1: SUBROUTINE DLATBS( UPLO, TRANS, DIAG, NORMIN, N, KD, AB, LDAB, X,
2: $ SCALE, CNORM, INFO )
3: *
4: * -- LAPACK auxiliary routine (version 3.2) --
5: * -- LAPACK is a software package provided by Univ. of Tennessee, --
6: * -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..--
7: * November 2006
8: *
9: * .. Scalar Arguments ..
10: CHARACTER DIAG, NORMIN, TRANS, UPLO
11: INTEGER INFO, KD, LDAB, N
12: DOUBLE PRECISION SCALE
13: * ..
14: * .. Array Arguments ..
15: DOUBLE PRECISION AB( LDAB, * ), CNORM( * ), X( * )
16: * ..
17: *
18: * Purpose
19: * =======
20: *
21: * DLATBS solves one of the triangular systems
22: *
23: * A *x = s*b or A'*x = s*b
24: *
25: * with scaling to prevent overflow, where A is an upper or lower
26: * triangular band matrix. Here A' denotes the transpose of A, x and b
27: * are n-element vectors, and s is a scaling factor, usually less than
28: * or equal to 1, chosen so that the components of x will be less than
29: * the overflow threshold. If the unscaled problem will not cause
30: * overflow, the Level 2 BLAS routine DTBSV is called. If the matrix A
31: * is singular (A(j,j) = 0 for some j), then s is set to 0 and a
32: * non-trivial solution to A*x = 0 is returned.
33: *
34: * Arguments
35: * =========
36: *
37: * UPLO (input) CHARACTER*1
38: * Specifies whether the matrix A is upper or lower triangular.
39: * = 'U': Upper triangular
40: * = 'L': Lower triangular
41: *
42: * TRANS (input) CHARACTER*1
43: * Specifies the operation applied to A.
44: * = 'N': Solve A * x = s*b (No transpose)
45: * = 'T': Solve A'* x = s*b (Transpose)
46: * = 'C': Solve A'* x = s*b (Conjugate transpose = Transpose)
47: *
48: * DIAG (input) CHARACTER*1
49: * Specifies whether or not the matrix A is unit triangular.
50: * = 'N': Non-unit triangular
51: * = 'U': Unit triangular
52: *
53: * NORMIN (input) CHARACTER*1
54: * Specifies whether CNORM has been set or not.
55: * = 'Y': CNORM contains the column norms on entry
56: * = 'N': CNORM is not set on entry. On exit, the norms will
57: * be computed and stored in CNORM.
58: *
59: * N (input) INTEGER
60: * The order of the matrix A. N >= 0.
61: *
62: * KD (input) INTEGER
63: * The number of subdiagonals or superdiagonals in the
64: * triangular matrix A. KD >= 0.
65: *
66: * AB (input) DOUBLE PRECISION array, dimension (LDAB,N)
67: * The upper or lower triangular band matrix A, stored in the
68: * first KD+1 rows of the array. The j-th column of A is stored
69: * in the j-th column of the array AB as follows:
70: * if UPLO = 'U', AB(kd+1+i-j,j) = A(i,j) for max(1,j-kd)<=i<=j;
71: * if UPLO = 'L', AB(1+i-j,j) = A(i,j) for j<=i<=min(n,j+kd).
72: *
73: * LDAB (input) INTEGER
74: * The leading dimension of the array AB. LDAB >= KD+1.
75: *
76: * X (input/output) DOUBLE PRECISION array, dimension (N)
77: * On entry, the right hand side b of the triangular system.
78: * On exit, X is overwritten by the solution vector x.
79: *
80: * SCALE (output) DOUBLE PRECISION
81: * The scaling factor s for the triangular system
82: * A * x = s*b or A'* x = s*b.
83: * If SCALE = 0, the matrix A is singular or badly scaled, and
84: * the vector x is an exact or approximate solution to A*x = 0.
85: *
86: * CNORM (input or output) DOUBLE PRECISION array, dimension (N)
87: *
88: * If NORMIN = 'Y', CNORM is an input argument and CNORM(j)
89: * contains the norm of the off-diagonal part of the j-th column
90: * of A. If TRANS = 'N', CNORM(j) must be greater than or equal
91: * to the infinity-norm, and if TRANS = 'T' or 'C', CNORM(j)
92: * must be greater than or equal to the 1-norm.
