예제 #1
0
extern "C" magma_int_t
magma_cheevx(char jobz, char range, char uplo, magma_int_t n,
             magmaFloatComplex *a, magma_int_t lda, float vl, float vu,
             magma_int_t il, magma_int_t iu, float abstol, magma_int_t *m,
             float *w, magmaFloatComplex *z, magma_int_t ldz, magmaFloatComplex *work, magma_int_t lwork,
             float *rwork, magma_int_t *iwork, magma_int_t *ifail, magma_int_t *info)
{
/*  -- MAGMA (version 1.4.1) --
       Univ. of Tennessee, Knoxville
       Univ. of California, Berkeley
       Univ. of Colorado, Denver
       December 2013

    Purpose
    =======
    CHEEVX computes selected eigenvalues and, optionally, eigenvectors
    of a complex Hermitian matrix A.  Eigenvalues and eigenvectors 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.

    A       (input/output) COMPLEX array, dimension (LDA, N)
            On entry, the Hermitian matrix A.  If UPLO = 'U', the
            leading N-by-N upper triangular part of A contains the
            upper triangular part of the matrix A.  If UPLO = 'L',
            the leading N-by-N lower triangular part of A contains
            the lower triangular part of the matrix A.
            On exit, the lower triangle (if UPLO='L') or the upper
            triangle (if UPLO='U') of A, including the diagonal, is
            destroyed.

    LDA     (input) INTEGER
            The leading dimension of the array A.  LDA >= max(1,N).

    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 A 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)
            On normal exit, 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 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/output) COMPLEX array, dimension (LWORK)
            On exit, if INFO = 0, WORK(1) returns the optimal LWORK.

    LWORK   (input) INTEGER
            The length of the array WORK.  LWORK >= max(1,2*N-1).
            For optimal efficiency, LWORK >= (NB+1)*N,
            where NB is the max of the blocksize for CHETRD and for
            CUNMTR as returned by ILAENV.

            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) 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.
    =====================================================================     */
    
    char uplo_[2] = {uplo, 0};
    char jobz_[2] = {jobz, 0};
    char range_[2] = {range, 0};
    
    magma_int_t izero = 0;
    magma_int_t ione = 1;
    
    char order[1];
    magma_int_t indd, inde;
    magma_int_t imax;
    magma_int_t lopt, itmp1, indee;
    magma_int_t lower, wantz;
    magma_int_t i, j, jj, i__1;
    magma_int_t alleig, valeig, indeig;
    magma_int_t iscale, indibl;
    magma_int_t indiwk, indisp, indtau;
    magma_int_t indrwk, indwrk;
    magma_int_t llwork, nsplit;
    magma_int_t lquery;
    magma_int_t iinfo;
    float safmin;
    float bignum;
    float smlnum;
    float eps, tmp1;
    float anrm;
    float sigma, d__1;
    float rmin, rmax;
    
    /* Function Body */
    lower = lapackf77_lsame(uplo_, MagmaLowerStr);
    wantz = lapackf77_lsame(jobz_, MagmaVecStr);
    alleig = lapackf77_lsame(range_, "A");
    valeig = lapackf77_lsame(range_, "V");
    indeig = lapackf77_lsame(range_, "I");
    lquery = lwork == -1;
    
    *info = 0;
    if (! (wantz || lapackf77_lsame(jobz_, MagmaNoVecStr))) {
        *info = -1;
    } else if (! (alleig || valeig || indeig)) {
        *info = -2;
    } else if (! (lower || lapackf77_lsame(uplo_, MagmaUpperStr))) {
        *info = -3;
    } else if (n < 0) {
        *info = -4;
    } else if (lda < max(1,n)) {
        *info = -6;
    } else if (ldz < 1 || (wantz && ldz < n)) {
        *info = -15;
    } else {
        if (valeig) {
            if (n > 0 && vu <= vl) {
                *info = -8;
            }
        } else if (indeig) {
            if (il < 1 || il > max(1,n)) {
                *info = -9;
            } else if (iu < min(n,il) || iu > n) {
                *info = -10;
            }
        }
    }
    
    magma_int_t nb = magma_get_chetrd_nb(n);
    
    lopt = n * (nb + 1);
    
    work[0] = MAGMA_C_MAKE( lopt, 0 );
    
    if (lwork < lopt && ! lquery) {
        *info = -17;
    }
    
    if (*info != 0) {
        magma_xerbla( __func__, -(*info));
        return *info;
    } else if (lquery) {
        return *info;
    }
    
    *m = 0;
    /* Check if matrix is very small then just call LAPACK on CPU, no need for GPU */
    if (n <= 128) {
        #ifdef ENABLE_DEBUG
        printf("--------------------------------------------------------------\n");
        printf("  warning matrix too small N=%d NB=%d, calling lapack on CPU  \n", (int) n, (int) nb);
        printf("--------------------------------------------------------------\n");
        #endif
        lapackf77_cheevx(jobz_, range_, uplo_,
                         &n, a, &lda, &vl, &vu, &il, &iu, &abstol, m,
                         w, z, &ldz, work, &lwork,
                         rwork, iwork, ifail, info);
        return *info;
    }
    
    --w;
    --work;
    --rwork;
    --iwork;
    --ifail;
    
    /* Get machine constants. */
    safmin = lapackf77_slamch("Safe minimum");
    eps = lapackf77_slamch("Precision");
    smlnum = safmin / eps;
    bignum = 1. / smlnum;
    rmin = magma_ssqrt(smlnum);
    rmax = magma_ssqrt(bignum);
    
    /* Scale matrix to allowable range, if necessary. */
    anrm = lapackf77_clanhe("M", uplo_, &n, a, &lda, &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) {
        d__1 = 1.;
        lapackf77_clascl(uplo_, &izero, &izero, &d__1, &sigma, &n, &n, a,
                         &lda, info);
        
        if (abstol > 0.) {
            abstol *= sigma;
        }
        if (valeig) {
            vl *= sigma;
            vu *= sigma;
        }
    }
    
    /* Call CHETRD to reduce Hermitian matrix to tridiagonal form. */
    indd = 1;
    inde = indd + n;
    indrwk = inde + n;
    indtau = 1;
    indwrk = indtau + n;
    llwork = lwork - indwrk + 1;
    
    magma_chetrd(uplo, n, a, lda, &rwork[indd], &rwork[inde], &work[indtau], &work[indwrk], llwork, &iinfo);
    
    lopt = n + (magma_int_t)MAGMA_C_REAL(work[indwrk]);
    
    /* If all eigenvalues are desired and ABSTOL is less than or equal to
       zero, then call SSTERF or CUNGTR and CSTEQR.  If this fails for
       some eigenvalue, then try SSTEBZ. */
    if ((alleig || (indeig && il == 1 && iu == n)) && abstol <= 0.) {
        blasf77_scopy(&n, &rwork[indd], &ione, &w[1], &ione);
        indee = indrwk + 2*n;
        if (! wantz) {
            i__1 = n - 1;
            blasf77_scopy(&i__1, &rwork[inde], &ione, &rwork[indee], &ione);
            lapackf77_ssterf(&n, &w[1], &rwork[indee], info);
        }
        else {
            lapackf77_clacpy("A", &n, &n, a, &lda, z, &ldz);
            lapackf77_cungtr(uplo_, &n, z, &ldz, &work[indtau], &work[indwrk], &llwork, &iinfo);
            i__1 = n - 1;
            blasf77_scopy(&i__1, &rwork[inde], &ione, &rwork[indee], &ione);
            lapackf77_csteqr(jobz_, &n, &w[1], &rwork[indee], z, &ldz, &rwork[indrwk], info);
            if (*info == 0) {
                for (i = 1; i <= n; ++i) {
                    ifail[i] = 0;
                }
            }
        }
        if (*info == 0) {
            *m = n;
        }
    }
    
    /* Otherwise, call SSTEBZ and, if eigenvectors are desired, CSTEIN. */
    if (*m == 0) {
        *info = 0;
        if (wantz) {
            *(unsigned char *)order = 'B';
        } else {
            *(unsigned char *)order = 'E';
        }
        indibl = 1;
        indisp = indibl + n;
        indiwk = indisp + n;
        lapackf77_sstebz(range_, order, &n, &vl, &vu, &il, &iu, &abstol, &rwork[indd], &rwork[inde], m,
                         &nsplit, &w[1], &iwork[indibl], &iwork[indisp], &rwork[indrwk], &iwork[indiwk], info);
        
        if (wantz) {
            lapackf77_cstein(&n, &rwork[indd], &rwork[inde], m, &w[1], &iwork[indibl], &iwork[indisp],
                             z, &ldz, &rwork[indrwk], &iwork[indiwk], &ifail[1], info);
            
            /* Apply unitary matrix used in reduction to tridiagonal
               form to eigenvectors returned by CSTEIN. */
            magma_cunmtr(MagmaLeft, uplo, MagmaNoTrans, n, *m, a, lda, &work[indtau],
                         z, ldz, &work[indwrk], llwork, &iinfo);
        }
    }
    /* If matrix was scaled, then rescale eigenvalues appropriately. */
    if (iscale == 1) {
        if (*info == 0) {
            imax = *m;
        } else {
            imax = *info - 1;
        }
        d__1 = 1. / sigma;
        blasf77_sscal(&imax, &d__1, &w[1], &ione);
    }
    
    /* If eigenvalues are not in order, then sort them, along with
       eigenvectors. */
    if (wantz) {
        for (j = 1; j <= *m-1; ++j) {
            i = 0;
            tmp1 = w[j];
            for (jj = j + 1; jj <= *m; ++jj) {
                if (w[jj] < tmp1) {
                    i = jj;
                    tmp1 = w[jj];
                }
            }
            
            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;
                blasf77_cswap(&n, z + (i-1)*ldz, &ione, z + (j-1)*ldz, &ione);
                if (*info != 0) {
                    itmp1 = ifail[i];
                    ifail[i] = ifail[j];
                    ifail[j] = itmp1;
                }
            }
        }
    }
    
    /* Set WORK(1) to optimal complex workspace size. */
    work[1] = MAGMA_C_MAKE( lopt, 0 );
    
    return *info;
    
} /* magma_cheevx */
예제 #2
0
파일: cheevx.cpp 프로젝트: cjy7117/FT-MAGMA
/**
    Purpose
    -------
    CHEEVX computes selected eigenvalues and, optionally, eigenvectors
    of a complex Hermitian matrix A.  Eigenvalues and eigenvectors can
    be selected by specifying either a range of values or a range of
    indices for the desired eigenvalues.

