Debian Codegraph @

inst/ChangeLog

raw
1 This directory contains the development version of the sparse matrix
2 `package SparseM. This is an informal revision history.
3
4 The initial version 0.01 was sent to David Meyer in Vienna June 10, 2002.
5
6 The new version is in SparseM and incorporates the Harwell Boeing stuff.
7
8 The version 0.05 was completed Oct 11 and was sent to CRAN for Kurt Hornik's
9 blessing prior to submission to JSS. There were several problems the most
10 serious being that the fortran versions of the hb routines weren't sufficiently
11 portable.
12
13 The version 0.09 was completed Oct 19 2002 incorporating the following changes:
14
15 o read.hb and write.hb are now in C and more portable this was the
16 major problem identified by Kurt in the initial posting
17 o the generic issues were all resolved thanks to the hint in Sec 6.1
18 of the Writing R-Exts which explains how to add ... to the
19 argument list of new generics using formals().
20 o a bug in matrix.csr was fixed to accomodate vector arguments
21 and zero matrices
22 o PACKAGE args were added to all .Fortran and .C calls
23 o Lots of minor inconsistencies in the man pages were fixed.
24
25 The version 0.10 was completed Nov 5 2002 and incorporates the following:
26
27 o remade the pdf version of the vignette in type 1 ps fonts using
28 dvips -Pwww SparseM
29 distill SparseM.ps
30 At the suggestion of Brian Ripley.
31 o altered matrix.csr to generalize the way it handled vector args.
32 o removed the unused variables from readwrite.c.
33 o moved lsq.rra from the "data" subdirectory to "inst/HBdata"; added
34 lsq.R in "data" so that the content can be read in using
35 "data(lsq)" in R; added "lsq.Rd" in "man"; and updated the
36 00Index.dcf file in "inst/doc"; updated the INDEX file.
37 o changed all the ".find.package" in the Rd files to "system.file".
38 o changed the call to read.matrix.hb to use "system.file" in
39 SparseM.Rnw
40
41 The version 0.11 was begun Jan 1 2003 and incorporates the following:
42
43 o the malloc.h header requirements in iohb.c and iohb.h were
44 removed...they are redundant given stdlib.h and caused problems
45 on my mac
46
47 The version 0.12 was begun Jan 20 2003 and incorporates the following:
48
49 o conversion from S3 to S4 method dispatch
50
51 The version 0.20 was completed Jan 24 2003 and incorporates the following:
52
53 o conversion from S3 to S4 method dispatch and extensive documentation changes
54
55 The version 0.21 was begun Mar 7 2003 and incorporates the following:
56
57 o matrix valued response in the slm formula
58
59 o a fix of a "design infelicity" in the multiplication of a csr times a
60 vector which created a matrix object without dimension attribute --
61 identified by John Chambers.
62
63 o matrix multiplication now allows csr times dense multiplication by coercion
64 of the dense matrix to sparse form, so is returned in sparse form.
65
66 The version 0.22 was begun Mar 11 2003 and incorporates the following:
67
68 o added a class matrix.coo for coordinate format sparse matrices
69
70 o [<-.matrix.csr and [.matrix.csr rewritten to use the .coo class
71
72 o validity checking added to both .csr and .coo classes. See the setClass
73 definitions and their methods for "initialize". This required quite a lot
74 of editing to insure that ja,ia, and dim attributes were really integer.
75
76 The version 0.24 was begun about April 1 2003
77
78 o commented out 'value <- as.matrix.coo(value)' in '[<-.matrix.csr' to move the class
79 transformation of value inside '[<-.matrix.coo. This is a workaround for
80 misbehavior of missing() in our opinion.
81
82 o added 'setMethod("t","matrix.coo",function(x) as.matrix.coo(t(as.matrix.csr(x)))' in
83 SparseM_Methods.R to define "t" for class "matrix.coo".
84
85 o uncommented
86 "setMethod("%*%",signature(x="matrix",y="matrix.csr"),.matmul.matrix.csr)"
87 "setMethod("%*%",signature(x="numeric",y="matrix.csr"),.matmul.matrix.csr)"
88 so that "%*%" will perform matrix times matrix.csr and vector times
89 matrix.csr multiplication
90
91 o removed the definition of "ncol.matrix.csr" and "nrow.matrix.csr" from SparseM.R
92
93 o added "dim.matrix.coo" to SparseM.R and
94 "setMethod("dim","matrix.coo",dim.matrix.coo)" to SparseM_Methods.R
95
96 o created matrix.coo-class.Rd
97
98 o fixed some of the inconsistency on matrix.csc-class.Rd, etc.
