LAPACK 3.3.1
Linear Algebra PACKage
|
00001 SUBROUTINE DQLT03( M, N, K, AF, C, CC, Q, LDA, TAU, WORK, LWORK, 00002 $ RWORK, RESULT ) 00003 * 00004 * -- LAPACK test routine (version 3.1) -- 00005 * Univ. of Tennessee, Univ. of California Berkeley and NAG Ltd.. 00006 * November 2006 00007 * 00008 * .. Scalar Arguments .. 00009 INTEGER K, LDA, LWORK, M, N 00010 * .. 00011 * .. Array Arguments .. 00012 DOUBLE PRECISION AF( LDA, * ), C( LDA, * ), CC( LDA, * ), 00013 $ Q( LDA, * ), RESULT( * ), RWORK( * ), TAU( * ), 00014 $ WORK( LWORK ) 00015 * .. 00016 * 00017 * Purpose 00018 * ======= 00019 * 00020 * DQLT03 tests DORMQL, which computes Q*C, Q'*C, C*Q or C*Q'. 00021 * 00022 * DQLT03 compares the results of a call to DORMQL with the results of 00023 * forming Q explicitly by a call to DORGQL and then performing matrix 00024 * multiplication by a call to DGEMM. 00025 * 00026 * Arguments 00027 * ========= 00028 * 00029 * M (input) INTEGER 00030 * The order of the orthogonal matrix Q. M >= 0. 00031 * 00032 * N (input) INTEGER 00033 * The number of rows or columns of the matrix C; C is m-by-n if 00034 * Q is applied from the left, or n-by-m if Q is applied from 00035 * the right. N >= 0. 00036 * 00037 * K (input) INTEGER 00038 * The number of elementary reflectors whose product defines the 00039 * orthogonal matrix Q. M >= K >= 0. 00040 * 00041 * AF (input) DOUBLE PRECISION array, dimension (LDA,N) 00042 * Details of the QL factorization of an m-by-n matrix, as 00043 * returned by DGEQLF. See SGEQLF for further details. 00044 * 00045 * C (workspace) DOUBLE PRECISION array, dimension (LDA,N) 00046 * 00047 * CC (workspace) DOUBLE PRECISION array, dimension (LDA,N) 00048 * 00049 * Q (workspace) DOUBLE PRECISION array, dimension (LDA,M) 00050 * 00051 * LDA (input) INTEGER 00052 * The leading dimension of the arrays AF, C, CC, and Q. 00053 * 00054 * TAU (input) DOUBLE PRECISION array, dimension (min(M,N)) 00055 * The scalar factors of the elementary reflectors corresponding 00056 * to the QL factorization in AF. 00057 * 00058 * WORK (workspace) DOUBLE PRECISION array, dimension (LWORK) 00059 * 00060 * LWORK (input) INTEGER 00061 * The length of WORK. LWORK must be at least M, and should be 00062 * M*NB, where NB is the blocksize for this environment. 00063 * 00064 * RWORK (workspace) DOUBLE PRECISION array, dimension (M) 00065 * 00066 * RESULT (output) DOUBLE PRECISION array, dimension (4) 00067 * The test ratios compare two techniques for multiplying a 00068 * random matrix C by an m-by-m orthogonal matrix Q. 00069 * RESULT(1) = norm( Q*C - Q*C ) / ( M * norm(C) * EPS ) 00070 * RESULT(2) = norm( C*Q - C*Q ) / ( M * norm(C) * EPS ) 00071 * RESULT(3) = norm( Q'*C - Q'*C )/ ( M * norm(C) * EPS ) 00072 * RESULT(4) = norm( C*Q' - C*Q' )/ ( M * norm(C) * EPS ) 00073 * 00074 * ===================================================================== 00075 * 00076 * .. Parameters .. 00077 DOUBLE PRECISION ZERO, ONE 00078 PARAMETER ( ZERO = 0.0D0, ONE = 1.0D0 ) 00079 DOUBLE PRECISION ROGUE 00080 PARAMETER ( ROGUE = -1.0D+10 ) 00081 * .. 00082 * .. Local Scalars .. 00083 CHARACTER SIDE, TRANS 00084 INTEGER INFO, ISIDE, ITRANS, J, MC, MINMN, NC 00085 DOUBLE PRECISION CNORM, EPS, RESID 00086 * .. 00087 * .. External Functions .. 00088 LOGICAL LSAME 00089 DOUBLE PRECISION DLAMCH, DLANGE 00090 EXTERNAL LSAME, DLAMCH, DLANGE 00091 * .. 00092 * .. External Subroutines .. 00093 EXTERNAL DGEMM, DLACPY, DLARNV, DLASET, DORGQL, DORMQL 00094 * .. 00095 * .. Local Arrays .. 00096 INTEGER ISEED( 4 ) 00097 * .. 00098 * .. Intrinsic Functions .. 00099 INTRINSIC DBLE, MAX, MIN 00100 * .. 00101 * .. Scalars in Common .. 00102 CHARACTER*32 SRNAMT 00103 * .. 00104 * .. Common blocks .. 00105 COMMON / SRNAMC / SRNAMT 00106 * .. 00107 * .. Data statements .. 00108 DATA ISEED / 1988, 1989, 1990, 1991 / 00109 * .. 00110 * .. Executable Statements .. 00111 * 00112 EPS = DLAMCH( 'Epsilon' ) 00113 MINMN = MIN( M, N ) 00114 * 00115 * Quick return if possible 00116 * 00117 IF( MINMN.EQ.0 ) THEN 00118 RESULT( 1 ) = ZERO 00119 RESULT( 2 ) = ZERO 00120 RESULT( 3 ) = ZERO 00121 RESULT( 4 ) = ZERO 00122 RETURN 00123 END IF 00124 * 00125 * Copy the last k columns of the factorization to the array Q 00126 * 00127 CALL DLASET( 'Full', M, M, ROGUE, ROGUE, Q, LDA ) 00128 IF( K.GT.0 .AND. M.GT.K ) 00129 $ CALL DLACPY( 'Full', M-K, K, AF( 1, N-K+1 ), LDA, 00130 $ Q( 1, M-K+1 ), LDA ) 00131 IF( K.GT.1 ) 00132 $ CALL DLACPY( 'Upper', K-1, K-1, AF( M-K+1, N-K+2 ), LDA, 00133 $ Q( M-K+1, M-K+2 ), LDA ) 00134 * 00135 * Generate the m-by-m matrix Q 00136 * 00137 SRNAMT = 'DORGQL' 00138 CALL DORGQL( M, M, K, Q, LDA, TAU( MINMN-K+1 ), WORK, LWORK, 00139 $ INFO ) 00140 * 00141 DO 30 ISIDE = 1, 2 00142 IF( ISIDE.EQ.1 ) THEN 00143 SIDE = 'L' 00144 MC = M 00145 NC = N 00146 ELSE 00147 SIDE = 'R' 00148 MC = N 00149 NC = M 00150 END IF 00151 * 00152 * Generate MC by NC matrix C 00153 * 00154 DO 10 J = 1, NC 00155 CALL DLARNV( 2, ISEED, MC, C( 1, J ) ) 00156 10 CONTINUE 00157 CNORM = DLANGE( '1', MC, NC, C, LDA, RWORK ) 00158 IF( CNORM.EQ.0.0D0 ) 00159 $ CNORM = ONE 00160 * 00161 DO 20 ITRANS = 1, 2 00162 IF( ITRANS.EQ.1 ) THEN 00163 TRANS = 'N' 00164 ELSE 00165 TRANS = 'T' 00166 END IF 00167 * 00168 * Copy C 00169 * 00170 CALL DLACPY( 'Full', MC, NC, C, LDA, CC, LDA ) 00171 * 00172 * Apply Q or Q' to C 00173 * 00174 SRNAMT = 'DORMQL' 00175 IF( K.GT.0 ) 00176 $ CALL DORMQL( SIDE, TRANS, MC, NC, K, AF( 1, N-K+1 ), LDA, 00177 $ TAU( MINMN-K+1 ), CC, LDA, WORK, LWORK, 00178 $ INFO ) 00179 * 00180 * Form explicit product and subtract 00181 * 00182 IF( LSAME( SIDE, 'L' ) ) THEN 00183 CALL DGEMM( TRANS, 'No transpose', MC, NC, MC, -ONE, Q, 00184 $ LDA, C, LDA, ONE, CC, LDA ) 00185 ELSE 00186 CALL DGEMM( 'No transpose', TRANS, MC, NC, NC, -ONE, C, 00187 $ LDA, Q, LDA, ONE, CC, LDA ) 00188 END IF 00189 * 00190 * Compute error in the difference 00191 * 00192 RESID = DLANGE( '1', MC, NC, CC, LDA, RWORK ) 00193 RESULT( ( ISIDE-1 )*2+ITRANS ) = RESID / 00194 $ ( DBLE( MAX( 1, M ) )*CNORM*EPS ) 00195 * 00196 20 CONTINUE 00197 30 CONTINUE 00198 * 00199 RETURN 00200 * 00201 * End of DQLT03 00202 * 00203 END