    Arguments
    ---------
    @param[in]
    jobz    magma_vec_t
      -     = MagmaNoVec:  Compute eigenvalues only;
      -     = MagmaVec:    Compute eigenvalues and eigenvectors.

    @param[in]
    range   magma_range_t
      -     = MagmaRangeAll: all eigenvalues will be found.
      -     = MagmaRangeV:   all eigenvalues in the half-open interval (VL,VU]
                   will be found.
      -     = MagmaRangeI:   the IL-th through IU-th eigenvalues will be found.

    @param[in]
    uplo    magma_uplo_t
      -     = MagmaUpper:  Upper triangle of A is stored;
      -     = MagmaLower:  Lower triangle of A is stored.

    @param[in]
    n       INTEGER
            The order of the matrix A.  N >= 0.

    @param[in,out]
    A       COMPLEX array, dimension (LDA, N)
            On entry, the Hermitian matrix A.  If UPLO = MagmaUpper, the
            leading N-by-N upper triangular part of A contains the
            upper triangular part of the matrix A.  If UPLO = MagmaLower,
            the leading N-by-N lower triangular part of A contains
            the lower triangular part of the matrix A.
            On exit, the lower triangle (if UPLO=MagmaLower) or the upper
            triangle (if UPLO=MagmaUpper) of A, including the diagonal, is
            destroyed.

    @param[in]
    lda     INTEGER
            The leading dimension of the array A.  LDA >= max(1,N).

    @param[in]
    vl      REAL
    @param[in]
    vu      REAL
            If RANGE=MagmaRangeV, the lower and upper bounds of the interval to
            be searched for eigenvalues. VL < VU.
            Not referenced if RANGE = MagmaRangeAll or MagmaRangeI.

    @param[in]
    il      INTEGER
    @param[in]
    iu      INTEGER
            If RANGE=MagmaRangeI, 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 = MagmaRangeAll or MagmaRangeV.

    @param[in]
    abstol  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| ),
    \n
            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 A to tridiagonal form.
    \n
            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').
    \n
            See "Computing Small Singular Values of Bidiagonal Matrices
            with Guaranteed High Relative Accuracy," by Demmel and
            Kahan, LAPACK Working Note #3.

    @param[out]
    m       INTEGER
            The total number of eigenvalues found.  0 <= M <= N.
            If RANGE = MagmaRangeAll, M = N, and if RANGE = MagmaRangeI, M = IU-IL+1.

    @param[out]
    w       REAL array, dimension (N)
            On normal exit, the first M elements contain the selected
            eigenvalues in ascending order.

    @param[out]
    Z       COMPLEX array, dimension (LDZ, max(1,M))
            If JOBZ = MagmaVec, 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 = MagmaNoVec, 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 = MagmaRangeV, the exact value of M
            is not known in advance and an upper bound must be used.

    @param[in]
    ldz     INTEGER
            The leading dimension of the array Z.  LDZ >= 1, and if
            JOBZ = MagmaVec, LDZ >= max(1,N).

    @param[out]
    work    (workspace) COMPLEX array, dimension (LWORK)
            On exit, if INFO = 0, WORK[0] returns the optimal LWORK.

    @param[in]
    lwork   INTEGER
            The length of the array WORK.  LWORK >= max(1,2*N-1).
            For optimal efficiency, LWORK >= (NB+1)*N,
            where NB is the max of the blocksize for CHETRD and for
            CUNMTR as returned by ILAENV.
    \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.

    @param
    rwork   (workspace) REAL array, dimension (7*N)

    @param
    iwork   (workspace) INTEGER array, dimension (5*N)

    @param[out]
    ifail   INTEGER array, dimension (N)
            If JOBZ = MagmaVec, 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 = MagmaNoVec, then IFAIL is not referenced.

    @param[out]
    info    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.

    @ingroup magma_cheev_driver
    ********************************************************************/
extern "C" magma_int_t
magma_cheevx(
    magma_vec_t jobz, magma_range_t range, magma_uplo_t uplo, magma_int_t n,
    magmaFloatComplex *A, magma_int_t lda, float vl, float vu,
    magma_int_t il, magma_int_t iu, float abstol, magma_int_t *m,
    float *w, magmaFloatComplex *Z, magma_int_t ldz, magmaFloatComplex *work, magma_int_t lwork,
    float *rwork, magma_int_t *iwork, magma_int_t *ifail,
    magma_int_t *info)
{
    const char* uplo_  = lapack_uplo_const( uplo  );
    const char* jobz_  = lapack_vec_const( jobz  );
    const char* range_ = lapack_range_const( range );
    
    magma_int_t izero = 0;
    magma_int_t ione = 1;
    
    const char* order_;
    magma_int_t indd, inde;
    magma_int_t imax;
    magma_int_t lopt, itmp1, indee;
    magma_int_t lower, wantz;
    magma_int_t i, j, jj, i__1;
    magma_int_t alleig, valeig, indeig;
    magma_int_t iscale, indibl;
    magma_int_t indiwk, indisp, indtau;
    magma_int_t indrwk, indwrk;
    magma_int_t llwork, nsplit;
    magma_int_t lquery;
    magma_int_t iinfo;
    float safmin;
    float bignum;
    float smlnum;
    float eps, tmp1;
    float anrm;
    float sigma, d__1;
    float rmin, rmax;
    
    /* Function Body */
    lower  = (uplo  == MagmaLower);
    wantz  = (jobz  == MagmaVec);
    alleig = (range == MagmaRangeAll);
    valeig = (range == MagmaRangeV);
    indeig = (range == MagmaRangeI);
    lquery = (lwork == -1);
    
    *info = 0;
    if (! (wantz || (jobz == MagmaNoVec))) {
        *info = -1;
    } else if (! (alleig || valeig || indeig)) {
        *info = -2;
    } else if (! (lower || (uplo == MagmaUpper))) {
        *info = -3;
    } else if (n < 0) {
        *info = -4;
    } else if (lda < max(1,n)) {
        *info = -6;
    } else if (ldz < 1 || (wantz && ldz < n)) {
        *info = -15;
    } else {
        if (valeig) {
            if (n > 0 && vu <= vl) {
                *info = -8;
            }
        } else if (indeig) {
            if (il < 1 || il > max(1,n)) {
                *info = -9;
            } else if (iu < min(n,il) || iu > n) {
                *info = -10;
            }
        }
    }
    
    magma_int_t nb = magma_get_chetrd_nb(n);
    
    lopt = n * (nb + 1);
    
    work[0] = MAGMA_C_MAKE( lopt, 0 );
    
    if (lwork < lopt && ! lquery) {
        *info = -17;
    }
    
    if (*info != 0) {
        magma_xerbla( __func__, -(*info));
        return *info;
    } else if (lquery) {
        return *info;
    }
    
    *m = 0;
    /* Check if matrix is very small then just call LAPACK on CPU, no need for GPU */
    if (n <= 128) {
        #ifdef ENABLE_DEBUG
        printf("--------------------------------------------------------------\n");
        printf("  warning matrix too small N=%d NB=%d, calling lapack on CPU  \n", (int) n, (int) nb);
        printf("--------------------------------------------------------------\n");
        #endif
        lapackf77_cheevx(jobz_, range_, uplo_,
                         &n, A, &lda, &vl, &vu, &il, &iu, &abstol, m,
                         w, Z, &ldz, work, &lwork,
                         rwork, iwork, ifail, info);
        return *info;
    }
    
    --w;
    --work;
    --rwork;
    --iwork;
    --ifail;
    
    /* Get machine constants. */
    safmin = lapackf77_slamch("Safe minimum");
    eps = lapackf77_slamch("Precision");
    smlnum = safmin / eps;
    bignum = 1. / smlnum;
    rmin = magma_ssqrt(smlnum);
    rmax = magma_ssqrt(bignum);
    
    /* Scale matrix to allowable range, if necessary. */
    anrm = lapackf77_clanhe("M", uplo_, &n, A, &lda, &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) {
        d__1 = 1.;
        lapackf77_clascl(uplo_, &izero, &izero, &d__1, &sigma, &n, &n, A,
                         &lda, info);
        
        if (abstol > 0.) {
            abstol *= sigma;
        }
        if (valeig) {
            vl *= sigma;
            vu *= sigma;
        }
    }
    
