COMPLEX*16 array, dimension (LDA, N) On entry, the N-by-N upper Hessenberg matrix A. Elements below the subdiagonal must be zero. If JOB='S', then on exit A and B will have been simultaneously reduced to upper triangular form. If JOB='E', then on exit A will have been destroyed. LDA (input) INTEGER The leading dimension of the array A. LDA >= max( 1, N ). B (input/output) COMPLEX*16 array, dimension (LDB, N) On entry, the N-by-N upper triangular matrix B. Elements below the diagonal must be zero. If JOB='S', then on exit A and B will have been simultaneously reduced to upper triangular form. If JOB='E', then on exit B will have been destroyed. LDB (input) INTEGER The leading dimension of the array B. LDB >= max( 1, N ). ALPHA (output) COMPLEX*16 array, dimension (N) The diagonal elements of A when the pair (A,B) has been reduced to Schur form. ALPHA(i)/BETA(i) i=1,...,N are the generalized eigenvalues. BETA (output) COMPLEX*16 array, dimension (N) The diagonal elements of B when the pair (A,B) has been reduced to Schur form. ALPHA(i)/BETA(i) i=1,...,N are the generalized eigenvalues. A and B are normalized so that BETA(1),...,BETA(N) are non-negative real numbers. Q (input/output) COMPLEX*16 array, dimension (LDQ, N) If COMPQ='N', then Q will not be referenced. If COMPQ='V' or 'I', then the conjugate transpose of the unitary transformations which are applied to A and B on the left will be applied to the array Q on the right. LDQ (input) INTEGER The leading dimension of the array Q. LDQ >= 1. If COMPQ='V' or 'I', then LDQ >= N. Z (input/output) COMPLEX*16 array, dimension (LDZ, N) If COMPZ='N', then Z will not be referenced. If COMPZ='V' or 'I', then the unitary transformations which are applied to A and B on the right will be applied to the array Z on the right. LDZ (input) INTEGER The leading dimension of the array Z. LDZ >= 1. If COMPZ='V' or 'I', then LDZ >= N. WORK (workspace/output) COMPLEX*16 array, dimension (LWORK) On exit, if INFO >= 0, WORK(1) returns the optimal LWORK. LWORK (input) INTEGER The dimension of the array WORK. LWORK >= max(1,N). If LWORK = -1, then a workspace query is assumed; the routine only calculates the optimal size of the WORK array, returns this value as the first entry of the WORK array, and no error message related to LWORK is issued by XERBLA. RWORK (workspace) DOUBLE PRECISION array, dimension (N) INFO (output) INTEGER = 0: successful exit < 0: if INFO = -i, the i-th argument had an illegal value = 1,...,N: the QZ iteration did not converge. (A,B) is not in Schur form, but ALPHA(i) and BETA(i), i=INFO+1,...,N should be correct. = N+1,...,2*N: the shift calculation failed. (A,B) is not in Schur form, but ALPHA(i) and BETA(i), i=INFO-N+1,...,N should be correct. > 2*N: various "impossible" errors. Further Details =============== We assume that complex ABS works as long as its value is less than overflow. ===================================================================== ----------------------- Begin Timing Code ------------------------ Parameter adjustments */ a_dim1 = *lda; a_offset = 1 + a_dim1 * 1; a -= a_offset; b_dim1 = *ldb; b_offset = 1 + b_dim1 * 1; b -= b_offset; --alpha; --beta; q_dim1 = *ldq; q_offset = 1 + q_dim1 * 1; q -= q_offset; z_dim1 = *ldz; z_offset = 1 + z_dim1 * 1; z__ -= z_offset; --work; --rwork; /* Function Body */ latime_1.itcnt = 0.; /* ------------------------ End Timing Code ------------------------- Decode JOB, COMPQ, COMPZ */ if (lsame_(job, "E")) { ilschr = FALSE_; ischur = 1; } else if (lsame_(job, "S")) { ilschr = TRUE_; ischur = 2; } else { ischur = 0; } if (lsame_(compq, "N")) { ilq = FALSE_; icompq = 1; nq = 0; } else if (lsame_(compq, "V")) { ilq = TRUE_; icompq = 2; nq = *n; } else if (lsame_(compq, "I")) { ilq = TRUE_; icompq = 3; nq = *n; } else { icompq = 0; } if (lsame_(compz, "N")) { ilz = FALSE_; icompz = 1; nz = 0; } else if (lsame_(compz, "V")) { ilz = TRUE_; icompz = 2; nz = *n; } else if (lsame_(compz, "I")) { ilz = TRUE_; icompz = 3; nz = *n; } else { icompz = 0; } /* Check Argument Values */ *info = 0; i__1 = max(1,*n); work[1].r = (doublereal) i__1, work[1].i = 0.; lquery = *lwork == -1; if (ischur == 0) { *info = -1; } else if (icompq == 0) { *info = -2; } else if (icompz == 0) { *info = -3; } else if (*n < 0) { *info = -4; } else if (*ilo < 1) { *info = -5; } else if (*ihi > *n || *ihi < *ilo - 1) { *info = -6; } else if (*lda < *n) { *info = -8; } else if (*ldb < *n) { *info = -10; } else if (*ldq < 1 || ilq && *ldq < *n) { *info = -14; } else if (*ldz < 1 || ilz && *ldz < *n) { *info = -16; } else if (*lwork < max(1,*n) && ! lquery) { *info = -18; } if (*info != 0) { i__1 = -(*info); xerbla_("ZHGEQZ", &i__1); return 0; } else if (lquery) { return 0; } /* Quick return if possible WORK( 1 ) = CMPLX( 1 ) */ if (*n <= 0) { work[1].r = 1., work[1].i = 0.; return 0; } /* Initialize Q and Z */ if (icompq == 3) { zlaset_("Full", n, n, &c_b1, &c_b2, &q[q_offset], ldq); } if (icompz == 3) { zlaset_("Full", n, n, &c_b1, &c_b2, &z__[z_offset], ldz); } /* Machine Constants */ in = *ihi + 1 - *ilo; safmin = dlamch_("S"); ulp = dlamch_("E") * dlamch_("B"); anorm = zlanhs_("F", &in, &a_ref(*ilo, *ilo), lda, &rwork[1]); bnorm = zlanhs_("F", &in, &b_ref(*ilo, *ilo), ldb, &rwork[1]); /* Computing MAX */ d__1 = safmin, d__2 = ulp * anorm; atol = max(d__1,d__2); /* Computing MAX */ d__1 = safmin, d__2 = ulp * bnorm; btol = max(d__1,d__2); ascale = 1. / max(safmin,anorm); bscale = 1. / max(safmin,bnorm); /* ---------------------- Begin Timing Code ------------------------- Count ops for norms, etc. */ opst = 0.; /* Computing 2nd power */ i__1 = *n; latime_1.ops += (doublereal) ((i__1 * i__1 << 2) + *n * 12 - 5); /* ----------------------- End Timing Code -------------------------- Set Eigenvalues IHI+1:N */ i__1 = *n; for (j = *ihi + 1; j <= i__1; ++j) { absb = z_abs(&b_ref(j, j)); if (absb > safmin) { i__2 = b_subscr(j, j); z__2.r = b[i__2].r / absb, z__2.i = b[i__2].i / absb; d_cnjg(&z__1, &z__2); signbc.r = z__1.r, signbc.i = z__1.i; i__2 = b_subscr(j, j); b[i__2].r = absb, b[i__2].i = 0.; if (ilschr) { i__2 = j - 1; zscal_(&i__2, &signbc, &b_ref(1, j), &c__1); zscal_(&j, &signbc, &a_ref(1, j), &c__1); /* ----------------- Begin Timing Code --------------------- */ opst += (doublereal) ((j - 1) * 12); /* ------------------ End Timing Code ---------------------- */ } else { i__2 = a_subscr(j, j); i__3 = a_subscr(j, j); z__1.r = a[i__3].r * signbc.r - a[i__3].i * signbc.i, z__1.i = a[i__3].r * signbc.i + a[i__3].i * signbc.r; a[i__2].r = z__1.r, a[i__2].i = z__1.i; } if (ilz) { zscal_(n, &signbc, &z___ref(1, j), &c__1); } /* ------------------- Begin Timing Code ---------------------- */ opst += (doublereal) (nz * 6 + 13); /* -------------------- End Timing Code ----------------------- */ } else { i__2 = b_subscr(j, j); b[i__2].r = 0., b[i__2].i = 0.; } i__2 = j; i__3 = a_subscr(j, j); alpha[i__2].r = a[i__3].r, alpha[i__2].i = a[i__3].i; i__2 = j; i__3 = b_subscr(j, j); beta[i__2].r = b[i__3].r, beta[i__2].i = b[i__3].i; /* L10: */ } /* If IHI < ILO, skip QZ steps */ if (*ihi < *ilo) { goto L190; } /* MAIN QZ ITERATION LOOP Initialize dynamic indices Eigenvalues ILAST+1:N have been found. Column operations modify rows IFRSTM:whatever Row operations modify columns whatever:ILASTM If only eigenvalues are being computed, then IFRSTM is the row of the last splitting row above row ILAST; this is always at least ILO. IITER counts iterations since the last eigenvalue was found, to tell when to use an extraordinary shift. MAXIT is the maximum number of QZ sweeps allowed. */ ilast = *ihi; if (ilschr) { ifrstm = 1; ilastm = *n; } else { ifrstm = *ilo; ilastm = *ihi; } iiter = 0; eshift.r = 0., eshift.i = 0.; maxit = (*ihi - *ilo + 1) * 30; i__1 = maxit; for (jiter = 1; jiter <= i__1; ++jiter) { /* Check for too many iterations. */ if (jiter > maxit) { goto L180; } /* Split the matrix if possible. Two tests: 1: A(j,j-1)=0 or j=ILO 2: B(j,j)=0 Special case: j=ILAST */ if (ilast == *ilo) { goto L60; } else { i__2 = a_subscr(ilast, ilast - 1); if ((d__1 = a[i__2].r, abs(d__1)) + (d__2 = d_imag(&a_ref(ilast, ilast - 1)), abs(d__2)) <= atol) { i__2 = a_subscr(ilast, ilast - 1); a[i__2].r = 0., a[i__2].i = 0.; goto L60; } } if (z_abs(&b_ref(ilast, ilast)) <= btol) { i__2 = b_subscr(ilast, ilast); b[i__2].r = 0., b[i__2].i = 0.; goto L50; } /* General case: j<ILAST */ i__2 = *ilo; for (j = ilast - 1; j >= i__2; --j) { /* Test 1: for A(j,j-1)=0 or j=ILO */ if (j == *ilo) { ilazro = TRUE_; } else { i__3 = a_subscr(j, j - 1); if ((d__1 = a[i__3].r, abs(d__1)) + (d__2 = d_imag(&a_ref(j, j - 1)), abs(d__2)) <= atol) { i__3 = a_subscr(j, j - 1); a[i__3].r = 0., a[i__3].i = 0.; ilazro = TRUE_; } else { ilazro = FALSE_; } } /* Test 2: for B(j,j)=0 */ if (z_abs(&b_ref(j, j)) < btol) { i__3 = b_subscr(j, j); b[i__3].r = 0., b[i__3].i = 0.; /* Test 1a: Check for 2 consecutive small subdiagonals in A */ ilazr2 = FALSE_; if (! ilazro) { i__3 = a_subscr(j, j - 1); i__4 = a_subscr(j + 1, j); i__5 = a_subscr(j, j); if (((d__1 = a[i__3].r, abs(d__1)) + (d__2 = d_imag(& a_ref(j, j - 1)), abs(d__2))) * (ascale * ((d__3 = a[i__4].r, abs(d__3)) + (d__4 = d_imag(&a_ref(j + 1, j)), abs(d__4)))) <= ((d__5 = a[i__5].r, abs( d__5)) + (d__6 = d_imag(&a_ref(j, j)), abs(d__6))) * (ascale * atol)) { ilazr2 = TRUE_; } } /* If both tests pass (1 & 2), i.e., the leading diagonal element of B in the block is zero, split a 1x1 block off at the top. (I.e., at the J-th row/column) The leading diagonal element of the remainder can also be zero, so this may have to be done repeatedly. */ if (ilazro || ilazr2) { i__3 = ilast - 1; for (jch = j; jch <= i__3; ++jch) { i__4 = a_subscr(jch, jch); ctemp.r = a[i__4].r, ctemp.i = a[i__4].i; zlartg_(&ctemp, &a_ref(jch + 1, jch), &c__, &s, & a_ref(jch, jch)); i__4 = a_subscr(jch + 1, jch); a[i__4].r = 0., a[i__4].i = 0.; i__4 = ilastm - jch; zrot_(&i__4, &a_ref(jch, jch + 1), lda, &a_ref(jch + 1, jch + 1), lda, &c__, &s); i__4 = ilastm - jch; zrot_(&i__4, &b_ref(jch, jch + 1), ldb, &b_ref(jch + 1, jch + 1), ldb, &c__, &s); if (ilq) { d_cnjg(&z__1, &s); zrot_(n, &q_ref(1, jch), &c__1, &q_ref(1, jch + 1) , &c__1, &c__, &z__1); } if (ilazr2) { i__4 = a_subscr(jch, jch - 1); i__5 = a_subscr(jch, jch - 1); z__1.r = c__ * a[i__5].r, z__1.i = c__ * a[i__5] .i; a[i__4].r = z__1.r, a[i__4].i = z__1.i; } ilazr2 = FALSE_; /* --------------- Begin Timing Code ----------------- */ opst += (doublereal) ((ilastm - jch) * 40 + 32 + nq * 20); /* ---------------- End Timing Code ------------------ */ i__4 = b_subscr(jch + 1, jch + 1); if ((d__1 = b[i__4].r, abs(d__1)) + (d__2 = d_imag(& b_ref(jch + 1, jch + 1)), abs(d__2)) >= btol) { if (jch + 1 >= ilast) { goto L60; } else { ifirst = jch + 1; goto L70; } } i__4 = b_subscr(jch + 1, jch + 1); b[i__4].r = 0., b[i__4].i = 0.; /* L20: */ } goto L50; } else { /* Only test 2 passed -- chase the zero to B(ILAST,ILAST) Then process as in the case B(ILAST,ILAST)=0 */ i__3 = ilast - 1; for (jch = j; jch <= i__3; ++jch) { i__4 = b_subscr(jch, jch + 1); ctemp.r = b[i__4].r, ctemp.i = b[i__4].i; zlartg_(&ctemp, &b_ref(jch + 1, jch + 1), &c__, &s, & b_ref(jch, jch + 1)); i__4 = b_subscr(jch + 1, jch + 1); b[i__4].r = 0., b[i__4].i = 0.; if (jch < ilastm - 1) { i__4 = ilastm - jch - 1; zrot_(&i__4, &b_ref(jch, jch + 2), ldb, &b_ref( jch + 1, jch + 2), ldb, &c__, &s); } i__4 = ilastm - jch + 2; zrot_(&i__4, &a_ref(jch, jch - 1), lda, &a_ref(jch + 1, jch - 1), lda, &c__, &s); if (ilq) { d_cnjg(&z__1, &s); zrot_(n, &q_ref(1, jch), &c__1, &q_ref(1, jch + 1) , &c__1, &c__, &z__1); } i__4 = a_subscr(jch + 1, jch); ctemp.r = a[i__4].r, ctemp.i = a[i__4].i; zlartg_(&ctemp, &a_ref(jch + 1, jch - 1), &c__, &s, & a_ref(jch + 1, jch)); i__4 = a_subscr(jch + 1, jch - 1); a[i__4].r = 0., a[i__4].i = 0.; i__4 = jch + 1 - ifrstm; zrot_(&i__4, &a_ref(ifrstm, jch), &c__1, &a_ref( ifrstm, jch - 1), &c__1, &c__, &s); i__4 = jch - ifrstm; zrot_(&i__4, &b_ref(ifrstm, jch), &c__1, &b_ref( ifrstm, jch - 1), &c__1, &c__, &s); if (ilz) { zrot_(n, &z___ref(1, jch), &c__1, &z___ref(1, jch - 1), &c__1, &c__, &s); } /* L30: */ } /* ---------------- Begin Timing Code ------------------- */ opst += (doublereal) ((ilastm + 1 - ifrstm) * 40 + 64 + ( nq + nz) * 20) * (doublereal) (ilast - j); /* ----------------- End Timing Code -------------------- */ goto L50; } } else if (ilazro) { /* Only test 1 passed -- work on J:ILAST */ ifirst = j; goto L70; } /* Neither test passed -- try next J L40: */ } /* (Drop-through is "impossible") */ *info = (*n << 1) + 1; goto L210; /* B(ILAST,ILAST)=0 -- clear A(ILAST,ILAST-1) to split off a 1x1 block. */ L50: i__2 = a_subscr(ilast, ilast); ctemp.r = a[i__2].r, ctemp.i = a[i__2].i; zlartg_(&ctemp, &a_ref(ilast, ilast - 1), &c__, &s, &a_ref(ilast, ilast)); i__2 = a_subscr(ilast, ilast - 1); a[i__2].r = 0., a[i__2].i = 0.; i__2 = ilast - ifrstm; zrot_(&i__2, &a_ref(ifrstm, ilast), &c__1, &a_ref(ifrstm, ilast - 1), &c__1, &c__, &s); i__2 = ilast - ifrstm; zrot_(&i__2, &b_ref(ifrstm, ilast), &c__1, &b_ref(ifrstm, ilast - 1), &c__1, &c__, &s); if (ilz) { zrot_(n, &z___ref(1, ilast), &c__1, &z___ref(1, ilast - 1), &c__1, &c__, &s); } /* --------------------- Begin Timing Code ----------------------- */ opst += (doublereal) ((ilast - ifrstm) * 40 + 32 + nz * 20); /* ---------------------- End Timing Code ------------------------ A(ILAST,ILAST-1)=0 -- Standardize B, set ALPHA and BETA */ L60: absb = z_abs(&b_ref(ilast, ilast)); if (absb > safmin) { i__2 = b_subscr(ilast, ilast); z__2.r = b[i__2].r / absb, z__2.i = b[i__2].i / absb; d_cnjg(&z__1, &z__2); signbc.r = z__1.r, signbc.i = z__1.i; i__2 = b_subscr(ilast, ilast); b[i__2].r = absb, b[i__2].i = 0.; if (ilschr) { i__2 = ilast - ifrstm; zscal_(&i__2, &signbc, &b_ref(ifrstm, ilast), &c__1); i__2 = ilast + 1 - ifrstm; zscal_(&i__2, &signbc, &a_ref(ifrstm, ilast), &c__1); /* ----------------- Begin Timing Code --------------------- */ opst += (doublereal) ((ilast - ifrstm) * 12); /* ------------------ End Timing Code ---------------------- */ } else { i__2 = a_subscr(ilast, ilast); i__3 = a_subscr(ilast, ilast); z__1.r = a[i__3].r * signbc.r - a[i__3].i * signbc.i, z__1.i = a[i__3].r * signbc.i + a[i__3].i * signbc.r; a[i__2].r = z__1.r, a[i__2].i = z__1.i; } if (ilz) { zscal_(n, &signbc, &z___ref(1, ilast), &c__1); } /* ------------------- Begin Timing Code ---------------------- */ opst += (doublereal) (nz * 6 + 13); /* -------------------- End Timing Code ----------------------- */ } else { i__2 = b_subscr(ilast, ilast); b[i__2].r = 0., b[i__2].i = 0.; } i__2 = ilast; i__3 = a_subscr(ilast, ilast); alpha[i__2].r = a[i__3].r, alpha[i__2].i = a[i__3].i; i__2 = ilast; i__3 = b_subscr(ilast, ilast); beta[i__2].r = b[i__3].r, beta[i__2].i = b[i__3].i; /* Go to next block -- exit if finished. */ --ilast; if (ilast < *ilo) { goto L190; } /* Reset counters */ iiter = 0; eshift.r = 0., eshift.i = 0.; if (! ilschr) { ilastm = ilast; if (ifrstm > ilast) { ifrstm = *ilo; } } goto L160; /* QZ step This iteration only involves rows/columns IFIRST:ILAST. We assume IFIRST < ILAST, and that the diagonal of B is non-zero. */ L70: ++iiter; if (! ilschr) { ifrstm = ifirst; } /* Compute the Shift. At this point, IFIRST < ILAST, and the diagonal elements of B(IFIRST:ILAST,IFIRST,ILAST) are larger than BTOL (in magnitude) */ if (iiter / 10 * 10 != iiter) { /* The Wilkinson shift (AEP p.512), i.e., the eigenvalue of the bottom-right 2x2 block of A inv(B) which is nearest to the bottom-right element. We factor B as U*D, where U has unit diagonals, and compute (A*inv(D))*inv(U). */ i__2 = b_subscr(ilast - 1, ilast); z__2.r = bscale * b[i__2].r, z__2.i = bscale * b[i__2].i; i__3 = b_subscr(ilast, ilast); z__3.r = bscale * b[i__3].r, z__3.i = bscale * b[i__3].i; z_div(&z__1, &z__2, &z__3); u12.r = z__1.r, u12.i = z__1.i; i__2 = a_subscr(ilast - 1, ilast - 1); z__2.r = ascale * a[i__2].r, z__2.i = ascale * a[i__2].i; i__3 = b_subscr(ilast - 1, ilast - 1); z__3.r = bscale * b[i__3].r, z__3.i = bscale * b[i__3].i; z_div(&z__1, &z__2, &z__3); ad11.r = z__1.r, ad11.i = z__1.i; i__2 = a_subscr(ilast, ilast - 1); z__2.r = ascale * a[i__2].r, z__2.i = ascale * a[i__2].i; i__3 = b_subscr(ilast - 1, ilast - 1); z__3.r = bscale * b[i__3].r, z__3.i = bscale * b[i__3].i; z_div(&z__1, &z__2, &z__3); ad21.r = z__1.r, ad21.i = z__1.i; i__2 = a_subscr(ilast - 1, ilast); z__2.r = ascale * a[i__2].r, z__2.i = ascale * a[i__2].i; i__3 = b_subscr(ilast, ilast); z__3.r = bscale * b[i__3].r, z__3.i = bscale * b[i__3].i; z_div(&z__1, &z__2, &z__3); ad12.r = z__1.r, ad12.i = z__1.i; i__2 = a_subscr(ilast, ilast); z__2.r = ascale * a[i__2].r, z__2.i = ascale * a[i__2].i; i__3 = b_subscr(ilast, ilast); z__3.r = bscale * b[i__3].r, z__3.i = bscale * b[i__3].i; z_div(&z__1, &z__2, &z__3); ad22.r = z__1.r, ad22.i = z__1.i; z__2.r = u12.r * ad21.r - u12.i * ad21.i, z__2.i = u12.r * ad21.i + u12.i * ad21.r; z__1.r = ad22.r - z__2.r, z__1.i = ad22.i - z__2.i; abi22.r = z__1.r, abi22.i = z__1.i; z__2.r = ad11.r + abi22.r, z__2.i = ad11.i + abi22.i; z__1.r = z__2.r * .5, z__1.i = z__2.i * .5; t.r = z__1.r, t.i = z__1.i; pow_zi(&z__4, &t, &c__2); z__5.r = ad12.r * ad21.r - ad12.i * ad21.i, z__5.i = ad12.r * ad21.i + ad12.i * ad21.r; z__3.r = z__4.r + z__5.r, z__3.i = z__4.i + z__5.i; z__6.r = ad11.r * ad22.r - ad11.i * ad22.i, z__6.i = ad11.r * ad22.i + ad11.i * ad22.r; z__2.r = z__3.r - z__6.r, z__2.i = z__3.i - z__6.i; z_sqrt(&z__1, &z__2); rtdisc.r = z__1.r, rtdisc.i = z__1.i; z__1.r = t.r - abi22.r, z__1.i = t.i - abi22.i; z__2.r = t.r - abi22.r, z__2.i = t.i - abi22.i; temp = z__1.r * rtdisc.r + d_imag(&z__2) * d_imag(&rtdisc); if (temp <= 0.) { z__1.r = t.r + rtdisc.r, z__1.i = t.i + rtdisc.i; shift.r = z__1.r, shift.i = z__1.i; } else { z__1.r = t.r - rtdisc.r, z__1.i = t.i - rtdisc.i; shift.r = z__1.r, shift.i = z__1.i; } /* ------------------- Begin Timing Code ---------------------- */ opst += 116.; /* -------------------- End Timing Code ----------------------- */ } else { /* Exceptional shift. Chosen for no particularly good reason. */ i__2 = a_subscr(ilast - 1, ilast); z__4.r = ascale * a[i__2].r, z__4.i = ascale * a[i__2].i; i__3 = b_subscr(ilast - 1, ilast - 1); z__5.r = bscale * b[i__3].r, z__5.i = bscale * b[i__3].i; z_div(&z__3, &z__4, &z__5); d_cnjg(&z__2, &z__3); z__1.r = eshift.r + z__2.r, z__1.i = eshift.i + z__2.i; eshift.r = z__1.r, eshift.i = z__1.i; shift.r = eshift.r, shift.i = eshift.i; /* ------------------- Begin Timing Code ---------------------- */ opst += 15.; /* -------------------- End Timing Code ----------------------- */ } /* Now check for two consecutive small subdiagonals. */ i__2 = ifirst + 1; for (j = ilast - 1; j >= i__2; --j) { istart = j; i__3 = a_subscr(j, j); z__2.r = ascale * a[i__3].r, z__2.i = ascale * a[i__3].i; i__4 = b_subscr(j, j); z__4.r = bscale * b[i__4].r, z__4.i = bscale * b[i__4].i; z__3.r = shift.r * z__4.r - shift.i * z__4.i, z__3.i = shift.r * z__4.i + shift.i * z__4.r; z__1.r = z__2.r - z__3.r, z__1.i = z__2.i - z__3.i; ctemp.r = z__1.r, ctemp.i = z__1.i; temp = (d__1 = ctemp.r, abs(d__1)) + (d__2 = d_imag(&ctemp), abs( d__2)); i__3 = a_subscr(j + 1, j); temp2 = ascale * ((d__1 = a[i__3].r, abs(d__1)) + (d__2 = d_imag(& a_ref(j + 1, j)), abs(d__2))); tempr = max(temp,temp2); if (tempr < 1. && tempr != 0.) { temp /= tempr; temp2 /= tempr; } i__3 = a_subscr(j, j - 1); if (((d__1 = a[i__3].r, abs(d__1)) + (d__2 = d_imag(&a_ref(j, j - 1)), abs(d__2))) * temp2 <= temp * atol) { goto L90; } /* L80: */ } istart = ifirst; i__2 = a_subscr(ifirst, ifirst); z__2.r = ascale * a[i__2].r, z__2.i = ascale * a[i__2].i; i__3 = b_subscr(ifirst, ifirst); z__4.r = bscale * b[i__3].r, z__4.i = bscale * b[i__3].i; z__3.r = shift.r * z__4.r - shift.i * z__4.i, z__3.i = shift.r * z__4.i + shift.i * z__4.r; z__1.r = z__2.r - z__3.r, z__1.i = z__2.i - z__3.i; ctemp.r = z__1.r, ctemp.i = z__1.i; /* --------------------- Begin Timing Code ----------------------- */ opst += -6.; /* ---------------------- End Timing Code ------------------------ */ L90: /* Do an implicit-shift QZ sweep. Initial Q */ i__2 = a_subscr(istart + 1, istart); z__1.r = ascale * a[i__2].r, z__1.i = ascale * a[i__2].i; ctemp2.r = z__1.r, ctemp2.i = z__1.i; /* --------------------- Begin Timing Code ----------------------- */ opst += (doublereal) ((ilast - istart) * 18 + 2); /* ---------------------- End Timing Code ------------------------ */ zlartg_(&ctemp, &ctemp2, &c__, &s, &ctemp3); /* Sweep */ i__2 = ilast - 1; for (j = istart; j <= i__2; ++j) { if (j > istart) { i__3 = a_subscr(j, j - 1); ctemp.r = a[i__3].r, ctemp.i = a[i__3].i; zlartg_(&ctemp, &a_ref(j + 1, j - 1), &c__, &s, &a_ref(j, j - 1)); i__3 = a_subscr(j + 1, j - 1); a[i__3].r = 0., a[i__3].i = 0.; } i__3 = ilastm; for (jc = j; jc <= i__3; ++jc) { i__4 = a_subscr(j, jc); z__2.r = c__ * a[i__4].r, z__2.i = c__ * a[i__4].i; i__5 = a_subscr(j + 1, jc); z__3.r = s.r * a[i__5].r - s.i * a[i__5].i, z__3.i = s.r * a[ i__5].i + s.i * a[i__5].r; z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i; ctemp.r = z__1.r, ctemp.i = z__1.i; i__4 = a_subscr(j + 1, jc); d_cnjg(&z__4, &s); z__3.r = -z__4.r, z__3.i = -z__4.i; i__5 = a_subscr(j, jc); z__2.r = z__3.r * a[i__5].r - z__3.i * a[i__5].i, z__2.i = z__3.r * a[i__5].i + z__3.i * a[i__5].r; i__6 = a_subscr(j + 1, jc); z__5.r = c__ * a[i__6].r, z__5.i = c__ * a[i__6].i; z__1.r = z__2.r + z__5.r, z__1.i = z__2.i + z__5.i; a[i__4].r = z__1.r, a[i__4].i = z__1.i; i__4 = a_subscr(j, jc); a[i__4].r = ctemp.r, a[i__4].i = ctemp.i; i__4 = b_subscr(j, jc); z__2.r = c__ * b[i__4].r, z__2.i = c__ * b[i__4].i; i__5 = b_subscr(j + 1, jc); z__3.r = s.r * b[i__5].r - s.i * b[i__5].i, z__3.i = s.r * b[ i__5].i + s.i * b[i__5].r; z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i; ctemp2.r = z__1.r, ctemp2.i = z__1.i; i__4 = b_subscr(j + 1, jc); d_cnjg(&z__4, &s); z__3.r = -z__4.r, z__3.i = -z__4.i; i__5 = b_subscr(j, jc); z__2.r = z__3.r * b[i__5].r - z__3.i * b[i__5].i, z__2.i = z__3.r * b[i__5].i + z__3.i * b[i__5].r; i__6 = b_subscr(j + 1, jc); z__5.r = c__ * b[i__6].r, z__5.i = c__ * b[i__6].i; z__1.r = z__2.r + z__5.r, z__1.i = z__2.i + z__5.i; b[i__4].r = z__1.r, b[i__4].i = z__1.i; i__4 = b_subscr(j, jc); b[i__4].r = ctemp2.r, b[i__4].i = ctemp2.i; /* L100: */ } if (ilq) { i__3 = *n; for (jr = 1; jr <= i__3; ++jr) { i__4 = q_subscr(jr, j); z__2.r = c__ * q[i__4].r, z__2.i = c__ * q[i__4].i; d_cnjg(&z__4, &s); i__5 = q_subscr(jr, j + 1); z__3.r = z__4.r * q[i__5].r - z__4.i * q[i__5].i, z__3.i = z__4.r * q[i__5].i + z__4.i * q[i__5].r; z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i; ctemp.r = z__1.r, ctemp.i = z__1.i; i__4 = q_subscr(jr, j + 1); z__3.r = -s.r, z__3.i = -s.i; i__5 = q_subscr(jr, j); z__2.r = z__3.r * q[i__5].r - z__3.i * q[i__5].i, z__2.i = z__3.r * q[i__5].i + z__3.i * q[i__5].r; i__6 = q_subscr(jr, j + 1); z__4.r = c__ * q[i__6].r, z__4.i = c__ * q[i__6].i; z__1.r = z__2.r + z__4.r, z__1.i = z__2.i + z__4.i; q[i__4].r = z__1.r, q[i__4].i = z__1.i; i__4 = q_subscr(jr, j); q[i__4].r = ctemp.r, q[i__4].i = ctemp.i; /* L110: */ } } i__3 = b_subscr(j + 1, j + 1); ctemp.r = b[i__3].r, ctemp.i = b[i__3].i; zlartg_(&ctemp, &b_ref(j + 1, j), &c__, &s, &b_ref(j + 1, j + 1)); i__3 = b_subscr(j + 1, j); b[i__3].r = 0., b[i__3].i = 0.; /* Computing MIN */ i__4 = j + 2; i__3 = min(i__4,ilast); for (jr = ifrstm; jr <= i__3; ++jr) { i__4 = a_subscr(jr, j + 1); z__2.r = c__ * a[i__4].r, z__2.i = c__ * a[i__4].i; i__5 = a_subscr(jr, j); z__3.r = s.r * a[i__5].r - s.i * a[i__5].i, z__3.i = s.r * a[ i__5].i + s.i * a[i__5].r; z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i; ctemp.r = z__1.r, ctemp.i = z__1.i; i__4 = a_subscr(jr, j); d_cnjg(&z__4, &s); z__3.r = -z__4.r, z__3.i = -z__4.i; i__5 = a_subscr(jr, j + 1); z__2.r = z__3.r * a[i__5].r - z__3.i * a[i__5].i, z__2.i = z__3.r * a[i__5].i + z__3.i * a[i__5].r; i__6 = a_subscr(jr, j); z__5.r = c__ * a[i__6].r, z__5.i = c__ * a[i__6].i; z__1.r = z__2.r + z__5.r, z__1.i = z__2.i + z__5.i; a[i__4].r = z__1.r, a[i__4].i = z__1.i; i__4 = a_subscr(jr, j + 1); a[i__4].r = ctemp.r, a[i__4].i = ctemp.i; /* L120: */ } i__3 = j; for (jr = ifrstm; jr <= i__3; ++jr) { i__4 = b_subscr(jr, j + 1); z__2.r = c__ * b[i__4].r, z__2.i = c__ * b[i__4].i; i__5 = b_subscr(jr, j); z__3.r = s.r * b[i__5].r - s.i * b[i__5].i, z__3.i = s.r * b[ i__5].i + s.i * b[i__5].r; z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i; ctemp.r = z__1.r, ctemp.i = z__1.i; i__4 = b_subscr(jr, j); d_cnjg(&z__4, &s); z__3.r = -z__4.r, z__3.i = -z__4.i; i__5 = b_subscr(jr, j + 1); z__2.r = z__3.r * b[i__5].r - z__3.i * b[i__5].i, z__2.i = z__3.r * b[i__5].i + z__3.i * b[i__5].r; i__6 = b_subscr(jr, j); z__5.r = c__ * b[i__6].r, z__5.i = c__ * b[i__6].i; z__1.r = z__2.r + z__5.r, z__1.i = z__2.i + z__5.i; b[i__4].r = z__1.r, b[i__4].i = z__1.i; i__4 = b_subscr(jr, j + 1); b[i__4].r = ctemp.r, b[i__4].i = ctemp.i; /* L130: */ } if (ilz) { i__3 = *n; for (jr = 1; jr <= i__3; ++jr) { i__4 = z___subscr(jr, j + 1); z__2.r = c__ * z__[i__4].r, z__2.i = c__ * z__[i__4].i; i__5 = z___subscr(jr, j); z__3.r = s.r * z__[i__5].r - s.i * z__[i__5].i, z__3.i = s.r * z__[i__5].i + s.i * z__[i__5].r; z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i; ctemp.r = z__1.r, ctemp.i = z__1.i; i__4 = z___subscr(jr, j); d_cnjg(&z__4, &s); z__3.r = -z__4.r, z__3.i = -z__4.i; i__5 = z___subscr(jr, j + 1); z__2.r = z__3.r * z__[i__5].r - z__3.i * z__[i__5].i, z__2.i = z__3.r * z__[i__5].i + z__3.i * z__[i__5] .r; i__6 = z___subscr(jr, j); z__5.r = c__ * z__[i__6].r, z__5.i = c__ * z__[i__6].i; z__1.r = z__2.r + z__5.r, z__1.i = z__2.i + z__5.i; z__[i__4].r = z__1.r, z__[i__4].i = z__1.i; i__4 = z___subscr(jr, j + 1); z__[i__4].r = ctemp.r, z__[i__4].i = ctemp.i; /* L140: */ } } /* L150: */ } /* --------------------- Begin Timing Code ----------------------- */ opst += (doublereal) (ilast - istart) * (doublereal) ((ilastm - ifrstm) * 40 + 184 + (nq + nz) * 20) - 20; /* ---------------------- End Timing Code ------------------------ */ L160: /* --------------------- Begin Timing Code ----------------------- End of iteration -- add in "small" contributions. */ latime_1.ops += opst; opst = 0.; /* ---------------------- End Timing Code ------------------------ L170: */ } /* Drop-through = non-convergence */ L180: *info = ilast; /* ---------------------- Begin Timing Code ------------------------- */ latime_1.ops += opst; opst = 0.; /* ----------------------- End Timing Code -------------------------- */ goto L210; /* Successful completion of all QZ steps */ L190: /* Set Eigenvalues 1:ILO-1 */ i__1 = *ilo - 1; for (j = 1; j <= i__1; ++j) { absb = z_abs(&b_ref(j, j)); if (absb > safmin) { i__2 = b_subscr(j, j); z__2.r = b[i__2].r / absb, z__2.i = b[i__2].i / absb; d_cnjg(&z__1, &z__2); signbc.r = z__1.r, signbc.i = z__1.i; i__2 = b_subscr(j, j); b[i__2].r = absb, b[i__2].i = 0.; if (ilschr) { i__2 = j - 1; zscal_(&i__2, &signbc, &b_ref(1, j), &c__1); zscal_(&j, &signbc, &a_ref(1, j), &c__1); /* ----------------- Begin Timing Code --------------------- */ opst += (doublereal) ((j - 1) * 12); /* ------------------ End Timing Code ---------------------- */ } else { i__2 = a_subscr(j, j); i__3 = a_subscr(j, j); z__1.r = a[i__3].r * signbc.r - a[i__3].i * signbc.i, z__1.i = a[i__3].r * signbc.i + a[i__3].i * signbc.r; a[i__2].r = z__1.r, a[i__2].i = z__1.i; } if (ilz) { zscal_(n, &signbc, &z___ref(1, j), &c__1); } /* ------------------- Begin Timing Code ---------------------- */ opst += (doublereal) (nz * 6 + 13); /* -------------------- End Timing Code ----------------------- */ } else { i__2 = b_subscr(j, j); b[i__2].r = 0., b[i__2].i = 0.; } i__2 = j; i__3 = a_subscr(j, j); alpha[i__2].r = a[i__3].r, alpha[i__2].i = a[i__3].i; i__2 = j; i__3 = b_subscr(j, j); beta[i__2].r = b[i__3].r, beta[i__2].i = b[i__3].i; /* L200: */ } /* Normal Termination */ *info = 0; /* Exit (other than argument error) -- return optimal workspace size */ L210: /* ---------------------- Begin Timing Code ------------------------- */ latime_1.ops += opst; opst = 0.; latime_1.itcnt = (doublereal) jiter; /* ----------------------- End Timing Code -------------------------- */ z__1.r = (doublereal) (*n), z__1.i = 0.; work[1].r = z__1.r, work[1].i = z__1.i; return 0; /* End of ZHGEQZ */ } /* zhgeqz_ */
/* Subroutine */ int zpteqr_(char *compz, integer *n, doublereal *d__, doublereal *e, doublecomplex *z__, integer *ldz, doublereal *work, integer *info) { /* System generated locals */ integer z_dim1, z_offset, i__1; /* Builtin functions */ double sqrt(doublereal); /* Local variables */ static doublecomplex c__[1] /* was [1][1] */; static integer i__; extern logical lsame_(char *, char *); static doublecomplex vt[1] /* was [1][1] */; extern /* Subroutine */ int xerbla_(char *, integer *); static integer icompz; extern /* Subroutine */ int zlaset_(char *, integer *, integer *, doublecomplex *, doublecomplex *, doublecomplex *, integer *), dpttrf_(integer *, doublereal *, doublereal *, integer *) , zbdsqr_(char *, integer *, integer *, integer *, integer *, doublereal *, doublereal *, doublecomplex *, integer *, doublecomplex *, integer *, doublecomplex *, integer *, doublereal *, integer *); static integer nru; #define z___subscr(a_1,a_2) (a_2)*z_dim1 + a_1 #define z___ref(a_1,a_2) z__[z___subscr(a_1,a_2)] /* -- LAPACK routine (instrumented to count operations, version 3.0) -- Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., Courant Institute, Argonne National Lab, and Rice University October 31, 1999 Common block to return operation count and iteration count ITCNT is initialized to 0, OPS is only incremented Purpose ======= ZPTEQR computes all eigenvalues and, optionally, eigenvectors of a symmetric positive definite tridiagonal matrix by first factoring the matrix using DPTTRF and then calling ZBDSQR to compute the singular values of the bidiagonal factor. This routine computes the eigenvalues of the positive definite tridiagonal matrix to high relative accuracy. This means that if the eigenvalues range over many orders of magnitude in size, then the small eigenvalues and corresponding eigenvectors will be computed more accurately than, for example, with the standard QR method. The eigenvectors of a full or band positive definite Hermitian matrix can also be found if ZHETRD, ZHPTRD, or ZHBTRD has been used to reduce this matrix to tridiagonal form. (The reduction to tridiagonal form, however, may preclude the possibility of obtaining high relative accuracy in the small eigenvalues of the original matrix, if these eigenvalues range over many orders of magnitude.) Arguments ========= COMPZ (input) CHARACTER*1 = 'N': Compute eigenvalues only. = 'V': Compute eigenvectors of original Hermitian matrix also. Array Z contains the unitary matrix used to reduce the original matrix to tridiagonal form. = 'I': Compute eigenvectors of tridiagonal matrix also. N (input) INTEGER The order of the matrix. N >= 0. D (input/output) DOUBLE PRECISION array, dimension (N) On entry, the n diagonal elements of the tridiagonal matrix. On normal exit, D contains the eigenvalues, in descending order. E (input/output) DOUBLE PRECISION array, dimension (N-1) On entry, the (n-1) subdiagonal elements of the tridiagonal matrix. On exit, E has been destroyed. Z (input/output) COMPLEX*16 array, dimension (LDZ, N) On entry, if COMPZ = 'V', the unitary matrix used in the reduction to tridiagonal form. On exit, if COMPZ = 'V', the orthonormal eigenvectors of the original Hermitian matrix; if COMPZ = 'I', the orthonormal eigenvectors of the tridiagonal matrix. If INFO > 0 on exit, Z contains the eigenvectors associated with only the stored eigenvalues. If COMPZ = 'N', then Z is not referenced. LDZ (input) INTEGER The leading dimension of the array Z. LDZ >= 1, and if COMPZ = 'V' or 'I', LDZ >= max(1,N). WORK (workspace) DOUBLE PRECISION array, dimension (4*N) INFO (output) INTEGER = 0: successful exit. < 0: if INFO = -i, the i-th argument had an illegal value. > 0: if INFO = i, and i is: <= N the Cholesky factorization of the matrix could not be performed because the i-th principal minor was not positive definite. > N the SVD algorithm failed to converge; if INFO = N+i, i off-diagonal elements of the bidiagonal factor did not converge to zero. ==================================================================== Test the input parameters. Parameter adjustments */ --d__; --e; z_dim1 = *ldz; z_offset = 1 + z_dim1 * 1; z__ -= z_offset; --work; /* Function Body */ *info = 0; if (lsame_(compz, "N")) { icompz = 0; } else if (lsame_(compz, "V")) { icompz = 1; } else if (lsame_(compz, "I")) { icompz = 2; } else { icompz = -1; } if (icompz < 0) { *info = -1; } else if (*n < 0) { *info = -2; } else if (*ldz < 1 || icompz > 0 && *ldz < max(1,*n)) { *info = -6; } if (*info != 0) { i__1 = -(*info); xerbla_("ZPTEQR", &i__1); return 0; } /* Quick return if possible */ if (*n == 0) { return 0; } if (*n == 1) { if (icompz > 0) { i__1 = z___subscr(1, 1); z__[i__1].r = 1., z__[i__1].i = 0.; } return 0; } if (icompz == 2) { zlaset_("Full", n, n, &c_b1, &c_b2, &z__[z_offset], ldz); } /* Call DPTTRF to factor the matrix. */ latime_1.ops = latime_1.ops + *n * 5 - 4; dpttrf_(n, &d__[1], &e[1], info); if (*info != 0) { return 0; } i__1 = *n; for (i__ = 1; i__ <= i__1; ++i__) { d__[i__] = sqrt(d__[i__]); /* L10: */ } i__1 = *n - 1; for (i__ = 1; i__ <= i__1; ++i__) { e[i__] *= d__[i__]; /* L20: */ } /* Call ZBDSQR to compute the singular values/vectors of the bidiagonal factor. */ if (icompz > 0) { nru = *n; } else { nru = 0; } zbdsqr_("Lower", n, &c__0, &nru, &c__0, &d__[1], &e[1], vt, &c__1, &z__[ z_offset], ldz, c__, &c__1, &work[1], info); /* Square the singular values. */ if (*info == 0) { latime_1.ops += *n; i__1 = *n; for (i__ = 1; i__ <= i__1; ++i__) { d__[i__] *= d__[i__]; /* L30: */ } } else { *info = *n + *info; } return 0; /* End of ZPTEQR */ } /* zpteqr_ */