99
100 o added matrix subscripting to '[.matrix.coo' and '[<-.matrix.coo' so that
101 one can perform, e.g., as.matrix.csr(A)[cbind(c(1,2,3),c(2,3,1))]
102 and as.matrix.csr(A)[cbind(c(1,2,3),c(2,3,1))] <- 100
103
104 o added matrix.csr class subscripting to '[.matrix.coo' and '[<-.matrix.coo'
105 so that one can perform, e.g., as.matrix.csr(A)[as.matrix.csr(A)>0]
106 and as.matrix.csr(A)[as.matrix.csr(A)>0] <- 100 for whatever it's worth.
107
108 o added logical subscripting to [.matrix.coo and [.matrix.csr
109
110 o fixed some bugs in group Ops.
111
112 o added some commentary in man and in vignette to encourage use of new()
113 in the definition of new .csr and .coo objects so that validation is done.
114
115 o simplified diag.assign.matrix.csr so that it looked just like the default
116 version -- this eliminates the need for a couple of the fortran functions.
117
118 o fixed several places where as.integer() was needed to wrap a definition of
119 a dimension attribute of a sparse object.
120
121
122 The version 0.25 was begun about April 16 2003
123
124 o fixed a bug in chol.matrix.csr that incorrectly allocated
125 space in large problems
126
127 The version 0.26 was done April 26 2003
128
129 o fixed a bug in * and %*% that returned an error when the answer
130 was a matrix or vector of all zeros
131
132 o refixed the malloc.h problem
133
134 The version 0.27 was done April 28 2003
135
136 o removed some debugging print statements from chol.f
137 o removed the uncompressed files from inst/doc
138
139 The version 0.28 was begun May 28 2003
140
141 o the .R files were consolidated at the suggestion of John Chambers after we
142 discovered that different machines ordered them differently and
143 the methods code needs to be read AFTER the function definitions.
144
145 o a bug in as.matrix.coo that mistook vector arguments for scalars was fixed
146
147 o a kronecker product operator %x% was added.
148
149 o return statements checked for the dreaded deprecated unlisted form
150
151 o updated some setIs() calls to conform to JC's new setClassUnion() form.
152 this is implemented with a version check to accomodate 1.7.0 users.
153
154 The version 0.30 was completed August 10 2003
155
156 o initially intended to conform with Kurt's changed of %x% to insist
157 that it was a binary operator, but this revealed a number of problems
158 with documentation that were attributable to an incomplete conversion
159 from S3 to S4 methods.
160
161 o Many S3 functions were incorporated directly into the setMethod calls so
162 that now there are fewer, but still many S3 method calls, these should
163 be gradually removed when more time and energy permits. For now, I hope
164 that the changes that have been made will permit a smooth transition to
165 1.8.0 when it is released. At the moment it does pass R CMD check on
166 the (almost) current version of R-devel. In particular, One
167 thing that eventually needs to be fixed is that all of the ontology
168 stuff should be brought up todate so that you can say:
169
170 as(x,"matrix.csr")
171
172 rather than
173
174 as.matrix.csr(x)
175
176 The version 0.31 was completed August 25 2003
177
178 o At the suggestion of Kurt we removed the explicit reading of the hb
179 data and replaced it with an R dataset loading. Comments are left
180 in the examples section indicating how to do the reading, but the
181 official view seems to be that the old approach was insufficiently
182 robust.
183
184 o The diag() update. Prior versions didn't really handle the coercion
185 of scalars and vectors to diag csr matrices very well. This was
186 fixed by introducing a new subclass matrix.diag.csr of which matrix.csr
187 is the superclass (meaning that they have the same rights and responsibilities)
188 so now you can create an n-dimensional identity by as(n,"matrix.diag.csr")
189 and a matrix with diagonal v by as(v,"matrix.diag.csr"). In the process
190 some problems with the default matrix.csr object were discovered and fixed.