    /* Call CHETRD to reduce Hermitian matrix to tridiagonal form. */
    indd = 1;
    inde = indd + n;
    indrwk = inde + n;
    indtau = 1;
    indwrk = indtau + n;
    llwork = lwork - indwrk + 1;
    
    magma_chetrd(uplo, n, A, lda, &rwork[indd], &rwork[inde], &work[indtau], &work[indwrk], llwork, &iinfo);
    
    lopt = n + (magma_int_t)MAGMA_C_REAL(work[indwrk]);
    
    /* If all eigenvalues are desired and ABSTOL is less than or equal to
       zero, then call SSTERF or CUNGTR and CSTEQR.  If this fails for
       some eigenvalue, then try SSTEBZ. */
    if ((alleig || (indeig && il == 1 && iu == n)) && abstol <= 0.) {
        blasf77_scopy(&n, &rwork[indd], &ione, &w[1], &ione);
        indee = indrwk + 2*n;
        if (! wantz) {
            i__1 = n - 1;
            blasf77_scopy(&i__1, &rwork[inde], &ione, &rwork[indee], &ione);
            lapackf77_ssterf(&n, &w[1], &rwork[indee], info);
        }
        else {
            lapackf77_clacpy("A", &n, &n, A, &lda, Z, &ldz);
            lapackf77_cungtr(uplo_, &n, Z, &ldz, &work[indtau], &work[indwrk], &llwork, &iinfo);
            i__1 = n - 1;
            blasf77_scopy(&i__1, &rwork[inde], &ione, &rwork[indee], &ione);
            lapackf77_csteqr(jobz_, &n, &w[1], &rwork[indee], Z, &ldz, &rwork[indrwk], info);
            if (*info == 0) {
                for (i = 1; i <= n; ++i) {
                    ifail[i] = 0;
                }
            }
        }
        if (*info == 0) {
            *m = n;
        }
    }
    
    /* Otherwise, call SSTEBZ and, if eigenvectors are desired, CSTEIN. */
    if (*m == 0) {
        *info = 0;
        if (wantz) {
            order_ = "B";
        } else {
            order_ = "E";
        }
        indibl = 1;
        indisp = indibl + n;
        indiwk = indisp + n;
        lapackf77_sstebz(range_, order_, &n, &vl, &vu, &il, &iu, &abstol, &rwork[indd], &rwork[inde], m,
                         &nsplit, &w[1], &iwork[indibl], &iwork[indisp], &rwork[indrwk], &iwork[indiwk], info);
        
        if (wantz) {
            lapackf77_cstein(&n, &rwork[indd], &rwork[inde], m, &w[1], &iwork[indibl], &iwork[indisp],
                             Z, &ldz, &rwork[indrwk], &iwork[indiwk], &ifail[1], info);
            
            /* Apply unitary matrix used in reduction to tridiagonal
               form to eigenvectors returned by CSTEIN. */
            magma_cunmtr(MagmaLeft, uplo, MagmaNoTrans, n, *m, A, lda, &work[indtau],
                         Z, ldz, &work[indwrk], llwork, &iinfo);
        }
    }
    /* If matrix was scaled, then rescale eigenvalues appropriately. */
    if (iscale == 1) {
        if (*info == 0) {
            imax = *m;
        } else {
            imax = *info - 1;
        }
        d__1 = 1. / sigma;
        blasf77_sscal(&imax, &d__1, &w[1], &ione);
    }
    
    /* If eigenvalues are not in order, then sort them, along with
       eigenvectors. */
    if (wantz) {
        for (j = 1; j <= *m-1; ++j) {
            i = 0;
            tmp1 = w[j];
            for (jj = j + 1; jj <= *m; ++jj) {
                if (w[jj] < tmp1) {
                    i = jj;
                    tmp1 = w[jj];
                }
            }
            
            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;
                blasf77_cswap(&n, Z + (i-1)*ldz, &ione, Z + (j-1)*ldz, &ione);
                if (*info != 0) {
                    itmp1 = ifail[i];
                    ifail[i] = ifail[j];
                    ifail[j] = itmp1;
                }
            }
        }
    }
    
    /* Set WORK[0] to optimal complex workspace size. */
    work[1] = MAGMA_C_MAKE( lopt, 0 );
    
    return *info;
    
} /* magma_cheevx */
예제 #3
0
파일: cheevdx.cpp 프로젝트: soulsheng/magma
extern "C" magma_int_t
magma_cheevdx(char jobz, char range, char uplo,
              magma_int_t n,
              magmaFloatComplex *a, magma_int_t lda,
              float vl, float vu, magma_int_t il, magma_int_t iu,
              magma_int_t *m, float *w,
              magmaFloatComplex *work, magma_int_t lwork,
              float *rwork, magma_int_t lrwork,
              magma_int_t *iwork, magma_int_t liwork,
              magma_int_t *info)
{
/*  -- MAGMA (version 1.4.0) --
       Univ. of Tennessee, Knoxville
       Univ. of California, Berkeley
       Univ. of Colorado, Denver
       August 2013

    Purpose
    =======
    CHEEVDX computes selected eigenvalues and, optionally, eigenvectors
    of a complex Hermitian matrix A. Eigenvalues and eigenvectors can
    be selected by specifying either a range of values or a range of
    indices for the desired eigenvalues.
    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.

    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.

    A       (input/output) COMPLEX array, dimension (LDA, N)
            On entry, the Hermitian matrix A.  If UPLO = 'U', the
            leading N-by-N upper triangular part of A contains the
            upper triangular part of the matrix A.  If UPLO = 'L',
            the leading N-by-N lower triangular part of A contains
            the lower triangular part of the matrix A.
            On exit, if JOBZ = 'V', then if INFO = 0, the first m columns
            of A contains the required
            orthonormal eigenvectors of the matrix A.
            If JOBZ = 'N', then on exit the lower triangle (if UPLO='L')
            or the upper triangle (if UPLO='U') of A, including the
            diagonal, is destroyed.

    LDA     (input) INTEGER
            The leading dimension of the array A.  LDA >= max(1,N).

    VL      (input) DOUBLE PRECISION
    VU      (input) DOUBLE PRECISION
            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'.

    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) DOUBLE PRECISION array, dimension (N)
            If INFO = 0, the required m eigenvalues in ascending order.

    WORK    (workspace/output) COMPLEX array, dimension (MAX(1,LWORK))
            On exit, if INFO = 0, WORK[0] returns the optimal LWORK.

    LWORK   (input) INTEGER
            The length of the array WORK.
            If N <= 1,                LWORK >= 1.
            If JOBZ  = 'N' and N > 1, LWORK >= N + N*NB.
            If JOBZ  = 'V' and N > 1, LWORK >= max( N + N*NB, 2*N + N**2 ).
            NB can be obtained through magma_get_chetrd_nb(N).

            If LWORK = -1, then a workspace query is assumed; the routine
            only calculates the optimal sizes of the WORK, RWORK and
            IWORK arrays, returns these values as the first entries of
            the WORK, RWORK and IWORK arrays, and no error message
            related to LWORK or LRWORK or LIWORK is issued by XERBLA.

    RWORK   (workspace/output) DOUBLE PRECISION array,
                                           dimension (LRWORK)
            On exit, if INFO = 0, RWORK[0] returns the optimal LRWORK.

    LRWORK  (input) INTEGER
            The dimension of the array RWORK.
            If N <= 1,                LRWORK >= 1.
            If JOBZ  = 'N' and N > 1, LRWORK >= N.
            If JOBZ  = 'V' and N > 1, LRWORK >= 1 + 5*N + 2*N**2.

            If LRWORK = -1, then a workspace query is assumed; the
            routine only calculates the optimal sizes of the WORK, RWORK
            and IWORK arrays, returns these values as the first entries
            of the WORK, RWORK and IWORK arrays, and no error message
            related to LWORK or LRWORK or LIWORK is issued by XERBLA.

    IWORK   (workspace/output) INTEGER array, dimension (MAX(1,LIWORK))
            On exit, if INFO = 0, IWORK[0] returns the optimal LIWORK.

    LIWORK  (input) INTEGER
            The dimension of the array IWORK.
            If N <= 1,                LIWORK >= 1.
            If JOBZ  = 'N' and N > 1, LIWORK >= 1.
            If JOBZ  = 'V' and N > 1, LIWORK >= 3 + 5*N.

            If LIWORK = -1, then a workspace query is assumed; the
            routine only calculates the optimal sizes of the WORK, RWORK
            and IWORK arrays, returns these values as the first entries
            of the WORK, RWORK and IWORK arrays, and no error message
            related to LWORK or LRWORK or 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 and JOBZ = 'N', then the algorithm failed
                  to converge; i off-diagonal elements of an intermediate
                  tridiagonal form did not converge to zero;
                  if INFO = i and JOBZ = 'V', then the algorithm failed
                  to compute an eigenvalue while working on the submatrix
                  lying in rows and columns INFO/(N+1) through
                  mod(INFO,N+1).