93: *
94: * If NORMIN = 'N', CNORM is an output argument and CNORM(j)
95: * returns the 1-norm of the offdiagonal part of the j-th column
96: * of A.
97: *
98: * INFO (output) INTEGER
99: * = 0: successful exit
100: * < 0: if INFO = -k, the k-th argument had an illegal value
101: *
102: * Further Details
103: * ======= =======
104: *
105: * A rough bound on x is computed; if that is less than overflow, DTBSV
106: * is called, otherwise, specific code is used which checks for possible
107: * overflow or divide-by-zero at every operation.
108: *
109: * A columnwise scheme is used for solving A*x = b. The basic algorithm
110: * if A is lower triangular is
111: *
112: * x[1:n] := b[1:n]
113: * for j = 1, ..., n
114: * x(j) := x(j) / A(j,j)
115: * x[j+1:n] := x[j+1:n] - x(j) * A[j+1:n,j]
116: * end
117: *
118: * Define bounds on the components of x after j iterations of the loop:
119: * M(j) = bound on x[1:j]
120: * G(j) = bound on x[j+1:n]
121: * Initially, let M(0) = 0 and G(0) = max{x(i), i=1,...,n}.
122: *
123: * Then for iteration j+1 we have
124: * M(j+1) <= G(j) / | A(j+1,j+1) |
125: * G(j+1) <= G(j) + M(j+1) * | A[j+2:n,j+1] |
126: * <= G(j) ( 1 + CNORM(j+1) / | A(j+1,j+1) | )
127: *
128: * where CNORM(j+1) is greater than or equal to the infinity-norm of
129: * column j+1 of A, not counting the diagonal. Hence
130: *
131: * G(j) <= G(0) product ( 1 + CNORM(i) / | A(i,i) | )
132: * 1<=i<=j
133: * and
134: *
135: * |x(j)| <= ( G(0) / |A(j,j)| ) product ( 1 + CNORM(i) / |A(i,i)| )
136: * 1<=i< j
137: *
138: * Since |x(j)| <= M(j), we use the Level 2 BLAS routine DTBSV if the
139: * reciprocal of the largest M(j), j=1,..,n, is larger than
140: * max(underflow, 1/overflow).
141: *
142: * The bound on x(j) is also used to determine when a step in the
143: * columnwise method can be performed without fear of overflow. If
144: * the computed bound is greater than a large constant, x is scaled to
145: * prevent overflow, but if the bound overflows, x is set to 0, x(j) to
146: * 1, and scale to 0, and a non-trivial solution to A*x = 0 is found.
147: *
148: * Similarly, a row-wise scheme is used to solve A'*x = b. The basic
149: * algorithm for A upper triangular is
150: *
151: * for j = 1, ..., n
152: * x(j) := ( b(j) - A[1:j-1,j]' * x[1:j-1] ) / A(j,j)
153: * end
154: *
155: * We simultaneously compute two bounds
156: * G(j) = bound on ( b(i) - A[1:i-1,i]' * x[1:i-1] ), 1<=i<=j
157: * M(j) = bound on x(i), 1<=i<=j
158: *
159: * The initial values are G(0) = 0, M(0) = max{b(i), i=1,..,n}, and we
160: * add the constraint G(j) >= G(j-1) and M(j) >= M(j-1) for j >= 1.
161: * Then the bound on x(j) is
162: *
163: * M(j) <= M(j-1) * ( 1 + CNORM(j) ) / | A(j,j) |
164: *
165: * <= M(0) * product ( ( 1 + CNORM(i) ) / |A(i,i)| )
166: * 1<=i<=j
167: *
168: * and we can safely call DTBSV if 1/M(n) and 1/G(n) are both greater
169: * than max(underflow, 1/overflow).
170: *
171: * =====================================================================
172: *
173: * .. Parameters ..
174: DOUBLE PRECISION ZERO, HALF, ONE
175: PARAMETER ( ZERO = 0.0D+0, HALF = 0.5D+0, ONE = 1.0D+0 )
176: * ..