/* Subroutine */ int chpevx_(char *jobz, char *range, char *uplo, integer *n, complex *ap, real *vl, real *vu, integer *il, integer *iu, real * abstol, integer *m, real *w, complex *z__, integer *ldz, complex * work, real *rwork, integer *iwork, integer *ifail, integer *info) { /* -- LAPACK driver routine (version 3.0) -- Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., Courant Institute, Argonne National Lab, and Rice University June 30, 1999 Purpose ======= CHPEVX computes selected eigenvalues and, optionally, eigenvectors of a complex Hermitian matrix A in packed storage. Eigenvalues/vectors can be selected by specifying either a range of values or a range of indices for the desired eigenvalues. Arguments ========= JOBZ (input) CHARACTER*1 = 'N': Compute eigenvalues only; = 'V': Compute eigenvalues and eigenvectors. RANGE (input) CHARACTER*1 = 'A': all eigenvalues will be found; = 'V': all eigenvalues in the half-open interval (VL,VU] will be found; = 'I': the IL-th through IU-th eigenvalues will be found. UPLO (input) CHARACTER*1 = 'U': Upper triangle of A is stored; = 'L': Lower triangle of A is stored. N (input) INTEGER The order of the matrix A. N >= 0. AP (input/output) COMPLEX array, dimension (N*(N+1)/2) On entry, the upper or lower triangle of the Hermitian matrix A, packed columnwise in a linear array. The j-th column of A is stored in the array AP as follows: if UPLO = 'U', AP(i + (j-1)*j/2) = A(i,j) for 1<=i<=j; if UPLO = 'L', AP(i + (j-1)*(2*n-j)/2) = A(i,j) for j<=i<=n. On exit, AP is overwritten by values generated during the reduction to tridiagonal form. If UPLO = 'U', the diagonal and first superdiagonal of the tridiagonal matrix T overwrite the corresponding elements of A, and if UPLO = 'L', the diagonal and first subdiagonal of T overwrite the corresponding elements of A. VL (input) REAL VU (input) REAL If RANGE='V', the lower and upper bounds of the interval to be searched for eigenvalues. VL < VU. Not referenced if RANGE = 'A' or 'I'. IL (input) INTEGER IU (input) INTEGER If RANGE='I', the indices (in ascending order) of the smallest and largest eigenvalues to be returned. 1 <= IL <= IU <= N, if N > 0; IL = 1 and IU = 0 if N = 0. Not referenced if RANGE = 'A' or 'V'. ABSTOL (input) REAL The absolute error tolerance for the eigenvalues. An approximate eigenvalue is accepted as converged when it is determined to lie in an interval [a,b] of width less than or equal to ABSTOL + EPS * max( |a|,|b| ) , where EPS is the machine precision. If ABSTOL is less than or equal to zero, then EPS*|T| will be used in its place, where |T| is the 1-norm of the tridiagonal matrix obtained by reducing AP to tridiagonal form. Eigenvalues will be computed most accurately when ABSTOL is set to twice the underflow threshold 2*SLAMCH('S'), not zero. If this routine returns with INFO>0, indicating that some eigenvectors did not converge, try setting ABSTOL to 2*SLAMCH('S'). See "Computing Small Singular Values of Bidiagonal Matrices with Guaranteed High Relative Accuracy," by Demmel and Kahan, LAPACK Working Note #3. M (output) INTEGER The total number of eigenvalues found. 0 <= M <= N. If RANGE = 'A', M = N, and if RANGE = 'I', M = IU-IL+1. W (output) REAL array, dimension (N) If INFO = 0, the selected eigenvalues in ascending order. Z (output) COMPLEX array, dimension (LDZ, max(1,M)) If JOBZ = 'V', then if INFO = 0, the first M columns of Z contain the orthonormal eigenvectors of the matrix A corresponding to the selected eigenvalues, with the i-th column of Z holding the eigenvector associated with W(i). If an eigenvector fails to converge, then that column of Z contains the latest approximation to the eigenvector, and the index of the eigenvector is returned in IFAIL. If JOBZ = 'N', then Z is not referenced. Note: the user must ensure that at least max(1,M) columns are supplied in the array Z; if RANGE = 'V', the exact value of M is not known in advance and an upper bound must be used. LDZ (input) INTEGER The leading dimension of the array Z. LDZ >= 1, and if JOBZ = 'V', LDZ >= max(1,N). WORK (workspace) COMPLEX array, dimension (2*N) RWORK (workspace) REAL array, dimension (7*N) IWORK (workspace) INTEGER array, dimension (5*N) IFAIL (output) INTEGER array, dimension (N) If JOBZ = 'V', then if INFO = 0, the first M elements of IFAIL are zero. If INFO > 0, then IFAIL contains the indices of the eigenvectors that failed to converge. If JOBZ = 'N', then IFAIL is not referenced. INFO (output) INTEGER = 0: successful exit < 0: if INFO = -i, the i-th argument had an illegal value > 0: if INFO = i, then i eigenvectors failed to converge. Their indices are stored in array IFAIL. ===================================================================== Test the input parameters. Parameter adjustments */ /* Table of constant values */ static integer c__1 = 1; /* System generated locals */ integer z_dim1, z_offset, i__1, i__2; real r__1, r__2; /* Builtin functions */ double sqrt(doublereal); /* Local variables */ static integer indd, inde; static real anrm; static integer imax; static real rmin, rmax; static integer itmp1, i__, j, indee; static real sigma; extern logical lsame_(char *, char *); static integer iinfo; extern /* Subroutine */ int sscal_(integer *, real *, real *, integer *); static char order[1]; extern /* Subroutine */ int cswap_(integer *, complex *, integer *, complex *, integer *), scopy_(integer *, real *, integer *, real * , integer *); static logical wantz; static integer jj; static logical alleig, indeig; static integer iscale, indibl; extern doublereal clanhp_(char *, char *, integer *, complex *, real *); static logical valeig; extern doublereal slamch_(char *); extern /* Subroutine */ int csscal_(integer *, real *, complex *, integer *); static real safmin; extern /* Subroutine */ int xerbla_(char *, integer *); static real abstll, bignum; static integer indiwk, indisp, indtau; extern /* Subroutine */ int chptrd_(char *, integer *, complex *, real *, real *, complex *, integer *), cstein_(integer *, real *, real *, integer *, real *, integer *, integer *, complex *, integer *, real *, integer *, integer *, integer *); static integer indrwk, indwrk; extern /* Subroutine */ int csteqr_(char *, integer *, real *, real *, complex *, integer *, real *, integer *), cupgtr_(char *, integer *, complex *, complex *, complex *, integer *, complex *, integer *), ssterf_(integer *, real *, real *, integer *); static integer nsplit; extern /* Subroutine */ int cupmtr_(char *, char *, char *, integer *, integer *, complex *, complex *, complex *, integer *, complex *, integer *); static real smlnum; extern /* Subroutine */ int sstebz_(char *, char *, integer *, real *, real *, integer *, integer *, real *, real *, real *, integer *, integer *, real *, integer *, integer *, real *, integer *, integer *); static real eps, vll, vuu, tmp1; #define z___subscr(a_1,a_2) (a_2)*z_dim1 + a_1 #define z___ref(a_1,a_2) z__[z___subscr(a_1,a_2)] --ap; --w; z_dim1 = *ldz; z_offset = 1 + z_dim1 * 1; z__ -= z_offset; --work; --rwork; --iwork; --ifail; /* Function Body */ wantz = lsame_(jobz, "V"); alleig = lsame_(range, "A"); valeig = lsame_(range, "V"); indeig = lsame_(range, "I"); *info = 0; if (! (wantz || lsame_(jobz, "N"))) { *info = -1; } else if (! (alleig || valeig || indeig)) { *info = -2; } else if (! (lsame_(uplo, "L") || lsame_(uplo, "U"))) { *info = -3; } else if (*n < 0) { *info = -4; } else { if (valeig) { if (*n > 0 && *vu <= *vl) { *info = -7; } } else if (indeig) { if (*il < 1 || *il > max(1,*n)) { *info = -8; } else if (*iu < min(*n,*il) || *iu > *n) { *info = -9; } } } if (*info == 0) { if (*ldz < 1 || wantz && *ldz < *n) { *info = -14; } } if (*info != 0) { i__1 = -(*info); xerbla_("CHPEVX", &i__1); return 0; } /* Quick return if possible */ *m = 0; if (*n == 0) { return 0; } if (*n == 1) { if (alleig || indeig) { *m = 1; w[1] = ap[1].r; } else { if (*vl < ap[1].r && *vu >= ap[1].r) { *m = 1; w[1] = ap[1].r; } } if (wantz) { i__1 = z___subscr(1, 1); z__[i__1].r = 1.f, z__[i__1].i = 0.f; } return 0; } /* Get machine constants. */ safmin = slamch_("Safe minimum"); eps = slamch_("Precision"); smlnum = safmin / eps; bignum = 1.f / smlnum; rmin = sqrt(smlnum); /* Computing MIN */ r__1 = sqrt(bignum), r__2 = 1.f / sqrt(sqrt(safmin)); rmax = dmin(r__1,r__2); /* Scale matrix to allowable range, if necessary. */ iscale = 0; abstll = *abstol; if (valeig) { vll = *vl; vuu = *vu; } else { vll = 0.f; vuu = 0.f; } anrm = clanhp_("M", uplo, n, &ap[1], &rwork[1]); if (anrm > 0.f && anrm < rmin) { iscale = 1; sigma = rmin / anrm; } else if (anrm > rmax) { iscale = 1; sigma = rmax / anrm; } if (iscale == 1) { i__1 = *n * (*n + 1) / 2; csscal_(&i__1, &sigma, &ap[1], &c__1); if (*abstol > 0.f) { abstll = *abstol * sigma; } if (valeig) { vll = *vl * sigma; vuu = *vu * sigma; } } /* Call CHPTRD to reduce Hermitian packed matrix to tridiagonal form. */ indd = 1; inde = indd + *n; indrwk = inde + *n; indtau = 1; indwrk = indtau + *n; chptrd_(uplo, n, &ap[1], &rwork[indd], &rwork[inde], &work[indtau], & iinfo); /* If all eigenvalues are desired and ABSTOL is less than or equal to zero, then call SSTERF or CUPGTR and CSTEQR. If this fails for some eigenvalue, then try SSTEBZ. */ if ((alleig || indeig && *il == 1 && *iu == *n) && *abstol <= 0.f) { scopy_(n, &rwork[indd], &c__1, &w[1], &c__1); indee = indrwk + (*n << 1); if (! wantz) { i__1 = *n - 1; scopy_(&i__1, &rwork[inde], &c__1, &rwork[indee], &c__1); ssterf_(n, &w[1], &rwork[indee], info); } else { cupgtr_(uplo, n, &ap[1], &work[indtau], &z__[z_offset], ldz, & work[indwrk], &iinfo); i__1 = *n - 1; scopy_(&i__1, &rwork[inde], &c__1, &rwork[indee], &c__1); csteqr_(jobz, n, &w[1], &rwork[indee], &z__[z_offset], ldz, & rwork[indrwk], info); if (*info == 0) { i__1 = *n; for (i__ = 1; i__ <= i__1; ++i__) { ifail[i__] = 0; /* L10: */ } } } if (*info == 0) { *m = *n; goto L20; } *info = 0; } /* Otherwise, call SSTEBZ and, if eigenvectors are desired, CSTEIN. */ if (wantz) { *(unsigned char *)order = 'B'; } else { *(unsigned char *)order = 'E'; } indibl = 1; indisp = indibl + *n; indiwk = indisp + *n; sstebz_(range, order, n, &vll, &vuu, il, iu, &abstll, &rwork[indd], & rwork[inde], m, &nsplit, &w[1], &iwork[indibl], &iwork[indisp], & rwork[indrwk], &iwork[indiwk], info); if (wantz) { cstein_(n, &rwork[indd], &rwork[inde], m, &w[1], &iwork[indibl], & iwork[indisp], &z__[z_offset], ldz, &rwork[indrwk], &iwork[ indiwk], &ifail[1], info); /* Apply unitary matrix used in reduction to tridiagonal form to eigenvectors returned by CSTEIN. */ indwrk = indtau + *n; cupmtr_("L", uplo, "N", n, m, &ap[1], &work[indtau], &z__[z_offset], ldz, &work[indwrk], info); } /* If matrix was scaled, then rescale eigenvalues appropriately. */ L20: if (iscale == 1) { if (*info == 0) { imax = *m; } else { imax = *info - 1; } r__1 = 1.f / sigma; sscal_(&imax, &r__1, &w[1], &c__1); } /* If eigenvalues are not in order, then sort them, along with eigenvectors. */ if (wantz) { i__1 = *m - 1; for (j = 1; j <= i__1; ++j) { i__ = 0; tmp1 = w[j]; i__2 = *m; for (jj = j + 1; jj <= i__2; ++jj) { if (w[jj] < tmp1) { i__ = jj; tmp1 = w[jj]; } /* L30: */ } if (i__ != 0) { itmp1 = iwork[indibl + i__ - 1]; w[i__] = w[j]; iwork[indibl + i__ - 1] = iwork[indibl + j - 1]; w[j] = tmp1; iwork[indibl + j - 1] = itmp1; cswap_(n, &z___ref(1, i__), &c__1, &z___ref(1, j), &c__1); if (*info != 0) { itmp1 = ifail[i__]; ifail[i__] = ifail[j]; ifail[j] = itmp1; } } /* L40: */ } } return 0; /* End of CHPEVX */ } /* chpevx_ */
/* Subroutine */ int cstegr_(char *jobz, char *range, integer *n, real *d__, real *e, real *vl, real *vu, integer *il, integer *iu, real *abstol, integer *m, real *w, complex *z__, integer *ldz, integer *isuppz, real *work, integer *lwork, integer *iwork, integer *liwork, integer * info) { /* System generated locals */ integer z_dim1, z_offset, i__1, i__2; real r__1, r__2; /* Builtin functions */ double sqrt(doublereal); /* Local variables */ static integer iend; static real rmin, rmax; static integer itmp; static real tnrm; static integer i__, j; static real scale; extern logical lsame_(char *, char *); static integer iinfo; extern /* Subroutine */ int sscal_(integer *, real *, real *, integer *), cswap_(integer *, complex *, integer *, complex *, integer *); static integer lwmin; static logical wantz; static integer jj; static logical alleig, indeig; static integer ibegin, iindbl; static logical valeig; extern doublereal slamch_(char *); extern /* Subroutine */ int claset_(char *, integer *, integer *, complex *, complex *, complex *, integer *); static real safmin; extern /* Subroutine */ int xerbla_(char *, integer *); static real bignum; static integer iindwk, indgrs, indwof; extern /* Subroutine */ int clarrv_(integer *, real *, real *, integer *, integer *, real *, integer *, real *, real *, complex *, integer * , integer *, real *, integer *, integer *), slarre_(integer *, real *, real *, real *, integer *, integer *, integer *, real *, real *, real *, real *, integer *); static real thresh; static integer iinspl, indwrk, liwmin; extern doublereal slanst_(char *, integer *, real *, real *); static integer nsplit; static real smlnum; static logical lquery; static real eps, tol, tmp; #define z___subscr(a_1,a_2) (a_2)*z_dim1 + a_1 #define z___ref(a_1,a_2) z__[z___subscr(a_1,a_2)] /* -- LAPACK computational routine (version 3.0) -- Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., Courant Institute, Argonne National Lab, and Rice University October 31, 1999 Purpose ======= CSTEGR computes selected eigenvalues and, optionally, eigenvectors of a real symmetric tridiagonal matrix T. Eigenvalues and eigenvectors can be selected by specifying either a range of values or a range of indices for the desired eigenvalues. The eigenvalues are computed by the dqds algorithm, while orthogonal eigenvectors are computed from various ``good'' L D L^T representations (also known as Relatively Robust Representations). Gram-Schmidt orthogonalization is avoided as far as possible. More specifically, the various steps of the algorithm are as follows. For the i-th unreduced block of T, (a) Compute T - sigma_i = L_i D_i L_i^T, such that L_i D_i L_i^T is a relatively robust representation, (b) Compute the eigenvalues, lambda_j, of L_i D_i L_i^T to high relative accuracy by the dqds algorithm, (c) If there is a cluster of close eigenvalues, "choose" sigma_i close to the cluster, and go to step (a), (d) Given the approximate eigenvalue lambda_j of L_i D_i L_i^T, compute the corresponding eigenvector by forming a rank-revealing twisted factorization. The desired accuracy of the output can be specified by the input parameter ABSTOL. For more details, see "A new O(n^2) algorithm for the symmetric tridiagonal eigenvalue/eigenvector problem", by Inderjit Dhillon, Computer Science Division Technical Report No. UCB/CSD-97-971, UC Berkeley, May 1997. Note 1 : Currently CSTEGR is only set up to find ALL the n eigenvalues and eigenvectors of T in O(n^2) time Note 2 : Currently the routine CSTEIN is called when an appropriate sigma_i cannot be chosen in step (c) above. CSTEIN invokes modified Gram-Schmidt when eigenvalues are close. Note 3 : CSTEGR works only on machines which follow ieee-754 floating-point standard in their handling of infinities and NaNs. Normal execution of CSTEGR may create NaNs and infinities and hence may abort due to a floating point exception in environments which do not conform to the ieee standard. Arguments ========= JOBZ (input) CHARACTER*1 = 'N': Compute eigenvalues only; = 'V': Compute eigenvalues and eigenvectors. RANGE (input) CHARACTER*1 = 'A': all eigenvalues will be found. = 'V': all eigenvalues in the half-open interval (VL,VU] will be found. = 'I': the IL-th through IU-th eigenvalues will be found. ********* Only RANGE = 'A' is currently supported ********************* N (input) INTEGER The order of the matrix. N >= 0. D (input/output) REAL array, dimension (N) On entry, the n diagonal elements of the tridiagonal matrix T. On exit, D is overwritten. E (input/output) REAL array, dimension (N) On entry, the (n-1) subdiagonal elements of the tridiagonal matrix T in elements 1 to N-1 of E; E(N) need not be set. On exit, E is overwritten. VL (input) REAL VU (input) REAL If RANGE='V', the lower and upper bounds of the interval to be searched for eigenvalues. VL < VU. Not referenced if RANGE = 'A' or 'I'. IL (input) INTEGER IU (input) INTEGER If RANGE='I', the indices (in ascending order) of the smallest and largest eigenvalues to be returned. 