    Further Details
    ===============
    Based on contributions by
       Jeff Rutter, Computer Science Division, University of California
       at Berkeley, USA

    Modified description of INFO. Sven, 16 Feb 05.
    =====================================================================   */

    char uplo_[2] = {uplo, 0};
    char jobz_[2] = {jobz, 0};
    char range_[2] = {range, 0};
    magma_int_t ione = 1;
    magma_int_t izero = 0;
    float d_one = 1.;

    float d__1;

    float eps;
    magma_int_t inde;
    float anrm;
    magma_int_t imax;
    float rmin, rmax;
    float sigma;
    magma_int_t iinfo, lwmin;
    magma_int_t lower;
    magma_int_t llrwk;
    magma_int_t wantz;
    magma_int_t indwk2, llwrk2;
    magma_int_t iscale;
    float safmin;
    float bignum;
    magma_int_t indtau;
    magma_int_t indrwk, indwrk, liwmin;
    magma_int_t lrwmin, llwork;
    float smlnum;
    magma_int_t lquery;
    magma_int_t alleig, valeig, indeig;

    float* dwork;

    wantz = lapackf77_lsame(jobz_, MagmaVecStr);
    lower = lapackf77_lsame(uplo_, MagmaLowerStr);

    alleig = lapackf77_lsame( range_, "A" );
    valeig = lapackf77_lsame( range_, "V" );
    indeig = lapackf77_lsame( range_, "I" );

    lquery = lwork == -1 || lrwork == -1 || liwork == -1;

    *info = 0;
    if (! (wantz || lapackf77_lsame(jobz_, MagmaNoVecStr))) {
        *info = -1;
    } else if (! (alleig || valeig || indeig)) {
        *info = -2;
    } else if (! (lower || lapackf77_lsame(uplo_, MagmaUpperStr))) {
        *info = -3;
    } else if (n < 0) {
        *info = -4;
    } else if (lda < max(1,n)) {
        *info = -6;
    } else {
        if (valeig) {
            if (n > 0 && vu <= vl) {
                *info = -8;
            }
        } else if (indeig) {
            if (il < 1 || il > max(1,n)) {
                *info = -9;
            } else if (iu < min(n,il) || iu > n) {
                *info = -10;
            }
        }
    }

    magma_int_t nb = magma_get_chetrd_nb( n );
    if ( n <= 1 ) {
        lwmin  = 1;
        lrwmin = 1;
        liwmin = 1;
    }
    else if ( wantz ) {
        lwmin  = max( n + n*nb, 2*n + n*n );
        lrwmin = 1 + 5*n + 2*n*n;
        liwmin = 3 + 5*n;
    }
    else {
        lwmin  = n + n*nb;
        lrwmin = n;
        liwmin = 1;
    }
    // multiply by 1+eps to ensure length gets rounded up,
    // if it cannot be exactly represented in floating point.
    work[0]  = MAGMA_C_MAKE( lwmin * (1. + lapackf77_slamch("Epsilon")), 0.);
    rwork[0] = lrwmin * (1. + lapackf77_slamch("Epsilon"));
    iwork[0] = liwmin;

    if ((lwork < lwmin) && !lquery) {
        *info = -14;
    } else if ((lrwork < lrwmin) && ! lquery) {
        *info = -16;
    } else if ((liwork < liwmin) && ! lquery) {
        *info = -18;
    }

    if (*info != 0) {
        magma_xerbla( __func__, -(*info) );
        return *info;
    }
    else if (lquery) {
        return *info;
    }

    /* Quick return if possible */
    if (n == 0) {
        return *info;
    }

    if (n == 1) {
        w[0] = MAGMA_C_REAL(a[0]);
        if (wantz) {
            a[0] = MAGMA_C_ONE;
        }
        return *info;
    }
    /* Check if matrix is very small then just call LAPACK on CPU, no need for GPU */
    if (n <= 128){
        #ifdef ENABLE_DEBUG
        printf("--------------------------------------------------------------\n");
        printf("  warning matrix too small N=%d NB=%d, calling lapack on CPU  \n", (int) n, (int) nb);
        printf("--------------------------------------------------------------\n");
        #endif
        lapackf77_cheevd(jobz_, uplo_,
                         &n, a, &lda,
                         w, work, &lwork,
#if defined(PRECISION_z) || defined(PRECISION_c)
                         rwork, &lrwork, 
#endif  
                         iwork, &liwork, info);
        return *info;
    }
    /* Get machine constants. */
    safmin = lapackf77_slamch("Safe minimum");
    eps    = lapackf77_slamch("Precision");
    smlnum = safmin / eps;
    bignum = 1. / smlnum;
    rmin = magma_ssqrt(smlnum);
    rmax = magma_ssqrt(bignum);

    /* Scale matrix to allowable range, if necessary. */
    anrm = lapackf77_clanhe("M", uplo_, &n, a, &lda, rwork);
    iscale = 0;
    if (anrm > 0. && anrm < rmin) {
        iscale = 1;
        sigma = rmin / anrm;
    } else if (anrm > rmax) {
        iscale = 1;
        sigma = rmax / anrm;
    }
    if (iscale == 1) {
        lapackf77_clascl(uplo_, &izero, &izero, &d_one, &sigma, &n, &n, a,
                         &lda, info);
    }

    /* Call CHETRD to reduce Hermitian matrix to tridiagonal form. */
    // chetrd rwork: e (n)
    // cstedx rwork: e (n) + llrwk (1 + 4*N + 2*N**2)  ==>  1 + 5n + 2n^2
    inde   = 0;
    indrwk = inde + n;
    llrwk  = lrwork - indrwk;

    // chetrd work: tau (n) + llwork (n*nb)  ==>  n + n*nb
    // cstedx work: tau (n) + z (n^2)
    // cunmtr work: tau (n) + z (n^2) + llwrk2 (n or n*nb)  ==>  2n + n^2, or n + n*nb + n^2
    indtau = 0;
    indwrk = indtau + n;
    indwk2 = indwrk + n*n;
    llwork = lwork - indwrk;
    llwrk2 = lwork - indwk2;

//
#ifdef ENABLE_TIMER
    magma_timestr_t start, end;
    start = get_current_time();
#endif

    magma_chetrd(uplo_[0], n, a, lda, w, &rwork[inde],
                 &work[indtau], &work[indwrk], llwork, &iinfo);

#ifdef ENABLE_TIMER
    end = get_current_time();
    printf("time chetrd = %6.2f\n", GetTimerValue(start,end)/1000.);
#endif

    /* For eigenvalues only, call SSTERF.  For eigenvectors, first call
     CSTEDC to generate the eigenvector matrix, WORK(INDWRK), of the
     tridiagonal matrix, then call CUNMTR to multiply it to the Householder
     transformations represented as Householder vectors in A. */
    if (! wantz) {
        lapackf77_ssterf(&n, w, &rwork[inde], info);

        magma_smove_eig(range, n, w, &il, &iu, vl, vu, m);

    } else {

#ifdef ENABLE_TIMER
        start = get_current_time();
#endif

        if (MAGMA_SUCCESS != magma_smalloc( &dwork, 3*n*(n/2 + 1) )) {
            *info = MAGMA_ERR_DEVICE_ALLOC;
            return *info;
        }

        magma_cstedx(range, n, vl, vu, il, iu, w, &rwork[inde],
                     &work[indwrk], n, &rwork[indrwk],
                     llrwk, iwork, liwork, dwork, info);

        magma_free( dwork );

#ifdef ENABLE_TIMER
        end = get_current_time();
        printf("time cstedx = %6.2f\n", GetTimerValue(start,end)/1000.);
        start = get_current_time();
#endif

        magma_smove_eig(range, n, w, &il, &iu, vl, vu, m);

        magma_cunmtr(MagmaLeft, uplo, MagmaNoTrans, n, *m, a, lda, &work[indtau],
                     &work[indwrk + n * (il-1) ], n, &work[indwk2], llwrk2, &iinfo);

        lapackf77_clacpy("A", &n, m, &work[indwrk + n * (il-1)] , &n, a, &lda);

#ifdef ENABLE_TIMER
        end = get_current_time();
        printf("time cunmtr + copy = %6.2f\n", GetTimerValue(start,end)/1000.);
#endif

    }

    /* If matrix was scaled, then rescale eigenvalues appropriately. */
    if (iscale == 1) {
        if (*info == 0) {
            imax = n;
        } else {
            imax = *info - 1;
        }
        d__1 = 1. / sigma;
        blasf77_sscal(&imax, &d__1, w, &ione);
    }

    work[0]  = MAGMA_C_MAKE( lwmin * (1. + lapackf77_slamch("Epsilon")), 0.);  // round up
    rwork[0] = lrwmin * (1. + lapackf77_slamch("Epsilon"));
    iwork[0] = liwmin;

    return *info;
} /* magma_cheevdx */
예제 #4
0
extern "C" magma_int_t
magma_cheevd(magma_vec_t jobz, magma_uplo_t uplo,
             magma_int_t n,
             magmaFloatComplex *a, magma_int_t lda,
             float *w,
             magmaFloatComplex *work, magma_int_t lwork,
             float *rwork, magma_int_t lrwork,
             magma_int_t *iwork, magma_int_t liwork,
             magma_int_t *info, magma_queue_t queue)
{
/*  -- clMAGMA (version 1.1.0) --
       Univ. of Tennessee, Knoxville
       Univ. of California, Berkeley
       Univ. of Colorado, Denver
       @date January 2014

    Purpose
    =======
    CHEEVD computes all eigenvalues and, optionally, eigenvectors of a
    complex Hermitian matrix A.  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.