177: * .. Local Scalars ..
178: LOGICAL NOTRAN, NOUNIT, UPPER
179: INTEGER I, IMAX, J, JFIRST, JINC, JLAST, JLEN, MAIND
180: DOUBLE PRECISION BIGNUM, GROW, REC, SMLNUM, SUMJ, TJJ, TJJS,
181: $ TMAX, TSCAL, USCAL, XBND, XJ, XMAX
182: * ..
183: * .. External Functions ..
184: LOGICAL LSAME
185: INTEGER IDAMAX
186: DOUBLE PRECISION DASUM, DDOT, DLAMCH
187: EXTERNAL LSAME, IDAMAX, DASUM, DDOT, DLAMCH
188: * ..
189: * .. External Subroutines ..
190: EXTERNAL DAXPY, DSCAL, DTBSV, XERBLA
191: * ..
192: * .. Intrinsic Functions ..
193: INTRINSIC ABS, MAX, MIN
194: * ..
195: * .. Executable Statements ..
196: *
197: INFO = 0
198: UPPER = LSAME( UPLO, 'U' )
199: NOTRAN = LSAME( TRANS, 'N' )
200: NOUNIT = LSAME( DIAG, 'N' )
201: *
202: * Test the input parameters.
203: *
204: IF( .NOT.UPPER .AND. .NOT.LSAME( UPLO, 'L' ) ) THEN
205: INFO = -1
206: ELSE IF( .NOT.NOTRAN .AND. .NOT.LSAME( TRANS, 'T' ) .AND. .NOT.
207: $ LSAME( TRANS, 'C' ) ) THEN
208: INFO = -2
209: ELSE IF( .NOT.NOUNIT .AND. .NOT.LSAME( DIAG, 'U' ) ) THEN
210: INFO = -3
211: ELSE IF( .NOT.LSAME( NORMIN, 'Y' ) .AND. .NOT.
212: $ LSAME( NORMIN, 'N' ) ) THEN
213: INFO = -4
214: ELSE IF( N.LT.0 ) THEN
215: INFO = -5
216: ELSE IF( KD.LT.0 ) THEN
217: INFO = -6
218: ELSE IF( LDAB.LT.KD+1 ) THEN
219: INFO = -8
220: END IF
221: IF( INFO.NE.0 ) THEN
222: CALL XERBLA( 'DLATBS', -INFO )
223: RETURN
224: END IF
225: *
226: * Quick return if possible
227: *
228: IF( N.EQ.0 )
229: $ RETURN
230: *
231: * Determine machine dependent parameters to control overflow.
232: *
233: SMLNUM = DLAMCH( 'Safe minimum' ) / DLAMCH( 'Precision' )
234: BIGNUM = ONE / SMLNUM
235: SCALE = ONE
236: *
237: IF( LSAME( NORMIN, 'N' ) ) THEN
238: *
239: * Compute the 1-norm of each column, not including the diagonal.
240: *
241: IF( UPPER ) THEN
242: *
243: * A is upper triangular.
244: *
245: DO 10 J = 1, N
246: JLEN = MIN( KD, J-1 )
247: CNORM( J ) = DASUM( JLEN, AB( KD+1-JLEN, J ), 1 )
248: 10 CONTINUE
249: ELSE
250: *
251: * A is lower triangular.
252: *
253: DO 20 J = 1, N
254: JLEN = MIN( KD, N-J )
255: IF( JLEN.GT.0 ) THEN
256: CNORM( J ) = DASUM( JLEN, AB( 2, J ), 1 )
257: ELSE
258: CNORM( J ) = ZERO
259: END IF
260: 20 CONTINUE
261: END IF
262: END IF
263: *
264: * Scale the column norms by TSCAL if the maximum element in CNORM is
265: * greater than BIGNUM.
266: *
267: IMAX = IDAMAX( N, CNORM, 1 )
268: TMAX = CNORM( IMAX )
269: IF( TMAX.LE.BIGNUM ) THEN
270: TSCAL = ONE
271: ELSE
272: TSCAL = ONE / ( SMLNUM*TMAX )
273: CALL DSCAL( N, TSCAL, CNORM, 1 )
274: END IF
275: *
276: * Compute a bound on the computed solution vector to see if the
277: * Level 2 BLAS routine DTBSV can be used.
