LAPACK 3.3.1
Linear Algebra PACKage
|
00001 REAL FUNCTION CLA_HERPVGRW( UPLO, N, INFO, A, LDA, AF, LDAF, IPIV, 00002 $ WORK ) 00003 * 00004 * -- LAPACK routine (version 3.2.2) -- 00005 * -- Contributed by James Demmel, Deaglan Halligan, Yozo Hida and -- 00006 * -- Jason Riedy of Univ. of California Berkeley. -- 00007 * -- June 2010 -- 00008 * 00009 * -- LAPACK is a software package provided by Univ. of Tennessee, -- 00010 * -- Univ. of California Berkeley and NAG Ltd. -- 00011 * 00012 IMPLICIT NONE 00013 * .. 00014 * .. Scalar Arguments .. 00015 CHARACTER*1 UPLO 00016 INTEGER N, INFO, LDA, LDAF 00017 * .. 00018 * .. Array Arguments .. 00019 INTEGER IPIV( * ) 00020 COMPLEX A( LDA, * ), AF( LDAF, * ) 00021 REAL WORK( * ) 00022 * .. 00023 * 00024 * Purpose 00025 * ======= 00026 * 00027 * CLA_HERPVGRW computes the reciprocal pivot growth factor 00028 * norm(A)/norm(U). The "max absolute element" norm is used. If this is 00029 * much less than 1, the stability of the LU factorization of the 00030 * (equilibrated) matrix A could be poor. This also means that the 00031 * solution X, estimated condition numbers, and error bounds could be 00032 * unreliable. 00033 * 00034 * Arguments 00035 * ========= 00036 * 00037 * UPLO (input) CHARACTER*1 00038 * = 'U': Upper triangle of A is stored; 00039 * = 'L': Lower triangle of A is stored. 00040 * 00041 * N (input) INTEGER 00042 * The number of linear equations, i.e., the order of the 00043 * matrix A. N >= 0. 00044 * 00045 * INFO (input) INTEGER 00046 * The value of INFO returned from SSYTRF, .i.e., the pivot in 00047 * column INFO is exactly 0. 00048 * 00049 * NCOLS (input) INTEGER 00050 * The number of columns of the matrix A. NCOLS >= 0. 00051 * 00052 * A (input) COMPLEX array, dimension (LDA,N) 00053 * On entry, the N-by-N matrix A. 00054 * 00055 * LDA (input) INTEGER 00056 * The leading dimension of the array A. LDA >= max(1,N). 00057 * 00058 * AF (input) COMPLEX array, dimension (LDAF,N) 00059 * The block diagonal matrix D and the multipliers used to 00060 * obtain the factor U or L as computed by CHETRF. 00061 * 00062 * LDAF (input) INTEGER 00063 * The leading dimension of the array AF. LDAF >= max(1,N). 00064 * 00065 * IPIV (input) INTEGER array, dimension (N) 00066 * Details of the interchanges and the block structure of D 00067 * as determined by CHETRF. 00068 * 00069 * WORK (input) COMPLEX array, dimension (2*N) 00070 * 00071 * ===================================================================== 00072 * 00073 * .. Local Scalars .. 00074 INTEGER NCOLS, I, J, K, KP 00075 REAL AMAX, UMAX, RPVGRW, TMP 00076 LOGICAL UPPER, LSAME 00077 COMPLEX ZDUM 00078 * .. 00079 * .. External Functions .. 00080 EXTERNAL LSAME, CLASET 00081 * .. 00082 * .. Intrinsic Functions .. 00083 INTRINSIC ABS, REAL, AIMAG, MAX, MIN 00084 * .. 00085 * .. Statement Functions .. 00086 REAL CABS1 00087 * .. 00088 * .. Statement Function Definitions .. 00089 CABS1( ZDUM ) = ABS( REAL ( ZDUM ) ) + ABS( AIMAG ( ZDUM ) ) 00090 * .. 00091 * .. Executable Statements .. 00092 * 00093 UPPER = LSAME( 'Upper', UPLO ) 00094 IF ( INFO.EQ.0 ) THEN 00095 IF (UPPER) THEN 00096 NCOLS = 1 00097 ELSE 00098 NCOLS = N 00099 END IF 00100 ELSE 00101 NCOLS = INFO 00102 END IF 00103 00104 RPVGRW = 1.0 00105 DO I = 1, 2*N 00106 WORK( I ) = 0.0 00107 END DO 00108 * 00109 * Find the max magnitude entry of each column of A. Compute the max 00110 * for all N columns so we can apply the pivot permutation while 00111 * looping below. Assume a full factorization is the common case. 00112 * 00113 IF ( UPPER ) THEN 00114 DO J = 1, N 00115 DO I = 1, J 00116 WORK( N+I ) = MAX( CABS1( A( I,J ) ), WORK( N+I ) ) 00117 WORK( N+J ) = MAX( CABS1( A( I,J ) ), WORK( N+J ) ) 00118 END DO 00119 END DO 00120 ELSE 00121 DO J = 1, N 00122 DO I = J, N 00123 WORK( N+I ) = MAX( CABS1( A( I, J ) ), WORK( N+I ) ) 00124 WORK( N+J ) = MAX( CABS1( A( I, J ) ), WORK( N+J ) ) 00125 END DO 00126 END DO 00127 END IF 00128 * 00129 * Now find the max magnitude entry of each column of U or L. Also 00130 * permute the magnitudes of A above so they're in the same order as 00131 * the factor. 