    A       (input/output) COMPLEX array, dimension (LDA, N)
            On entry, the Hermitian matrix A.  If UPLO = 'U', the
            leading N-by-N upper triangular part of A contains the
            upper triangular part of the matrix A.  If UPLO = 'L',
            the leading N-by-N lower triangular part of A contains
            the lower triangular part of the matrix A.
            On exit, if JOBZ = 'V', then if INFO = 0, A contains the
            orthonormal eigenvectors of the matrix A.
            If JOBZ = 'N', then on exit the lower triangle (if UPLO='L')
            or the upper triangle (if UPLO='U') of A, including the
            diagonal, is destroyed.

    LDA     (input) INTEGER
            The leading dimension of the array A.  LDA >= max(1,N).

    W       (output) REAL array, dimension (N)
            If INFO = 0, the eigenvalues in ascending order.

    WORK    (workspace/output) COMPLEX array, dimension (MAX(1,LWORK))
            On exit, if INFO = 0, WORK[0] returns the optimal LWORK.

    LWORK   (input) INTEGER
            The length of the array WORK.
            If N <= 1,                LWORK must be at least 1.
            If JOBZ  = 'N' and N > 1, LWORK must be at least N + N*NB.
            If JOBZ  = 'V' and N > 1, LWORK must be at least 2*N + N**2.
            NB can be obtained through magma_get_chetrd_nb(N).

            If LWORK = -1, then a workspace query is assumed; the routine
            only calculates the optimal sizes of the WORK, RWORK and
            IWORK arrays, returns these values as the first entries of
            the WORK, RWORK and IWORK arrays, and no error message
            related to LWORK or LRWORK or LIWORK is issued by XERBLA.

    RWORK   (workspace/output) REAL array, dimension (LRWORK)
            On exit, if INFO = 0, RWORK[0] returns the optimal LRWORK.

    LRWORK  (input) INTEGER
            The dimension of the 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 sizes of the WORK, RWORK
            and IWORK arrays, returns these values as the first entries
            of the WORK, RWORK and IWORK arrays, and no error message
            related to LWORK or LRWORK or LIWORK is issued by XERBLA.

    IWORK   (workspace/output) INTEGER array, dimension (MAX(1,LIWORK))
            On exit, if INFO = 0, IWORK[0] returns the optimal LIWORK.

    LIWORK  (input) INTEGER
            The dimension of the array IWORK.
            If N <= 1,                LIWORK must be at least 1.
            If JOBZ  = 'N' and 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 sizes of the WORK, RWORK
            and IWORK arrays, returns these values as the first entries
            of the WORK, RWORK and IWORK arrays, and no error message
            related to LWORK or LRWORK or 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 and JOBZ = 'N', then the algorithm failed
                  to converge; i off-diagonal elements of an intermediate
                  tridiagonal form did not converge to zero;
                  if INFO = i and JOBZ = 'V', then the algorithm failed
                  to compute an eigenvalue while working on the submatrix
                  lying in rows and columns INFO/(N+1) through
                  mod(INFO,N+1).

    Further Details
    ===============
    Based on contributions by
       Jeff Rutter, Computer Science Division, University of California
       at Berkeley, USA

    Modified description of INFO. Sven, 16 Feb 05.
    =====================================================================   */

    magma_uplo_t uplo_ = uplo;
    magma_vec_t jobz_ = jobz;
    magma_int_t ione = 1;
    magma_int_t izero = 0;
    float d_one = 1.;
    
    float d__1;
    
    float eps;
    magma_int_t inde;
    float anrm;
    magma_int_t imax;
    float rmin, rmax;
    float sigma;
    magma_int_t iinfo, lwmin;
    magma_int_t lower;
    magma_int_t llrwk;
    magma_int_t wantz;
    magma_int_t indwk2, llwrk2;
    magma_int_t iscale;
    float safmin;
    float bignum;
    magma_int_t indtau;
    magma_int_t indrwk, indwrk, liwmin;
    magma_int_t lrwmin, llwork;
    float smlnum;
    magma_int_t lquery;
    
    magmaFloat_ptr dwork;
    
    wantz = lapackf77_lsame(lapack_const(jobz_), MagmaVecStr);
    lower = lapackf77_lsame(lapack_const(uplo_), MagmaLowerStr);
    lquery = lwork == -1 || lrwork == -1 || liwork == -1;
    
    *info = 0;
    if (! (wantz || lapackf77_lsame(lapack_const(jobz_), MagmaNoVecStr))) {
        *info = -1;
    } else if (! (lower || lapackf77_lsame(lapack_const(uplo_), MagmaUpperStr))) {
        *info = -2;
    } else if (n < 0) {
        *info = -3;
    } else if (lda < max(1,n)) {
        *info = -5;
    }
    
    magma_int_t nb = magma_get_chetrd_nb( n );
    if ( n <= 1 ) {
        lwmin  = 1;
        lrwmin = 1;
        liwmin = 1;
    }
    else if ( wantz ) {
        lwmin  = 2*n + n*n;
        lrwmin = 1 + 5*n + 2*n*n;
        liwmin = 3 + 5*n;
    }
    else {
        lwmin  = n + n*nb;
        lrwmin = n;
        liwmin = 1;
    }
    // multiply by 1+eps to ensure length gets rounded up,
    // if it cannot be exactly represented in floating point.
    work[0]  = MAGMA_C_MAKE( lwmin * (1. + lapackf77_slamch("Epsilon")), 0.);
    rwork[0] = lrwmin * (1. + lapackf77_slamch("Epsilon"));
    iwork[0] = liwmin;
    
    if ((lwork < lwmin) && !lquery) {
        *info = -8;
    } else if ((lrwork < lrwmin) && ! lquery) {
        *info = -10;
    } else if ((liwork < liwmin) && ! lquery) {
        *info = -12;
    }
    
    if (*info != 0) {
        magma_xerbla( __func__, -(*info) );
        return *info;
    }
    else if (lquery) {
        return *info;
    }
    
    /* Quick return if possible */
    if (n == 0) {
        return *info;
    }
    
    if (n == 1) {
        w[0] = MAGMA_C_REAL(a[0]);
        if (wantz) {
            a[0] = MAGMA_C_ONE;
        }
        return *info;
    }
    
    /* Get machine constants. */
    safmin = lapackf77_slamch("Safe minimum");
    eps    = lapackf77_slamch("Precision");
    smlnum = safmin / eps;
    bignum = 1. / smlnum;
    rmin = magma_ssqrt(smlnum);
    rmax = magma_ssqrt(bignum);
    
    /* Scale matrix to allowable range, if necessary. */
    anrm = lapackf77_clanhe("M", lapack_const(uplo_), &n, a, &lda, rwork);
    iscale = 0;
    if (anrm > 0. && anrm < rmin) {
        iscale = 1;
        sigma = rmin / anrm;
    } else if (anrm > rmax) {
        iscale = 1;
        sigma = rmax / anrm;
    }
    if (iscale == 1) {
        lapackf77_clascl(lapack_const(uplo_), &izero, &izero, &d_one, &sigma, &n, &n, a,
                         &lda, info);
    }
    
    /* Call CHETRD to reduce Hermitian matrix to tridiagonal form. */
    // chetrd rwork: e (n)
    // cstedx rwork: e (n) + llrwk (1 + 4*N + 2*N**2)  ==>  1 + 5n + 2n^2
    inde   = 0;
    indrwk = inde + n;
    llrwk  = lrwork - indrwk;
    
    // chetrd work: tau (n) + llwork (n*nb)  ==>  n + n*nb
    // cstedx work: tau (n) + z (n^2)
    // cunmtr work: tau (n) + z (n^2) + llwrk2 (n or n*nb)  ==>  2n + n^2, or n + n*nb + n^2
    indtau = 0;
    indwrk = indtau + n;
    indwk2 = indwrk + n*n;
    llwork = lwork - indwrk;
    llwrk2 = lwork - indwk2;
    
//#define ENABLE_TIMER
#ifdef ENABLE_TIMER
    magma_timestr_t start, end;
    start = get_current_time();
#endif
    
    magma_chetrd(lapack_const(uplo)[0], n, a, lda, w, &rwork[inde],
                 &work[indtau], &work[indwrk], llwork, &iinfo, queue);
    
#ifdef ENABLE_TIMER
    end = get_current_time();
    printf("time chetrd = %6.2f\n", GetTimerValue(start,end)/1000.);
#endif
    
    /* For eigenvalues only, call SSTERF.  For eigenvectors, first call
     CSTEDC to generate the eigenvector matrix, WORK(INDWRK), of the
     tridiagonal matrix, then call CUNMTR to multiply it to the Householder
     transformations represented as Householder vectors in A. */
    if (! wantz) {
        lapackf77_ssterf(&n, w, &rwork[inde], info);
    } else {
        
#ifdef ENABLE_TIMER
        start = get_current_time();
#endif
        
        if (MAGMA_SUCCESS != magma_smalloc( &dwork, (3*n*(n/2 + 1) ) )) {
            *info = MAGMA_ERR_DEVICE_ALLOC;
            return *info;
        }
        
        magma_cstedx(MagmaAllVec, n, 0., 0., 0, 0, w, &rwork[inde],
                     &work[indwrk], n, &rwork[indrwk],
                     llrwk, iwork, liwork, dwork, info, queue);
        
        magma_free( dwork );
        
#ifdef ENABLE_TIMER
        end = get_current_time();
        printf("time cstedx = %6.2f\n", GetTimerValue(start,end)/1000.);
        
        start = get_current_time();
#endif
        
        magma_cunmtr(MagmaLeft, uplo, MagmaNoTrans, n, n, a, lda, &work[indtau],
                     &work[indwrk], n, &work[indwk2], llwrk2, &iinfo, queue);
        
        lapackf77_clacpy("A", &n, &n, &work[indwrk], &n, a, &lda);
        
#ifdef ENABLE_TIMER
        end = get_current_time();
        printf("time cunmtr + copy = %6.2f\n", GetTimerValue(start,end)/1000.);
#endif
    }
    
    /* If matrix was scaled, then rescale eigenvalues appropriately. */
    if (iscale == 1) {
        if (*info == 0) {
            imax = n;
        } else {
            imax = *info - 1;
        }
        d__1 = 1. / sigma;
        blasf77_sscal(&imax, &d__1, w, &ione);
    }
    
    work[0]  = MAGMA_C_MAKE( lwmin * (1. + lapackf77_slamch("Epsilon")), 0.);  // round up
    rwork[0] = lrwmin * (1. + lapackf77_slamch("Epsilon"));
    iwork[0] = liwmin;
    
    return *info;
} /* magma_cheevd */
예제 #5
0
파일: cheevd.cpp 프로젝트: cjy7117/FT-MAGMA
/**
    Purpose
    -------
    CHEEVD computes all eigenvalues and, optionally, eigenvectors of a
    complex Hermitian matrix A.  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
    ---------
    @param[in]
    jobz    magma_vec_t
      -     = MagmaNoVec:  Compute eigenvalues only;
      -     = MagmaVec:    Compute eigenvalues and eigenvectors.

    @param[in]
    uplo    magma_uplo_t
      -     = MagmaUpper:  Upper triangle of A is stored;
      -     = MagmaLower:  Lower triangle of A is stored.

    @param[in]
    n       INTEGER
            The order of the matrix A.  N >= 0.

    @param[in,out]
    A       COMPLEX array, dimension (LDA, N)
            On entry, the Hermitian matrix A.  If UPLO = MagmaUpper, the
            leading N-by-N upper triangular part of A contains the
            upper triangular part of the matrix A.  If UPLO = MagmaLower,
            the leading N-by-N lower triangular part of A contains
            the lower triangular part of the matrix A.
            On exit, if JOBZ = MagmaVec, then if INFO = 0, A contains the
            orthonormal eigenvectors of the matrix A.
            If JOBZ = MagmaNoVec, then on exit the lower triangle (if UPLO=MagmaLower)
            or the upper triangle (if UPLO=MagmaUpper) of A, including the
            diagonal, is destroyed.

    @param[in]
    lda     INTEGER
            The leading dimension of the array A.  LDA >= max(1,N).

    @param[out]
    w       REAL array, dimension (N)
            If INFO = 0, the eigenvalues in ascending order.

    @param[out]
    work    (workspace) COMPLEX array, dimension (MAX(1,LWORK))
            On exit, if INFO = 0, WORK[0] returns the optimal LWORK.

    @param[in]
    lwork   INTEGER
            The length of the array WORK.
            If N <= 1,                      LWORK >= 1.
            If JOBZ = MagmaNoVec and N > 1, LWORK >= N + N*NB.
            If JOBZ = MagmaVec   and N > 1, LWORK >= max( N + N*NB, 2*N + N**2 ).
            NB can be obtained through magma_get_chetrd_nb(N).
    \n
            If LWORK = -1, then a workspace query is assumed; the routine
            only calculates the optimal sizes of the WORK, RWORK and
            IWORK arrays, returns these values as the first entries of
            the WORK, RWORK and IWORK arrays, and no error message
            related to LWORK or LRWORK or LIWORK is issued by XERBLA.

    @param[out]
    rwork   (workspace) REAL array, dimension (LRWORK)
            On exit, if INFO = 0, RWORK[0] returns the optimal LRWORK.

    @param[in]
    lrwork  INTEGER
            The dimension of the array RWORK.
            If N <= 1,                      LRWORK >= 1.
            If JOBZ = MagmaNoVec and N > 1, LRWORK >= N.
            If JOBZ = MagmaVec   and N > 1, LRWORK >= 1 + 5*N + 2*N**2.
    \n
            If LRWORK = -1, then a workspace query is assumed; the
            routine only calculates the optimal sizes of the WORK, RWORK
            and IWORK arrays, returns these values as the first entries
            of the WORK, RWORK and IWORK arrays, and no error message
            related to LWORK or LRWORK or LIWORK is issued by XERBLA.

    @param[out]
    iwork   (workspace) INTEGER array, dimension (MAX(1,LIWORK))
            On exit, if INFO = 0, IWORK[0] returns the optimal LIWORK.

    @param[in]
    liwork  INTEGER
            The dimension of the array IWORK.
            If N <= 1,                      LIWORK >= 1.
            If JOBZ = MagmaNoVec and N > 1, LIWORK >= 1.
            If JOBZ = MagmaVec   and N > 1, LIWORK >= 3 + 5*N.
    \n
            If LIWORK = -1, then a workspace query is assumed; the
            routine only calculates the optimal sizes of the WORK, RWORK
            and IWORK arrays, returns these values as the first entries
            of the WORK, RWORK and IWORK arrays, and no error message
            related to LWORK or LRWORK or LIWORK is issued by XERBLA.

    @param[out]
    info    INTEGER
      -     = 0:  successful exit
      -     < 0:  if INFO = -i, the i-th argument had an illegal value
      -     > 0:  if INFO = i and JOBZ = MagmaNoVec, then the algorithm failed
                  to converge; i off-diagonal elements of an intermediate
                  tridiagonal form did not converge to zero;
                  if INFO = i and JOBZ = MagmaVec, then the algorithm failed
                  to compute an eigenvalue while working on the submatrix
                  lying in rows and columns INFO/(N+1) through
                  mod(INFO,N+1).

    Further Details
    ---------------
    Based on contributions by
       Jeff Rutter, Computer Science Division, University of California
       at Berkeley, USA

    Modified description of INFO. Sven, 16 Feb 05.

    @ingroup magma_cheev_driver
    ********************************************************************/
extern "C" magma_int_t
magma_cheevd(
    magma_vec_t jobz, magma_uplo_t uplo,
    magma_int_t n,
    magmaFloatComplex *A, magma_int_t lda,
    float *w,
    magmaFloatComplex *work, magma_int_t lwork,
    #ifdef COMPLEX
    float *rwork, magma_int_t lrwork,
    #endif
    magma_int_t *iwork, magma_int_t liwork,
    magma_int_t *info)
{
    const char* uplo_ = lapack_uplo_const( uplo );
    const char* jobz_ = lapack_vec_const( jobz );
    magma_int_t ione = 1;
    magma_int_t izero = 0;
    float d_one = 1.;

    float d__1;

    float eps;
    magma_int_t inde;
    float anrm;
    magma_int_t imax;
    float rmin, rmax;
    float sigma;
    magma_int_t iinfo, lwmin;
    magma_int_t lower;
    magma_int_t llrwk;
    magma_int_t wantz;
    magma_int_t indwk2, llwrk2;
    magma_int_t iscale;
    float safmin;
    float bignum;
    magma_int_t indtau;
    magma_int_t indrwk, indwrk, liwmin;
    magma_int_t lrwmin, llwork;
    float smlnum;
    magma_int_t lquery;

    float* dwork;

    wantz = (jobz == MagmaVec);
    lower = (uplo == MagmaLower);
    lquery = (lwork == -1 || lrwork == -1 || liwork == -1);

    *info = 0;

    if (! (wantz || (jobz == MagmaNoVec))) {
        *info = -1;
    } else if (! (lower || (uplo == MagmaUpper))) {
        *info = -2;
    } else if (n < 0) {
        *info = -3;
    } else if (lda < max(1,n)) {
        *info = -5;
    }

    magma_int_t nb = magma_get_chetrd_nb( n );
    if ( n <= 1 ) {
        lwmin  = 1;
        lrwmin = 1;
        liwmin = 1;
    }
    else if ( wantz ) {
        lwmin  = max( n + n*nb, 2*n + n*n );
        lrwmin = 1 + 5*n + 2*n*n;
        liwmin = 3 + 5*n;
    }
    else {
        lwmin  = n + n*nb;
        lrwmin = n;
        liwmin = 1;
    }
    
    // multiply by 1+eps (in Double!) to ensure length gets rounded up,
    // if it cannot be exactly represented in floating point.
    real_Double_t one_eps = 1. + lapackf77_slamch("Epsilon");
    work[0]  = MAGMA_C_MAKE( lwmin * one_eps, 0 );
    rwork[0] = lrwmin * one_eps;
    iwork[0] = liwmin;

    if ((lwork < lwmin) && !lquery) {
        *info = -8;
    } else if ((lrwork < lrwmin) && ! lquery) {
        *info = -10;
    } else if ((liwork < liwmin) && ! lquery) {
        *info = -12;
    }

    if (*info != 0) {
        magma_xerbla( __func__, -(*info) );
        return *info;
    }
    else if (lquery) {
        return *info;
    }

    /* Quick return if possible */
    if (n == 0) {
        return *info;
    }

    if (n == 1) {
        w[0] = MAGMA_C_REAL( A[0] );
        if (wantz) {
            A[0] = MAGMA_C_ONE;
        }
        return *info;
    }