278: *
279: J = IDAMAX( N, X, 1 )
280: XMAX = ABS( X( J ) )
281: XBND = XMAX
282: IF( NOTRAN ) THEN
283: *
284: * Compute the growth in A * x = b.
285: *
286: IF( UPPER ) THEN
287: JFIRST = N
288: JLAST = 1
289: JINC = -1
290: MAIND = KD + 1
291: ELSE
292: JFIRST = 1
293: JLAST = N
294: JINC = 1
295: MAIND = 1
296: END IF
297: *
298: IF( TSCAL.NE.ONE ) THEN
299: GROW = ZERO
300: GO TO 50
301: END IF
302: *
303: IF( NOUNIT ) THEN
304: *
305: * A is non-unit triangular.
306: *
307: * Compute GROW = 1/G(j) and XBND = 1/M(j).
308: * Initially, G(0) = max{x(i), i=1,...,n}.
309: *
310: GROW = ONE / MAX( XBND, SMLNUM )
311: XBND = GROW
312: DO 30 J = JFIRST, JLAST, JINC
313: *
314: * Exit the loop if the growth factor is too small.
315: *
316: IF( GROW.LE.SMLNUM )
317: $ GO TO 50
318: *
319: * M(j) = G(j-1) / abs(A(j,j))
320: *
321: TJJ = ABS( AB( MAIND, J ) )
322: XBND = MIN( XBND, MIN( ONE, TJJ )*GROW )
323: IF( TJJ+CNORM( J ).GE.SMLNUM ) THEN
324: *
325: * G(j) = G(j-1)*( 1 + CNORM(j) / abs(A(j,j)) )
326: *
327: GROW = GROW*( TJJ / ( TJJ+CNORM( J ) ) )
328: ELSE
329: *
330: * G(j) could overflow, set GROW to 0.
331: *
332: GROW = ZERO
333: END IF
334: 30 CONTINUE
335: GROW = XBND
336: ELSE
337: *
338: * A is unit triangular.
339: *
340: * Compute GROW = 1/G(j), where G(0) = max{x(i), i=1,...,n}.
341: *
342: GROW = MIN( ONE, ONE / MAX( XBND, SMLNUM ) )
343: DO 40 J = JFIRST, JLAST, JINC
344: *
345: * Exit the loop if the growth factor is too small.
346: *
347: IF( GROW.LE.SMLNUM )
348: $ GO TO 50
349: *
350: * G(j) = G(j-1)*( 1 + CNORM(j) )
351: *
352: GROW = GROW*( ONE / ( ONE+CNORM( J ) ) )
353: 40 CONTINUE
354: END IF
355: 50 CONTINUE
356: *
357: ELSE
358: *
359: * Compute the growth in A' * x = b.
360: *
361: IF( UPPER ) THEN
362: JFIRST = 1
363: JLAST = N
364: JINC = 1
365: MAIND = KD + 1
366: ELSE
367: JFIRST = N
368: JLAST = 1
369: JINC = -1
370: MAIND = 1
371: END IF
372: *
373: IF( TSCAL.NE.ONE ) THEN
374: GROW = ZERO
375: GO TO 80
376: END IF
377: *
378: IF( NOUNIT ) THEN
379: *
380: * A is non-unit triangular.
381: *
382: * Compute GROW = 1/G(j) and XBND = 1/M(j).
383: * Initially, M(0) = max{x(i), i=1,...,n}.
384: *
385: GROW = ONE / MAX( XBND, SMLNUM )
386: XBND = GROW
387: DO 60 J = JFIRST, JLAST, JINC
388: *
389: * Exit the loop if the growth factor is too small.