00132 * 00133 * The iteration orders and permutations were copied from csytrs. 00134 * Calls to SSWAP would be severe overkill. 00135 * 00136 IF ( UPPER ) THEN 00137 K = N 00138 DO WHILE ( K .LT. NCOLS .AND. K.GT.0 ) 00139 IF ( IPIV( K ).GT.0 ) THEN 00140 ! 1x1 pivot 00141 KP = IPIV( K ) 00142 IF ( KP .NE. K ) THEN 00143 TMP = WORK( N+K ) 00144 WORK( N+K ) = WORK( N+KP ) 00145 WORK( N+KP ) = TMP 00146 END IF 00147 DO I = 1, K 00148 WORK( K ) = MAX( CABS1( AF( I, K ) ), WORK( K ) ) 00149 END DO 00150 K = K - 1 00151 ELSE 00152 ! 2x2 pivot 00153 KP = -IPIV( K ) 00154 TMP = WORK( N+K-1 ) 00155 WORK( N+K-1 ) = WORK( N+KP ) 00156 WORK( N+KP ) = TMP 00157 DO I = 1, K-1 00158 WORK( K ) = MAX( CABS1( AF( I, K ) ), WORK( K ) ) 00159 WORK( K-1 ) = 00160 $ MAX( CABS1( AF( I, K-1 ) ), WORK( K-1 ) ) 00161 END DO 00162 WORK( K ) = MAX( CABS1( AF( K, K ) ), WORK( K ) ) 00163 K = K - 2 00164 END IF 00165 END DO 00166 K = NCOLS 00167 DO WHILE ( K .LE. N ) 00168 IF ( IPIV( K ).GT.0 ) THEN 00169 KP = IPIV( K ) 00170 IF ( KP .NE. K ) THEN 00171 TMP = WORK( N+K ) 00172 WORK( N+K ) = WORK( N+KP ) 00173 WORK( N+KP ) = TMP 00174 END IF 00175 K = K + 1 00176 ELSE 00177 KP = -IPIV( K ) 00178 TMP = WORK( N+K ) 00179 WORK( N+K ) = WORK( N+KP ) 00180 WORK( N+KP ) = TMP 00181 K = K + 2 00182 END IF 00183 END DO 00184 ELSE 00185 K = 1 00186 DO WHILE ( K .LE. NCOLS ) 00187 IF ( IPIV( K ).GT.0 ) THEN 00188 ! 1x1 pivot 00189 KP = IPIV( K ) 00190 IF ( KP .NE. K ) THEN 00191 TMP = WORK( N+K ) 00192 WORK( N+K ) = WORK( N+KP ) 00193 WORK( N+KP ) = TMP 00194 END IF 00195 DO I = K, N 00196 WORK( K ) = MAX( CABS1( AF( I, K ) ), WORK( K ) ) 00197 END DO 00198 K = K + 1 00199 ELSE 00200 ! 2x2 pivot 00201 KP = -IPIV( K ) 00202 TMP = WORK( N+K+1 ) 00203 WORK( N+K+1 ) = WORK( N+KP ) 00204 WORK( N+KP ) = TMP 00205 DO I = K+1, N 00206 WORK( K ) = MAX( CABS1( AF( I, K ) ), WORK( K ) ) 00207 WORK( K+1 ) = 00208 $ MAX( CABS1( AF( I, K+1 ) ) , WORK( K+1 ) ) 00209 END DO 00210 WORK(K) = MAX( CABS1( AF( K, K ) ), WORK( K ) ) 00211 K = K + 2 00212 END IF 00213 END DO 00214 K = NCOLS 00215 DO WHILE ( K .GE. 1 ) 00216 IF ( IPIV( K ).GT.0 ) THEN 00217 KP = IPIV( K ) 00218 IF ( KP .NE. K ) THEN 00219 TMP = WORK( N+K ) 00220 WORK( N+K ) = WORK( N+KP ) 00221 WORK( N+KP ) = TMP 00222 END IF 00223 K = K - 1 00224 ELSE 00225 KP = -IPIV( K ) 00226 TMP = WORK( N+K ) 00227 WORK( N+K ) = WORK( N+KP ) 00228 WORK( N+KP ) = TMP 00229 K = K - 2 00230 ENDIF 00231 END DO 00232 END IF 00233 * 00234 * Compute the *inverse* of the max element growth factor. Dividing 00235 * by zero would imply the largest entry of the factor's column is 00236 * zero. Than can happen when either the column of A is zero or 00237 * massive pivots made the factor underflow to zero. Neither counts 00238 * as growth in itself, so simply ignore terms with zero 00239 * denominators. 00240 * 00241 IF ( UPPER ) THEN 00242 DO I = NCOLS, N 00243 UMAX = WORK( I ) 00244 AMAX = WORK( N+I ) 00245 IF ( UMAX /= 0.0 ) THEN 00246 RPVGRW = MIN( AMAX / UMAX, RPVGRW ) 00247 END IF 00248 END DO 00249 ELSE 00250 DO I = 1, NCOLS 00251 UMAX = WORK( I ) 00252 AMAX = WORK( N+I ) 00253 IF ( UMAX /= 0.0 ) THEN 00254 RPVGRW = MIN( AMAX / UMAX, RPVGRW ) 00255 END IF 00256 END DO 00257 END IF 00258 00259 CLA_HERPVGRW = RPVGRW 00260 END