Home My Page Projects Code Snippets Project Openings SML/NJ
Summary Activity Forums Tracker Lists Tasks Docs Surveys News SCM Files

SCM Repository

[smlnj] Diff of /sml/trunk/HISTORY
ViewVC logotype

Diff of /sml/trunk/HISTORY

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

revision 977, Wed Nov 14 16:53:16 2001 UTC revision 1238, Mon Jun 10 20:35:28 2002 UTC
# Line 8  Line 8 
8  The form of an entry should be:  The form of an entry should be:
9    
10  Name:  Name:
11  Date:  Date: yyyy/mm/dd
12  Tag: <post-commit CVS tag>  Tag: <post-commit CVS tag>
13  Description:  Description:
14    
15  ----------------------------------------------------------------------  ----------------------------------------------------------------------
16  Name: Matthias Blume  Name: Matthias Blume
17    Date: 2002/06/10 16:35:00 EDT
18    Tag: blume-20020610-ieeereal
19    Description:
20    
21    I back-ported my implementation of IEEEReal.fromString from the
22    idlbasis-devel branch so that we can test it.
23    
24    Another small change is that ppDec tries to give more information
25    than just "<sig>" in the case of functors.  However, this code is
26    broken in some mysterious way if the functor's body's signature
27    has not been declared by ascription but gets inferred from the
28    implementation.  This needs fixing...
29    
30    ----------------------------------------------------------------------
31    Name: Matthias Blume
32    Date: 2002/05/31
33    Tag: blume-20020531-btrace-mode
34    Description:
35    
36    Resurrected SMLofNJ.Internals.BTrace.mode.  (It accidentally fell by
37    the wayside when I switched over to using Controls everywhere.)
38    
39    ----------------------------------------------------------------------
40    Name: Lal George
41    Date: 2002/05/23 12:21:40 EDT
42    Tag: george-20020523-visual-labels
43    Description:
44    
45    Labels are now displayed in the graphical output to make
46    the fall-through and target blocks obvious.
47    
48    ----------------------------------------------------------------------
49    Name: Matthias Blume
50    Date: 2002/05/22 11:03:00 EDT
51    Tag: blume-20020522-shrink
52    Description:
53    
54    John tweaked yesterday's fix for 1131 to handle an out-of-memory
55    situation that comes up when allocating huge arrays.
56    
57    ----------------------------------------------------------------------
58    Name: Matthias Blume
59    Date: 2002/05/21 16:00:00 EDT
60    Tag: Release_110_40
61    Description:
62    
63    New working release (110.40).  New bootfiles.
64    
65    [Also: John Reppy fixed GC bug 1131.]
66    
67    ----------------------------------------------------------------------
68    Name: Matthias Blume
69    Date: 2002/05/21 12:35:00 EDT
70    Tag: blume-20020521-cmdoc
71    Description:
72    
73    CM documentation update.
74    
75    ----------------------------------------------------------------------
76    Name: Matthias Blume
77    Date: 2002/05/21 10:55:00 EDT
78    Tag: blume-20020521-misc
79    Description:
80    
81    - John tweaked runtime to be silent on heap export (except when
82      GC messages are on).
83    - I added a few more things (cross-compiling versions of CMB) to
84      config/preloads (as suggestions).
85    
86    ----------------------------------------------------------------------
87    Name: Matthias Blume
88    Date: 2002/05/20 22:25:00 EDT
89    Tag: blume-20020520-controls
90    Description:
91    
92    - Added ControlUtil structure to control-lib.cm.
93    - Use it throughout.
94    - Used Controls facility to define MLRISC controls (as opposed to
95      registering MLRISC control ref cells with Controls after the fact)
96    - Fixed messed-up controls priorities.
97    
98    * Removed again all the stuff from config/preloads that one wouldn't
99      be able to preload at the time the initial heap image is built.
100      (Many libraries, e.g., CML, do not exist yet at this time.  The
101       only libraries that can be preloaded via config/preloads are those
102       that come bundled with the bootfiles.)
103    
104    ----------------------------------------------------------------------
105    Name: Matthias Blume
106    Date: 2002/05/20 10:59:00 EDT
107    Tag: blume-20020520-preloads
108    Description:
109    
110    Added a lot of commented-out suggestions for things to be included
111    in config/preloads.
112    
113    ----------------------------------------------------------------------
114    Name: Allen Leung
115    Date: 2002/05/18 14:20:00 EDT
116    Tag: leunga-20020518-mdl
117    Description:
118    
119    o Made the mdl tool stuff compile and run again.
120    o I've disabled all the stuff that depends on RTL specifications; they
121      are all badly broken anyway.
122    
123    ----------------------------------------------------------------------
124    Name: Matthias Blume
125    Date: 2002/05/17 16:49:00 EDT
126    Tag: blume-20020517-controls
127    Description:
128    
129    0. John Reppy made several modifications to the SML/NJ library.