191 And the code for .matrix.csr.elemul was changed to reflect these changes.
192
193 The version 0.32 was completed December 20 2003
194
195 o Fixed a defunct call to print.coefmat
196
197 The version 0.33 was completed January 20 2004
198
199 o fixed bug in scalar division...identified by Renald Buter
200
201 The version 0.35 was completed January 20 2004
202
203 o fixed bug in matrix.diag.csr coercion ... identified by Nicholas Lewin-Koh
204
205 The version 0.36 was completed May 12 2004
206
207 o fixed a version testing problem for R-devel 2.0.0
208 o added Depends R(>=1.8) to Description at suggestion of BDRipley
209 o added errfig in the .C() call in read.matrix.hb
210 o commented out some GTIMER, ETIME commands that caused AIX problems
211
212 The version 0.40 was completed July 17, 2004
213
214 o added a check for symmetry before the Cholesky computation.
215 o added a norm() function to do the symmetry checking
216 o added a det() function to compute the determinant of both
217 matrix.csr and matrix.csr.chol objects.
218
219 The version 0.50 was completed July 17, 2004
220
221 o the vignette was uncompressed so that it was accessible
222 via vignette().
223 o the call to read.matrix.hb was commented out and replaced
224 as in the other documentation calls with calls to
225 data(lsq) ...
226 o a comment was added to the solve documentation indicating
227 the existence of the det object in chol structures.
228
229 The version 0.52
230
231 o An example of Steven Farber revealed that when the
232 dimensions of matrix.csr and matrix.coo objects
233 are very large the validity checking in setClass
234 failed due to integer overflow. The product
235 nrow*ncol was replaced by prod(object@dimensions)
236 which apparently coerces the result to real.
237 o A misplaced } was fixed in slm.Rd found by Kurt's
238 pre 2.0.0 new documentation checking.
239 o version checking was modified to eliminate a last.warning
240 in 2.0.0.
241
242 The version 0.53
243
244 o elementwise multiplication failed for A*B with both A and B
245 of class matrix.csr, as noted by Suzzane Heim --
246
247 o added a [<-.matrix.diag.csr function, inadvertently omitted from
248 the prior revision of this class. (Produced a mysterious
249 "Error: incorrect number of subscripts on matrix" message.
250 This also fixed a subsetting-assignments-to-zero problem for
251 this class.
252
253 The version 0.54
254
255 o After an extended discussion in (and off) R-devel, Roger Bivand very
256 kindly provided a NAMESPACE file and some associated modifications
257 of the SparseM.R file. These changes allow among other
258 things S4 method dispatch to find functions in SparseM
259 that were hidden from view when it didn't have a NAMESPACE.
260 The original problem child was det() which SparseM declared
261 generic, but this wasn't visible to Bivand's spatial package.
262 Use requires various import directives in the NAMESPACE
263 of the importing package, however quantreg which uses SparseM
264 for certain nonparametric regression problems worked without
265 modification with the new version of SparseM. Bivand's
266 modifications of the R code are indicated with comments "#RSB"
267 in the SparseM.R file.
268
269 The version 0.55
270
271 o Added lazyloading to description file at the request of BDR.
272
273 o Jon Stearley (Sandia) noted that R segfaulted on his attempts
274 to use write.matrix.hb. The segmentation fault was caused
275 by the "NULL" value of "guesol" that was being passed into
276 readwrite.c. Pin replaced the NULL assignment with an
277 empty string "" assignment. I've also updated the help file
278 SparseM.hb.Rd to include an example of write.matrix.hb.
279
280 The version 0.56
281
282 o the example added in 0.55 caused the windows version to segfault
283 Uwe Ligges kindly determined that adding ZipData: No to the
284 Description file was enough to fix this.
285 The version 0.57
286
287 o NAMESPACE and getMethod added to acommodate an asymmetry in the
288 way that the stats package handles model.matrix and model.response
289 fix kindly provided by Roger Bivand.
290
291 The version 0.58
292
293 o several "assign num to lab" statements in sparskit.f were updated
294 to anticipate gnu fortran compilers that have apparently
295 decided that such syntax is obsolete. Thanks to Jan de Leeuw
296 for pointing this out based on gfortran-4.0 on osx 10.34.