390: *
391: IF( GROW.LE.SMLNUM )
392: $ GO TO 80
393: *
394: * G(j) = max( G(j-1), M(j-1)*( 1 + CNORM(j) ) )
395: *
396: XJ = ONE + CNORM( J )
397: GROW = MIN( GROW, XBND / XJ )
398: *
399: * M(j) = M(j-1)*( 1 + CNORM(j) ) / abs(A(j,j))
400: *
401: TJJ = ABS( AB( MAIND, J ) )
402: IF( XJ.GT.TJJ )
403: $ XBND = XBND*( TJJ / XJ )
404: 60 CONTINUE
405: GROW = MIN( GROW, XBND )
406: ELSE
407: *
408: * A is unit triangular.
409: *
410: * Compute GROW = 1/G(j), where G(0) = max{x(i), i=1,...,n}.
411: *
412: GROW = MIN( ONE, ONE / MAX( XBND, SMLNUM ) )
413: DO 70 J = JFIRST, JLAST, JINC
414: *
415: * Exit the loop if the growth factor is too small.
416: *
417: IF( GROW.LE.SMLNUM )
418: $ GO TO 80
419: *
420: * G(j) = ( 1 + CNORM(j) )*G(j-1)
421: *
422: XJ = ONE + CNORM( J )
423: GROW = GROW / XJ
424: 70 CONTINUE
425: END IF
426: 80 CONTINUE
427: END IF
428: *
429: IF( ( GROW*TSCAL ).GT.SMLNUM ) THEN
430: *
431: * Use the Level 2 BLAS solve if the reciprocal of the bound on
432: * elements of X is not too small.
433: *
434: CALL DTBSV( UPLO, TRANS, DIAG, N, KD, AB, LDAB, X, 1 )
435: ELSE
436: *
437: * Use a Level 1 BLAS solve, scaling intermediate results.
438: *
439: IF( XMAX.GT.BIGNUM ) THEN
440: *
441: * Scale X so that its components are less than or equal to
442: * BIGNUM in absolute value.
443: *
444: SCALE = BIGNUM / XMAX
445: CALL DSCAL( N, SCALE, X, 1 )
446: XMAX = BIGNUM
447: END IF
448: *
449: IF( NOTRAN ) THEN
450: *
451: * Solve A * x = b
452: *
453: DO 110 J = JFIRST, JLAST, JINC
454: *
455: * Compute x(j) = b(j) / A(j,j), scaling x if necessary.
456: *
457: XJ = ABS( X( J ) )
458: IF( NOUNIT ) THEN
459: TJJS = AB( MAIND, J )*TSCAL
460: ELSE
461: TJJS = TSCAL
462: IF( TSCAL.EQ.ONE )
463: $ GO TO 100
464: END IF
465: TJJ = ABS( TJJS )
466: IF( TJJ.GT.SMLNUM ) THEN
467: *
468: * abs(A(j,j)) > SMLNUM:
469: *
470: IF( TJJ.LT.ONE ) THEN
471: IF( XJ.GT.TJJ*BIGNUM ) THEN
472: *
473: * Scale x by 1/b(j).
474: *
475: REC = ONE / XJ
476: CALL DSCAL( N, REC, X, 1 )
477: SCALE = SCALE*REC
478: XMAX = XMAX*REC
479: END IF
480: END IF
481: X( J ) = X( J ) / TJJS
482: XJ = ABS( X( J ) )
483: ELSE IF( TJJ.GT.ZERO ) THEN
484: *
485: * 0 < abs(A(j,j)) <= SMLNUM:
486: *
487: IF( XJ.GT.TJJ*BIGNUM ) THEN
488: *
489: * Scale x by (1/abs(x(j)))*abs(A(j,j))*BIGNUM
490: * to avoid overflow when dividing by A(j,j).
491: *
492: REC = ( TJJ*BIGNUM ) / XJ
493: IF( CNORM( J ).GT.ONE ) THEN
494: *
495: * Scale by 1/CNORM(j) to avoid overflow when
496: * multiplying x(j) times column j.
497: *
498: REC = REC / CNORM( J )
499: END IF
500: CALL DSCAL( N, REC, X, 1 )
501: SCALE = SCALE*REC
502: XMAX = XMAX*REC
503: END IF
504: X( J ) = X( J ) / TJJS
505: XJ = ABS( X( J ) )
506: ELSE
507: *
508: * A(j,j) = 0: Set x(1:n) = 0, x(j) = 1, and
509: * scale = 0, and compute a solution to A*x = 0.