130       In particular, there is a shiny new controls-lib.cm.
131    
132    1. Pushed new controls interface through compiler so that everything
133       compiles again.
134    
135    2. Added FormatComb and FORMAT_COMB to the CML version of the
136       SML/NJ library (so that CML compiles again).
137    
138    3. Modified init scripts because XXX_DEFAULT environment variables
139       are no longer with us.  (Boot-time initialization is now done
140       using the same environment variables that are also used for
141       startup-time initialization of controls.)
142    
143    ----------------------------------------------------------------------
144    Name: Lal George
145    Date: 2002/05/15 09:20:10 EDT
146    Tag: george-20020515-pseudo-op-decls
147    Description:
148    
149            All pseudo-ops emitted before the first segment declaration
150    such as TEXT, DATA, and BSS directives are assumed to be global
151    declarations and are emitted first in the assembly file. This is
152    useful in a number of situations where one has pseudo-ops that are not
153    specific to any segment, and also works around the constraint that one
154    cannot have client pseudo-ops in the TEXT segment.
155    
156            Because no segment is associated with these declarations it is
157    an error to allocate any space or objects before the first segment
158    directive and an exception will be raised. However, we cannot make
159    this check for client pseudo-ops.
160    
161            These top level declarations are a field in the CFG graph_info.
162    In theory you can continue to add to this field after the CFG has been
163    built -- provided you know what you are doing;-)
164    
165    ----------------------------------------------------------------------
166    Name: Matthias Blume
167    Date: 2002/05/13 16:40:00 EDT
168    Tag: blume-20020513-pp-etc
169    Description:
170    
171    A few minor bugfixes:
172    
173      - Stopgap measure for bug recently reported by Elsa Gunter (ppDec).
174        (Bogus printouts for redefined bindings still occur.  Compiler
175        bug should no longer occur now.  We need to redo the prettyprinter
176        from scratch.)
177    
178      - CM pathname printer now also adds escape sequences for ( and )
179    
180      - commend and docu fixes for ml-nlffi
181    
182    ----------------------------------------------------------------------
183    Name: Matthias Blume
184    Date: 2002/05/10 16:40:00 EDT
185    Tag: blume-20020510-erg-textio
186    Description:
187    
188    Applied the following bugfix provided by Emden Gansner:
189    
190        Output is corrupted when outputSubstr is used rather than output.
191    
192        The problem occurs when a substring
193    
194            ss = (s, dataStart, dataLen)
195    
196        where dataStart > 0, fills a stream buffer with avail bytes left.
197        avail bytes of s, starting at index dataStart, are copied into the
198        buffer, the buffer is flushed, and then the remaining dataLen-avail
199        bytes of ss are copied into the beginning of the buffer. Instead of
200        starting this copy at index dataStart+avail in s, the current code
201        starts the copy at index avail.
202    
203        Fix:
204        In text-io-fn.sml, change line 695 from
205             val needsFlush = copyVec(v, avail, dataLen-avail, buf, 0)
206        to
207             val needsFlush = copyVec(v, dataStart+avail, dataLen-avail, buf, 0)
208    
209    ----------------------------------------------------------------------
210    Name: Matthias Blume
211    Date: 2002/04/12 13:55:00 EDT
212    Tag: blume-20020412-assyntax
213    Description:
214    
215    1. Grabbed newer assyntax.h from the XFree86 project.
216    2. Fiddled with how to compile X86.prim.asm without warnings.
217    3. (Very) Minor cleanup in CM.
218    
219    ----------------------------------------------------------------------
220    Name: Matthias Blume
221    Date: 2002/04/01 (no joke!) 17:07:00 EST
222    Tag: blume-20020401-x86div
223    Description:
224    
225    Added full support for div/mod/rem/quot on the x86, using the machine
226    instruction's two results (without clumsily recomputing the remainder)
227    directly where appropriate.
228    
229    Some more extensive power-of-two support was added to the x86 instruction
230    selector (avoiding expensive divs, mods, and muls where they can be
231    replaced with cheaper shifts and masks).  However, this sort of thing
232    ought to be done earlier, e.g., within the CPS optimizer so that
233    all architectures benefit from it.
234    
235    The compiler compiles to a fixed point, but changes might be somewhat
236    fragile nevertheless.  Please, report any strange things that you might
237    see wrt. div/mod/quot/rem...
238    
239    ----------------------------------------------------------------------
240    Name: Matthias Blume
241    Date: 2002/03/29 17:22:00
242    Tag: blume-20020329-div
243    Description:
244    
245    Fixed my broken div/mod logic.  Unfortunately, this means that the
246    inline code for div/mod now has one more comparison than before.
247    Fast paths (quotient > 0 or remainder = 0) are not affected, though.
248    The problem was with quotient = 0, because that alone does not tell
249    us which way the rounding went.  One then has to look at whether
250    remainder and divisor have the same sign...  :(
251    
252    Anyway, I replaced the bootfiles with fresh ones...
253    
254    ----------------------------------------------------------------------
255    Name: Matthias Blume
256    Date: 2002/03/29 14:10:00 EST
257    Tag: blume-20020329-inlprims
258    Description:
259    
260    NEW BOOTFILES!!!    Version number bumped to 110.39.3.
261    
262    Primops have changed. This means that the bin/boot-file formats have
263    changed as well.
264    
265    To make sure that there is no confusion, I made a new version.
266    
267    
268    CHANGES:
269    
270    * removed REMT from mltree (remainder should never overflow).
271    
272    * added primops to deal with divisions of all flavors to the frontend
273    
274    * handled these primops all the way through so they map to their respective
275      MLRISC support
276    
277    * used these primops in the implementation of Int, Int32, Word, Word32
278    
279    * removed INLDIV, INLMOD, and INLREM as they are no longer necessary
280    
281    * parameterized INLMIN, INLMAX, and INLABS by a numkind
282    
283    * translate.sml now deals with all flavors of INL{MIN,MAX,ABS}, including
284      floating point
285    
286    * used INL{MIN,MAX,ABS} in the implementation of Int, Int32, Word, Word32,
287      and Real (but Real.abs maps to a separate floating-point-only primop)
288    
289    
290    TODO items:
291    
292    * Hacked Alpha32 instruction selection, disabling the selection of REMx
293      instructions because the machine instruction encoder cannot handle
294      them.  (Hppa, PPC, and Sparc instruction selection did not handle
295      REM in the first place, and REM is supported by the x86 machine coder.)
296    
297    * Handle DIV and MOD with DIV_TO_NEGINF directly in the x86 instruction
298      selection phase.  (The two can be streamlined because the hardware
299      delivers both quotient and remainder at the same time anyway.)
300    
301    * Think about what to do with "valOf(Int32.minInt) div ~1" and friends.
302      (Currently the behavior is inconsistent both across architectures and
303      wrt. the draft Basis spec.)
304    
305    * Word8 should eventually be handled natively, too.
306    
307    * There seems to be one serious bug in mltree-gen.sml.  It appears, though,
308      as if there currently is no execution path that could trigger it in
309      SML/NJ.  (The assumptions underlying functions arith and promotable do not
310      hold for things like multiplication and division.)
311    
312    ----------------------------------------------------------------------
313    Name: Matthias Blume
314    Date: 2002/03/27 16:27:00 EST
315    Tag: blume-20020327-mlrisc-divisions
316    Description:
317    
318    Added support for all four division operations (ML's div, mod, quot,
319    and rem) to MLRISC.  In the course of doing so, I also rationalized
320    the naming (no more annoying switch-around of DIV and QUOT), by
321    parameterizing the operation by div_rounding_mode (which can be either
322    DIV_TO_ZERO or DIV_TO_NEGINF).
323    
324    The generic MLTreeGen functor takes care of compiling all four
325    operations down to only round-to-zero div.
326    
327    Missing pieces:
328    
329      * Doing something smarter than relying on MLTreeGen on architectures
330        like, e.g., the x86 where hardware division delivers both quotient and
331        remainder at the same time.  With this, the implementation of the
332        round-to-neginf operations could be further streamlined.
333    
334      * Remove inlining support for div/mod/rem from the frontend and replace it
335        with primops that get carried through to the backend.  Do this for all
336        int and word types.
337    
338    ----------------------------------------------------------------------
339    Name: Matthias Blume
340    Date: 2002/03/25 17:25:00 EST
341    Tag: blume-20020325-divmod
342    Description:
343    
344    I improved (hopefully without breaking them) the implementation of Int.div,
345    Int.mod, and Int.rem.   For this, the code in translate.sml now takes
346    advantage of the following observations:
347    
348      Let  q = x quot y      r = x rem y
349           d = x div  y      m = x mod y
350    
351    where "quot" is the round-to-zero version of integer division that
352    hardware usually provides.  Then we have:
353    
354         r = x - q * y        where neither the * nor the - will overflow
355         d = if q >= 0 orelse x = q * y then q else q - 1
356                              where neither the * nor the - will overflow
357         m = if q >= 0 orelse r = 0 then r else r + y
358                              where the + will not overflow
359    
360    This results in substantial simplification of the generated code.
361    The following table shows the number of CFG nodes and edges generated
362    for
363            fun f (x, y) = x OPER y
364            (* with OPER \in div, mod, quot, rem *)
365    
366    
367        OPER | nodes(old) | edges(old) | nodes(new) | edges(new)
368        --------------------------------------------------------
369         div |         24 |         39 |         12 |         16
370         mod |         41 |         71 |         12 |         16
371        quot |          8 |         10 |          8 |         10
372         rem |         10 |         14 |          8 |         10
373    
374    
375    ----------------------------------------------------------------------
376    Name: Matthias Blume
377    Date: 2002/03/25 22:06:00 EST
378    Tag: blume-20020325-cprotobug
379    Description:
380    
381    Fixed a bug in cproto (c prototype decoder).
382    
383    ----------------------------------------------------------------------
384    Name: Matthias Blume
385    Date: 2002/03/25 16:00:00 EST
386    Tag: blume-20020325-raw-primops
387    Description:
388    
389    I did some cleanup to Allen's new primop code and
390    replaced yesterday's bootfiles with new ones.
391    (But they are stored in the same place.)
392    
393    ----------------------------------------------------------------------
394    Name: Matthias Blume
395    Date: 2002/03/24 22:40:00 EST
396    Tag: blume-20020324-bootfiles
397    Description:
398    
399    Made the bootfiles that Allen asked for.
400    
401    ----------------------------------------------------------------------
402    Name: Allen Leung
403    Date: 2002/03/23 15:50:00 EST
404    Tag: leunga-20020323-flint-cps-rcc-primops
405    Description:
406    
407      1. Changes to FLINT primops:
408    
409        (* make a call to a C-function;
410         * The primop carries C function prototype information and specifies
411         * which of its (ML-) arguments are floating point. C prototype
412         * information is for use by the backend, ML information is for
413         * use by the CPS converter. *)
414      | RAW_CCALL of { c_proto: CTypes.c_proto,
415                       ml_args: ccall_type list,
416                       ml_res_opt: ccall_type option,
417                       reentrant : bool
418                     } option
419       (* Allocate uninitialized storage on the heap.
420        * The record is meant to hold short-lived C objects, i.e., they
421        * are not ML pointers.  With the tag, the representation is
422        * the same as RECORD with tag tag_raw32 (sz=4), or tag_fblock (sz=8)
423        *)
424      | RAW_RECORD of {tag:bool,sz:int}
425      and ccall_type = CCALL_INT32 | CCALL_REAL64 | CCALL_ML_PTR
426    
427      2.  These CPS primops are now overloaded:
428    
429           rawload of {kind:numkind}
430           rawstore of {kind:numkind}
431    
432          The one argument form is:
433    
434             rawload {kind} address
435    
436          The two argument form is:
437    
438             rawload {kind} [ml object, byte-offset]
439    
440      3. RAW_CCALL/RCC now takes two extra arguments:
441    
442         a. The first is whether the C call is reentrant, i.e., whether
443            ML state should be saved and restored.
444         b. The second argument is a string argument specifying the name of
445            library and the C function.
446    
447         These things are currently not handled in the code generator, yet.
448    
449      4. In CProto,
450    
451         An encoding type of "bool" means "ml object" and is mapped into
452         C prototype of PTR.  Note that "bool" is different than "string",
453         even though "string" is also mapped into PTR, because "bool"
454         is assigned an CPS type of BOGt, while "string" is assigned INT32t.
455    
456      5. Pickler/unpicker
457    
458         Changed to handle RAW_RECORD and newest RAW_CCALL
459    
460      6. MLRiscGen,
461    
462         1. Changed to handle the new rawload/rawstore/rawrecord operators.
463         2. Code for handling C Calls has been moved to a new module CPSCCalls,
464            in the file CodeGen/cpscompile/cps-c-calls.sml
465    
466      7. Added the conditional move operator
467    
468             condmove of branch
469    
470         to cps.  Generation of this is still buggy so it is currently
471         disabled.
472    
473    ----------------------------------------------------------------------
474    Name: Lal George
475    Date: 2002/03/22 14:18:25 EST
476    Tag: george-20020322-cps-branch-prob
477    Description:
478    
479    Implemented the Ball-Larus branch prediction-heuristics, and
480    incorporated graphical viewers for control flow graphs.
481    
482    Ball-Larus Heuristics:
483    ---------------------
484    See the file compiler/CodeGen/cpscompile/cpsBranchProb.sml.
485    
486    By design it uses the Dempster-Shafer theory for combining
487    probabilities.  For example, in the function:
488    
489        fun f(n,acc) = if n = 0 then acc else f(n-1, n*acc)
490    
491    the ball-larus heuristics predicts that the n=0 is unlikely
492    (OH-heuristic), and the 'then' branch is unlikely because of the
493    RH-heuristic -- giving the 'then' branch an even lower combined
494    probability using the Dempster-Shafer theory.
495    
496    Finally, John Reppy's loop analysis in MLRISC, further lowers the
497    probability of the 'then' branch because of the loop in the else
498    branch.
499    
500    
501    Graphical Viewing:
502    ------------------
503    I merely plugged in Allen's graphical viewers into the compiler. The
504    additional code is not much. At the top level, saying:
505    
506            Control.MLRISC.getFlag "cfg-graphical-view" := true;
507    
508    will display the graphical view of the control flow graph just before
509    back-patching.  daVinci must be in your path for this to work. If
510    daVinci is not available, then the default viewer can be changed
511    using:
512    
513            Control.MLRISC.getString "viewer"
514    
515    which can be set to "dot" or "vcg" for the corresponding viewers. Of
516    course, these viewers must be in your path.