297
298 The version 0.59
299
300 o the subroutine prtmt intended to print matrices in Harwell Boeing
301 format also caused problems for the new gfortran-4.0 compiler
302 since this routine (in sparskit) was never called it was
303 simply removed from sparskit.f. Jan reports that this allows
304 SparseM to build on his configuration.
305
306 The version 0.60
307
308 o Jeffrey Chang pointed out that indexing failed to work properly
309 when there were repeated indices. This has now been fixed,
310 but the resulting code is very slow due to looping on an
311 rbind/cbind operation. This would be a useful feature to
312 speed up, if one were really into resampling.
313
314 o In the process of making the changes mentioned above, it was
315 noted that [.matrix.coo should be exported via NAMESPACE.
316
317 The version 0.61
318
319 o Jon Stearley (Sandia) noted that R segfaulted on some attempts
320 to use write.matrix.hb. The segmentation fault was caused
321 by inappropriate use of default format statements that led
322 to incorrectly written indices. Pin modified the default
323 formats to reduce the probability of such occurances.
324
325 o A Makefile was added to the vignette directory to rationalize
326 things a bit.
327 The version 0.62
328
329 o Brian Ripley pointed out a bug in iohb.c which caused a segfault
330 in some linux systems with gcc4 and was idenntified more
331 precisely using valgrind. Some googling found a corrected
332 version at:
333 http://www.cs.utk.edu/~eijkhout/salsa-salsa-docs/iohb_8c.html#a6
334
335 o Tom Blackwell (UMich) pointed out a problem with elemult which
336 turned out to be an unfortunate sequence of if/else's that
337 omitted to treat the default case properly.
338 The version 0.63
339 o The functionality of reading and writing Harwell-Boeing files
340 has been temporarily suspended pending reevaluation of
341 replacing iohb.c with code that doesn't cause memory
342 faults. See Doug Bates Matrix package for a limited
343 alternative approach.
344 The version 0.64
345 o The function read.matrix.hb is replaced by a pure R function, purging
346 iohb.c to read Harwell-Boeing formatted files. The new function
347 shamelessly plagarizes from Doug Bates's readHB function in the
348 Matrix package. The main alterations were to add functionality for rhs's,
349 and adapt the function to SparseM matrix classes. As of this moment
350 we have no plans to produce a parallel function for writing
351 Harwell-Boeing format.
352
353 The version 0.65
354 o The way that chol.default, and the setMethod calls for det and chol
355 for the class matrix were changed to conform to R 2.2, following
356 advice of Brian Ripley and Duncan Murdoch.
357 The version 0.66
358 o Bug found by Gregg Lind that made diag assigments fail when the
359 matrix was all zeros. Modified [<-.matrix.coo to handle
360 NAs from the match() call.
361 The version 0.67
362 o Typos in section headings of .Rd files noted by Kurt.
363
364 The version 0.68
365 o Bug in slm.wfit fixed, thanks to i.m.s. white (edinburgh).
366 The version 0.69
367 o Bug in slm.fit.csr -- calls to chol and backsolve failed to
368 pass along the ... arguments. Thanks to Jeff Racine.
369 o Added code for extractAIC.slm and deviance.slm to permit
370 use of step() on fitted slm objects, code contributed
371 by Jeff Racine.
372 o added setMethod for as.matrix to convert matrix.csc objects
373 that had been overlooked for a long time.
374 o changed the printing onLoad at the suggestion of Martin M.
375
376 The version 0.70
377 o The setGeneric for chol was changed in accordance with another
378 suggestion by Martin M. This seems to fix a conflict
379 with Matrix as described in an R-devel message by David
380 Pleydell (20 Feb 2006).
381 The version 0.72
382 o A ... argument was added to the as.matrix.[foo] functions to
383 accomodate a change in base R.
384 The version 0.73
385 o Thanks to contributions of Ben Shaby (Stat, Cornell) it is
386 now possible to coerce matrix.csr.chol objects back
387 to matrix.csr form, separate backsolve, and forwardsolve
388 functions are provided in addition to the original backsolve
389 that (still) does the double-backsolve, and it is now also
390 possible to get a log det value from chol instead of det.
391 The version 0.74
392 o Changed warning in as.matrix.csr to error condition when nnz != z$nnz,
393 and changed defintion of nnz to by sum(abs(x) >= eps) from >
394 to agree with fortran "csr" function. Bug report from Mark
395 Robinson (wehi.edu.au)
396 The version 0.76
397 o Updated lsq.rda so that it was a proper s4 object as requested by BR
398 to conform with the current R-devel. Removed SaveImage line
399 from DESCRIPTION file.
