001:       SUBROUTINE ZCGESV( N, NRHS, A, LDA, IPIV, B, LDB, X, LDX, WORK,
002:      +                   SWORK, RWORK, ITER, INFO )
003: *
004: *  -- LAPACK PROTOTYPE driver routine (version 3.2) --
005: *  -- LAPACK is a software package provided by Univ. of Tennessee,    --
006: *  -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
007: *     January 2007
008: *
009: *     ..
010: *     .. Scalar Arguments ..
011:       INTEGER            INFO, ITER, LDA, LDB, LDX, N, NRHS
012: *     ..
013: *     .. Array Arguments ..
014:       INTEGER            IPIV( * )
015:       DOUBLE PRECISION   RWORK( * )
016:       COMPLEX            SWORK( * )
017:       COMPLEX*16         A( LDA, * ), B( LDB, * ), WORK( N, * ),
018:      +                   X( LDX, * )
019: *     ..
020: *
021: *  Purpose
022: *  =======
023: *
024: *  ZCGESV computes the solution to a complex system of linear equations
025: *     A * X = B,
026: *  where A is an N-by-N matrix and X and B are N-by-NRHS matrices.
027: *
028: *  ZCGESV first attempts to factorize the matrix in COMPLEX and use this
029: *  factorization within an iterative refinement procedure to produce a
030: *  solution with COMPLEX*16 normwise backward error quality (see below).
031: *  If the approach fails the method switches to a COMPLEX*16
032: *  factorization and solve.
033: *
034: *  The iterative refinement is not going to be a winning strategy if
035: *  the ratio COMPLEX performance over COMPLEX*16 performance is too
036: *  small. A reasonable strategy should take the number of right-hand
037: *  sides and the size of the matrix into account. This might be done
038: *  with a call to ILAENV in the future. Up to now, we always try
039: *  iterative refinement.
040: *
041: *  The iterative refinement process is stopped if
042: *      ITER > ITERMAX
043: *  or for all the RHS we have:
044: *      RNRM < SQRT(N)*XNRM*ANRM*EPS*BWDMAX
045: *  where
046: *      o ITER is the number of the current iteration in the iterative
047: *        refinement process
048: *      o RNRM is the infinity-norm of the residual
049: *      o XNRM is the infinity-norm of the solution
050: *      o ANRM is the infinity-operator-norm of the matrix A
051: *      o EPS is the machine epsilon returned by DLAMCH('Epsilon')
052: *  The value ITERMAX and BWDMAX are fixed to 30 and 1.0D+00
053: *  respectively.
054: *
055: *  Arguments
056: *  =========
057: *
058: *  N       (input) INTEGER
059: *          The number of linear equations, i.e., the order of the
060: *          matrix A.  N >= 0.
061: *
062: *  NRHS    (input) INTEGER
063: *          The number of right hand sides, i.e., the number of columns
064: *          of the matrix B.  NRHS >= 0.
065: *
066: *  A       (input or input/ouptut) COMPLEX*16 array,
067: *          dimension (LDA,N)
068: *          On entry, the N-by-N coefficient matrix A.
069: *          On exit, if iterative refinement has been successfully used
070: *          (INFO.EQ.0 and ITER.GE.0, see description below), then A is
071: *          unchanged, if double precision factorization has been used
072: *          (INFO.EQ.0 and ITER.LT.0, see description below), then the
073: *          array A contains the factors L and U from the factorization
074: *          A = P*L*U; the unit diagonal elements of L are not stored.
075: *
076: *  LDA     (input) INTEGER
077: *          The leading dimension of the array A.  LDA >= max(1,N).
078: *
079: *  IPIV    (output) INTEGER array, dimension (N)
080: *          The pivot indices that define the permutation matrix P;
081: *          row i of the matrix was interchanged with row IPIV(i).
082: *          Corresponds either to the single precision factorization
083: *          (if INFO.EQ.0 and ITER.GE.0) or the double precision
084: *          factorization (if INFO.EQ.0 and ITER.LT.0).
085: *
086: *  B       (input) COMPLEX*16 array, dimension (LDB,NRHS)
087: *          The N-by-NRHS right hand side matrix B.
088: *
089: *  LDB     (input) INTEGER
090: *          The leading dimension of the array B.  LDB >= max(1,N).
091: *
092: *  X       (output) COMPLEX*16 array, dimension (LDX,NRHS)
093: *          If INFO = 0, the N-by-NRHS solution matrix X.
094: *
095: *  LDX     (input) INTEGER
096: *          The leading dimension of the array X.  LDX >= max(1,N).
097: *
098: *  WORK    (workspace) COMPLEX*16 array, dimension (N*NRHS)
099: *          This array is used to hold the residual vectors.
100: *
101: *  SWORK   (workspace) COMPLEX array, dimension (N*(N+NRHS))
102: *          This array is used to use the single precision matrix and the
103: *          right-hand sides or solutions in single precision.