517    
518    The above will display the compilation unit at the level of clusters,
519    many of which are small, boring, and un-interesting. Also setting:
520    
521            Control.MLRISC.getInt "cfg-graphical-view_size"
522    
523    will display clusters that are larger than the value set by the above.
524    
525    
526    ----------------------------------------------------------------------
527    Name: Matthias Blume
528    Date: 2002/03/21 22:20:00 EST
529    Tag: blume-20020321-kmp-bugfix
530    Description:
531    
532    Changed the interface to the KMP routine in PreString and fixed
533    a minor bug in one place where it was used.
534    
535    ----------------------------------------------------------------------
536    Name: Allen Leung
537    Date: 2002/03/21 20:30:00 EST
538    Tag: leunga-20020321-cfg
539    Description:
540    
541      Fixed a potential problem in cfg edge splitting.
542    
543    ----------------------------------------------------------------------
544    Name: Allen Leung
545    Date: 2002/03/21 17:15:00 EST
546    Tag: leunga-20020321-x86-fp-cfg
547    Description:
548    
549      1. Recoded the buggy parts of x86-fp.
550    
551         a. All the block reordering code has been removed.
552            We now depend on the block placement phases to do this work.
553    
554         b. Critical edge splitting code has been simplified and moved into the
555            CFG modules, as where they belong.
556    
557         Both of these were quite buggy and complex.  The code is now much, much
558         simpler.
559    
560      2. X86 backend.
561    
562         a. Added instructions for 64-bit support.  Instruction selection for
563            64-bit has not been committed, however, since that
564            requires changes to MLTREE which haven't been approved by
565            Lal and John.
566    
567         b. Added support for FUCOMI and FUCOMIP when generating code for
568            PentiumPro and above.  We only generate these instructions in
569            the fast-fp mode.
570    
571         c. Added cases for JP and JNP in X86FreqProps.
572    
573      3. CFG
574    
575         CFG now has a bunch of methods for edge splitting and merging.
576    
577      4. Machine description.
578    
579         John's simplification of MLTREE_BASIS.fcond broke a few machine
580         description things:
581    
582         rtl-build.{sig,sml} and hppa.mdl fixed.
583    
584         NOTE: the machine description stuff in the repository is still broken.
585               Again, I can't put my fixes in because that involves
586               changes to MLTREE.
587    
588    ----------------------------------------------------------------------
589    Name: Matthias Blume
590    Date: 2002/03/20 15:55:00 EST
591    Tag: blume-20020320-kmp
592    Description:
593    
594    Implemented Knuth-Morris-Pratt string matching in PreString and used
595    it for String.isSubstring, Substring.isSubstring, and
596    Substring.position.
597    
598    (Might need some stress-testing.  Simple examples worked fine.)
599    
600    ----------------------------------------------------------------------
601    Name: Matthias Blume
602    Date: 2002/03/19 16:37:00 EST
603    Tag: blume-20020319-witnesses
604    Description:
605    
606    Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
607    
608    This implements a generic mechanism for changing constness qualifiers
609    anywhere within big C types without resorting to outright "casts".
610    (So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
611    modify the constness at the outermost level.)
612    The implementation of "convert" is based on the idea of "witness"
613    values -- values that are not used by the operation but whose types
614    "testify" to their applicability.  On the implementation side, "convert"
615    is simply a projection (returning its second curried argument).  With
616    cross-module inlining, it should not result in any machine code being
617    generated.
618    
619    ----------------------------------------------------------------------
620    Name: Matthias Blume
621    Date: 2002/03/15 16:40:00 EST
622    Tag: blume-20020315-basis
623    Description:
624    
625    Provided (preliminary?) implementations for
626    
627      {String,Substring}.{concatWith,isSuffix,isSubstring}
628    
629    and
630    
631      Substring.full
632    
633    Those are in the Basis spec but they were missing in SML/NJ.
634    
635    ----------------------------------------------------------------------
636    Name: Matthias Blume
637    Date: 2002/03/14 21:30:00 EST
638    Tag: blume-20020314-controls
639    Description:
640    
641    Controls:
642    ---------
643    
644    1. Factored out the recently-added Controls : CONTROLS stuff and put
645       it into its own library $/controls-lib.cm.  The source tree for
646       this is under src/smlnj-lib/Controls.
647    
648    2. Changed the names of types and functions in this interface, so they
649       make a bit more "sense":
650    
651          module -> registry
652          'a registry -> 'a group
653    
654    3. The interface now deals in ref cells only.  The getter/setter interface
655       is (mostly) gone.
656    
657    4. Added a function that lets one register an already-existing ref cell.