400 The version 0.77
401 o Updated triogramX.R to triogramX.rda similarly...
402 The version 0.78
403 o Fixed bug in [<-.matrix.coo". When the rw was a two column matrix
404 there was an implicit assumption that x[-s] = x when s was empty
405 but this turned out to be wrong.
406 The version 0.79
407 o I've redefine the class "matrix.csc.hb" in SparseM.R to include NULL
408 in the slots rhs.ra, rhs.dim and rhs.mode.
409 o read.matrix.hb will return NULL for rhs.ra, rhs.dim and rhs.mode if
410 the rhs is missing.
411 o bug in matrix.csr.elemul that returned an error when the result of
412 the multiplication was a matrix consisting entirely of zeros.
413 Thanks to Ian Erickson for this report.
414 The version 0.80
415 o At the suggestion of BDR the onLoad message was changed from cat()
416 to packageStartMessage()
417
418 The version 0.82
419 o default tmpmax in chol was increased from 10 * nrows to 30 * nrows.
420
421 The version 0.83
422 o setGeneric code for backsolve fixed so that it passes arguments to
423 the default method in base::backsolve. Thanks to Achim Zeileis for
424 the report of a problem encountered when using lars, and to John
425 Chambers for patiently explaining what was going wrong.
426
427 The version 0.84
428 o Several organisational changes of method definition, with the
429 main goal of improved cohabitation with the Matrix package:
430 - using less setGeneric() calls,
431 e.g., chol() and norm() now should still work with Matrix present;
432 - defining methods for kronecker() instead of "%x%" which now works
433 automagically via kronecker();
434 - defining methods for determinant(), in order to eventually
435 deprecate those for det().
436 o default value of tmpmax increased to 50*nrow(x) with luck this will
437 make it less likely to get complaints about needed to increase it.
438 The version 0.86
439 o Changed License in DESCRIPTION to file License and added a somewhat
440 inflammatory comment to the Licence file that was sent to R-devel in
441 2009.
442 The version 0.89
443 o added a setAs for matrix to matrix.csr conversion
444 The version 0.90
445 o Conditionally removed assigninNamespace() in onLoad as suggested
446 by BDR see mail from Oct 28 2011.
447 The version 0.96
448 1. Removed the LICENSE file that had described some uncertainties with respect
449 to the license status of the code in src/cholesky.f. These uncertainties have
450 now been resolved by the original authors, Esmond Ng and Barry Peyton,
451 declaring their code to be open source, and thus the quantreg package can now
452 be considered fully GPL. See the README file for further details.
453
454 2. fixed some vestigial calls to fprintf IOHBTerminate etc that
455 contained un-R-like printing or exit calls.
456
457 3. Added a .Rinstignore file to deal with the vignette Makefile
458 complaint from R CMD check.
459
460 4. Modified the .onLoad call to include a call to assignInNamespace()
461 to deal with a problem with kronecker. (This is provisional until
462 2.15 is released.) Thanks to BDR for helpful advice on this and the
463 other items in this revision.
464
465 The version 0.96
466 1. Fixed bug in element by element exponentiation when one of the
467 arguments was scalar.
468
469 The version 0.97
470 1. Removed iohb.c and readwrite.c which were deprecated much earlier
471 and produced various warning flags from clang in the initial testing
472 of R 3.0 beta.
473
474 The version 0.98
475 1. Fixed various dimension problems identified by the new R CMD check
476 protocol.
477
478 The Version 1.00
479 1. Added a paragraph in SparseM.solve.Rd about passed values of the
480 mysterious chol() storage allocation parameters.
481
482 The version 1.01
483 1. Removed the onLoad() call which seemed to cause problems for Uwe's
484 build of the windoz version due to some problem with the way %x% was
485 treated there. Strange since whatever was happening was only supposed
486 to happen for R v2.15 and lesser.
487
488 The version 1.03
489 1. Moved SparseM.Rnw to a new vignettes directory at the suggestion
490 of Brian Ripley and removed the Makefile for the vignette at the
491 suggestion of Uwe Ligges.