510: *
511: DO 90 I = 1, N
512: X( I ) = ZERO
513: 90 CONTINUE
514: X( J ) = ONE
515: XJ = ONE
516: SCALE = ZERO
517: XMAX = ZERO
518: END IF
519: 100 CONTINUE
520: *
521: * Scale x if necessary to avoid overflow when adding a
522: * multiple of column j of A.
523: *
524: IF( XJ.GT.ONE ) THEN
525: REC = ONE / XJ
526: IF( CNORM( J ).GT.( BIGNUM-XMAX )*REC ) THEN
527: *
528: * Scale x by 1/(2*abs(x(j))).
529: *
530: REC = REC*HALF
531: CALL DSCAL( N, REC, X, 1 )
532: SCALE = SCALE*REC
533: END IF
534: ELSE IF( XJ*CNORM( J ).GT.( BIGNUM-XMAX ) ) THEN
535: *
536: * Scale x by 1/2.
537: *
538: CALL DSCAL( N, HALF, X, 1 )
539: SCALE = SCALE*HALF
540: END IF
541: *
542: IF( UPPER ) THEN
543: IF( J.GT.1 ) THEN
544: *
545: * Compute the update
546: * x(max(1,j-kd):j-1) := x(max(1,j-kd):j-1) -
547: * x(j)* A(max(1,j-kd):j-1,j)
548: *
549: JLEN = MIN( KD, J-1 )
550: CALL DAXPY( JLEN, -X( J )*TSCAL,
551: $ AB( KD+1-JLEN, J ), 1, X( J-JLEN ), 1 )
552: I = IDAMAX( J-1, X, 1 )
553: XMAX = ABS( X( I ) )
554: END IF
555: ELSE IF( J.LT.N ) THEN
556: *
557: * Compute the update
558: * x(j+1:min(j+kd,n)) := x(j+1:min(j+kd,n)) -
559: * x(j) * A(j+1:min(j+kd,n),j)
560: *
561: JLEN = MIN( KD, N-J )
562: IF( JLEN.GT.0 )
563: $ CALL DAXPY( JLEN, -X( J )*TSCAL, AB( 2, J ), 1,
564: $ X( J+1 ), 1 )
565: I = J + IDAMAX( N-J, X( J+1 ), 1 )
566: XMAX = ABS( X( I ) )
567: END IF
568: 110 CONTINUE
569: *
570: ELSE
571: *
572: * Solve A' * x = b
573: *
574: DO 160 J = JFIRST, JLAST, JINC
575: *
576: * Compute x(j) = b(j) - sum A(k,j)*x(k).
577: * k<>j
578: *
579: XJ = ABS( X( J ) )
580: USCAL = TSCAL
581: REC = ONE / MAX( XMAX, ONE )
582: IF( CNORM( J ).GT.( BIGNUM-XJ )*REC ) THEN
583: *
584: * If x(j) could overflow, scale x by 1/(2*XMAX).
585: *
586: REC = REC*HALF
587: IF( NOUNIT ) THEN
588: TJJS = AB( MAIND, J )*TSCAL
589: ELSE
590: TJJS = TSCAL
591: END IF
592: TJJ = ABS( TJJS )
593: IF( TJJ.GT.ONE ) THEN
594: *
595: * Divide by A(j,j) when scaling x if A(j,j) > 1.
596: *
597: REC = MIN( ONE, REC*TJJ )
598: USCAL = USCAL / TJJS
599: END IF
600: IF( REC.LT.ONE ) THEN
601: CALL DSCAL( N, REC, X, 1 )
602: SCALE = SCALE*REC
603: XMAX = XMAX*REC
604: END IF
605: END IF
606: *
607: SUMJ = ZERO
608: IF( USCAL.EQ.ONE ) THEN
609: *
610: * If the scaling needed for A in the dot product is 1,
611: * call DDOT to perform the dot product.