658    
659    5. Made the corresponding modifications to the rest of the code so that
660       everything compiles again.
661    
662    6. Changed the implementation of Controls.MLRISC back to something closer
663       to the original.  In particular, this module (and therefore MLRISC)
664       does not depend on Controls.  There now is some link-time code in
665       int-sys.sml that registers the MLRISC controls with the Controls
666       module.
667    
668    CM:
669    ---
670    
671      * One can now specify the lambda-split aggressiveness in init.cmi.
672    
673    ----------------------------------------------------------------------
674    Name: Allen Leung
675    Date: 2002/03/13 17:30:00 EST
676    Tag: leunga-20020313-x86-fp-unary
677    Description:
678    
679    Bug fix for:
680    
681    > leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
682    > Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
683    > - fun f(x,(y,z)) = Real.~ y;
684    > [autoloading]
685    > [autoloading done]
686    >       fchsl   (%eax), 184(%esp)
687    > Error: MLRisc bug: X86MCEmitter.emitInstr
688    >
689    > uncaught exception Error
690    >   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
691    
692    The problem was that the code generator did not generate any fp registers
693    in this case, and the ra didn't know that it needed to run the X86FP phase to
694    translate the pseudo fp instruction.   This only happened with unary fp
695    operators in certain situations.
696    
697    ----------------------------------------------------------------------
698    Name: Matthias Blume
699    Date: 2002/03/13 14:00:00 EST
700    Tag: blume-20020313-overload-etc
701    Description:
702    
703    1. Added _overload as a synonym for overload for backward compatibility.
704       (Control.overloadKW must be true for either version to be accepted.)
705    
706    2. Fixed bug in install script that caused more things to be installed
707       than what was requested in config/targets.
708    
709    3. Made CM aware of the (_)overload construct so that autoloading
710       works.
711    
712    ----------------------------------------------------------------------
713    Name: Matthias Blume
714    Date: 2002/03/12 22:03:00 EST
715    Tag: blume-20020312-url
716    Description:
717    
718    Forgot to update BOOT and srcarchiveurl.
719    
720    ----------------------------------------------------------------------
721    Name: Matthias Blume
722    Date: 2002/03/12 17:30:00 EST
723    Tag: blume-20020312-version110392
724    Description:
725    
726    Yet another version number bump (because of small changes to the
727    binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
728    
729    Changes:
730    
731      The new pid generation scheme described a few weeks ago was overly
732      complicated.  I implemented a new mechanism that is simpler and
733      provides a bit more "stability":  Once CM has seen a compilation
734      unit, it keeps its identity constant (as long as you do not delete
735      those crucial CM/GUID/* files).  This means that when you change
736      an interface, compile, then go back to the old interface, and
737      compile again, you arrive at the original pid.
738    
739      There now also is a mechanism that instructs CM to use the plain
740      environment hash as a module's pid (effectively making its GUID
741      the empty string).  For this, "noguid" must be specified as an
742      option to the .sml file in question within its .cm file.
743      This is most useful for code that is being generated by tools such
744      as ml-nlffigen (because during development programmers tend to
745      erase the tool's entire output directory tree including CM's cached
746      GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
747      revert to the old, broken behavior of SML/NJ, but in specific cases
748      where there is no danger of interface confusion, its use is ok
749      (I think).
750    
751      ml-nlffigen by default generates "noguid" annotations.  They can be
752      turned off by specifying -guid in its command line.
753    
754    ----------------------------------------------------------------------
755    Name: Lal George
756    Date: 2002/03/12 12 14:42:36 EST
757    Tag: george-20020312-frequency-computation
758    Description:
759    
760    Integrated jump chaining and static block frequency into the
761    compiler. More details and numbers later.
762    
763    ----------------------------------------------------------------------
764    Name: Lal George
765    Date: 2002/03/11 11 22:38:53 EST
766    Tag: george-20020311-jump-chain-elim
767    Description:
768    
769    Tested the jump chain elimination on all architectures (except the
770    hppa).  This is on by default right now and is profitable for the
771    alpha and x86, however, it may not be profitable for the sparc and ppc
772    when compiling the compiler.
773    
774    The gc test will typically jump to a label at the end of the cluster,
775    where there is another jump to an external cluster containing the actual
776    code to invoke gc. This is to allow factoring of common gc invocation
777    sequences. That is to say, we generate:
778    
779            f:
780               testgc
781               ja   L1      % jump if above to L1
782    
783            L1:
784               jmp L2
785    
786    
787    After jump chain elimination the 'ja L1' instructions is converted to
788    'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
789    up being implemented in their long form (if L2 is far away) using:
790    
791            jbe     L3      % jump if below or equal to L3
792            jmp     L2
793         L3:
794            ...