104: *
105: *  RWORK   (workspace) DOUBLE PRECISION array, dimension (N)
106: *
107: *  ITER    (output) INTEGER
108: *          < 0: iterative refinement has failed, COMPLEX*16
109: *               factorization has been performed
110: *               -1 : the routine fell back to full precision for
111: *                    implementation- or machine-specific reasons
112: *               -2 : narrowing the precision induced an overflow,
113: *                    the routine fell back to full precision
114: *               -3 : failure of CGETRF
115: *               -31: stop the iterative refinement after the 30th
116: *                    iterations
117: *          > 0: iterative refinement has been sucessfully used.
118: *               Returns the number of iterations
119: *
120: *  INFO    (output) INTEGER
121: *          = 0:  successful exit
122: *          < 0:  if INFO = -i, the i-th argument had an illegal value
123: *          > 0:  if INFO = i, U(i,i) computed in COMPLEX*16 is exactly
124: *                zero.  The factorization has been completed, but the
125: *                factor U is exactly singular, so the solution
126: *                could not be computed.
127: *
128: *  =========
129: *
130: *     .. Parameters ..
131:       LOGICAL            DOITREF
132:       PARAMETER          ( DOITREF = .TRUE. )
133: *
134:       INTEGER            ITERMAX
135:       PARAMETER          ( ITERMAX = 30 )
136: *
137:       DOUBLE PRECISION   BWDMAX
138:       PARAMETER          ( BWDMAX = 1.0E+00 )
139: *
140:       COMPLEX*16         NEGONE, ONE
141:       PARAMETER          ( NEGONE = ( -1.0D+00, 0.0D+00 ),
142:      +                   ONE = ( 1.0D+00, 0.0D+00 ) )
143: *
144: *     .. Local Scalars ..
145:       INTEGER            I, IITER, PTSA, PTSX
146:       DOUBLE PRECISION   ANRM, CTE, EPS, RNRM, XNRM
147:       COMPLEX*16         ZDUM
148: *
149: *     .. External Subroutines ..
150:       EXTERNAL           CGETRS, CGETRF, CLAG2Z, XERBLA, ZAXPY, ZGEMM,
151:      +                   ZLACPY, ZLAG2C
152: *     ..
153: *     .. External Functions ..
154:       INTEGER            IZAMAX
155:       DOUBLE PRECISION   DLAMCH, ZLANGE
156:       EXTERNAL           IZAMAX, DLAMCH, ZLANGE
157: *     ..
158: *     .. Intrinsic Functions ..
159:       INTRINSIC          ABS, DBLE, MAX, SQRT
160: *     ..
161: *     .. Statement Functions ..
162:       DOUBLE PRECISION   CABS1
163: *     ..
164: *     .. Statement Function definitions ..
165:       CABS1( ZDUM ) = ABS( DBLE( ZDUM ) ) + ABS( DIMAG( ZDUM ) )
166: *     ..
167: *     .. Executable Statements ..
168: *
169:       INFO = 0
170:       ITER = 0
171: *
172: *     Test the input parameters.
173: *
174:       IF( N.LT.0 ) THEN
175:          INFO = -1
176:       ELSE IF( NRHS.LT.0 ) THEN
177:          INFO = -2
178:       ELSE IF( LDA.LT.MAX( 1, N ) ) THEN
179:          INFO = -4
180:       ELSE IF( LDB.LT.MAX( 1, N ) ) THEN
181:          INFO = -7
182:       ELSE IF( LDX.LT.MAX( 1, N ) ) THEN
183:          INFO = -9
184:       END IF
185:       IF( INFO.NE.0 ) THEN
186:          CALL XERBLA( 'ZCGESV', -INFO )
187:          RETURN
188:       END IF
189: *
190: *     Quick return if (N.EQ.0).
191: *
192:       IF( N.EQ.0 )
193:      +   RETURN
194: *
195: *     Skip single precision iterative refinement if a priori slower
196: *     than double precision factorization.
197: *
198:       IF( .NOT.DOITREF ) THEN
199:          ITER = -1
200:          GO TO 40
201:       END IF
202: *
203: *     Compute some constants.
204: *
205:       ANRM = ZLANGE( 'I', N, N, A, LDA, RWORK )
206:       EPS = DLAMCH( 'Epsilon' )
207:       CTE = ANRM*EPS*SQRT( DBLE( N ) )*BWDMAX
208: *
209: *     Set the indices PTSA, PTSX for referencing SA and SX in SWORK.
210: *
211:       PTSA = 1
212:       PTSX = PTSA + N*N
213: *
214: *     Convert B from double precision to single precision and store the
215: *     result in SX.
216: *
217:       CALL ZLAG2C( N, NRHS, B, LDB, SWORK( PTSX ), N, INFO )
218: *
219:       IF( INFO.NE.0 ) THEN
220:          ITER = -2
221:          GO TO 40
222:       END IF
223: *
224: *     Convert A from double precision to single precision and store the
225: *     result in SA.
226: *
227:       CALL ZLAG2C( N, N, A, LDA, SWORK( PTSA ), N, INFO )
228: *
229:       IF( INFO.NE.0 ) THEN
230:          ITER = -2
231:          GO TO 40
232:       END IF
233: *
234: *     Compute the LU factorization of SA.