    /* If matrix is very small, then just call LAPACK on CPU, no need for GPU */
    if (n <= 128) {
        lapackf77_cheevd( jobz_, uplo_,
                          &n, A, &lda,
                          w, work, &lwork,
                          #ifdef COMPLEX
                          rwork, &lrwork,
                          #endif
                          iwork, &liwork, info );
        return *info;
    }

    /* Get machine constants. */
    safmin = lapackf77_slamch("Safe minimum");
    eps    = lapackf77_slamch("Precision");
    smlnum = safmin / eps;
    bignum = 1. / smlnum;
    rmin = magma_ssqrt( smlnum );
    rmax = magma_ssqrt( bignum );

    /* Scale matrix to allowable range, if necessary. */
    anrm = lapackf77_clanhe( "M", uplo_, &n, A, &lda, rwork );
    iscale = 0;
    if (anrm > 0. && anrm < rmin) {
        iscale = 1;
        sigma = rmin / anrm;
    } else if (anrm > rmax) {
        iscale = 1;
        sigma = rmax / anrm;
    }
    if (iscale == 1) {
        lapackf77_clascl( uplo_, &izero, &izero, &d_one, &sigma, &n, &n, A, &lda, info );
    }

    /* Call CHETRD to reduce Hermitian matrix to tridiagonal form. */
    // chetrd rwork: e (n)
    // cstedx rwork: e (n) + llrwk (1 + 4*N + 2*N**2)  ==>  1 + 5n + 2n^2
    inde   = 0;
    indrwk = inde + n;
    llrwk  = lrwork - indrwk;

    // chetrd work: tau (n) + llwork (n*nb)  ==>  n + n*nb
    // cstedx work: tau (n) + z (n^2)
    // cunmtr work: tau (n) + z (n^2) + llwrk2 (n or n*nb)  ==>  2n + n^2, or n + n*nb + n^2
    indtau = 0;
    indwrk = indtau + n;
    indwk2 = indwrk + n*n;
    llwork = lwork - indwrk;
    llwrk2 = lwork - indwk2;

    magma_timer_t time=0;
    timer_start( time );

    magma_chetrd( uplo, n, A, lda, w, &rwork[inde],
                  &work[indtau], &work[indwrk], llwork, &iinfo );

    timer_stop( time );
    timer_printf( "time chetrd = %6.2f\n", time );

    /* For eigenvalues only, call SSTERF.  For eigenvectors, first call
     * CSTEDC to generate the eigenvector matrix, WORK(INDWRK), of the
     * tridiagonal matrix, then call CUNMTR to multiply it to the Householder
     * transformations represented as Householder vectors in A. */
    if (! wantz) {
        lapackf77_ssterf( &n, w, &rwork[inde], info );
    }
    else {
        timer_start( time );

        if (MAGMA_SUCCESS != magma_smalloc( &dwork, 3*n*(n/2 + 1) )) {
            *info = MAGMA_ERR_DEVICE_ALLOC;
            return *info;
        }

        magma_cstedx( MagmaRangeAll, n, 0., 0., 0, 0, w, &rwork[inde],
                      &work[indwrk], n, &rwork[indrwk], llrwk,
                      iwork, liwork, dwork, info );

        magma_free( dwork );

        timer_stop( time );
        timer_printf( "time cstedx = %6.2f\n", time );
        timer_start( time );

        magma_cunmtr( MagmaLeft, uplo, MagmaNoTrans, n, n, A, lda, &work[indtau],
                      &work[indwrk], n, &work[indwk2], llwrk2, &iinfo );

        lapackf77_clacpy( "A", &n, &n, &work[indwrk], &n, A, &lda );

        timer_stop( time );
        timer_printf( "time cunmtr + copy = %6.2f\n", time );
    }

    /* If matrix was scaled, then rescale eigenvalues appropriately. */
    if (iscale == 1) {
        if (*info == 0) {
            imax = n;
        } else {
            imax = *info - 1;
        }
        d__1 = 1. / sigma;
        blasf77_sscal( &imax, &d__1, w, &ione );
    }

    work[0]  = MAGMA_C_MAKE( lwmin * one_eps, 0 );  // round up
    rwork[0] = lrwmin * one_eps;
    iwork[0] = liwmin;

    return *info;
} /* magma_cheevd */
예제 #6
0
파일: cheevdx.cpp 프로젝트: xulunfan/magma
/**
    Purpose
    -------
    CHEEVDX computes selected eigenvalues and, optionally, eigenvectors
    of a complex Hermitian matrix A. Eigenvalues and eigenvectors can
    be selected by specifying either a range of values or a range of
    indices for the desired eigenvalues.
    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
    ---------
    @param[in]
    jobz    magma_vec_t
      -     = MagmaNoVec:  Compute eigenvalues only;
      -     = MagmaVec:    Compute eigenvalues and eigenvectors.

    @param[in]
    range   magma_range_t
      -     = MagmaRangeAll: all eigenvalues will be found.
      -     = MagmaRangeV:   all eigenvalues in the half-open interval (VL,VU]
                   will be found.
      -     = MagmaRangeI:   the IL-th through IU-th eigenvalues will be found.

    @param[in]
    uplo    magma_uplo_t
      -     = MagmaUpper:  Upper triangle of A is stored;
      -     = MagmaLower:  Lower triangle of A is stored.

    @param[in]
    n       INTEGER
            The order of the matrix A.  N >= 0.

    @param[in,out]
    A       COMPLEX array, dimension (LDA, N)
            On entry, the Hermitian matrix A.  If UPLO = MagmaUpper, the
            leading N-by-N upper triangular part of A contains the
            upper triangular part of the matrix A.  If UPLO = MagmaLower,
            the leading N-by-N lower triangular part of A contains
            the lower triangular part of the matrix A.
            On exit, if JOBZ = MagmaVec, then if INFO = 0, the first m columns
            of A contains the required
            orthonormal eigenvectors of the matrix A.
            If JOBZ = MagmaNoVec, then on exit the lower triangle (if UPLO=MagmaLower)
            or the upper triangle (if UPLO=MagmaUpper) of A, including the
            diagonal, is destroyed.

    @param[in]
    lda     INTEGER
            The leading dimension of the array A.  LDA >= max(1,N).

    @param[in]
    vl      REAL
    @param[in]
    vu      REAL
            If RANGE=MagmaRangeV, the lower and upper bounds of the interval to
            be searched for eigenvalues. VL < VU.
            Not referenced if RANGE = MagmaRangeAll or MagmaRangeI.

    @param[in]
    il      INTEGER
    @param[in]
    iu      INTEGER
            If RANGE=MagmaRangeI, 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 = MagmaRangeAll or MagmaRangeV.

    @param[out]
    m       INTEGER
            The total number of eigenvalues found.  0 <= M <= N.
            If RANGE = MagmaRangeAll, M = N, and if RANGE = MagmaRangeI, M = IU-IL+1.

    @param[out]
    w       REAL array, dimension (N)
            If INFO = 0, the required m eigenvalues in ascending order.

    @param[out]
    work    (workspace) COMPLEX array, dimension (MAX(1,LWORK))
            On exit, if INFO = 0, WORK[0] returns the optimal LWORK.

    @param[in]
    lwork   INTEGER
            The length of the array WORK.
            If N <= 1,                      LWORK >= 1.
            If JOBZ = MagmaNoVec and N > 1, LWORK >= N + N*NB.
            If JOBZ = MagmaVec   and N > 1, LWORK >= max( N + N*NB, 2*N + N**2 ).
            NB can be obtained through magma_get_chetrd_nb(N).
    \n
            If LWORK = -1, then a workspace query is assumed; the routine
            only calculates the optimal sizes of the WORK, RWORK and
            IWORK arrays, returns these values as the first entries of
            the WORK, RWORK and IWORK arrays, and no error message
            related to LWORK or LRWORK or LIWORK is issued by XERBLA.

    @param[out]
    rwork   (workspace) REAL array,
                                           dimension (LRWORK)
            On exit, if INFO = 0, RWORK[0] returns the optimal LRWORK.

    @param[in]
    lrwork  INTEGER
            The dimension of the array RWORK.
            If N <= 1,                      LRWORK >= 1.
            If JOBZ = MagmaNoVec and N > 1, LRWORK >= N.
            If JOBZ = MagmaVec   and N > 1, LRWORK >= 1 + 5*N + 2*N**2.
    \n
            If LRWORK = -1, then a workspace query is assumed; the
            routine only calculates the optimal sizes of the WORK, RWORK
            and IWORK arrays, returns these values as the first entries
            of the WORK, RWORK and IWORK arrays, and no error message
            related to LWORK or LRWORK or LIWORK is issued by XERBLA.

    @param[out]
    iwork   (workspace) INTEGER array, dimension (MAX(1,LIWORK))
            On exit, if INFO = 0, IWORK[0] returns the optimal LIWORK.

    @param[in]
    liwork  INTEGER
            The dimension of the array IWORK.
            If N <= 1,                      LIWORK >= 1.
            If JOBZ = MagmaNoVec and N > 1, LIWORK >= 1.
            If JOBZ = MagmaVec   and N > 1, LIWORK >= 3 + 5*N.
    \n
            If LIWORK = -1, then a workspace query is assumed; the
            routine only calculates the optimal sizes of the WORK, RWORK
            and IWORK arrays, returns these values as the first entries
            of the WORK, RWORK and IWORK arrays, and no error message
            related to LWORK or LRWORK or LIWORK is issued by XERBLA.