795    
796    
797    For large compilation units L2  may be far away.
798    
799    
800    ----------------------------------------------------------------------
801    Name: Matthias Blume
802    Date: 2002/03/11 13:30:00 EST
803    Tag: blume-20020311-mltreeeval
804    Description:
805    
806    A functor parameter was missing.
807    
808    ----------------------------------------------------------------------
809    Name: Allen Leung
810    Date: 2002/03/11 10:30:00 EST
811    Tag: leunga-20020311-runtime-string0
812    Description:
813    
814       The representation of the empty string now points to a
815    legal null terminated C string instead of unit.  It is now possible
816    to convert an ML string into C string with InlineT.CharVector.getData.
817    This compiles into one single machine instruction.
818    
819    ----------------------------------------------------------------------
820    Name: Allen Leung
821    Date: 2002/03/10 23:55:00 EST
822    Tag: leunga-20020310-x86-call
823    Description:
824    
825       Added machine generation for CALL instruction (relative displacement mode)
826    
827    ----------------------------------------------------------------------
828    Name: Matthias Blume
829    Date: 2002/03/08 16:05:00
830    Tag: blume-20020308-entrypoints
831    Description:
832    
833    Version number bumped to 110.39.1.  NEW BOOTFILES!
834    
835    Entrypoints: non-zero offset into a code object where execution should begin.
836    
837    - Added the notion of an entrypoint to CodeObj.
838    - Added reading/writing of entrypoint info to Binfile.
839    - Made runtime system bootloader aware of entrypoints.
840    - Use the address of the label of the first function given to mlriscGen
841      as the entrypoint.  This address is currently always 0, but it will
842      not be 0 once we turn on block placement.
843    - Removed the linkage cluster code (which was The Other Way(tm) of dealing
844      with entry points) from mlriscGen.
845    
846    ----------------------------------------------------------------------
847    Name: Allen Leung
848    Date: 2002/03/07 20:45:00 EST
849    Tag: leunga-20020307-x86-cmov
850    Description:
851    
852       Bug fixes for CMOVcc on x86.
853    
854       1. Added machine code generation for CMOVcc
855       2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
856       3. CMOVcc cannot have an immediate operand as argument.
857    
858    ----------------------------------------------------------------------
859    Name: Matthias Blume
860    Date: 2002/03/07 16:15:00 EST
861    Tag: blume-20020307-controls
862    Description:
863    
864    This is a very large but mostly boring patch which makes (almost)
865    every tuneable compiler knob (i.e., pretty much everything under
866    Control.* plus a few other things) configurable via both the command
867    line and environment variables in the style CM did its configuration
868    until now.
869    
870    Try starting sml with '-h' (or, if you are brave, '-H')
871    
872    To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
873    implements the underlying generic mechanism.
874    
875    The interface to some of the existing such facilities has changed somewhat.
876    For example, the MLRiscControl module now provides mkFoo instead of getFoo.
877    (The getFoo interface is still there for backward-compatibility, but its
878    use is deprecated.)
879    
880    The ml-build script passes -Cxxx=yyy command-line arguments through so
881    that one can now twiddle the compiler settings when using this "batch"
882    compiler.
883    
884    TODO items:
885    
886    We should go through and throw out all controls that are no longer
887    connected to anything.  Moreover, we should go through and provide
888    meaningful (and correct!) documentation strings for those controls
889    that still are connected.
890    
891    Currently, multiple calls to Controls.new are accepted (only the first
892    has any effect).  Eventually we should make sure that every control
893    is being made (via Controls.new) exactly once.  Future access can then
894    be done using Controls.acc.
895    
896    Finally, it would probably be a good idea to use the getter-setter
897    interface to controls rather than ref cells.  For the time being, both
898    styles are provided by the Controls module, but getter-setter pairs are
899    better if thread-safety is of any concern because they can be wrapped.
900    
901    *****************************************
902    
903    One bug fix: The function blockPlacement in three of the MLRISC
904    backpatch files used to be hard-wired to one of two possibilities at
905    link time (according to the value of the placementFlag).  But (I
906    think) it should rather sense the flag every time.
907    
908    *****************************************
909    
910    Other assorted changes (by other people who did not supply a HISTORY entry):
911    
912    1. the cross-module inliner now works much better (Monnier)
913    2. representation of weights, frequencies, and probabilities in MLRISC
914       changed in preparation of using those for weighted block placement
915       (Reppy, George)
916    
917    ----------------------------------------------------------------------
918    Name: Lal George
919    Date: 2002/03/07 14:44:24 EST 2002
920    Tag: george-20020307-weighted-block-placement
921    
922    Tested the weighted block placement optimization on all architectures
923    (except the hppa) using AMPL to generate the block and edge frequencies.
924    Changes were required in the machine properties to correctly
925    categorize trap instructions. There is an MLRISC flag
926    "weighted-block-placement" that can be used to enable weighted block
927    placement, but this will be ineffective without block/edge
928    frequencies (coming soon).