1 <= IL <= IU <= N, if N > 0; IL = 1 and IU = 0 if N = 0. Not referenced if RANGE = 'A' or 'V'. ABSTOL (input) REAL The absolute error tolerance for the eigenvalues/eigenvectors. IF JOBZ = 'V', the eigenvalues and eigenvectors output have residual norms bounded by ABSTOL, and the dot products between different eigenvectors are bounded by ABSTOL. If ABSTOL is less than N*EPS*|T|, then N*EPS*|T| will be used in its place, where EPS is the machine precision and |T| is the 1-norm of the tridiagonal matrix. The eigenvalues are computed to an accuracy of EPS*|T| irrespective of ABSTOL. If high relative accuracy is important, set ABSTOL to DLAMCH( 'Safe minimum' ). See Barlow and Demmel "Computing Accurate Eigensystems of Scaled Diagonally Dominant Matrices", LAPACK Working Note #7 for a discussion of which matrices define their eigenvalues to high relative accuracy. M (output) INTEGER The total number of eigenvalues found. 0 <= M <= N. If RANGE = 'A', M = N, and if RANGE = 'I', M = IU-IL+1. W (output) REAL array, dimension (N) The first M elements contain the selected eigenvalues in ascending order. Z (output) COMPLEX array, dimension (LDZ, max(1,M) ) If JOBZ = 'V', then if INFO = 0, the first M columns of Z contain the orthonormal eigenvectors of the matrix T corresponding to the selected eigenvalues, with the i-th column of Z holding the eigenvector associated with W(i). If JOBZ = 'N', then Z is not referenced. Note: the user must ensure that at least max(1,M) columns are supplied in the array Z; if RANGE = 'V', the exact value of M is not known in advance and an upper bound must be used. LDZ (input) INTEGER The leading dimension of the array Z. LDZ >= 1, and if JOBZ = 'V', LDZ >= max(1,N). ISUPPZ (output) INTEGER ARRAY, dimension ( 2*max(1,M) ) The support of the eigenvectors in Z, i.e., the indices indicating the nonzero elements in Z. The i-th eigenvector is nonzero only in elements ISUPPZ( 2*i-1 ) through ISUPPZ( 2*i ). WORK (workspace/output) REAL array, dimension (LWORK) On exit, if INFO = 0, WORK(1) returns the optimal (and minimal) LWORK. LWORK (input) INTEGER The dimension of the array WORK. LWORK >= max(1,18*N) If LWORK = -1, then a workspace query is assumed; the routine only calculates the optimal size of the WORK array, returns this value as the first entry of the WORK array, and no error message related to LWORK is issued by XERBLA. IWORK (workspace/output) INTEGER array, dimension (LIWORK) On exit, if INFO = 0, IWORK(1) returns the optimal LIWORK. LIWORK (input) INTEGER The dimension of the array IWORK. LIWORK >= max(1,10*N) If LIWORK = -1, then a workspace query is assumed; the routine only calculates the optimal size of the IWORK array, returns this value as the first entry of the IWORK array, and no error message related to LIWORK is issued by XERBLA. INFO (output) INTEGER = 0: successful exit < 0: if INFO = -i, the i-th argument had an illegal value > 0: if INFO = 1, internal error in SLARRE, if INFO = 2, internal error in CLARRV. Further Details =============== Based on contributions by Inderjit Dhillon, IBM Almaden, USA Osni Marques, LBNL/NERSC, USA Ken Stanley, Computer Science Division, University of California at Berkeley, USA ===================================================================== Test the input parameters. Parameter adjustments */ --d__; --e; --w; z_dim1 = *ldz; z_offset = 1 + z_dim1 * 1; z__ -= z_offset; --isuppz; --work; --iwork; /* Function Body */ wantz = lsame_(jobz, "V"); alleig = lsame_(range, "A"); valeig = lsame_(range, "V"); indeig = lsame_(range, "I"); lquery = *lwork == -1 || *liwork == -1; lwmin = *n * 18; liwmin = *n * 10; *info = 0; if (! (wantz || lsame_(jobz, "N"))) { *info = -1; } else if (! (alleig || valeig || indeig)) { *info = -2; /* The following two lines need to be removed once the RANGE = 'V' and RANGE = 'I' options are provided. */ } else if (valeig || indeig) { *info = -2; } else if (*n < 0) { *info = -3; } else if (valeig && *n > 0 && *vu <= *vl) { *info = -7; } else if (indeig && *il < 1) { *info = -8; /* The following change should be made in DSTEVX also, otherwise IL can be specified as N+1 and IU as N. ELSE IF( INDEIG .AND. ( IU.LT.MIN( N, IL ) .OR. IU.GT.N ) ) THEN */ } else if (indeig && (*iu < *il || *iu > *n)) { *info = -9; } else if (*ldz < 1 || wantz && *ldz < *n) { *info = -14; } else if (*lwork < lwmin && ! lquery) { *info = -17; } else if (*liwork < liwmin && ! lquery) { *info = -19; } if (*info == 0) { work[1] = (real) lwmin; iwork[1] = liwmin; } if (*info != 0) { i__1 = -(*info); xerbla_("CSTEGR", &i__1); return 0; } else if (lquery) { return 0; } /* Quick return if possible */ *m = 0; if (*n == 0) { return 0; } if (*n == 1) { if (alleig || indeig) { *m = 1; w[1] = d__[1]; } else { if (*vl < d__[1] && *vu >= d__[1]) { *m = 1; w[1] = d__[1]; } } if (wantz) { i__1 = z___subscr(1, 1); z__[i__1].r = 1.f, z__[i__1].i = 0.f; } return 0; } /* Get machine constants. */ safmin = slamch_("Safe minimum"); eps = slamch_("Precision"); smlnum = safmin / eps; bignum = 1.f / smlnum; rmin = sqrt(smlnum); /* Computing MIN */ r__1 = sqrt(bignum), r__2 = 1.f / sqrt(sqrt(safmin)); rmax = dmin(r__1,r__2); /* Scale matrix to allowable range, if necessary. */ scale = 1.f; tnrm = slanst_("M", n, &d__[1], &e[1]); if (tnrm > 0.f && tnrm < rmin) { scale = rmin / tnrm; } else if (tnrm > rmax) { scale = rmax / tnrm; } if (scale != 1.f) { sscal_(n, &scale, &d__[1], &c__1); i__1 = *n - 1; sscal_(&i__1, &scale, &e[1], &c__1); tnrm *= scale; } indgrs = 1; indwof = (*n << 1) + 1; indwrk = *n * 3 + 1; iinspl = 1; iindbl = *n + 1; iindwk = (*n << 1) + 1; claset_("Full", n, n, &c_b1, &c_b1, &z__[z_offset], ldz); /* Compute the desired eigenvalues of the tridiagonal after splitting into smaller subblocks if the corresponding of-diagonal elements are small */ thresh = eps * tnrm; slarre_(n, &d__[1], &e[1], &thresh, &nsplit, &iwork[iinspl], m, &w[1], & work[indwof], &work[indgrs], &work[indwrk], &iinfo); if (iinfo != 0) { *info = 1; return 0; } if (wantz) { /* Compute the desired eigenvectors corresponding to the computed eigenvalues Computing MAX */ r__1 = *abstol, r__2 = (real) (*n) * thresh; tol = dmax(r__1,r__2); ibegin = 1; i__1 = nsplit; for (i__ = 1; i__ <= i__1; ++i__) { iend = iwork[iinspl + i__ - 1]; i__2 = iend; for (j = ibegin; j <= i__2; ++j) { iwork[iindbl + j - 1] = i__; /* L10: */ } ibegin = iend + 1; /* L20: */ } clarrv_(n, &d__[1], &e[1], &iwork[iinspl], m, &w[1], &iwork[iindbl], & work[indgrs], &tol, &z__[z_offset], ldz, &isuppz[1], &work[ indwrk], &iwork[iindwk], &iinfo); if (iinfo != 0) { *info = 2; return 0; } } ibegin = 1; i__1 = nsplit; for (i__ = 1; i__ <= i__1; ++i__) { iend = iwork[iinspl + i__ - 1]; i__2 = iend; for (j = ibegin; j <= i__2; ++j) { w[j] += work[indwof + i__ - 1]; /* L30: */ } ibegin = iend + 1; /* L40: */ } /* If matrix was scaled, then rescale eigenvalues appropriately. */ if (scale != 1.f) { r__1 = 1.f / scale; sscal_(m, &r__1, &w[1], &c__1); } /* If eigenvalues are not in order, then sort them, along with eigenvectors. */ if (nsplit > 1) { i__1 = *m - 1; for (j = 1; j <= i__1; ++j) { i__ = 0; tmp = w[j]; i__2 = *m; for (jj = j + 1; jj <= i__2; ++jj) { if (w[jj] < tmp) { i__ = jj; tmp = w[jj]; } /* L50: */ } if (i__ != 0) { w[i__] = w[j]; w[j] = tmp; if (wantz) { cswap_(n, &z___ref(1, i__), &c__1, &z___ref(1, j), &c__1); itmp = isuppz[(i__ << 1) - 1]; isuppz[(i__ << 1) - 1] = isuppz[(j << 1) - 1]; isuppz[(j << 1) - 1] = itmp; itmp = isuppz[i__ * 2]; isuppz[i__ * 2] = isuppz[j * 2]; isuppz[j * 2] = itmp; } } /* L60: */ } } work[1] = (real) lwmin; iwork[1] = liwmin; return 0; /* End of CSTEGR */ } /* cstegr_ */
/* Subroutine */ int zhpevd_(char *jobz, char *uplo, integer *n, doublecomplex *ap, doublereal *w, doublecomplex *z__, integer *ldz, doublecomplex *work, integer *lwork, doublereal *rwork, integer * lrwork, integer *iwork, integer *liwork, integer *info) { /* -- LAPACK driver routine (version 3.0) -- Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., Courant Institute, Argonne National Lab, and Rice University June 30, 1999 Purpose ======= ZHPEVD computes all the eigenvalues and, optionally, eigenvectors of a complex Hermitian matrix A in packed storage. If eigenvectors are desired, it uses a divide and conquer algorithm. The divide and conquer algorithm makes very mild assumptions about floating point arithmetic. It will work on machines with a guard digit in add/subtract, or on those binary machines without guard digits which subtract like the Cray X-MP, Cray Y-MP, Cray C-90, or Cray-2. It could conceivably fail on hexadecimal or decimal machines without guard digits, but we know of none. Arguments ========= JOBZ (input) CHARACTER*1 = 'N': Compute eigenvalues only; = 'V': Compute eigenvalues and eigenvectors. UPLO (input) CHARACTER*1 = 'U': Upper triangle of A is stored; = 'L': Lower triangle of A is stored. N (input) INTEGER The order of the matrix A. N >= 0. AP (input/output) COMPLEX*16 array, dimension (N*(N+1)/2) On entry, the upper or lower triangle of the Hermitian matrix A, packed columnwise in a linear array. The j-th column of A is stored in the array AP as follows: if UPLO = 'U', AP(i + (j-1)*j/2) = A(i,j) for 1<=i<=j; if UPLO = 'L', AP(i + (j-1)*(2*n-j)/2) = A(i,j) for j<=i<=n. On exit, AP is overwritten by values generated during the reduction to tridiagonal form. If UPLO = 'U', the diagonal and first superdiagonal of the tridiagonal matrix T overwrite the corresponding elements of A, and if UPLO = 'L', the diagonal and first subdiagonal of T overwrite the corresponding elements of A. W (output) DOUBLE PRECISION array, dimension (N) If INFO = 0, the eigenvalues in ascending order. Z (output) COMPLEX*16 array, dimension (LDZ, N) If JOBZ = 'V', then if INFO = 0, Z contains the orthonormal eigenvectors of the matrix A, with the i-th column of Z holding the eigenvector associated with W(i). If JOBZ = 'N', then Z is not referenced. LDZ (input) INTEGER The leading dimension of the array Z. LDZ >= 1, and if JOBZ = 'V', LDZ >= max(1,N). WORK (workspace/output) COMPLEX*16 array, dimension (LWORK) On exit, if INFO = 0, WORK(1) returns the optimal LWORK. LWORK (input) INTEGER The dimension of array WORK. If N <= 1, LWORK must be at least 1. If JOBZ = 'N' and N > 1, LWORK must be at least N. If JOBZ = 'V' and N > 1, LWORK must be at least 2*N. If LWORK = -1, then a workspace query is assumed; the routine only calculates the optimal size of the WORK array, returns this value as the first entry of the WORK array, and no error message related to LWORK is issued by XERBLA. RWORK (workspace/output) DOUBLE PRECISION array, dimension (LRWORK) On exit, if INFO = 0, RWORK(1) returns the optimal LRWORK. LRWORK (input) INTEGER The dimension of array RWORK. If N <= 1, LRWORK must be at least 1. If JOBZ = 'N' and N > 1, LRWORK must be at least N. If JOBZ = 'V' and N > 1, LRWORK must be at least 1 + 5*N + 2*N**2. If LRWORK = -1, then a workspace query is assumed; the routine only calculates the optimal size of the RWORK array, returns this value as the first entry of the RWORK array, and no error message related to LRWORK is issued by XERBLA. IWORK (workspace/output) INTEGER array, dimension (LIWORK) On exit, if INFO = 0, IWORK(1) returns the optimal LIWORK. LIWORK (input) INTEGER The dimension of array IWORK. If JOBZ = 'N' or N <= 1, LIWORK must be at least 1. If JOBZ = 'V' and N > 1, LIWORK must be at least 3 + 5*N. If LIWORK = -1, then a workspace query is assumed; the routine only calculates the optimal size of the IWORK array, returns this value as the first entry of the IWORK array, and no error message related to LIWORK is issued by XERBLA. INFO (output) INTEGER = 0: successful exit < 0: if INFO = -i, the i-th argument had an illegal value. > 0: if INFO = i, the algorithm failed to converge; i off-diagonal elements of an intermediate tridiagonal form did not converge to zero. ===================================================================== Test the input parameters. Parameter adjustments */ /* Table of constant values */ static integer c__1 = 1; /* System generated locals */ integer z_dim1, z_offset, i__1; doublereal d__1; /* Builtin functions */ double sqrt(doublereal); /* Local variables */ static integer inde; static doublereal anrm; static integer imax; static doublereal rmin, rmax; extern /* Subroutine */ int dscal_(integer *, doublereal *, doublereal *, integer *); static doublereal sigma; extern logical lsame_(char *, char *); static integer iinfo, lwmin, llrwk, llwrk; static logical wantz; extern doublereal dlamch_(char *); static integer iscale; static doublereal safmin; extern /* Subroutine */ int xerbla_(char *, integer *), zdscal_( integer *, doublereal *, doublecomplex *, integer *); static doublereal bignum; static integer indtau; extern /* Subroutine */ int dsterf_(integer *, doublereal *, doublereal *, integer *); extern doublereal zlanhp_(char *, char *, integer *, doublecomplex *, doublereal *); extern /* Subroutine */ int zstedc_(char *, integer *, doublereal *, doublereal *, doublecomplex *, integer *, doublecomplex *, integer *, doublereal *, integer *, integer *, integer *, integer *); static integer indrwk, indwrk, liwmin, lrwmin; static doublereal smlnum; extern /* Subroutine */ int zhptrd_(char *, integer *, doublecomplex *, doublereal *, doublereal *, doublecomplex *, integer *); static logical lquery; extern /* Subroutine */ int zupmtr_(char *, char *, char *, integer *, integer *, doublecomplex *, doublecomplex *, doublecomplex *, integer *, doublecomplex *, integer *); static doublereal eps; #define z___subscr(a_1,a_2) (a_2)*z_dim1 + a_1 #define z___ref(a_1,a_2) z__[z___subscr(a_1,a_2)] --ap; --w; z_dim1 = *ldz; z_offset = 1 + z_dim1 * 1; z__ -= z_offset; --work; --rwork; --iwork; /* Function Body */ wantz = lsame_(jobz, "V"); lquery = *lwork == -1 || *lrwork == -1 || *liwork == -1; *info = 0; if (*n <= 1) { lwmin = 1; liwmin = 1; lrwmin = 1; } else { if (wantz) { lwmin = *n << 1; /* Computing 2nd power */ i__1 = *n; lrwmin = *n * 5 + 1 + (i__1 * i__1 << 1); liwmin = *n * 5 + 3; } else { lwmin = *n; lrwmin = *n; liwmin = 1; } } if (! (wantz || lsame_(jobz, "N"))) { *info = -1; } else if (! (lsame_(uplo, "L") || lsame_(uplo, "U"))) { *info = -2; } else if (*n < 0) { *info = -3; } else if (*ldz < 1 || wantz && *ldz < *n) { *info = -7; } else if (*lwork < lwmin && ! lquery) { *info = -9; } else if (*lrwork < lrwmin && ! lquery) { *info = -11; } else if (*liwork < liwmin && ! lquery) { *info = -13; } if (*info == 0) { work[1].r = (doublereal) lwmin, work[1].i = 0.; rwork[1] = (doublereal) lrwmin; iwork[1] = liwmin; } if (*info != 0) { i__1 = -(*info); xerbla_("ZHPEVD", &i__1); return 0; } /* Quick return if possible */ if (*n == 0) { return 0; } if (*n == 1) { w[1] = ap[1].r; if (wantz) { i__1 = z___subscr(1, 1); z__[i__1].r = 1., z__[i__1].i = 0.; } return 0; } /* Get machine constants. */ safmin = dlamch_("Safe minimum"); eps = dlamch_("Precision"); smlnum = safmin / eps; bignum = 1. / smlnum; rmin = sqrt(smlnum); rmax = sqrt(bignum); /* Scale matrix to allowable range, if necessary. */ anrm = zlanhp_("M", uplo, n, &ap[1], &rwork[1]); iscale = 0; if (anrm > 0. && anrm < rmin) { iscale = 1; sigma = rmin / anrm; } else if (anrm > rmax) { iscale = 1; sigma = rmax / anrm; } if (iscale == 1) { i__1 = *n * (*n + 1) / 2; zdscal_(&i__1, &sigma, &ap[1], &c__1); } /* Call ZHPTRD to reduce Hermitian packed matrix to tridiagonal form. */ inde = 1; indtau = 1; indrwk = inde + *n; indwrk = indtau + *n; llwrk = *lwork - indwrk + 1; llrwk = *lrwork - indrwk + 1; zhptrd_(uplo, n, &ap[1], &w[1], &rwork[inde], &work[indtau], &iinfo); /* For eigenvalues only, call DSTERF. For eigenvectors, first call ZUPGTR to generate the orthogonal matrix, then call ZSTEDC. */ if (! wantz) { dsterf_(n, &w[1], &rwork[inde], info); } else { zstedc_("I", n, &w[1], &rwork[inde], &z__[z_offset], ldz, &work[ indwrk], &llwrk, &rwork[indrwk], &llrwk, &iwork[1], liwork, info); zupmtr_("L", uplo, "N", n, n, &ap[1], &work[indtau], &z__[z_offset], ldz, &work[indwrk], &iinfo); } /* If matrix was scaled, then rescale eigenvalues appropriately. */ if (iscale == 1) { if (*info == 0) { imax = *n; } else { imax = *info - 1; } d__1 = 1. / sigma; dscal_(&imax, &d__1, &w[1], &c__1); } work[1].r = (doublereal) lwmin, work[1].i = 0.; rwork[1] = (doublereal) lrwmin; iwork[1] = liwmin; return 0; /* End of ZHPEVD */ } /* zhpevd_ */
/* Subroutine */ int cdrvgg_(integer *nsizes, integer *nn, integer *ntypes, logical *dotype, integer *iseed, real *thresh, real *thrshn, integer * nounit, complex *a, integer *lda, complex *b, complex *s, complex *t, complex *s2, complex *t2, complex *q, integer *ldq, complex *z__, complex *alpha1, complex *beta1, complex *alpha2, complex *beta2, complex *vl, complex *vr, complex *work, integer *lwork, real *rwork, real *result, integer *info) { /* Initialized data */ static integer kclass[26] = { 1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2, 2,2,2,3 }; static integer kbmagn[26] = { 1,1,1,1,1,1,1,1,3,2,3,2,2,3,1,1,1,1,1,1,1,3, 2,3,2,1 }; static integer ktrian[26] = { 0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1,1, 1,1,1,1 }; static logical lasign[26] = { FALSE_,FALSE_,FALSE_,FALSE_,FALSE_,FALSE_, TRUE_,FALSE_,TRUE_,TRUE_,FALSE_,FALSE_,TRUE_,TRUE_,TRUE_,FALSE_, TRUE_,FALSE_,FALSE_,FALSE_,TRUE_,TRUE_,TRUE_,TRUE_,TRUE_,FALSE_ }; static logical lbsign[26] = { FALSE_,FALSE_,FALSE_,FALSE_,FALSE_,FALSE_, FALSE_,TRUE_,FALSE_,FALSE_,TRUE_,TRUE_,FALSE_,FALSE_,TRUE_,FALSE_, TRUE_,FALSE_,FALSE_,FALSE_,FALSE_,FALSE_,FALSE_,FALSE_,FALSE_, FALSE_ }; static integer kz1[6] = { 0,1,2,1,3,3 }; static integer kz2[6] = { 0,0,1,2,1,1 }; static integer kadd[6] = { 0,0,0,0,3,2 }; static integer katype[26] = { 0,1,0,1,2,3,4,1,4,4,1,1,4,4,4,2,4,5,8,7,9,4, 4,4,4,0 }; static integer kbtype[26] = { 0,0,1,1,2,-3,1,4,1,1,4,4,1,1,-4,2,-4,8,8,8, 8,8,8,8,8,0 }; static integer kazero[26] = { 1,1,1,1,1,1,2,1,2,2,1,1,2,2,3,1,3,5,5,5,5,3, 3,3,3,1 }; static integer kbzero[26] = { 1,1,1,1,1,1,1,2,1,1,2,2,1,1,4,1,4,6,6,6,6,4, 4,4,4,1 }; static integer kamagn[26] = { 1,1,1,1,1,1,1,1,2,3,2,3,2,3,1,1,1,1,1,1,1,2, 3,3,2,1 }; /* Format strings */ static char fmt_9999[] = "(\002 CDRVGG: \002,a,\002 returned INFO=\002,i" "6,\002.\002,/9x,\002N=\002,i6,\002, JTYPE=\002,i6,\002, ISEED=" "(\002,3(i5,\002,\002),i5,\002)\002)"; static char fmt_9998[] = "(\002 CDRVGG: \002,a,\002 Eigenvectors from" " \002,a,\002 incorrectly \002,\002normalized.\002,/\002 Bits of " "error=\002,0p,g10.3,\002,\002,9x,\002N=\002,i6,\002, JTYPE=\002," "i6,\002, ISEED=(\002,3(i5,\002,\002),i5,\002)\002)"; static char fmt_9997[] = "(/1x,a3,\002 -- Complex Generalized eigenvalue" " problem driver\002)"; static char fmt_9996[] = "(\002 Matrix types (see CDRVGG for details):" " \002)"; static char fmt_9995[] = "(\002 Special Matrices:\002,23x,\002(J'=transp" "osed Jordan block)\002,/\002 1=(0,0) 2=(I,0) 3=(0,I) 4=(I,I" ") 5=(J',J') \002,\0026=(diag(J',I), diag(I,J'))\002,/\002 Diag" "onal Matrices: ( \002,\002D=diag(0,1,2,...) )\002,/\002 7=(D," "I) 9=(large*D, small*I\002,\002) 11=(large*I, small*D) 13=(l" "arge*D, large*I)\002,/\002 8=(I,D) 10=(small*D, large*I) 12=" "(small*I, large*D) \002,\002 14=(small*D, small*I)\002,/\002 15" "=(D, reversed D)\002)"; static char fmt_9994[] = "(\002 Matrices Rotated by Random \002,a,\002 M" "atrices U, V:\002,/\002 16=Transposed Jordan Blocks " " 19=geometric \002,\002alpha, beta=0,1\002,/\002 17=arithm. alp" "ha&beta \002,\002 20=arithmetic alpha, beta=0," "1\002,/\002 18=clustered \002,\002alpha, beta=0,1 21" "=random alpha, beta=0,1\002,/\002 Large & Small Matrices:\002," "/\002 22=(large, small) \002,\00223=(small,large) 24=(smal" "l,small) 25=(large,large)\002,/\002 26=random O(1) matrices" ".\002)"; static char fmt_9993[] = "(/\002 Tests performed: (S is Schur, T is tri" "angular, \002,\002Q and Z are \002,a,\002,\002,/20x,\002l and r " "are the appropriate left and right\002,/19x,\002eigenvectors, re" "sp., a is alpha, b is beta, and\002,/19x,a,\002 means \002,a," "\002.)\002,/\002 1 = | A - Q S Z\002,a,\002 | / ( |A| n ulp ) " " 2 = | B - Q T Z\002,a,\002 | / ( |B| n ulp )\002,/\002 3 = | " "I - QQ\002,a,\002 | / ( n ulp ) 4 = | I - ZZ\002,a" ",\002 | / ( n ulp )\002,/\002 5 = difference between (alpha,beta" ") and diagonals of\002,\002 (S,T)\002,/\002 6 = max | ( b A - a " "B )\002,a,\002 l | / const. 7 = max | ( b A - a B ) r | / cons" "t.\002,/1x)"; static char fmt_9992[] = "(\002 Matrix order=\002,i5,\002, type=\002,i2" ",\002, seed=\002,4(i4,\002,\002),\002 result \002,i3,\002 is\002" ",0p,f8.2)"; static char fmt_9991[] = "(\002 Matrix order=\002,i5,\002, type=\002,i2" ",\002, seed=\002,4(i4,\002,\002),\002 result \002,i3,\002 is\002" ",1p,e10.3)"; /* System generated locals */ integer a_dim1, a_offset, b_dim1, b_offset, q_dim1, q_offset, s_dim1, s_offset, s2_dim1, s2_offset, t_dim1, t_offset, t2_dim1, t2_offset, vl_dim1, vl_offset, vr_dim1, vr_offset, z_dim1, z_offset, i__1, i__2, i__3, i__4, i__5, i__6, i__7, i__8, i__9, i__10, i__11; real r__1, r__2, r__3, r__4, r__5, r__6, r__7, r__8, r__9, r__10, r__11, r__12, r__13, r__14, r__15, r__16; complex q__1, q__2, q__3, q__4; /* Builtin functions */ double r_sign(real *, real *), c_abs(complex *); void r_cnjg(complex *, complex *); integer s_wsfe(cilist *), do_fio(integer *, char *, ftnlen), e_wsfe(void); double r_imag(complex *); /* Local variables */ static integer iadd, nmax; static real temp1, temp2; static integer j, n; static logical badnn; extern /* Subroutine */ int cgegs_(char *, char *, integer *, complex *, integer *, complex *, integer *, complex *, complex *, complex *, integer *, complex *, integer *, complex *, integer *, real *, integer *), cgegv_(char *, char *, integer *, complex *, integer *, complex *, integer *, complex *, complex *, complex *, integer *, complex *, integer *, complex *, integer *, real *, integer *), cget51_(integer *, integer *, complex *, integer *, complex *, integer *, complex *, integer *, complex *, integer *, complex *, real *, real *), cget52_(logical *, integer *, complex *, integer *, complex *, integer *, complex *, integer *, complex *, complex *, complex *, real *, real *); static real dumma[4]; static integer iinfo; static real rmagn[4]; static complex ctemp; static integer nmats, jsize, nerrs, i1, jtype, ntest, n1; extern /* Subroutine */ int clatm4_(integer *, integer *, integer *, integer *, logical *, real *, real *, real *, integer *, integer * , complex *, integer *), cunm2r_(char *, char *, integer *, integer *, integer *, complex *, integer *, complex *, complex *, integer *, complex *, integer *); static integer jc, nb; extern /* Subroutine */ int slabad_(real *, real *); static integer in, jr; extern /* Subroutine */ int clarfg_(integer *, complex *, complex *, integer *, complex *); static integer ns; extern /* Complex */ VOID clarnd_(complex *, integer *, integer *); extern doublereal slamch_(char *); extern /* Subroutine */ int clacpy_(char *, integer *, integer *, complex *, integer *, complex *, integer *); static real safmin, safmax; static integer ioldsd[4]; extern integer ilaenv_(integer *, char *, char *, integer *, integer *, integer *, integer *, ftnlen, ftnlen); extern /* Subroutine */ int alasvm_(char *, integer *, integer *, integer *, integer *), claset_(char *, integer *, integer *, complex *, complex *, complex *, integer *), xerbla_(char *, integer *); static real ulpinv; static integer lwkopt, mtypes, ntestt, nbz; static real ulp; /* Fortran I/O blocks */ static cilist io___43 = { 0, 0, 0, fmt_9999, 0 }; static cilist io___44 = { 0, 0, 0, fmt_9999, 0 }; static cilist io___47 = { 0, 0, 0, fmt_9999, 0 }; static cilist io___49 = { 0, 0, 0, fmt_9998, 0 }; static cilist io___50 = { 0, 0, 0, fmt_9998, 0 }; static cilist io___51 = { 0, 0, 0, fmt_9997, 0 }; static cilist io___52 = { 0, 0, 0, fmt_9996, 0 }; static cilist io___53 = { 0, 0, 0, fmt_9995, 0 }; static cilist io___54 = { 0, 0, 0, fmt_9994, 0 }; static cilist io___55 = { 0, 0, 0, fmt_9993, 0 }; static cilist io___56 = { 0, 0, 0, fmt_9992, 0 }; static cilist io___57 = { 0, 0, 0, fmt_9991, 0 }; #define a_subscr(a_1,a_2) (a_2)*a_dim1 + a_1 #define a_ref(a_1,a_2) a[a_subscr(a_1,a_2)] #define b_subscr(a_1,a_2) (a_2)*b_dim1 + a_1 #define b_ref(a_1,a_2) b[b_subscr(a_1,a_2)] #define q_subscr(a_1,a_2) (a_2)*q_dim1 + a_1 #define q_ref(a_1,a_2) q[q_subscr(a_1,a_2)] #define s_subscr(a_1,a_2) (a_2)*s_dim1 + a_1 #define s_ref(a_1,a_2) s[s_subscr(a_1,a_2)] #define t_subscr(a_1,a_2) (a_2)*t_dim1 + a_1 #define t_ref(a_1,a_2) t[t_subscr(a_1,a_2)] #define z___subscr(a_1,a_2) (a_2)*z_dim1 + a_1 #define z___ref(a_1,a_2) z__[z___subscr(a_1,a_2)] /* -- LAPACK test routine (version 3.0) -- Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., Courant Institute, Argonne National Lab, and Rice University September 30, 1994 Purpose ======= CDRVGG checks the nonsymmetric generalized eigenvalue driver routines. T T T CGEGS factors A and B as Q S Z and Q T Z , where means transpose, T is upper triangular, S is in generalized Schur form (upper triangular), and Q and Z are unitary. It also computes the generalized eigenvalues (alpha(1),beta(1)), ..., (alpha(n),beta(n)), where alpha(j)=S(j,j) and beta(j)=T(j,j) -- thus, w(j) = alpha(j)/beta(j) is a root of the generalized eigenvalue problem det( A - w(j) B ) = 0 and m(j) = beta(j)/alpha(j) is a root of the essentially equivalent problem det( m(j) A - B ) = 0 CGEGV computes the generalized eigenvalues (alpha(1),beta(1)), ..., (alpha(n),beta(n)), the matrix L whose columns contain the generalized left eigenvectors l, and the matrix R whose columns contain the generalized right eigenvectors r for the pair (A,B). When CDRVGG is called, a number of matrix "sizes" ("n's") and a number of matrix "types" are specified. For each size ("n") and each type of matrix, one matrix will be generated and used to test the nonsymmetric eigenroutines. For each matrix, 7 tests will be performed and compared with the threshhold THRESH: Results from CGEGS: H (1) | A - Q S Z | / ( |A| n ulp ) H (2) | B - Q T Z | / ( |B| n ulp ) H (3) | I - QQ | / ( n ulp ) H (4) | I - ZZ | / ( n ulp ) (5) maximum over j of D(j) where: |alpha(j) - S(j,j)| |beta(j) - T(j,j)| D(j) = ------------------------ + ----------------------- max(|alpha(j)|,|S(j,j)|) max(|beta(j)|,|T(j,j)|) Results from CGEGV: (6) max over all left eigenvalue/-vector pairs (beta/alpha,l) of | l**H * (beta A - alpha B) | / ( ulp max( |beta A|, |alpha B| ) ) where l**H is the conjugate tranpose of l. (7) max over all right eigenvalue/-vector pairs (beta/alpha,r) of | (beta A - alpha B) r | / ( ulp max( |beta A|, |alpha B| ) ) Test Matrices ---- -------- The sizes of the test matrices are specified by an array NN(1:NSIZES); the value of each element NN(j) specifies one size. The "types" are specified by a logical array DOTYPE( 1:NTYPES ); if DOTYPE(j) is .TRUE., then matrix type "j" will be generated. Currently, the list of possible types is: (1) ( 0, 0 ) (a pair of zero matrices) (2) ( I, 0 ) (an identity and a zero matrix) (3) ( 0, I ) (an identity and a zero matrix) (4) ( I, I ) (a pair of identity matrices) t t (5) ( J , J ) (a pair of transposed Jordan blocks) t ( I 0 ) (6) ( X, Y ) where X = ( J 0 ) and Y = ( t ) ( 0 I ) ( 0 J ) and I is a k x k identity and J a (k+1)x(k+1) Jordan block; k=(N-1)/2 (7) ( D, I ) where D is diag( 0, 1,..., N-1 ) (a diagonal matrix with those diagonal entries.) (8) ( I, D ) (9) ( big*D, small*I ) where "big" is near overflow and small=1/big (10) ( small*D, big*I ) (11) ( big*I, small*D ) (12) ( small*I, big*D ) (13) ( big*D, big*I ) (14) ( small*D, small*I ) (15) ( D1, D2 ) where D1 is diag( 0, 0, 1, ..., N-3, 0 ) and D2 is diag( 0, N-3, N-4,..., 1, 0, 0 ) t t (16) Q ( J , J ) Z where Q and Z are random unitary matrices. (17) Q ( T1, T2 ) Z where T1 and T2 are upper triangular matrices with random O(1) entries above the diagonal and diagonal entries diag(T1) = ( 0, 0, 1, ..., N-3, 0 ) and diag(T2) = ( 0, N-3, N-4,..., 1, 0, 0 ) (18) Q ( T1, T2 ) Z diag(T1) = ( 0, 0, 1, 1, s, ..., s, 0 ) diag(T2) = ( 0, 1, 0, 1,..., 1, 0 ) s = machine precision. (19) Q ( T1, T2 ) Z diag(T1)=( 0,0,1,1, 1-d, ..., 1-(N-5)*d=s, 0 ) diag(T2) = ( 0, 1, 0, 1, ..., 1, 0 ) N-5 (20) Q ( T1, T2 ) Z diag(T1)=( 0, 0, 1, 1, a, ..., a =s, 0 ) diag(T2) = ( 0, 1, 0, 1, ..., 1, 0, 0 ) (21) Q ( T1, T2 ) Z diag(T1)=( 0, 0, 1, r1, r2, ..., r(N-4), 0 ) diag(T2) = ( 0, 1, 0, 1, ..., 1, 0, 0 ) where r1,..., r(N-4) are random. (22) Q ( big*T1, small*T2 ) Z diag(T1) = ( 0, 0, 1, ..., N-3, 0 ) diag(T2) = ( 0, 1, ..., 1, 0, 0 ) (23) Q ( small*T1, big*T2 ) Z diag(T1) = ( 0, 0, 1, ..., N-3, 0 ) diag(T2) = ( 0, 1, ..., 1, 0, 0 ) (24) Q ( small*T1, small*T2 ) Z diag(T1) = ( 0, 0, 1, ..., N-3, 0 ) diag(T2) = ( 0, 1, ..., 1, 0, 0 ) (25) Q ( big*T1, big*T2 ) Z diag(T1) = ( 0, 0, 1, ..., N-3, 0 ) diag(T2) = ( 0, 1, ..., 1, 0, 0 ) (26) Q ( T1, T2 ) Z where T1 and T2 are random upper-triangular matrices. Arguments ========= NSIZES (input) INTEGER The number of sizes of matrices to use. If it is zero, CDRVGG does nothing. It must be at least zero. NN (input) INTEGER array, dimension (NSIZES) An array containing the sizes to be used for the matrices. Zero values will be skipped. The values must be at least zero. NTYPES (input) INTEGER The number of elements in DOTYPE. If it is zero, CDRVGG does nothing. It must be at least zero. If it is MAXTYP+1 and NSIZES is 1, then an additional type, MAXTYP+1 is defined, which is to use whatever matrix is in A. This is only useful if DOTYPE(1:MAXTYP) is .FALSE. and DOTYPE(MAXTYP+1) is .TRUE. . DOTYPE (input) LOGICAL array, dimension (NTYPES) If DOTYPE(j) is .TRUE., then for each size in NN a matrix of that size and of type j will be generated. If NTYPES is smaller than the maximum number of types defined (PARAMETER MAXTYP), then types NTYPES+1 through MAXTYP will not be generated. If NTYPES is larger than MAXTYP, DOTYPE(MAXTYP+1) through DOTYPE(NTYPES) will be ignored. ISEED (input/output) INTEGER array, dimension (4) On entry ISEED specifies the seed of the random number generator. The array elements should be between 0 and 4095; if not they will be reduced mod 4096. Also, ISEED(4) must be odd. The random number generator uses a linear congruential sequence limited to small integers, and so should produce machine independent random numbers. The values