    @param[out]
    info    INTEGER
      -     = 0:  successful exit
      -     < 0:  if INFO = -i, the i-th argument had an illegal value
      -     > 0:  if INFO = i and JOBZ = MagmaNoVec, then the algorithm failed
                  to converge; i off-diagonal elements of an intermediate
                  tridiagonal form did not converge to zero;
                  if INFO = i and JOBZ = MagmaVec, then the algorithm failed
                  to compute an eigenvalue while working on the submatrix
                  lying in rows and columns INFO/(N+1) through
                  mod(INFO,N+1).

    Further Details
    ---------------
    Based on contributions by
       Jeff Rutter, Computer Science Division, University of California
       at Berkeley, USA

    Modified description of INFO. Sven, 16 Feb 05.

    @ingroup magma_cheev_driver
    ********************************************************************/
extern "C" magma_int_t
magma_cheevdx(
    magma_vec_t jobz, magma_range_t range, magma_uplo_t uplo,
    magma_int_t n,
    magmaFloatComplex *A, magma_int_t lda,
    float vl, float vu, magma_int_t il, magma_int_t iu,
    magma_int_t *m, float *w,
    magmaFloatComplex *work, magma_int_t lwork,
    #ifdef COMPLEX
    float *rwork, magma_int_t lrwork,
    #endif
    magma_int_t *iwork, magma_int_t liwork,
    magma_int_t *info)
{
    const char* uplo_  = lapack_uplo_const( uplo  );
    const char* jobz_  = lapack_vec_const( jobz  );
    magma_int_t ione = 1;
    magma_int_t izero = 0;
    float d_one = 1.;

    float d__1;

    float eps;
    magma_int_t inde;
    float anrm;
    magma_int_t imax;
    float rmin, rmax;
    float sigma;
    magma_int_t iinfo, lwmin;
    magma_int_t lower;
    magma_int_t llrwk;
    magma_int_t wantz;
    magma_int_t indwk2, llwrk2;
    magma_int_t iscale;
    float safmin;
    float bignum;
    magma_int_t indtau;
    magma_int_t indrwk, indwrk, liwmin;
    magma_int_t lrwmin, llwork;
    float smlnum;
    magma_int_t lquery;
    magma_int_t alleig, valeig, indeig;

    float* dwork;

    wantz = (jobz == MagmaVec);
    lower = (uplo == MagmaLower);

    alleig = (range == MagmaRangeAll);
    valeig = (range == MagmaRangeV);
    indeig = (range == MagmaRangeI);

    lquery = (lwork == -1 || lrwork == -1 || liwork == -1);

    *info = 0;
    if (! (wantz || (jobz == MagmaNoVec))) {
        *info = -1;
    } else if (! (alleig || valeig || indeig)) {
        *info = -2;
    } else if (! (lower || (uplo == MagmaUpper))) {
        *info = -3;
    } else if (n < 0) {
        *info = -4;
    } else if (lda < max(1,n)) {
        *info = -6;
    } else {
        if (valeig) {
            if (n > 0 && vu <= vl) {
                *info = -8;
            }
        } else if (indeig) {
            if (il < 1 || il > max(1,n)) {
                *info = -9;
            } else if (iu < min(n,il) || iu > n) {
                *info = -10;
            }
        }
    }

    magma_int_t nb = magma_get_chetrd_nb( n );
    if ( n <= 1 ) {
        lwmin  = 1;
        lrwmin = 1;
        liwmin = 1;
    }
    else if ( wantz ) {
        lwmin  = max( n + n*nb, 2*n + n*n );
        lrwmin = 1 + 5*n + 2*n*n;
        liwmin = 3 + 5*n;
    }
    else {
        lwmin  = n + n*nb;
        lrwmin = n;
        liwmin = 1;
    }
    
    work[0]  = magma_cmake_lwork( lwmin );
    rwork[0] = magma_smake_lwork( lrwmin );
    iwork[0] = liwmin;

    if ((lwork < lwmin) && !lquery) {
        *info = -14;
    } else if ((lrwork < lrwmin) && ! lquery) {
        *info = -16;
    } else if ((liwork < liwmin) && ! lquery) {
        *info = -18;
    }

    if (*info != 0) {
        magma_xerbla( __func__, -(*info) );
        return *info;
    }
    else if (lquery) {
        return *info;
    }

    /* Quick return if possible */
    if (n == 0) {
        return *info;
    }

    if (n == 1) {
        w[0] = MAGMA_C_REAL(A[0]);
        if (wantz) {
            A[0] = MAGMA_C_ONE;
        }
        return *info;
    }
    /* Check if matrix is very small then just call LAPACK on CPU, no need for GPU */
    if (n <= 128) {
        #ifdef ENABLE_DEBUG
        printf("--------------------------------------------------------------\n");
        printf("  warning matrix too small N=%d NB=%d, calling lapack on CPU  \n", (int) n, (int) nb);
        printf("--------------------------------------------------------------\n");
        #endif
        lapackf77_cheevd(jobz_, uplo_,
                         &n, A, &lda,
                         w, work, &lwork,
                         #ifdef COMPLEX
                         rwork, &lrwork,
                         #endif
                         iwork, &liwork, info);
        return *info;
    }
    /* Get machine constants. */
    safmin = lapackf77_slamch("Safe minimum");
    eps    = lapackf77_slamch("Precision");
    smlnum = safmin / eps;
    bignum = 1. / smlnum;
    rmin = magma_ssqrt(smlnum);
    rmax = magma_ssqrt(bignum);

    /* Scale matrix to allowable range, if necessary. */
    anrm = lapackf77_clanhe("M", uplo_, &n, A, &lda, rwork);
    iscale = 0;
    if (anrm > 0. && anrm < rmin) {
        iscale = 1;
        sigma = rmin / anrm;
    } else if (anrm > rmax) {
        iscale = 1;
        sigma = rmax / anrm;
    }
    if (iscale == 1) {
        lapackf77_clascl(uplo_, &izero, &izero, &d_one, &sigma, &n, &n, A,
                         &lda, info);
    }

    /* Call CHETRD to reduce Hermitian matrix to tridiagonal form. */
    // chetrd rwork: e (n)
    // cstedx rwork: e (n) + llrwk (1 + 4*N + 2*N**2)  ==>  1 + 5n + 2n^2
    inde   = 0;
    indrwk = inde + n;
    llrwk  = lrwork - indrwk;

    // chetrd work: tau (n) + llwork (n*nb)  ==>  n + n*nb
    // cstedx work: tau (n) + z (n^2)
    // cunmtr work: tau (n) + z (n^2) + llwrk2 (n or n*nb)  ==>  2n + n^2, or n + n*nb + n^2
    indtau = 0;
    indwrk = indtau + n;
    indwk2 = indwrk + n*n;
    llwork = lwork - indwrk;
    llwrk2 = lwork - indwk2;

    magma_timer_t time=0;
    timer_start( time );

    magma_chetrd(uplo, n, A, lda, w, &rwork[inde],
                 &work[indtau], &work[indwrk], llwork, &iinfo);

    timer_stop( time );
    timer_printf( "time chetrd = %6.2f\n", time );

    /* For eigenvalues only, call SSTERF.  For eigenvectors, first call
     CSTEDC to generate the eigenvector matrix, WORK(INDWRK), of the
     tridiagonal matrix, then call CUNMTR to multiply it to the Householder
     transformations represented as Householder vectors in A. */
    if (! wantz) {
        lapackf77_ssterf(&n, w, &rwork[inde], info);

        magma_smove_eig(range, n, w, &il, &iu, vl, vu, m);
    }
    else {
        timer_start( time );

        if (MAGMA_SUCCESS != magma_smalloc( &dwork, 3*n*(n/2 + 1) )) {
            *info = MAGMA_ERR_DEVICE_ALLOC;
            return *info;
        }

        magma_cstedx(range, n, vl, vu, il, iu, w, &rwork[inde],
                     &work[indwrk], n, &rwork[indrwk],
                     llrwk, iwork, liwork, dwork, info);

        magma_free( dwork );

        timer_stop( time );
        timer_printf( "time cstedx = %6.2f\n", time );
        timer_start( time );

        magma_smove_eig(range, n, w, &il, &iu, vl, vu, m);

        magma_cunmtr(MagmaLeft, uplo, MagmaNoTrans, n, *m, A, lda, &work[indtau],
                     &work[indwrk + n * (il-1) ], n, &work[indwk2], llwrk2, &iinfo);

        lapackf77_clacpy("A", &n, m, &work[indwrk + n * (il-1)], &n, A, &lda);

        timer_stop( time );
        timer_printf( "time cunmtr + copy = %6.2f\n", time );
    }

    /* If matrix was scaled, then rescale eigenvalues appropriately. */
    if (iscale == 1) {
        if (*info == 0) {
            imax = n;
        } else {
            imax = *info - 1;
        }
        d__1 = 1. / sigma;
        blasf77_sscal(&imax, &d__1, w, &ione);
    }

    work[0]  = magma_cmake_lwork( lwmin );
    rwork[0] = magma_smake_lwork( lrwmin );
    iwork[0] = liwmin;

    return *info;
} /* magma_cheevdx */