492
493 The version 1.05
494 1. Attempted to fix the definition of det and associated methods to
495 conform with procedure used by the Matrix package. If I understand this
496 properly, the idea is that det can be defined as base::det which defaults
497 to the generic determinant() and then by defining the environment(det)
498 as environment() this ensures that calls to det() from SparseM will use
499 the methods appropriate to SparseM objects.
500
501
502 The version 1.7
503
504 1. Removed coerce from exports.
505
506 The version 1.71
507
508 1. Fixed the REAL*8 and INTEGER*4 occurences in accordance with Kurt's email.
509
510 The version 1.73
511
512 1. changed occurences of unix.time to system.time in accordance with Kurt's
513 directive.
514
515 The version 1.74
516
517 1. One more unix -> system
518
519 The version 1.75
520
521 1. package -> PACKAGE [per Kurt's email of 3/2/17]
522
523 The version 1.76
524
525 1. Used package_native_routine_registration_skeleton to remove
526 PACKAGE = SparseM [per Kurt's email of 3/2/17]
527
528 The version 1.77
529
530 1. Changed setMethod for diag (matrix.csr) in accordance with MM's suggestion
531
532 2. removed setGeneric for image also suggested by MM.
533
534 The version 1.78
535
536 1. Changed check for symmetry to use all.equal() in response to an error
537 reported by Brian Ripley.
538
539 The version 1.79
540
541 1. Removed a "f_" from the .Fortran call in chol2csr following bug report
542 by Alex Ballantynea (Melbourne U)
543
544 The version 1.82
545
546 1. Modernized the cholesky.f to remove arithmetic ifs, suggested by BDR.
547 2. Repaired some syntax errors in lsq.Rd.
548 3. Commented out S3Method lines at the end of NAMESPACE file, at suggestion
549 of Ivan Krylov, see mail of May 27, 2024.
550 4. This version was accepted by Uwe for CRAN, despite generating an error
551 with balanceTest() in the RItools package.
552 5. After some chaos involving namespace issues that were due to extraneous
553 conditions, the package passed R CMD check after adding a setGeneric(image)
554 command in SparseM.R. (This allowed me to restore the vignette in its Rnw
555 form.
556 6. This left things in the status quo ante with the problem in RItools, which
557 I'm hoping Jake Bowers and his collaborators can help debug.
558 7. Meanwhile there is a broken 1.82 on CRAN that I've asked Uwe to remove.
559 8. The problem with the RItools example turned out to be that chol was being
560 applied to a singular matrix. I've added a warning in the documentation that
561 is no pre-checking for PD matrices.
562 9. I also found that determinant wasn't acknowledged to be generic, so this
563 has been repaired, however there is still a problem with accessing the
564 documentation of this.
565
566 The version 1.84
567 (much by Martin Maechler who also adds himself to the Authors 'ctb')
568
569 1. chol(<matrix.csr) gets new argument `warn.only = FALSE` in order to
570 still return in case of errors (as this has happened in older
571 versions, at least coincidentally). Further new arguments for
572 previously hardwired "tuning" constants are
573 'tiny = 1e-30',
574 'Large = 1e128',
575 'cacheKb = 1024L',
576 'level = 8L'.
577 Note that `cacheKb` was hardwired to 64, clearly too small a
578 default for today's hardware.
579 Further, the chol() method is well documented ( ../man/SparseM.solve.Rd ).
580 An example has been added where _decreasing_ `tiny` was beneficial.
581
582 2. determinant(<matrix.csr>) allows to pass optional arguments for chol().
583 3. det() now works, automagically calling _our_ determinant().
584 4. Fix backsolve() and forwardsolve() methods documentation, adding an
585 `drop=TRUE` argument in one case.
586 4b. These tweaks have also eliminated the unpleasant message when SparseM is attached:
587
588 The following object is masked from ‘package:base’: \cr backsolve
589
590 5. A nice simple show() method now (auto-)prints our sparse matrices with
591 structural zeros shown as " . "
592 6. norm(<matrix.csr>, "I") now gives an error (instead of invisible(NULL)).
593 7. demo(LeastSquares) and demo(Solve) work now.
594
595 The version 1.84-1
596
597 - src/SparseM.h : use 'int_least32_t' instead of 'int' for Fortran 'logical', in two cases (fixing 'LTO').
598 - src/csr.f (etc): drop first *unused* argument 'ra' from nzero().
599 - src/cholesky.f: replace computed goto's in code by select case(.) case() ...