of ISEED are changed on exit, and can be used in the next call to CDRVGG to continue the same random number sequence. THRESH (input) REAL A test will count as "failed" if the "error", computed as described above, exceeds THRESH. Note that the error is scaled to be O(1), so THRESH should be a reasonably small multiple of 1, e.g., 10 or 100. In particular, it should not depend on the precision (single vs. double) or the size of the matrix. It must be at least zero. THRSHN (input) REAL Threshhold for reporting eigenvector normalization error. If the normalization of any eigenvector differs from 1 by more than THRSHN*ulp, then a special error message will be printed. (This is handled separately from the other tests, since only a compiler or programming error should cause an error message, at least if THRSHN is at least 5--10.) NOUNIT (input) INTEGER The FORTRAN unit number for printing out error messages (e.g., if a routine returns IINFO not equal to 0.) A (input/workspace) COMPLEX array, dimension (LDA, max(NN)) Used to hold the original A matrix. Used as input only if NTYPES=MAXTYP+1, DOTYPE(1:MAXTYP)=.FALSE., and DOTYPE(MAXTYP+1)=.TRUE. LDA (input) INTEGER The leading dimension of A, B, S, T, S2, and T2. It must be at least 1 and at least max( NN ). B (input/workspace) COMPLEX array, dimension (LDA, max(NN)) Used to hold the original B matrix. Used as input only if NTYPES=MAXTYP+1, DOTYPE(1:MAXTYP)=.FALSE., and DOTYPE(MAXTYP+1)=.TRUE. S (workspace) COMPLEX array, dimension (LDA, max(NN)) The upper triangular matrix computed from A by CGEGS. T (workspace) COMPLEX array, dimension (LDA, max(NN)) The upper triangular matrix computed from B by CGEGS. S2 (workspace) COMPLEX array, dimension (LDA, max(NN)) The matrix computed from A by CGEGV. This will be the Schur (upper triangular) form of some matrix related to A, but will not, in general, be the same as S. T2 (workspace) COMPLEX array, dimension (LDA, max(NN)) The matrix computed from B by CGEGV. This will be the Schur form of some matrix related to B, but will not, in general, be the same as T. Q (workspace) COMPLEX array, dimension (LDQ, max(NN)) The (left) unitary matrix computed by CGEGS. LDQ (input) INTEGER The leading dimension of Q, Z, VL, and VR. It must be at least 1 and at least max( NN ). Z (workspace) COMPLEX array, dimension (LDQ, max(NN)) The (right) unitary matrix computed by CGEGS. ALPHA1 (workspace) COMPLEX array, dimension (max(NN)) BETA1 (workspace) COMPLEX array, dimension (max(NN)) The generalized eigenvalues of (A,B) computed by CGEGS. ALPHA1(k) / BETA1(k) is the k-th generalized eigenvalue of the matrices in A and B. ALPHA2 (workspace) COMPLEX array, dimension (max(NN)) BETA2 (workspace) COMPLEX array, dimension (max(NN)) The generalized eigenvalues of (A,B) computed by CGEGV. ALPHA2(k) / BETA2(k) is the k-th generalized eigenvalue of the matrices in A and B. VL (workspace) COMPLEX array, dimension (LDQ, max(NN)) The (lower triangular) left eigenvector matrix for the matrices in A and B. VR (workspace) COMPLEX array, dimension (LDQ, max(NN)) The (upper triangular) right eigenvector matrix for the matrices in A and B. WORK (workspace) COMPLEX array, dimension (LWORK) LWORK (input) INTEGER The number of entries in WORK. This must be at least MAX( 2*N, N*(NB+1), (k+1)*(2*k+N+1) ), where "k" is the sum of the blocksize and number-of-shifts for CHGEQZ, and NB is the greatest of the blocksizes for CGEQRF, CUNMQR, and CUNGQR. (The blocksizes and the number-of-shifts are retrieved through calls to ILAENV.) RWORK (workspace) REAL array, dimension (8*N) RESULT (output) REAL array, dimension (7) The values computed by the tests described above. The values are currently limited to 1/ulp, to avoid overflow. INFO (output) INTEGER = 0: successful exit < 0: if INFO = -i, the i-th argument had an illegal value. > 0: A routine returned an error code. INFO is the absolute value of the INFO value returned. ===================================================================== Parameter adjustments */ --nn; --dotype; --iseed; t2_dim1 = *lda; t2_offset = 1 + t2_dim1 * 1; t2 -= t2_offset; s2_dim1 = *lda; s2_offset = 1 + s2_dim1 * 1; s2 -= s2_offset; t_dim1 = *lda; t_offset = 1 + t_dim1 * 1; t -= t_offset; s_dim1 = *lda; s_offset = 1 + s_dim1 * 1; s -= s_offset; b_dim1 = *lda; b_offset = 1 + b_dim1 * 1; b -= b_offset; a_dim1 = *lda; a_offset = 1 + a_dim1 * 1; a -= a_offset; vr_dim1 = *ldq; vr_offset = 1 + vr_dim1 * 1; vr -= vr_offset; vl_dim1 = *ldq; vl_offset = 1 + vl_dim1 * 1; vl -= vl_offset; z_dim1 = *ldq; z_offset = 1 + z_dim1 * 1; z__ -= z_offset; q_dim1 = *ldq; q_offset = 1 + q_dim1 * 1; q -= q_offset; --alpha1; --beta1; --alpha2; --beta2; --work; --rwork; --result; /* Function Body Check for errors */ *info = 0; badnn = FALSE_; nmax = 1; i__1 = *nsizes; for (j = 1; j <= i__1; ++j) { /* Computing MAX */ i__2 = nmax, i__3 = nn[j]; nmax = max(i__2,i__3); if (nn[j] < 0) { badnn = TRUE_; } /* L10: */ } /* Maximum blocksize and shift -- we assume that blocksize and number of shifts are monotone increasing functions of N. Computing MAX */ i__1 = 1, i__2 = ilaenv_(&c__1, "CGEQRF", " ", &nmax, &nmax, &c_n1, &c_n1, (ftnlen)6, (ftnlen)1), i__1 = max(i__1,i__2), i__2 = ilaenv_(& c__1, "CUNMQR", "LC", &nmax, &nmax, &nmax, &c_n1, (ftnlen)6, ( ftnlen)2), i__1 = max(i__1,i__2), i__2 = ilaenv_(&c__1, "CUNGQR", " ", &nmax, &nmax, &nmax, &c_n1, (ftnlen)6, (ftnlen)1); nb = max(i__1,i__2); nbz = ilaenv_(&c__1, "CHGEQZ", "SII", &nmax, &c__1, &nmax, &c__0, (ftnlen) 6, (ftnlen)3); ns = ilaenv_(&c__4, "CHGEQZ", "SII", &nmax, &c__1, &nmax, &c__0, (ftnlen) 6, (ftnlen)3); i1 = nbz + ns; /* Computing MAX */ i__1 = nmax << 1, i__2 = nmax * (nb + 1), i__1 = max(i__1,i__2), i__2 = (( i1 << 1) + nmax + 1) * (i1 + 1); lwkopt = max(i__1,i__2); /* Check for errors */ if (*nsizes < 0) { *info = -1; } else if (badnn) { *info = -2; } else if (*ntypes < 0) { *info = -3; } else if (*thresh < 0.f) { *info = -6; } else if (*lda <= 1 || *lda < nmax) { *info = -10; } else if (*ldq <= 1 || *ldq < nmax) { *info = -19; } else if (lwkopt > *lwork) { *info = -30; } if (*info != 0) { i__1 = -(*info); xerbla_("CDRVGG", &i__1); return 0; } /* Quick return if possible */ if (*nsizes == 0 || *ntypes == 0) { return 0; } ulp = slamch_("Precision"); safmin = slamch_("Safe minimum"); safmin /= ulp; safmax = 1.f / safmin; slabad_(&safmin, &safmax); ulpinv = 1.f / ulp; /* The values RMAGN(2:3) depend on N, see below. */ rmagn[0] = 0.f; rmagn[1] = 1.f; /* Loop over sizes, types */ ntestt = 0; nerrs = 0; nmats = 0; i__1 = *nsizes; for (jsize = 1; jsize <= i__1; ++jsize) { n = nn[jsize]; n1 = max(1,n); rmagn[2] = safmax * ulp / (real) n1; rmagn[3] = safmin * ulpinv * n1; if (*nsizes != 1) { mtypes = min(26,*ntypes); } else { mtypes = min(27,*ntypes); } i__2 = mtypes; for (jtype = 1; jtype <= i__2; ++jtype) { if (! dotype[jtype]) { goto L150; } ++nmats; ntest = 0; /* Save ISEED in case of an error. */ for (j = 1; j <= 4; ++j) { ioldsd[j - 1] = iseed[j]; /* L20: */ } /* Initialize RESULT */ for (j = 1; j <= 7; ++j) { result[j] = 0.f; /* L30: */ } /* Compute A and B Description of control parameters: KCLASS: =1 means w/o rotation, =2 means w/ rotation, =3 means random. KATYPE: the "type" to be passed to CLATM4 for computing A. KAZERO: the pattern of zeros on the diagonal for A: =1: ( xxx ), =2: (0, xxx ) =3: ( 0, 0, xxx, 0 ), =4: ( 0, xxx, 0, 0 ), =5: ( 0, 0, 1, xxx, 0 ), =6: ( 0, 1, 0, xxx, 0 ). (xxx means a string of non-zero entries.) KAMAGN: the magnitude of the matrix: =0: zero, =1: O(1), =2: large, =3: small. LASIGN: .TRUE. if the diagonal elements of A are to be multiplied by a random magnitude 1 number. KBTYPE, KBZERO, KBMAGN, IBSIGN: the same, but for B. KTRIAN: =0: don't fill in the upper triangle, =1: do. KZ1, KZ2, KADD: used to implement KAZERO and KBZERO. RMAGN: used to implement KAMAGN and KBMAGN. */ if (mtypes > 26) { goto L110; } iinfo = 0; if (kclass[jtype - 1] < 3) { /* Generate A (w/o rotation) */ if ((i__3 = katype[jtype - 1], abs(i__3)) == 3) { in = ((n - 1) / 2 << 1) + 1; if (in != n) { claset_("Full", &n, &n, &c_b1, &c_b1, &a[a_offset], lda); } } else { in = n; } clatm4_(&katype[jtype - 1], &in, &kz1[kazero[jtype - 1] - 1], &kz2[kazero[jtype - 1] - 1], &lasign[jtype - 1], & rmagn[kamagn[jtype - 1]], &ulp, &rmagn[ktrian[jtype - 1] * kamagn[jtype - 1]], &c__2, &iseed[1], &a[ a_offset], lda); iadd = kadd[kazero[jtype - 1] - 1]; if (iadd > 0 && iadd <= n) { i__3 = a_subscr(iadd, iadd); i__4 = kamagn[jtype - 1]; a[i__3].r = rmagn[i__4], a[i__3].i = 0.f; } /* Generate B (w/o rotation) */ if ((i__3 = kbtype[jtype - 1], abs(i__3)) == 3) { in = ((n - 1) / 2 << 1) + 1; if (in != n) { claset_("Full", &n, &n, &c_b1, &c_b1, &b[b_offset], lda); } } else { in = n; } clatm4_(&kbtype[jtype - 1], &in, &kz1[kbzero[jtype - 1] - 1], &kz2[kbzero[jtype - 1] - 1], &lbsign[jtype - 1], & rmagn[kbmagn[jtype - 1]], &c_b39, &rmagn[ktrian[jtype - 1] * kbmagn[jtype - 1]], &c__2, &iseed[1], &b[ b_offset], lda); iadd = kadd[kbzero[jtype - 1] - 1]; if (iadd != 0 && iadd <= n) { i__3 = b_subscr(iadd, iadd); i__4 = kbmagn[jtype - 1]; b[i__3].r = rmagn[i__4], b[i__3].i = 0.f; } if (kclass[jtype - 1] == 2 && n > 0) { /* Include rotations Generate Q, Z as Householder transformations times a diagonal matrix. */ i__3 = n - 1; for (jc = 1; jc <= i__3; ++jc) { i__4 = n; for (jr = jc; jr <= i__4; ++jr) { i__5 = q_subscr(jr, jc); clarnd_(&q__1, &c__3, &iseed[1]); q[i__5].r = q__1.r, q[i__5].i = q__1.i; i__5 = z___subscr(jr, jc); clarnd_(&q__1, &c__3, &iseed[1]); z__[i__5].r = q__1.r, z__[i__5].i = q__1.i; /* L40: */ } i__4 = n + 1 - jc; clarfg_(&i__4, &q_ref(jc, jc), &q_ref(jc + 1, jc), & c__1, &work[jc]); i__4 = (n << 1) + jc; i__5 = q_subscr(jc, jc); r__2 = q[i__5].r; r__1 = r_sign(&c_b39, &r__2); work[i__4].r = r__1, work[i__4].i = 0.f; i__4 = q_subscr(jc, jc); q[i__4].r = 1.f, q[i__4].i = 0.f; i__4 = n + 1 - jc; clarfg_(&i__4, &z___ref(jc, jc), &z___ref(jc + 1, jc), &c__1, &work[n + jc]); i__4 = n * 3 + jc; i__5 = z___subscr(jc, jc); r__2 = z__[i__5].r; r__1 = r_sign(&c_b39, &r__2); work[i__4].r = r__1, work[i__4].i = 0.f; i__4 = z___subscr(jc, jc); z__[i__4].r = 1.f, z__[i__4].i = 0.f; /* L50: */ } clarnd_(&q__1, &c__3, &iseed[1]); ctemp.r = q__1.r, ctemp.i = q__1.i; i__3 = q_subscr(n, n); q[i__3].r = 1.f, q[i__3].i = 0.f; i__3 = n; work[i__3].r = 0.f, work[i__3].i = 0.f; i__3 = n * 3; r__1 = c_abs(&ctemp); q__1.r = ctemp.r / r__1, q__1.i = ctemp.i / r__1; work[i__3].r = q__1.r, work[i__3].i = q__1.i; clarnd_(&q__1, &c__3, &iseed[1]); ctemp.r = q__1.r, ctemp.i = q__1.i; i__3 = z___subscr(n, n); z__[i__3].r = 1.f, z__[i__3].i = 0.f; i__3 = n << 1; work[i__3].r = 0.f, work[i__3].i = 0.f; i__3 = n << 2; r__1 = c_abs(&ctemp); q__1.r = ctemp.r / r__1, q__1.i = ctemp.i / r__1; work[i__3].r = q__1.r, work[i__3].i = q__1.i; /* Apply the diagonal matrices */ i__3 = n; for (jc = 1; jc <= i__3; ++jc) { i__4 = n; for (jr = 1; jr <= i__4; ++jr) { i__5 = a_subscr(jr, jc); i__6 = (n << 1) + jr; r_cnjg(&q__3, &work[n * 3 + jc]); q__2.r = work[i__6].r * q__3.r - work[i__6].i * q__3.i, q__2.i = work[i__6].r * q__3.i + work[i__6].i * q__3.r; i__7 = a_subscr(jr, jc); q__1.r = q__2.r * a[i__7].r - q__2.i * a[i__7].i, q__1.i = q__2.r * a[i__7].i + q__2.i * a[ i__7].r; a[i__5].r = q__1.r, a[i__5].i = q__1.i; i__5 = b_subscr(jr, jc); i__6 = (n << 1) + jr; r_cnjg(&q__3, &work[n * 3 + jc]); q__2.r = work[i__6].r * q__3.r - work[i__6].i * q__3.i, q__2.i = work[i__6].r * q__3.i + work[i__6].i * q__3.r; i__7 = b_subscr(jr, jc); q__1.r = q__2.r * b[i__7].r - q__2.i * b[i__7].i, q__1.i = q__2.r * b[i__7].i + q__2.i * b[ i__7].r; b[i__5].r = q__1.r, b[i__5].i = q__1.i; /* L60: */ } /* L70: */ } i__3 = n - 1; cunm2r_("L", "N", &n, &n, &i__3, &q[q_offset], ldq, &work[ 1], &a[a_offset], lda, &work[(n << 1) + 1], & iinfo); if (iinfo != 0) { goto L100; } i__3 = n - 1; cunm2r_("R", "C", &n, &n, &i__3, &z__[z_offset], ldq, & work[n + 1], &a[a_offset], lda, &work[(n << 1) + 1], &iinfo); if (iinfo != 0) { goto L100; } i__3 = n - 1; cunm2r_("L", "N", &n, &n, &i__3, &q[q_offset], ldq, &work[ 1], &b[b_offset], lda, &work[(n << 1) + 1], & iinfo); if (iinfo != 0) { goto L100; } i__3 = n - 1; cunm2r_("R", "C", &n, &n, &i__3, &z__[z_offset], ldq, & work[n + 1], &b[b_offset], lda, &work[(n << 1) + 1], &iinfo); if (iinfo != 0) { goto L100; } } } else { /* Random matrices */ i__3 = n; for (jc = 1; jc <= i__3; ++jc) { i__4 = n; for (jr = 1; jr <= i__4; ++jr) { i__5 = a_subscr(jr, jc); i__6 = kamagn[jtype - 1]; clarnd_(&q__2, &c__4, &iseed[1]); q__1.r = rmagn[i__6] * q__2.r, q__1.i = rmagn[i__6] * q__2.i; a[i__5].r = q__1.r, a[i__5].i = q__1.i; i__5 = b_subscr(jr, jc); i__6 = kbmagn[jtype - 1]; clarnd_(&q__2, &c__4, &iseed[1]); q__1.r = rmagn[i__6] * q__2.r, q__1.i = rmagn[i__6] * q__2.i; b[i__5].r = q__1.r, b[i__5].i = q__1.i; /* L80: */ } /* L90: */ } } L100: if (iinfo != 0) { io___43.ciunit = *nounit; s_wsfe(&io___43); do_fio(&c__1, "Generator", (ftnlen)9); do_fio(&c__1, (char *)&iinfo, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)); do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof(integer)); e_wsfe(); *info = abs(iinfo); return 0; } L110: /* Call CGEGS to compute H, T, Q, Z, alpha, and beta. */ clacpy_(" ", &n, &n, &a[a_offset], lda, &s[s_offset], lda); clacpy_(" ", &n, &n, &b[b_offset], lda, &t[t_offset], lda); ntest = 1; result[1] = ulpinv; cgegs_("V", "V", &n, &s[s_offset], lda, &t[t_offset], lda, & alpha1[1], &beta1[1], &q[q_offset], ldq, &z__[z_offset], ldq, &work[1], lwork, &rwork[1], &iinfo); if (iinfo != 0) { io___44.ciunit = *nounit; s_wsfe(&io___44); do_fio(&c__1, "CGEGS", (ftnlen)5); do_fio(&c__1, (char *)&iinfo, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)); do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof(integer)); e_wsfe(); *info = abs(iinfo); goto L130; } ntest = 4; /* Do tests 1--4 */ cget51_(&c__1, &n, &a[a_offset], lda, &s[s_offset], lda, &q[ q_offset], ldq, &z__[z_offset], ldq, &work[1], &rwork[1], &result[1]); cget51_(&c__1, &n, &b[b_offset], lda, &t[t_offset], lda, &q[ q_offset], ldq, &z__[z_offset], ldq, &work[1], &rwork[1], &result[2]); cget51_(&c__3, &n, &b[b_offset], lda, &t[t_offset], lda, &q[ q_offset], ldq, &q[q_offset], ldq, &work[1], &rwork[1], & result[3]); cget51_(&c__3, &n, &b[b_offset], lda, &t[t_offset], lda, &z__[ z_offset], ldq, &z__[z_offset], ldq, &work[1], &rwork[1], &result[4]); /* Do test 5: compare eigenvalues with diagonals. */ temp1 = 0.f; i__3 = n; for (j = 1; j <= i__3; ++j) { i__4 = j; i__5 = s_subscr(j, j); q__2.r = alpha1[i__4].r - s[i__5].r, q__2.i = alpha1[i__4].i - s[i__5].i; q__1.r = q__2.r, q__1.i = q__2.i; i__6 = j; i__7 = t_subscr(j, j); q__4.r = beta1[i__6].r - t[i__7].r, q__4.i = beta1[i__6].i - t[i__7].i; q__3.r = q__4.r, q__3.i = q__4.i; /* Computing MAX */ i__8 = j; i__9 = s_subscr(j, j); r__13 = safmin, r__14 = (r__1 = alpha1[i__8].r, dabs(r__1)) + (r__2 = r_imag(&alpha1[j]), dabs(r__2)), r__13 = max( r__13,r__14), r__14 = (r__3 = s[i__9].r, dabs(r__3)) + (r__4 = r_imag(&s_ref(j, j)), dabs(r__4)); /* Computing MAX */ i__10 = j; i__11 = t_subscr(j, j); r__15 = safmin, r__16 = (r__5 = beta1[i__10].r, dabs(r__5)) + (r__6 = r_imag(&beta1[j]), dabs(r__6)), r__15 = max( r__15,r__16), r__16 = (r__7 = t[i__11].r, dabs(r__7)) + (r__8 = r_imag(&t_ref(j, j)), dabs(r__8)); temp2 = (((r__9 = q__1.r, dabs(r__9)) + (r__10 = r_imag(&q__1) , dabs(r__10))) / dmax(r__13,r__14) + ((r__11 = q__3.r, dabs(r__11)) + (r__12 = r_imag(&q__3), dabs( r__12))) / dmax(r__15,r__16)) / ulp; temp1 = dmax(temp1,temp2); /* L120: */ } result[5] = temp1; /* Call CGEGV to compute S2, T2, VL, and VR, do tests. Eigenvalues and Eigenvectors */ clacpy_(" ", &n, &n, &a[a_offset], lda, &s2[s2_offset], lda); clacpy_(" ", &n, &n, &b[b_offset], lda, &t2[t2_offset], lda); ntest = 6; result[6] = ulpinv; cgegv_("V", "V", &n, &s2[s2_offset], lda, &t2[t2_offset], lda, & alpha2[1], &beta2[1], &vl[vl_offset], ldq, &vr[vr_offset], ldq, &work[1], lwork, &rwork[1], &iinfo); if (iinfo != 0) { io___47.ciunit = *nounit; s_wsfe(&io___47); do_fio(&c__1, "CGEGV", (ftnlen)5); do_fio(&c__1, (char *)&iinfo, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)); do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof(integer)); e_wsfe(); *info = abs(iinfo); goto L130; } ntest = 7; /* Do Tests 6 and 7 */ cget52_(&c_true, &n, &a[a_offset], lda, &b[b_offset], lda, &vl[ vl_offset], ldq, &alpha2[1], &beta2[1], &work[1], &rwork[ 1], dumma); result[6] = dumma[0]; if (dumma[1] > *thrshn) { io___49.ciunit = *nounit; s_wsfe(&io___49); do_fio(&c__1, "Left", (ftnlen)4); do_fio(&c__1, "CGEGV", (ftnlen)5); do_fio(&c__1, (char *)&dumma[1], (ftnlen)sizeof(real)); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)); do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof(integer)); e_wsfe(); } cget52_(&c_false, &n, &a[a_offset], lda, &b[b_offset], lda, &vr[ vr_offset], ldq, &alpha2[1], &beta2[1], &work[1], &rwork[ 1], dumma); result[7] = dumma[0]; if (dumma[1] > *thresh) { io___50.ciunit = *nounit; s_wsfe(&io___50); do_fio(&c__1, "Right", (ftnlen)5); do_fio(&c__1, "CGEGV", (ftnlen)5); do_fio(&c__1, (char *)&dumma[1], (ftnlen)sizeof(real)); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)); do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof(integer)); e_wsfe(); } /* End of Loop -- Check for RESULT(j) > THRESH */ L130: ntestt += ntest; /* Print out tests which fail. */ i__3 = ntest; for (jr = 1; jr <= i__3; ++jr) { if (result[jr] >= *thresh) { /* If this is the first test to fail, print a header to the data file. */ if (nerrs == 0) { io___51.ciunit = *nounit; s_wsfe(&io___51); do_fio(&c__1, "CGG", (ftnlen)3); e_wsfe(); /* Matrix types */ io___52.ciunit = *nounit; s_wsfe(&io___52); e_wsfe(); io___53.ciunit = *nounit; s_wsfe(&io___53); e_wsfe(); io___54.ciunit = *nounit; s_wsfe(&io___54); do_fio(&c__1, "Unitary", (ftnlen)7); e_wsfe(); /* Tests performed */ io___55.ciunit = *nounit; s_wsfe(&io___55); do_fio(&c__1, "unitary", (ftnlen)7); do_fio(&c__1, "*", (ftnlen)1); do_fio(&c__1, "conjugate transpose", (ftnlen)19); for (j = 1; j <= 5; ++j) { do_fio(&c__1, "*", (ftnlen)1); } e_wsfe(); } ++nerrs; if (result[jr] < 1e4f) { io___56.ciunit = *nounit; s_wsfe(&io___56); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)) ; do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof( integer)); do_fio(&c__1, (char *)&jr, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&result[jr], (ftnlen)sizeof( real)); e_wsfe(); } else { io___57.ciunit = *nounit; s_wsfe(&io___57); do_fio(&c__1, (char *)&n, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&jtype, (ftnlen)sizeof(integer)) ; do_fio(&c__4, (char *)&ioldsd[0], (ftnlen)sizeof( integer)); do_fio(&c__1, (char *)&jr, (ftnlen)sizeof(integer)); do_fio(&c__1, (char *)&result[jr], (ftnlen)sizeof( real)); e_wsfe(); } } /* L140: */ } L150: ; } /* L160: */ } /* Summary */ alasvm_("CGG", nounit, &nerrs, &ntestt, &c__0); return 0; /* End of CDRVGG */ } /* cdrvgg_ */
/* Subroutine */ int zlahqr_(logical *wantt, logical *wantz, integer *n, integer *ilo, integer *ihi, doublecomplex *h__, integer *ldh, doublecomplex *w, integer *iloz, integer *ihiz, doublecomplex *z__, integer *ldz, integer *info) { /* System generated locals */ integer h_dim1, h_offset, z_dim1, z_offset, i__1, i__2, i__3, i__4, i__5; doublereal d__1, d__2, d__3, d__4, d__5, d__6; doublecomplex z__1, z__2, z__3, z__4; /* Builtin functions */ double d_imag(doublecomplex *); void z_sqrt(doublecomplex *, doublecomplex *), d_cnjg(doublecomplex *, doublecomplex *); double z_abs(doublecomplex *); /* Local variables */ static doublecomplex temp; static doublereal opst; static integer i__, j, k, l, m; static doublereal s; static doublecomplex t, u, v[2], x, y; extern /* Subroutine */ int zscal_(integer *, doublecomplex *, doublecomplex *, integer *); static doublereal rtemp; static integer i1, i2; static doublereal rwork[1]; static doublecomplex t1; static doublereal t2; extern /* Subroutine */ int zcopy_(integer *, doublecomplex *, integer *, doublecomplex *, integer *); static doublecomplex v2; static doublereal h10; static doublecomplex h11; static doublereal h21; static doublecomplex h22; static integer nh; extern doublereal dlamch_(char *); static integer nz; extern /* Subroutine */ int zlarfg_(integer *, doublecomplex *, doublecomplex *, integer *, doublecomplex *); extern /* Double Complex */ VOID zladiv_(doublecomplex *, doublecomplex *, doublecomplex *); extern doublereal zlanhs_(char *, integer *, doublecomplex *, integer *, doublereal *); static doublereal smlnum; static doublecomplex h11s; static integer itn, its; static doublereal ulp; static doublecomplex sum; static doublereal tst1; #define h___subscr(a_1,a_2) (a_2)*h_dim1 + a_1 #define h___ref(a_1,a_2) h__[h___subscr(a_1,a_2)] #define z___subscr(a_1,a_2) (a_2)*z_dim1 + a_1 #define z___ref(a_1,a_2) z__[z___subscr(a_1,a_2)] /* -- LAPACK auxiliary routine (instrumented to count operations) -- Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., Courant Institute, Argonne National Lab, and Rice University June 30, 1999 Common block to return operation count. Purpose ======= ZLAHQR is an auxiliary routine called by ZHSEQR to update the eigenvalues and Schur decomposition already computed by ZHSEQR, by dealing with the Hessenberg submatrix in rows and columns ILO to IHI. Arguments ========= WANTT (input) LOGICAL = .TRUE. : the full Schur form T is required; = .FALSE.: only eigenvalues are required. WANTZ (input) LOGICAL = .TRUE. : the matrix of Schur vectors Z is required; = .FALSE.: Schur vectors are not required. N (input) INTEGER The order of the matrix H. N >= 0. ILO (input) INTEGER IHI (input) INTEGER It is assumed that H is already upper triangular in rows and columns IHI+1:N, and that H(ILO,ILO-1) = 0 (unless ILO = 1). ZLAHQR works primarily with the Hessenberg submatrix in rows and columns ILO to IHI, but applies transformations to all of H if WANTT is .TRUE.. 1 <= ILO <= max(1,IHI); IHI <= N. H (input/output) COMPLEX*16 array, dimension (LDH,N) On entry, the upper Hessenberg matrix H. On exit, if WANTT is .TRUE., H is upper triangular in rows and columns ILO:IHI, with any 2-by-2 diagonal blocks in standard form. If WANTT is .FALSE., the contents of H are unspecified on exit. LDH (input) INTEGER The leading dimension of the array H. LDH >= max(1,N). W (output) COMPLEX*16 array, dimension (N) The computed eigenvalues ILO to IHI are stored in the corresponding elements of W. If WANTT is .TRUE., the eigenvalues are stored in the same order as on the diagonal of the Schur form returned in H, with W(i) = H(i,i). ILOZ (input) INTEGER IHIZ (input) INTEGER Specify the rows of Z to which transformations must be applied if WANTZ is .TRUE.. 1 <= ILOZ <= ILO; IHI <= IHIZ <= N. Z (input/output) COMPLEX*16 array, dimension (LDZ,N) If WANTZ is .TRUE., on entry Z must contain the current matrix Z of transformations accumulated by ZHSEQR, and on exit Z has been updated; transformations are applied only to the submatrix Z(ILOZ:IHIZ,ILO:IHI). If WANTZ is .FALSE., Z is not referenced. LDZ (input) INTEGER The leading dimension of the array Z. LDZ >= max(1,N). INFO (output) INTEGER = 0: successful exit > 0: if INFO = i, ZLAHQR failed to compute all the eigenvalues ILO to IHI in a total of 30*(IHI-ILO+1) iterations; elements i+1:ihi of W contain those eigenvalues which have been successfully computed. ===================================================================== Parameter adjustments */ h_dim1 = *ldh; h_offset = 1 + h_dim1 * 1; h__ -= h_offset; --w; z_dim1 = *ldz; z_offset = 1 + z_dim1 * 1; z__ -= z_offset; /* Function Body */ *info = 0; /* ** Initialize */ opst = 0.; /* ** Quick return if possible */ if (*n == 0) { return 0; } if (*ilo == *ihi) { i__1 = *ilo; i__2 = h___subscr(*ilo, *ilo); w[i__1].r = h__[i__2].r, w[i__1].i = h__[i__2].i; return 0; } nh = *ihi - *ilo + 1; nz = *ihiz - *iloz + 1; /* Set machine-dependent constants for the stopping criterion. If norm(H) <= sqrt(OVFL), overflow should not occur. */ ulp = dlamch_("Precision"); smlnum = dlamch_("Safe minimum") / ulp; /* I1 and I2 are the indices of the first row and last column of H to which transformations must be applied. If eigenvalues only are being computed, I1 and I2 are set inside the main loop. */ if (*wantt) { i1 = 1; i2 = *n; } /* ITN is the total number of QR iterations allowed. */ itn = nh * 30; /* The main loop begins here. I is the loop index and decreases from IHI to ILO in steps of 1. Each iteration of the loop works with the active submatrix in rows and columns L to I. Eigenvalues I+1 to IHI have already converged. Either L = ILO, or H(L,L-1) is negligible so that the matrix splits. */ i__ = *ihi; L10: if (i__ < *ilo) { goto L130; } /* Perform QR iterations on rows and columns ILO to I until a submatrix of order 1 splits off at the bottom because a subdiagonal element has become negligible. */ l = *ilo; i__1 = itn; for (its = 0; its <= i__1; ++its) { /* Look for a single small subdiagonal element. */ i__2 = l + 1; for (k = i__; k >= i__2; --k) { i__3 = h___subscr(k - 1, k - 1); i__4 = h___subscr(k, k); tst1 = (d__1 = h__[i__3].r, abs(d__1)) + (d__2 = d_imag(&h___ref( k - 1, k - 1)), abs(d__2)) + ((d__3 = h__[i__4].r, abs( d__3)) + (d__4 = d_imag(&h___ref(k, k)), abs(d__4))); if (tst1 == 0.) { i__3 = i__ - l + 1; tst1 = zlanhs_("1", &i__3, &h___ref(l, l), ldh, rwork); /* ** Increment op count */ latime_1.ops += (i__ - l + 1) * 5 * (i__ - l) / 2; /* ** */ } i__3 = h___subscr(k, k - 1); /* Computing MAX */ d__2 = ulp * tst1; if ((d__1 = h__[i__3].r, abs(d__1)) <= max(d__2,smlnum)) { goto L30; } /* L20: */ } L30: l = k; /* ** Increment op count */ opst += (i__ - l + 1) * 5; /* ** */ if (l > *ilo) { /* H(L,L-1) is negligible */ i__2 = h___subscr(l, l - 1); h__[i__2].r = 0., h__[i__2].i = 0.; } /* Exit from loop if a submatrix of order 1 has split off. */ if (l >= i__) { goto L120; } /* Now the active submatrix is in rows and columns L to I. If eigenvalues only are being computed, only the active submatrix need be transformed. */ if (! (*wantt)) { i1 = l; i2 = i__; } if (its == 10 || its == 20) { /* Exceptional shift. */ i__2 = h___subscr(i__, i__ - 1); s = (d__1 = h__[i__2].r, abs(d__1)) * .75; i__2 = h___subscr(i__, i__); z__1.r = s + h__[i__2].r, z__1.i = h__[i__2].i; t.r = z__1.r, t.i = z__1.i; /* ** Increment op count */ opst += 1; /* ** */ } else { /* Wilkinson's shift. */ i__2 = h___subscr(i__, i__); t.r = h__[i__2].r, t.i = h__[i__2].i; i__2 = h___subscr(i__ - 1, i__); i__3 = h___subscr(i__, i__ - 1); d__1 = h__[i__3].r; z__1.r = d__1 * h__[i__2].r, z__1.i = d__1 * h__[i__2].i; u.r = z__1.r, u.i = z__1.i; /* ** Increment op count */ opst += 2; /* ** */ if (u.r != 0. || u.i != 0.) { i__2 = h___subscr(i__ - 1, i__ - 1); z__2.r = h__[i__2].r - t.r, z__2.i = h__[i__2].i - t.i; z__1.r = z__2.r * .5, z__1.i = z__2.i * .5; x.r = z__1.r, x.i = z__1.i; z__3.r = x.r * x.r - x.i * x.i, z__3.i = x.r * x.i + x.i * x.r; z__2.r = z__3.r + u.r, z__2.i = z__3.i + u.i; z_sqrt(&z__1, &z__2); y.r = z__1.r, y.i = z__1.i; if (x.r * y.r + d_imag(&x) * d_imag(&y) < 0.) { z__1.r = -y.r, z__1.i = -y.i; y.r = z__1.r, y.i = z__1.i; } z__3.r = x.r + y.r, z__3.i = x.i + y.i; zladiv_(&z__2, &u, &z__3); z__1.r = t.r - z__2.r, z__1.i = t.i - z__2.i; t.r = z__1.r, t.i = z__1.i; /* ** Increment op count */ opst += 20; /* ** */ } } /* Look for two consecutive small subdiagonal elements. */ i__2 = l + 1; for (m = i__ - 1; m >= i__2; --m) { /* Determine the effect of starting the single-shift QR iteration at row M, and see if this would make H(M,M-1) negligible. */ i__3 = h___subscr(m, m); h11.r = h__[i__3].r, h11.i = h__[i__3].i; i__3 = h___subscr(m + 1, m + 1); h22.r = h__[i__3].r, h22.i = h__[i__3].i; z__1.r = h11.r - t.r, z__1.i = h11.i - t.i; h11s.r = z__1.r, h11s.i = z__1.i; i__3 = h___subscr(m + 1, m); h21 = h__[i__3].r; s = (d__1 = h11s.r, abs(d__1)) + (d__2 = d_imag(&h11s), abs(d__2)) + abs(h21); z__1.r = h11s.r / s, z__1.i = h11s.i / s; h11s.r = z__1.r, h11s.i = z__1.i; h21 /= s; v[0].r = h11s.r, v[0].i = h11s.i; v[1].r = h21, v[1].i = 0.; i__3 = h___subscr(m, m - 1); h10 = h__[i__3].r; tst1 = ((d__1 = h11s.r, abs(d__1)) + (d__2 = d_imag(&h11s), abs( d__2))) * ((d__3 = h11.r, abs(d__3)) + (d__4 = d_imag(& h11), abs(d__4)) + ((d__5 = h22.r, abs(d__5)) + (d__6 = d_imag(&h22), abs(d__6)))); if ((d__1 = h10 * h21, abs(d__1)) <= ulp * tst1) { goto L50; } /* L40: */ } i__2 = h___subscr(l, l); h11.r = h__[i__2].r, h11.i = h__[i__2].i; i__2 = h___subscr(l + 1, l + 1); h22.r = h__[i__2].r, h22.i = h__[i__2].i; z__1.r = h11.r - t.r, z__1.i = h11.i - t.i; h11s.r = z__1.r, h11s.i = z__1.i; i__2 = h___subscr(l + 1, l); h21 = h__[i__2].r; s = (d__1 = h11s.r, abs(d__1)) + (d__2 = d_imag(&h11s), abs(d__2)) + abs(h21); z__1.r = h11s.r / s, z__1.i = h11s.i / s; h11s.r = z__1.r, h11s.i = z__1.i; h21 /= s; v[0].r = h11s.r, v[0].i = h11s.i; v[1].r = h21, v[1].i = 0.; L50: /* ** Increment op count */ opst += (i__ - m) * 14; /* ** Single-shift QR step */ i__2 = i__ - 1; for (k = m; k <= i__2; ++k) { /* The first iteration of this loop determines a reflection G from the vector V and applies it from left and right to H, thus creating a nonzero bulge below the subdiagonal. Each subsequent iteration determines a reflection G to restore the Hessenberg form in the (K-1)th column, and thus chases the bulge one step toward the bottom of the active submatrix. V(2) is always real before the call to ZLARFG, and hence after the call T2 ( = T1*V(2) ) is also real. */ if (k > m) { zcopy_(&c__2, &h___ref(k, k - 1), &c__1, v, &c__1); } zlarfg_(&c__2, v, &v[1], &c__1, &t1); /* ** Increment op count */ opst += 38; /* ** */ if (k > m) { i__3 = h___subscr(k, k - 1); h__[i__3].r = v[0].r, h__[i__3].i = v[0].i; i__3 = h___subscr(k + 1, k - 1); h__[i__3].r = 0., h__[i__3].i = 0.; } v2.r = v[1].r, v2.i = v[1].i; z__1.r = t1.r * v2.r - t1.i * v2.i, z__1.i = t1.r * v2.i + t1.i * v2.r; t2 = z__1.r; /* Apply G from the left to transform the rows of the matrix in columns K to I2. */ i__3 = i2; for (j = k; j <= i__3; ++j) { d_cnjg(&z__3, &t1); i__4 = h___subscr(k, j); z__2.r = z__3.r * h__[i__4].r - z__3.i * h__[i__4].i, z__2.i = z__3.r * h__[i__4].i + z__3.i * h__[i__4].r; i__5 = h___subscr(k + 1, j); z__4.r = t2 * h__[i__5].r, z__4.i = t2 * h__[i__5].i; z__1.r = z__2.r + z__4.r, z__1.i = z__2.i + z__4.i; sum.r = z__1.r, sum.i = z__1.i; i__4 = h___subscr(k, j); i__5 = h___subscr(k, j); z__1.r = h__[i__5].r - sum.r, z__1.i = h__[i__5].i - sum.i; h__[i__4].r = z__1.r, h__[i__4].i = z__1.i; i__4 = h___subscr(k + 1, j); i__5 = h___subscr(k + 1, j); z__2.r = sum.r * v2.r - sum.i * v2.i, z__2.i = sum.r * v2.i + sum.i * v2.r; z__1.r = h__[i__5].r - z__2.r, z__1.i = h__[i__5].i - z__2.i; h__[i__4].r = z__1.r, h__[i__4].i = z__1.i; /* L60: */ } /* Apply G from the right to transform the columns of the matrix in rows I1 to min(K+2,I). Computing MIN */ i__4 = k + 2; i__3 = min(i__4,i__); for (j = i1; j <= i__3; ++j) { i__4 = h___subscr(j, k); z__2.r = t1.r * h__[i__4].r - t1.i * h__[i__4].i, z__2.i = t1.r * h__[i__4].i + t1.i * h__[i__4].r; i__5 = h___subscr(j, k + 1); z__3.r = t2 * h__[i__5].r, z__3.i = t2 * h__[i__5].i; z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i; sum.r = z__1.r, sum.i = z__1.i; i__4 = h___subscr(j, k); i__5 = h___subscr(j, k); z__1.r = h__[i__5].r - sum.r, z__1.i = h__[i__5].i - sum.i; h__[i__4].r = z__1.r, h__[i__4].i = z__1.i; i__4 = h___subscr(j, k + 1); i__5 = h___subscr(j, k + 1); d_cnjg(&z__3, &v2); z__2.r = sum.r * z__3.r - sum.i * z__3.i, z__2.i = sum.r * z__3.i + sum.i * z__3.r; z__1.r = h__[i__5].r - z__2.r, z__1.i = h__[i__5].i - z__2.i; h__[i__4].r = z__1.r, h__[i__4].i = z__1.i; /* L70: */ } /* ** Increment op count Computing MIN */ i__3 = 2, i__4 = i__ - k; latime_1.ops += (i2 - i1 + 2 + min(i__3,i__4)) * 20; /* ** */ if (*wantz) { /* Accumulate transformations in the matrix Z */ i__3 = *ihiz; for (j = *iloz; j <= i__3; ++j) { i__4 = z___subscr(j, k); z__2.r = t1.r * z__[i__4].r - t1.i * z__[i__4].i, z__2.i = t1.r * z__[i__4].i + t1.i * z__[i__4].r; i__5 = z___subscr(j, k + 1); z__3.r = t2 * z__[i__5].r, z__3.i = t2 * z__[i__5].i; z__1.r = z__2.r + z__3.r, z__1.i = z__2.i + z__3.i; sum.r = z__1.r, sum.i = z__1.i; i__4 = z___subscr(j, k); i__5 = z___subscr(j, k); z__1.r = z__[i__5].r - sum.r, z__1.i = z__[i__5].i - sum.i; z__[i__4].r = z__1.r, z__[i__4].i = z__1.i; i__4 = z___subscr(j, k + 1); i__5 = z___subscr(j, k + 1); d_cnjg(&z__3, &v2); z__2.r = sum.r * z__3.r - sum.i * z__3.i, z__2.i = sum.r * z__3.i + sum.i * z__3.r; z__1.r = z__[i__5].r - z__2.r, z__1.i = z__[i__5].i - z__2.i; z__[i__4].r = z__1.r, z__[i__4].i = z__1.i; /* L80: */ } /* ** Increment op count */ latime_1.ops += nz * 20; /* ** */ } if (k == m && m > l) { /* If the QR step was started at row M > L because two consecutive small subdiagonals were found, then extra scaling must be performed to ensure that H(M,M-1) remains real. */ z__1.r = 1. - t1.r, z__1.i = 0. - t1.i; temp.r = z__1.r, temp.i = z__1.i; d__1 = z_abs(&temp); z__1.r = temp.r / d__1, z__1.i = temp.i / d__1; temp.r = z__1.r, temp.i = z__1.i; i__3 = h___subscr(m + 1, m); i__4 = h___subscr(m + 1, m); d_cnjg(&z__2, &temp); z__1.r = h__[i__4].r * z__2.r - h__[i__4].i * z__2.i, z__1.i = h__[i__4].r * z__2.i + h__[i__4].i * z__2.r; h__[i__3].r = z__1.r, h__[i__3].i = z__1.i; if (m + 2 <= i__) { i__3 = h___subscr(m + 2, m + 1); i__4 = h___subscr(m + 2, m + 1); z__1.r = h__[i__4].r * temp.r - h__[i__4].i * temp.i, z__1.i = h__[i__4].r * temp.i + h__[i__4].i * temp.r; h__[i__3].r = z__1.r, h__[i__3].i = z__1.i; } i__3 = i__; for (j = m; j <= i__3; ++j) { if (j != m + 1) { if (i2 > j) { i__4 = i2 - j; zscal_(&i__4, &temp, &h___ref(j, j + 1), ldh); } i__4 = j - i1; d_cnjg(&z__1, &temp); zscal_(&i__4, &z__1, &h___ref(i1, j), &c__1); /* ** Increment op count */ opst += (i2 - i1 + 3) * 6; /* ** */ if (*wantz) { d_cnjg(&z__1, &temp); zscal_(&nz, &z__1, &z___ref(*iloz, j), &c__1); /* ** Increment op count */ opst += nz * 6; /* ** */ } } /* L90: */ } } /* L100: */ } /* Ensure that H(I,I-1) is real. */ i__2 = h___subscr(i__, i__ - 1); temp.r = h__[i__2].r, temp.i = h__[i__2].i; if (d_imag(&temp) != 0.) { rtemp = z_abs(&temp); i__2 = h___subscr(i__, i__ - 1); h__[i__2].r = rtemp, h__[i__2].i = 0.; z__1.r = temp.r / rtemp, z__1.i = temp.i / rtemp; temp.r = z__1.r, temp.i = z__1.i; if (i2 > i__) { i__2 = i2 - i__; d_cnjg(&z__1, &temp); zscal_(&i__2, &z__1, &h___ref(i__, i__ + 1), ldh); } i__2 = i__ - i1; zscal_(&i__2, &temp, &h___ref(i1, i__), &c__1); /* ** Increment op count */ opst += (i2 - i1 + 1) * 6; /* ** */ if (*wantz) { zscal_(&nz, &temp, &z___ref(*iloz, i__), &c__1); /* ** Increment op count */ opst += nz * 6; /* ** */ } } /* L110: */ } /* Failure to converge in remaining number of iterations */ *info = i__; return 0; L120: /* H(I,I-1) is negligible: one eigenvalue has converged. */ i__1 = i__; i__2 = h___subscr(i__, i__); w[i__1].r = h__[i__2].r, w[i__1].i = h__[i__2].i; /* Decrement number of remaining iterations, and return to start of the main loop with new value of I. */ itn -= its; i__ = l - 1; goto L10; L130: /* ** Compute final op count */ latime_1.ops += opst; /* ** */ return 0; /* End of ZLAHQR */ } /* zlahqr_ */
/* Subroutine */ int clarrv_(integer *n, real *d__, real *l, integer *isplit, integer *m, real *w, integer *iblock, real *gersch, real *tol, complex *z__, integer *ldz, integer *isuppz, real *work, integer * iwork, integer *info) { /* System generated locals */ integer z_dim1, z_offset, i__1, i__2, i__3, i__4, i__5, i__6; real r__1, r__2; complex q__1, q__2; /* Builtin functions */ double sqrt(doublereal); /* Local variables */ static integer iend, jblk, iter, temp[1], ktot, itmp1, itmp2, i__, j, k, p, q, indld; static real sigma; static integer ndone, iinfo, iindr; static real resid; extern /* Complex */ VOID cdotu_(complex *, integer *, complex *, integer *, complex *, integer *); static integer nclus; extern /* Subroutine */ int caxpy_(integer *, complex *, complex *, integer *, complex *, integer *), scopy_(integer *, real *, integer *, real *, integer *); static integer iindc1, iindc2, indin1, indin2; extern /* Subroutine */ int clar1v_(integer *, integer *, integer *, real *, real *, real *, real *, real *, real *, complex *, real *, real *, integer *, integer *, real *); extern doublereal scnrm2_(integer *, complex *, integer *); static real lambda; static integer im, in, ibegin, indgap, indlld; extern doublereal slamch_(char *); static real mingma; extern /* Subroutine */ int csscal_(integer *, real *, complex *, integer *); static integer oldien, oldncl; static real relgap; extern /* Subroutine */ int claset_(char *, integer *, integer *, complex *, complex *, complex *, integer *); static integer oldcls, ndepth, inderr, iindwk; extern /* Subroutine */ int cstein_(integer *, real *, real *, integer *, real *, integer *, integer *, complex *, integer *, real *, integer *, integer *, integer *), slarrb_(integer *, real *, real *, real *, real *, integer *, integer *, real *, real *, real *, real *, real *, real *, integer *, integer *); static logical mgscls; static integer lsbdpt, newcls, oldfst; static real minrgp; static integer indwrk; extern /* Subroutine */ int slarrf_(integer *, real *, real *, real *, real *, integer *, integer *, real *, real *, real *, real *, integer *, integer *); static integer oldlst; static real reltol; static integer maxitr, newfrs, newftt; static real mgstol; static integer nsplit; static real nrminv, rqcorr; static integer newlst, newsiz; static real gap, eps, ztz, tmp1; #define z___subscr(a_1,a_2) (a_2)*z_dim1 + a_1 #define z___ref(a_1,a_2) z__[z___subscr(a_1,a_2)] /* -- LAPACK auxiliary routine (instru to count ops, version 3.0) -- Univ. of Tennessee, Univ. of California Berkeley, NAG Ltd., Courant Institute, Argonne National Lab, and Rice University June 30, 1999 Common block to return operation count .. Purpose ======= CLARRV computes the eigenvectors of the tridiagonal matrix T = L D L^T given L, D and the eigenvalues of L D L^T. The input eigenvalues should have high relative accuracy with respect to the entries of L and D. The desired accuracy of the output can be specified by the input parameter TOL. Arguments ========= N (input) INTEGER The order of the matrix. N >= 0. D (input/output) REAL array, dimension (N) On entry, the n diagonal elements of the diagonal matrix D. On exit, D may be overwritten. L (input/output) REAL array, dimension (N-1) On entry, the (n-1) subdiagonal elements of the unit bidiagonal matrix L in elements 1 to N-1 of L. L(N) need not be set. On exit, L is overwritten. ISPLIT (input) INTEGER array, dimension (N) The splitting points, at which T breaks up into submatrices. The first submatrix consists of rows/columns 1 to ISPLIT( 1 ), the second of rows/columns ISPLIT( 1 )+1 through ISPLIT( 2 ), etc. TOL (input) REAL The absolute error tolerance for the eigenvalues/eigenvectors. Errors in the input eigenvalues must be bounded by TOL. The eigenvectors output have residual norms bounded by TOL, and the dot products between different eigenvectors are bounded by TOL. TOL must be at least N*EPS*|T|, where EPS is the machine precision and |T| is the 1-norm of the tridiagonal matrix. M (input) INTEGER The total number of eigenvalues found. 0 <= M <= N. If RANGE = 'A', M = N, and if RANGE = 'I', M = IU-IL+1. W (input) REAL array, dimension (N) The first M elements of W contain the eigenvalues for which eigenvectors are to be computed. The eigenvalues should be grouped by split-off block and ordered from smallest to largest within the block ( The output array W from SLARRE is expected here ). Errors in W must be bounded by TOL (see above). IBLOCK (input) INTEGER array, dimension (N) The submatrix indices associated with the corresponding eigenvalues in W; IBLOCK(i)=1 if eigenvalue W(i) belongs to the first submatrix from the top, =2 if W(i) belongs to the second submatrix, etc. Z (output) COMPLEX array, dimension (LDZ, max(1,M) ) If JOBZ = 'V', then if INFO = 0, the first M columns of Z contain the orthonormal eigenvectors of the matrix T corresponding to the selected eigenvalues, with the i-th column of Z holding the eigenvector associated with W(i). If JOBZ = 'N', then Z is not referenced. Note: the user must ensure that at least max(1,M) columns are supplied in the array Z; if RANGE = 'V', the exact value of M is not known in advance and an upper bound must be used. LDZ (input) INTEGER The leading dimension of the array Z. LDZ >= 1, and if JOBZ = 'V', LDZ >= max(1,N). ISUPPZ (output) INTEGER ARRAY, dimension ( 2*max(1,M) ) The support of the eigenvectors in Z, i.e., the indices indicating the nonzero elements in Z. The i-th eigenvector is nonzero only in elements ISUPPZ( 2*i-1 ) through ISUPPZ( 2*i ). WORK (workspace) REAL array, dimension (13*N) IWORK (workspace) INTEGER array, dimension (6*N) INFO (output) INTEGER = 0: successful exit < 0: if INFO = -i, the i-th argument had an illegal value > 0: if INFO = 1, internal error in SLARRB if INFO = 2, internal error in CSTEIN Further Details =============== Based on contributions by Inderjit Dhillon, IBM Almaden, USA Osni Marques, LBNL/NERSC, USA Ken Stanley, Computer Science Division, University of California at Berkeley, USA ===================================================================== Test the input parameters. Parameter adjustments */ --d__; --l; --isplit; --w; --iblock; --gersch; z_dim1 = *ldz; z_offset = 1 + z_dim1 * 1; z__ -= z_offset; --isuppz; --work; --iwork; /* Function Body */ inderr = *n + 1; indld = *n << 1; indlld = *n * 3; indgap = *n << 2; indin1 = *n * 5 + 1; indin2 = *n * 6 + 1; indwrk = *n * 7 + 1; iindr = *n; iindc1 = *n << 1; iindc2 = *n * 3; iindwk = (*n << 2) + 1; eps = slamch_("Precision"); i__1 = *n << 1; for (i__ = 1; i__ <= i__1; ++i__) { iwork[i__] = 0; /* L10: */ } latime_1.ops += (real) (*m + 1); i__1 = *m; for (i__ = 1; i__ <= i__1; ++i__) { work[inderr + i__ - 1] = eps * (r__1 = w[i__], dabs(r__1)); /* L20: */ } claset_("Full", n, n, &c_b1, &c_b1, &z__[z_offset], ldz); mgstol = eps * 5.f; nsplit = iblock[*m]; ibegin = 1; i__1 = nsplit; for (jblk = 1; jblk <= i__1; ++jblk) { iend = isplit[jblk]; /* Find the eigenvectors of the submatrix indexed IBEGIN through IEND. */ if (ibegin == iend) { i__2 = z___subscr(ibegin, ibegin); z__[i__2].r = 1.f, z__[i__2].i = 0.f; isuppz[(ibegin << 1) - 1] = ibegin; isuppz[ibegin * 2] = ibegin; ibegin = iend + 1; goto L170; } oldien = ibegin - 1; in = iend - oldien; latime_1.ops += 1.f; /* Computing MIN */ r__1 = .01f, r__2 = 1.f / (real) in; reltol = dmin(r__1,r__2); im = in; scopy_(&im, &w[ibegin], &c__1, &work[1], &c__1); latime_1.ops += (real) (in - 1); i__2 = in - 1; for (i__ = 1; i__ <= i__2; ++i__) { work[indgap + i__] = work[i__ + 1] - work[i__]; /* L30: */ } /* Computing MAX */ r__2 = (r__1 = work[in], dabs(r__1)); work[indgap + in] = dmax(r__2,eps); ndone = 0; ndepth = 0; lsbdpt = 1; nclus = 1; iwork[iindc1 + 1] = 1; iwork[iindc1 + 2] = in; /* While( NDONE.LT.IM ) do */ L40: if (ndone < im) { oldncl = nclus; nclus = 0; lsbdpt = 1 - lsbdpt; i__2 = oldncl; for (i__ = 1; i__ <= i__2; ++i__) { if (lsbdpt == 0) { oldcls = iindc1; newcls = iindc2; } else { oldcls = iindc2; newcls = iindc1; } /* If NDEPTH > 1, retrieve the relatively robust representation (RRR) and perform limited bisection (if necessary) to get approximate eigenvalues. */ j = oldcls + (i__ << 1); oldfst = iwork[j - 1]; oldlst = iwork[j]; if (ndepth > 0) { j = oldien + oldfst; i__3 = in; for (k = 1; k <= i__3; ++k) { i__4 = z___subscr(ibegin + k - 1, oldien + oldfst); d__[ibegin + k - 1] = z__[i__4].r; i__4 = z___subscr(ibegin + k - 1, oldien + oldfst + 1) ; l[ibegin + k - 1] = z__[i__4].r; /* L45: */ } sigma = l[iend]; } k = ibegin; latime_1.ops += (real) (in - 1 << 1); i__3 = in - 1; for (j = 1; j <= i__3; ++j) { work[indld + j] = d__[k] * l[k]; work[indlld + j] = work[indld + j] * l[k]; ++k; /* L50: */ } if (ndepth > 0) { slarrb_(&in, &d__[ibegin], &l[ibegin], &work[indld + 1], & work[indlld + 1], &oldfst, &oldlst, &sigma, & reltol, &work[1], &work[indgap + 1], &work[inderr] , &work[indwrk], &iwork[iindwk], &iinfo); if (iinfo != 0) { *info = 1; return 0; } } /* Classify eigenvalues of the current representation (RRR) as (i) isolated, (ii) loosely clustered or (iii) tightly clustered */ newfrs = oldfst; i__3 = oldlst; for (j = oldfst; j <= i__3; ++j) { latime_1.ops += 1.f; if (j == oldlst || work[indgap + j] >= reltol * (r__1 = work[j], dabs(r__1))) { newlst = j; } else { /* continue (to the next loop) */ latime_1.ops += 1.f; relgap = work[indgap + j] / (r__1 = work[j], dabs( r__1)); if (j == newfrs) { minrgp = relgap; } else { minrgp = dmin(minrgp,relgap); } goto L140; } newsiz = newlst - newfrs + 1; maxitr = 10; newftt = oldien + newfrs; if (newsiz > 1) { mgscls = newsiz <= 20 && minrgp >= mgstol; if (! mgscls) { i__4 = in; for (k = 1; k <= i__4; ++k) { i__5 = z___subscr(ibegin + k - 1, newftt); work[indin1 + k - 1] = z__[i__5].r; i__5 = z___subscr(ibegin + k - 1, newftt + 1); work[indin2 + k - 1] = z__[i__5].r; /* L55: */ } slarrf_(&in, &d__[ibegin], &l[ibegin], &work[ indld + 1], &work[indlld + 1], &newfrs, & newlst, &work[1], &work[indin1], &work[ indin2], &work[indwrk], &iwork[iindwk], info); if (*info == 0) { ++nclus; k = newcls + (nclus << 1); iwork[k - 1] = newfrs; iwork[k] = newlst; } else { *info = 0; if (minrgp >= mgstol) { mgscls = TRUE_; } else { /* Call CSTEIN to process this tight cluster. This happens only if MINRGP <= MGSTOL and SLARRF returns INFO = 1. The latter means that a new RRR to "break" the cluster could not be found. */ work[indwrk] = d__[ibegin]; latime_1.ops += (real) (in - 1); i__4 = in - 1; for (k = 1; k <= i__4; ++k) { work[indwrk + k] = d__[ibegin + k] + work[indlld + k]; /* L60: */ } i__4 = newsiz; for (k = 1; k <= i__4; ++k) { iwork[iindwk + k - 1] = 1; /* L70: */ } i__4 = newlst; for (k = newfrs; k <= i__4; ++k) { isuppz[(ibegin + k << 1) - 3] = 1; isuppz[(ibegin + k << 1) - 2] = in; /* L80: */ } temp[0] = in; cstein_(&in, &work[indwrk], &work[indld + 1], &newsiz, &work[newfrs], & iwork[iindwk], temp, &z___ref( ibegin, newftt), ldz, &work[ indwrk + in], &iwork[iindwk + in], &iwork[iindwk + (in << 1)], & iinfo); if (iinfo != 0) { *info = 2; return 0; } ndone += newsiz; } } } } else { mgscls = FALSE_; } if (newsiz == 1 || mgscls) { ktot = newftt; i__4 = newlst; for (k = newfrs; k <= i__4; ++k) { iter = 0; L90: lambda = work[k]; clar1v_(&in, &c__1, &in, &lambda, &d__[ibegin], & l[ibegin], &work[indld + 1], &work[indlld + 1], &gersch[(oldien << 1) + 1], & z___ref(ibegin, ktot), &ztz, &mingma, & iwork[iindr + ktot], &isuppz[(ktot << 1) - 1], &work[indwrk]); latime_1.ops += 4.f; tmp1 = 1.f / ztz; nrminv = sqrt(tmp1); resid = dabs(mingma) * nrminv; rqcorr = mingma * tmp1; if (k == in) { gap = work[indgap + k - 1]; } else if (k == 1) { gap = work[indgap + k]; } else { /* Computing MIN */ r__1 = work[indgap + k - 1], r__2 = work[ indgap + k]; gap = dmin(r__1,r__2); } ++iter; latime_1.ops += 3.f; if (resid > *tol * gap && dabs(rqcorr) > eps * 4.f * dabs(lambda)) { latime_1.ops += 1.f; work[k] = lambda + rqcorr; if (iter < maxitr) { goto L90; } } iwork[ktot] = 1; if (newsiz == 1) { ++ndone; } latime_1.ops += (real) (in << 1); csscal_(&in, &nrminv, &z___ref(ibegin, ktot), & c__1); ++ktot; /* L100: */ } if (newsiz > 1) { itmp1 = isuppz[(newftt << 1) - 1]; itmp2 = isuppz[newftt * 2]; ktot = oldien + newlst; i__4 = ktot; for (p = newftt + 1; p <= i__4; ++p) { i__5 = p - 1; for (q = newftt; q <= i__5; ++q) { latime_1.ops += (real) (in * 10); cdotu_(&q__2, &in, &z___ref(ibegin, p), & c__1, &z___ref(ibegin, q), &c__1); q__1.r = -q__2.r, q__1.i = -q__2.i; tmp1 = q__1.r; q__1.r = tmp1, q__1.i = 0.f; caxpy_(&in, &q__1, &z___ref(ibegin, q), & c__1, &z___ref(ibegin, p), &c__1); /* L110: */ } latime_1.ops += (real) ((in << 3) + 1); tmp1 = 1.f / scnrm2_(&in, &z___ref(ibegin, p), &c__1); csscal_(&in, &tmp1, &z___ref(ibegin, p), & c__1); /* Computing MIN */ i__5 = itmp1, i__6 = isuppz[(p << 1) - 1]; itmp1 = min(i__5,i__6); /* Computing MAX */ i__5 = itmp2, i__6 = isuppz[p * 2]; itmp2 = max(i__5,i__6); /* L120: */ } i__4 = ktot; for (p = newftt; p <= i__4; ++p) { isuppz[(p << 1) - 1] = itmp1; isuppz[p * 2] = itmp2; /* L130: */ } ndone += newsiz; } } newfrs = j + 1; L140: ; } /* L150: */ } ++ndepth; goto L40; } j = ibegin << 1; i__2 = iend; for (i__ = ibegin; i__ <= i__2; ++i__) { isuppz[j - 1] += oldien; isuppz[j] += oldien; j += 2; /* L160: */ } ibegin = iend + 1; L170: ; } return 0; /* End of CLARRV */ } /* clarrv_ */