235: *
236:       CALL CGETRF( N, N, SWORK( PTSA ), N, IPIV, INFO )
237: *
238:       IF( INFO.NE.0 ) THEN
239:          ITER = -3
240:          GO TO 40
241:       END IF
242: *
243: *     Solve the system SA*SX = SB.
244: *
245:       CALL CGETRS( 'No transpose', N, NRHS, SWORK( PTSA ), N, IPIV,
246:      +             SWORK( PTSX ), N, INFO )
247: *
248: *     Convert SX back to double precision
249: *
250:       CALL CLAG2Z( N, NRHS, SWORK( PTSX ), N, X, LDX, INFO )
251: *
252: *     Compute R = B - AX (R is WORK).
253: *
254:       CALL ZLACPY( 'All', N, NRHS, B, LDB, WORK, N )
255: *
256:       CALL ZGEMM( 'No Transpose', 'No Transpose', N, NRHS, N, NEGONE, A,
257:      +            LDA, X, LDX, ONE, WORK, N )
258: *
259: *     Check whether the NRHS normwise backward errors satisfy the
260: *     stopping criterion. If yes, set ITER=0 and return.
261: *
262:       DO I = 1, NRHS
263:          XNRM = CABS1( X( IZAMAX( N, X( 1, I ), 1 ), I ) )
264:          RNRM = CABS1( WORK( IZAMAX( N, WORK( 1, I ), 1 ), I ) )
265:          IF( RNRM.GT.XNRM*CTE )
266:      +      GO TO 10
267:       END DO
268: *
269: *     If we are here, the NRHS normwise backward errors satisfy the
270: *     stopping criterion. We are good to exit.
271: *
272:       ITER = 0
273:       RETURN
274: *
275:    10 CONTINUE
276: *
277:       DO 30 IITER = 1, ITERMAX
278: *
279: *        Convert R (in WORK) from double precision to single precision
280: *        and store the result in SX.
281: *
282:          CALL ZLAG2C( N, NRHS, WORK, N, SWORK( PTSX ), N, INFO )
283: *
284:          IF( INFO.NE.0 ) THEN
285:             ITER = -2
286:             GO TO 40
287:          END IF
288: *
289: *        Solve the system SA*SX = SR.
290: *
291:          CALL CGETRS( 'No transpose', N, NRHS, SWORK( PTSA ), N, IPIV,
292:      +                SWORK( PTSX ), N, INFO )
293: *
294: *        Convert SX back to double precision and update the current
295: *        iterate.
296: *
297:          CALL CLAG2Z( N, NRHS, SWORK( PTSX ), N, WORK, N, INFO )
298: *
299:          DO I = 1, NRHS
300:             CALL ZAXPY( N, ONE, WORK( 1, I ), 1, X( 1, I ), 1 )
301:          END DO
302: *
303: *        Compute R = B - AX (R is WORK).
304: *
305:          CALL ZLACPY( 'All', N, NRHS, B, LDB, WORK, N )
306: *
307:          CALL ZGEMM( 'No Transpose', 'No Transpose', N, NRHS, N, NEGONE,
308:      +               A, LDA, X, LDX, ONE, WORK, N )
309: *
310: *        Check whether the NRHS normwise backward errors satisfy the
311: *        stopping criterion. If yes, set ITER=IITER>0 and return.
312: *
313:          DO I = 1, NRHS
314:             XNRM = CABS1( X( IZAMAX( N, X( 1, I ), 1 ), I ) )
315:             RNRM = CABS1( WORK( IZAMAX( N, WORK( 1, I ), 1 ), I ) )
316:             IF( RNRM.GT.XNRM*CTE )
317:      +         GO TO 20
318:          END DO
319: *
320: *        If we are here, the NRHS normwise backward errors satisfy the
321: *        stopping criterion, we are good to exit.
322: *
323:          ITER = IITER
324: *
325:          RETURN
326: *
327:    20    CONTINUE
328: *
329:    30 CONTINUE
330: *
331: *     If we are at this place of the code, this is because we have
332: *     performed ITER=ITERMAX iterations and never satisified the stopping
333: *     criterion, set up the ITER flag accordingly and follow up on double
334: *     precision routine.
335: *
336:       ITER = -ITERMAX - 1
337: *
338:    40 CONTINUE
339: *
340: *     Single-precision iterative refinement failed to converge to a
341: *     satisfactory solution, so we resort to double precision.
342: *
343:       CALL ZGETRF( N, N, A, LDA, IPIV, INFO )
344: *
345:       IF( INFO.NE.0 )
346:      +   RETURN
347: *
348:       CALL ZLACPY( 'All', N, NRHS, B, LDB, X, LDX )
349:       CALL ZGETRS( 'No transpose', N, NRHS, A, LDA, IPIV, X, LDX,
350:      +             INFO )
351: *
352:       RETURN
353: *
354: *     End of ZCGESV.
355: *
356:       END
357: