SUBROUTINE CLALSD( UPLO, SMLSIZ, N, NRHS, D, E, B, LDB, RCOND, $ RANK, WORK, RWORK, IWORK, INFO ) * * -- LAPACK routine (version 3.2.2) -- * -- LAPACK is a software package provided by Univ. of Tennessee, -- * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- * June 2010 * * .. Scalar Arguments .. CHARACTER UPLO INTEGER INFO, LDB, N, NRHS, RANK, SMLSIZ REAL RCOND * .. * .. Array Arguments .. INTEGER IWORK( * ) REAL D( * ), E( * ), RWORK( * ) COMPLEX B( LDB, * ), WORK( * ) * .. * * Purpose * ======= * * CLALSD uses the singular value decomposition of A to solve the least * squares problem of finding X to minimize the Euclidean norm of each * column of A*X-B, where A is N-by-N upper bidiagonal, and X and B * are N-by-NRHS. The solution X overwrites B. * * The singular values of A smaller than RCOND times the largest * singular value are treated as zero in solving the least squares * problem; in this case a minimum norm solution is returned. * The actual singular values are returned in D in ascending order. * * This code 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 XMP, Cray YMP, Cray C 90, or Cray 2. * It could conceivably fail on hexadecimal or decimal machines * without guard digits, but we know of none. * * Arguments * ========= * * UPLO (input) CHARACTER*1 * = 'U': D and E define an upper bidiagonal matrix. * = 'L': D and E define a lower bidiagonal matrix. * * SMLSIZ (input) INTEGER * The maximum size of the subproblems at the bottom of the * computation tree. * * N (input) INTEGER * The dimension of the bidiagonal matrix. N >= 0. * * NRHS (input) INTEGER * The number of columns of B. NRHS must be at least 1. * * D (input/output) REAL array, dimension (N) * On entry D contains the main diagonal of the bidiagonal * matrix. On exit, if INFO = 0, D contains its singular values. * * E (input/output) REAL array, dimension (N-1) * Contains the super-diagonal entries of the bidiagonal matrix. * On exit, E has been destroyed. * * B (input/output) COMPLEX array, dimension (LDB,NRHS) * On input, B contains the right hand sides of the least * squares problem. On output, B contains the solution X. * * LDB (input) INTEGER * The leading dimension of B in the calling subprogram. * LDB must be at least max(1,N). * * RCOND (input) REAL * The singular values of A less than or equal to RCOND times * the largest singular value are treated as zero in solving * the least squares problem. If RCOND is negative, * machine precision is used instead. * For example, if diag(S)*X=B were the least squares problem, * where diag(S) is a diagonal matrix of singular values, the * solution would be X(i) = B(i) / S(i) if S(i) is greater than * RCOND*max(S), and X(i) = 0 if S(i) is less than or equal to * RCOND*max(S). * * RANK (output) INTEGER * The number of singular values of A greater than RCOND times * the largest singular value. * * WORK (workspace) COMPLEX array, dimension (N * NRHS). * * RWORK (workspace) REAL array, dimension at least * (9*N + 2*N*SMLSIZ + 8*N*NLVL + 3*SMLSIZ*NRHS + * MAX( (SMLSIZ+1)**2, N*(1+NRHS) + 2*NRHS ), * where * NLVL = MAX( 0, INT( LOG_2( MIN( M,N )/(SMLSIZ+1) ) ) + 1 ) * * IWORK (workspace) INTEGER array, dimension (3*N*NLVL + 11*N). * * INFO (output) INTEGER * = 0: successful exit. * < 0: if INFO = -i, the i-th argument had an illegal value. * > 0: The algorithm failed to compute a singular value while * working on the submatrix lying in rows and columns * INFO/(N+1) through MOD(INFO,N+1). * * Further Details * =============== * * Based on contributions by * Ming Gu and Ren-Cang Li, Computer Science Division, University of * California at Berkeley, USA * Osni Marques, LBNL/NERSC, USA * * ===================================================================== * * .. Parameters .. REAL ZERO, ONE, TWO PARAMETER ( ZERO = 0.0E0, ONE = 1.0E0, TWO = 2.0E0 ) COMPLEX CZERO PARAMETER ( CZERO = ( 0.0E0, 0.0E0 ) ) * .. * .. Local Scalars .. INTEGER BX, BXST, C, DIFL, DIFR, GIVCOL, GIVNUM, $ GIVPTR, I, ICMPQ1, ICMPQ2, IRWB, IRWIB, IRWRB, $ IRWU, IRWVT, IRWWRK, IWK, J, JCOL, JIMAG, $ JREAL, JROW, K, NLVL, NM1, NRWORK, NSIZE, NSUB, $ PERM, POLES, S, SIZEI, SMLSZP, SQRE, ST, ST1, $ U, VT, Z REAL CS, EPS, ORGNRM, R, RCND, SN, TOL * .. * .. External Functions .. INTEGER ISAMAX REAL SLAMCH, SLANST EXTERNAL ISAMAX, SLAMCH, SLANST * .. * .. External Subroutines .. EXTERNAL CCOPY, CLACPY, CLALSA, CLASCL, CLASET, CSROT, $ SGEMM, SLARTG, SLASCL, SLASDA, SLASDQ, SLASET, $ SLASRT, XERBLA * .. * .. Intrinsic Functions .. INTRINSIC ABS, AIMAG, CMPLX, INT, LOG, REAL, SIGN * .. * .. Executable Statements .. * * Test the input parameters. * INFO = 0 * IF( N.LT.0 ) THEN INFO = -3 ELSE IF( NRHS.LT.1 ) THEN INFO = -4 ELSE IF( ( LDB.LT.1 ) .OR. ( LDB.LT.N ) ) THEN INFO = -8 END IF IF( INFO.NE.0 ) THEN CALL XERBLA( 'CLALSD', -INFO ) RETURN END IF * EPS = SLAMCH( 'Epsilon' ) * * Set up the tolerance. * IF( ( RCOND.LE.ZERO ) .OR. ( RCOND.GE.ONE ) ) THEN RCND = EPS ELSE RCND = RCOND END IF * RANK = 0 * * Quick return if possible. * IF( N.EQ.0 ) THEN RETURN ELSE IF( N.EQ.1 ) THEN IF( D( 1 ).EQ.ZERO ) THEN CALL CLASET( 'A', 1, NRHS, CZERO, CZERO, B, LDB ) ELSE RANK = 1 CALL CLASCL( 'G', 0, 0, D( 1 ), ONE, 1, NRHS, B, LDB, INFO ) D( 1 ) = ABS( D( 1 ) ) END IF RETURN END IF * * Rotate the matrix if it is lower bidiagonal. * IF( UPLO.EQ.'L' ) THEN DO 10 I = 1, N - 1 CALL SLARTG( D( I ), E( I ), CS, SN, R ) D( I ) = R E( I ) = SN*D( I+1 ) D( I+1 ) = CS*D( I+1 ) IF( NRHS.EQ.1 ) THEN CALL CSROT( 1, B( I, 1 ), 1, B( I+1, 1 ), 1, CS, SN ) ELSE RWORK( I*2-1 ) = CS RWORK( I*2 ) = SN END IF 10 CONTINUE IF( NRHS.GT.1 ) THEN DO 30 I = 1, NRHS DO 20 J = 1, N - 1 CS = RWORK( J*2-1 ) SN = RWORK( J*2 ) CALL CSROT( 1, B( J, I ), 1, B( J+1, I ), 1, CS, SN ) 20 CONTINUE 30 CONTINUE END IF END IF * * Scale. * NM1 = N - 1 ORGNRM = SLANST( 'M', N, D, E ) IF( ORGNRM.EQ.ZERO ) THEN CALL CLASET( 'A', N, NRHS, CZERO, CZERO, B, LDB ) RETURN END IF * CALL SLASCL( 'G', 0, 0, ORGNRM, ONE, N, 1, D, N, INFO ) CALL SLASCL( 'G', 0, 0, ORGNRM, ONE, NM1, 1, E, NM1, INFO ) * * If N is smaller than the minimum divide size SMLSIZ, then solve * the problem with another solver. * IF( N.LE.SMLSIZ ) THEN IRWU = 1 IRWVT = IRWU + N*N IRWWRK = IRWVT + N*N IRWRB = IRWWRK IRWIB = IRWRB + N*NRHS IRWB = IRWIB + N*NRHS CALL SLASET( 'A', N, N, ZERO, ONE, RWORK( IRWU ), N ) CALL SLASET( 'A', N, N, ZERO, ONE, RWORK( IRWVT ), N ) CALL SLASDQ( 'U', 0, N, N, N, 0, D, E, RWORK( IRWVT ), N, $ RWORK( IRWU ), N, RWORK( IRWWRK ), 1, $ RWORK( IRWWRK ), INFO ) IF( INFO.NE.0 ) THEN RETURN END IF * * In the real version, B is passed to SLASDQ and multiplied * internally by Q'. Here B is complex and that product is * computed below in two steps (real and imaginary parts). * J = IRWB - 1 DO 50 JCOL = 1, NRHS DO 40 JROW = 1, N J = J + 1 RWORK( J ) = REAL( B( JROW, JCOL ) ) 40 CONTINUE 50 CONTINUE CALL SGEMM( 'T', 'N', N, NRHS, N, ONE, RWORK( IRWU ), N, $ RWORK( IRWB ), N, ZERO, RWORK( IRWRB ), N ) J = IRWB - 1 DO 70 JCOL = 1, NRHS DO 60 JROW = 1, N J = J + 1 RWORK( J ) = AIMAG( B( JROW, JCOL ) ) 60 CONTINUE 70 CONTINUE CALL SGEMM( 'T', 'N', N, NRHS, N, ONE, RWORK( IRWU ), N, $ RWORK( IRWB ), N, ZERO, RWORK( IRWIB ), N ) JREAL = IRWRB - 1 JIMAG = IRWIB - 1 DO 90 JCOL = 1, NRHS DO 80 JROW = 1, N JREAL = JREAL + 1 JIMAG = JIMAG + 1 B( JROW, JCOL ) = CMPLX( RWORK( JREAL ), RWORK( JIMAG ) ) 80 CONTINUE 90 CONTINUE * TOL = RCND*ABS( D( ISAMAX( N, D, 1 ) ) ) DO 100 I = 1, N IF( D( I ).LE.TOL ) THEN CALL CLASET( 'A', 1, NRHS, CZERO, CZERO, B( I, 1 ), LDB ) ELSE CALL CLASCL( 'G', 0, 0, D( I ), ONE, 1, NRHS, B( I, 1 ), $ LDB, INFO ) RANK = RANK + 1 END IF 100 CONTINUE * * Since B is complex, the following call to SGEMM is performed * in two steps (real and imaginary parts). That is for V * B * (in the real version of the code V' is stored in WORK). * * CALL SGEMM( 'T', 'N', N, NRHS, N, ONE, WORK, N, B, LDB, ZERO, * $ WORK( NWORK ), N ) * J = IRWB - 1 DO 120 JCOL = 1, NRHS DO 110 JROW = 1, N J = J + 1 RWORK( J ) = REAL( B( JROW, JCOL ) ) 110 CONTINUE 120 CONTINUE CALL SGEMM( 'T', 'N', N, NRHS, N, ONE, RWORK( IRWVT ), N, $ RWORK( IRWB ), N, ZERO, RWORK( IRWRB ), N ) J = IRWB - 1 DO 140 JCOL = 1, NRHS DO 130 JROW = 1, N J = J + 1 RWORK( J ) = AIMAG( B( JROW, JCOL ) ) 130 CONTINUE 140 CONTINUE CALL SGEMM( 'T', 'N', N, NRHS, N, ONE, RWORK( IRWVT ), N, $ RWORK( IRWB ), N, ZERO, RWORK( IRWIB ), N ) JREAL = IRWRB - 1 JIMAG = IRWIB - 1 DO 160 JCOL = 1, NRHS DO 150 JROW = 1, N JREAL = JREAL + 1 JIMAG = JIMAG + 1 B( JROW, JCOL ) = CMPLX( RWORK( JREAL ), RWORK( JIMAG ) ) 150 CONTINUE 160 CONTINUE * * Unscale. * CALL SLASCL( 'G', 0, 0, ONE, ORGNRM, N, 1, D, N, INFO ) CALL SLASRT( 'D', N, D, INFO ) CALL CLASCL( 'G', 0, 0, ORGNRM, ONE, N, NRHS, B, LDB, INFO ) * RETURN END IF * * Book-keeping and setting up some constants. * NLVL = INT( LOG( REAL( N ) / REAL( SMLSIZ+1 ) ) / LOG( TWO ) ) + 1 * SMLSZP = SMLSIZ + 1 * U = 1 VT = 1 + SMLSIZ*N DIFL = VT + SMLSZP*N DIFR = DIFL + NLVL*N Z = DIFR + NLVL*N*2 C = Z + NLVL*N S = C + N POLES = S + N GIVNUM = POLES + 2*NLVL*N NRWORK = GIVNUM + 2*NLVL*N BX = 1 * IRWRB = NRWORK IRWIB = IRWRB + SMLSIZ*NRHS IRWB = IRWIB + SMLSIZ*NRHS * SIZEI = 1 + N K = SIZEI + N GIVPTR = K + N PERM = GIVPTR + N GIVCOL = PERM + NLVL*N IWK = GIVCOL + NLVL*N*2 * ST = 1 SQRE = 0 ICMPQ1 = 1 ICMPQ2 = 0 NSUB = 0 * DO 170 I = 1, N IF( ABS( D( I ) ).LT.EPS ) THEN D( I ) = SIGN( EPS, D( I ) ) END IF 170 CONTINUE * DO 240 I = 1, NM1 IF( ( ABS( E( I ) ).LT.EPS ) .OR. ( I.EQ.NM1 ) ) THEN NSUB = NSUB + 1 IWORK( NSUB ) = ST * * Subproblem found. First determine its size and then * apply divide and conquer on it. * IF( I.LT.NM1 ) THEN * * A subproblem with E(I) small for I < NM1. * NSIZE = I - ST + 1 IWORK( SIZEI+NSUB-1 ) = NSIZE ELSE IF( ABS( E( I ) ).GE.EPS ) THEN * * A subproblem with E(NM1) not too small but I = NM1. * NSIZE = N - ST + 1 IWORK( SIZEI+NSUB-1 ) = NSIZE ELSE * * A subproblem with E(NM1) small. This implies an * 1-by-1 subproblem at D(N), which is not solved * explicitly. * NSIZE = I - ST + 1 IWORK( SIZEI+NSUB-1 ) = NSIZE NSUB = NSUB + 1 IWORK( NSUB ) = N IWORK( SIZEI+NSUB-1 ) = 1 CALL CCOPY( NRHS, B( N, 1 ), LDB, WORK( BX+NM1 ), N ) END IF ST1 = ST - 1 IF( NSIZE.EQ.1 ) THEN * * This is a 1-by-1 subproblem and is not solved * explicitly. * CALL CCOPY( NRHS, B( ST, 1 ), LDB, WORK( BX+ST1 ), N ) ELSE IF( NSIZE.LE.SMLSIZ ) THEN * * This is a small subproblem and is solved by SLASDQ. * CALL SLASET( 'A', NSIZE, NSIZE, ZERO, ONE, $ RWORK( VT+ST1 ), N ) CALL SLASET( 'A', NSIZE, NSIZE, ZERO, ONE, $ RWORK( U+ST1 ), N ) CALL SLASDQ( 'U', 0, NSIZE, NSIZE, NSIZE, 0, D( ST ), $ E( ST ), RWORK( VT+ST1 ), N, RWORK( U+ST1 ), $ N, RWORK( NRWORK ), 1, RWORK( NRWORK ), $ INFO ) IF( INFO.NE.0 ) THEN RETURN END IF * * In the real version, B is passed to SLASDQ and multiplied * internally by Q'. Here B is complex and that product is * computed below in two steps (real and imaginary parts). * J = IRWB - 1 DO 190 JCOL = 1, NRHS DO 180 JROW = ST, ST + NSIZE - 1 J = J + 1 RWORK( J ) = REAL( B( JROW, JCOL ) ) 180 CONTINUE 190 CONTINUE CALL SGEMM( 'T', 'N', NSIZE, NRHS, NSIZE, ONE, $ RWORK( U+ST1 ), N, RWORK( IRWB ), NSIZE, $ ZERO, RWORK( IRWRB ), NSIZE ) J = IRWB - 1 DO 210 JCOL = 1, NRHS DO 200 JROW = ST, ST + NSIZE - 1 J = J + 1 RWORK( J ) = AIMAG( B( JROW, JCOL ) ) 200 CONTINUE 210 CONTINUE CALL SGEMM( 'T', 'N', NSIZE, NRHS, NSIZE, ONE, $ RWORK( U+ST1 ), N, RWORK( IRWB ), NSIZE, $ ZERO, RWORK( IRWIB ), NSIZE ) JREAL = IRWRB - 1 JIMAG = IRWIB - 1 DO 230 JCOL = 1, NRHS DO 220 JROW = ST, ST + NSIZE - 1 JREAL = JREAL + 1 JIMAG = JIMAG + 1 B( JROW, JCOL ) = CMPLX( RWORK( JREAL ), $ RWORK( JIMAG ) ) 220 CONTINUE 230 CONTINUE * CALL CLACPY( 'A', NSIZE, NRHS, B( ST, 1 ), LDB, $ WORK( BX+ST1 ), N ) ELSE * * A large problem. Solve it using divide and conquer. * CALL SLASDA( ICMPQ1, SMLSIZ, NSIZE, SQRE, D( ST ), $ E( ST ), RWORK( U+ST1 ), N, RWORK( VT+ST1 ), $ IWORK( K+ST1 ), RWORK( DIFL+ST1 ), $ RWORK( DIFR+ST1 ), RWORK( Z+ST1 ), $ RWORK( POLES+ST1 ), IWORK( GIVPTR+ST1 ), $ IWORK( GIVCOL+ST1 ), N, IWORK( PERM+ST1 ), $ RWORK( GIVNUM+ST1 ), RWORK( C+ST1 ), $ RWORK( S+ST1 ), RWORK( NRWORK ), $ IWORK( IWK ), INFO ) IF( INFO.NE.0 ) THEN RETURN END IF BXST = BX + ST1 CALL CLALSA( ICMPQ2, SMLSIZ, NSIZE, NRHS, B( ST, 1 ), $ LDB, WORK( BXST ), N, RWORK( U+ST1 ), N, $ RWORK( VT+ST1 ), IWORK( K+ST1 ), $ RWORK( DIFL+ST1 ), RWORK( DIFR+ST1 ), $ RWORK( Z+ST1 ), RWORK( POLES+ST1 ), $ IWORK( GIVPTR+ST1 ), IWORK( GIVCOL+ST1 ), N, $ IWORK( PERM+ST1 ), RWORK( GIVNUM+ST1 ), $ RWORK( C+ST1 ), RWORK( S+ST1 ), $ RWORK( NRWORK ), IWORK( IWK ), INFO ) IF( INFO.NE.0 ) THEN RETURN END IF END IF ST = I + 1 END IF 240 CONTINUE * * Apply the singular values and treat the tiny ones as zero. * TOL = RCND*ABS( D( ISAMAX( N, D, 1 ) ) ) * DO 250 I = 1, N * * Some of the elements in D can be negative because 1-by-1 * subproblems were not solved explicitly. * IF( ABS( D( I ) ).LE.TOL ) THEN CALL CLASET( 'A', 1, NRHS, CZERO, CZERO, WORK( BX+I-1 ), N ) ELSE RANK = RANK + 1 CALL CLASCL( 'G', 0, 0, D( I ), ONE, 1, NRHS, $ WORK( BX+I-1 ), N, INFO ) END IF D( I ) = ABS( D( I ) ) 250 CONTINUE * * Now apply back the right singular vectors. * ICMPQ2 = 1 DO 320 I = 1, NSUB ST = IWORK( I ) ST1 = ST - 1 NSIZE = IWORK( SIZEI+I-1 ) BXST = BX + ST1 IF( NSIZE.EQ.1 ) THEN CALL CCOPY( NRHS, WORK( BXST ), N, B( ST, 1 ), LDB ) ELSE IF( NSIZE.LE.SMLSIZ ) THEN * * Since B and BX are complex, the following call to SGEMM * is performed in two steps (real and imaginary parts). * * CALL SGEMM( 'T', 'N', NSIZE, NRHS, NSIZE, ONE, * $ RWORK( VT+ST1 ), N, RWORK( BXST ), N, ZERO, * $ B( ST, 1 ), LDB ) * J = BXST - N - 1 JREAL = IRWB - 1 DO 270 JCOL = 1, NRHS J = J + N DO 260 JROW = 1, NSIZE JREAL = JREAL + 1 RWORK( JREAL ) = REAL( WORK( J+JROW ) ) 260 CONTINUE 270 CONTINUE CALL SGEMM( 'T', 'N', NSIZE, NRHS, NSIZE, ONE, $ RWORK( VT+ST1 ), N, RWORK( IRWB ), NSIZE, ZERO, $ RWORK( IRWRB ), NSIZE ) J = BXST - N - 1 JIMAG = IRWB - 1 DO 290 JCOL = 1, NRHS J = J + N DO 280 JROW = 1, NSIZE JIMAG = JIMAG + 1 RWORK( JIMAG ) = AIMAG( WORK( J+JROW ) ) 280 CONTINUE 290 CONTINUE CALL SGEMM( 'T', 'N', NSIZE, NRHS, NSIZE, ONE, $ RWORK( VT+ST1 ), N, RWORK( IRWB ), NSIZE, ZERO, $ RWORK( IRWIB ), NSIZE ) JREAL = IRWRB - 1 JIMAG = IRWIB - 1 DO 310 JCOL = 1, NRHS DO 300 JROW = ST, ST + NSIZE - 1 JREAL = JREAL + 1 JIMAG = JIMAG + 1 B( JROW, JCOL ) = CMPLX( RWORK( JREAL ), $ RWORK( JIMAG ) ) 300 CONTINUE 310 CONTINUE ELSE CALL CLALSA( ICMPQ2, SMLSIZ, NSIZE, NRHS, WORK( BXST ), N, $ B( ST, 1 ), LDB, RWORK( U+ST1 ), N, $ RWORK( VT+ST1 ), IWORK( K+ST1 ), $ RWORK( DIFL+ST1 ), RWORK( DIFR+ST1 ), $ RWORK( Z+ST1 ), RWORK( POLES+ST1 ), $ IWORK( GIVPTR+ST1 ), IWORK( GIVCOL+ST1 ), N, $ IWORK( PERM+ST1 ), RWORK( GIVNUM+ST1 ), $ RWORK( C+ST1 ), RWORK( S+ST1 ), $ RWORK( NRWORK ), IWORK( IWK ), INFO ) IF( INFO.NE.0 ) THEN RETURN END IF END IF 320 CONTINUE * * Unscale and sort the singular values. * CALL SLASCL( 'G', 0, 0, ONE, ORGNRM, N, 1, D, N, INFO ) CALL SLASRT( 'D', N, D, INFO ) CALL CLASCL( 'G', 0, 0, ORGNRM, ONE, N, NRHS, B, LDB, INFO ) * RETURN * * End of CLALSD * END