612: *
613: IF( UPPER ) THEN
614: JLEN = MIN( KD, J-1 )
615: SUMJ = DDOT( JLEN, AB( KD+1-JLEN, J ), 1,
616: $ X( J-JLEN ), 1 )
617: ELSE
618: JLEN = MIN( KD, N-J )
619: IF( JLEN.GT.0 )
620: $ SUMJ = DDOT( JLEN, AB( 2, J ), 1, X( J+1 ), 1 )
621: END IF
622: ELSE
623: *
624: * Otherwise, use in-line code for the dot product.
625: *
626: IF( UPPER ) THEN
627: JLEN = MIN( KD, J-1 )
628: DO 120 I = 1, JLEN
629: SUMJ = SUMJ + ( AB( KD+I-JLEN, J )*USCAL )*
630: $ X( J-JLEN-1+I )
631: 120 CONTINUE
632: ELSE
633: JLEN = MIN( KD, N-J )
634: DO 130 I = 1, JLEN
635: SUMJ = SUMJ + ( AB( I+1, J )*USCAL )*X( J+I )
636: 130 CONTINUE
637: END IF
638: END IF
639: *
640: IF( USCAL.EQ.TSCAL ) THEN
641: *
642: * Compute x(j) := ( x(j) - sumj ) / A(j,j) if 1/A(j,j)
643: * was not used to scale the dotproduct.
644: *
645: X( J ) = X( J ) - SUMJ
646: XJ = ABS( X( J ) )
647: IF( NOUNIT ) THEN
648: *
649: * Compute x(j) = x(j) / A(j,j), scaling if necessary.
650: *
651: TJJS = AB( MAIND, J )*TSCAL
652: ELSE
653: TJJS = TSCAL
654: IF( TSCAL.EQ.ONE )
655: $ GO TO 150
656: END IF
657: TJJ = ABS( TJJS )
658: IF( TJJ.GT.SMLNUM ) THEN
659: *
660: * abs(A(j,j)) > SMLNUM:
661: *
662: IF( TJJ.LT.ONE ) THEN
663: IF( XJ.GT.TJJ*BIGNUM ) THEN
664: *
665: * Scale X by 1/abs(x(j)).
666: *
667: REC = ONE / XJ
668: CALL DSCAL( N, REC, X, 1 )
669: SCALE = SCALE*REC
670: XMAX = XMAX*REC
671: END IF
672: END IF
673: X( J ) = X( J ) / TJJS
674: ELSE IF( TJJ.GT.ZERO ) THEN
675: *
676: * 0 < abs(A(j,j)) <= SMLNUM:
677: *
678: IF( XJ.GT.TJJ*BIGNUM ) THEN
679: *
680: * Scale x by (1/abs(x(j)))*abs(A(j,j))*BIGNUM.
681: *
682: REC = ( TJJ*BIGNUM ) / XJ
683: CALL DSCAL( N, REC, X, 1 )
684: SCALE = SCALE*REC
685: XMAX = XMAX*REC
686: END IF
687: X( J ) = X( J ) / TJJS
688: ELSE
689: *
690: * A(j,j) = 0: Set x(1:n) = 0, x(j) = 1, and
691: * scale = 0, and compute a solution to A'*x = 0.
692: *
693: DO 140 I = 1, N
694: X( I ) = ZERO
695: 140 CONTINUE
696: X( J ) = ONE
697: SCALE = ZERO
698: XMAX = ZERO
699: END IF
700: 150 CONTINUE
701: ELSE
702: *
703: * Compute x(j) := x(j) / A(j,j) - sumj if the dot
704: * product has already been divided by 1/A(j,j).
705: *
706: X( J ) = X( J ) / TJJS - SUMJ
707: END IF
708: XMAX = MAX( XMAX, ABS( X( J ) ) )
709: 160 CONTINUE
710: END IF
711: SCALE = SCALE / TSCAL
712: END IF
713: *
714: * Scale the column norms by 1/TSCAL for return.
715: *
716: IF( TSCAL.NE.ONE ) THEN
717: CALL DSCAL( N, ONE / TSCAL, CNORM, 1 )
718: END IF
719: *
720: RETURN
721: *
722: * End of DLATBS
723: *
724: END
CVSweb interface <joel.bertrand@systella.fr>