929    
930    
931    ----------------------------------------------------------------------
932    Name: Lal George
933    Date: 2002/03/05 17:24:48 EST
934    Tag: george-20020305-linkage-cluster
935    
936    In order to support the block placement optimization, a new cluster
937    is generated as the very first cluster (called the linkage cluster).
938    It contains a single jump to the 'real' entry point for the compilation
939    unit. Block placement has no effect on the linkage cluster itself, but
940    all the other clusters  have full freedom in the manner in which they
941    reorder blocks or functions.
942    
943    On the x86 the typical linkage code that is generated is:
944       ----------------------
945            .align 2
946       L0:
947            addl    $L1-L0, 72(%esp)
948            jmp     L1
949    
950    
951            .align  2
952       L1:
953       ----------------------
954    
955    72(%esp) is the memory location for the stdlink register. This
956    must contain the address of the CPS function being called. In the
957    above example, it contains the address of  L0; before
958    calling L1 (the real entry point for the compilation unit), it
959    must contain the address for L1, and hence
960    
961            addl $L1-L0, 72(%esp)
962    
963    I have tested this on all architectures except the hppa.The increase
964    in code size is of course negligible
965    
966    ----------------------------------------------------------------------
967    Name: Allen Leung
968    Date: 2002/03/03 13:20:00 EST
969    Tag: leunga-20020303-mlrisc-tools
970    
971      Added #[ ... ] expressions to mlrisc tools
972    
973    ----------------------------------------------------------------------
974    Name: Matthias Blume
975    Date: 2002/02/27 12:29:00 EST
976    Tag: blume-20020227-cdebug
977    Description:
978    
979    - made types in structure C and C_Debug to be equal
980    - got rid of code duplication (c-int.sml vs. c-int-debug.sml)
981    - there no longer is a C_Int_Debug (C_Debug is directly derived from C)
982    
983    ----------------------------------------------------------------------
984    Name: Matthias Blume
985    Date: 2002/02/26 12:00:00 EST
986    Tag: blume-20020226-ffi
987    Description:
988    
989    1. Fixed a minor bug in CM's "noweb" tool:
990       If numbering is turned off, then truly don't number (i.e., do not
991       supply the -L option to noweb).  The previous behavior was to supply
992       -L'' -- which caused noweb to use the "default" line numbering scheme.
993       Thanks to Chris Richards for pointing this out (and supplying the fix).
994    
995    2. Once again, I reworked some aspects of the FFI:
996    
997       A. The incomplete/complete type business:
998    
999       - Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
1000         gone!
1001       - ML types representing an incomplete type are now *equal* to
1002         ML types representing their corresponding complete types (just like
1003         in C).  This is still safe because ml-nlffigen will not generate
1004         RTTI for incomplete types, nor will it generate functions that
1005         require access to such RTTI.   But when ML code generated from both
1006         incomplete and complete versions of the C type meet, the ML types
1007         are trivially interoperable.
1008    
1009         NOTE:  These changes restore the full generality of the translation
1010         (which was previously lost when I eliminated functorization)!
1011    
1012       B. Enum types:
1013    
1014       - Structure C now has a type constructor "enum" that is similar to
1015         how the "su" constructor works.  However, "enum" is not a phantom
1016         type because each "T enum" has values (and is isomorphic to
1017         MLRep.Signed.int).
1018       - There are generic access operations for enum objects (using
1019         MLRep.Signed.int).
1020       - ml-nlffigen will generate a structure E_foo for each "enum foo".
1021         * The structure contains the definition of type "mlrep" (the ML-side
1022         representation type of the enum).  Normally, mlrep is the same
1023         as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
1024         then mlrep will be defined as a datatype -- thus facilitating
1025         pattern matching on mlrep values.
1026         ("-ec" will be suppressed if there are duplicate values in an
1027          enumeration.)
1028         * Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
1029         will be generated for each C enum constant xxx.
1030         * Conversion functions m2i and i2m convert between mlrep and
1031         MLRep.Signed.int.  (Without "-ec", these functions are identities.)
1032         * Coversion functions c and ml convert between mlrep and "tag enum".
1033         * Access functions (get/set) fetch and store mlrep values.
1034       - By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
1035         enumerations are merged into one single enumeration represented by
1036         structure E_'.
1037    
1038    ----------------------------------------------------------------------
1039    Name: Allen Leung
1040    Date: 2002/02/25 04:45:00 EST
1041    Tag: leunga-20020225-cps-spill
1042    
1043    This is a new implementation of the CPS spill phase.
1044    The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
1045    In case of problems, replace it with the old file spill.sml
1046    
1047    The current compiler runs into some serious performance problems when
1048    constructing a large record.  This can happen when we try to compile a
1049    structure with many items.  Even a very simple structure like the following
1050    makes the compiler slow down.
1051    
1052        structure Foo = struct
1053           val x_1 = 0w1 : Word32.int
1054           val x_2 = 0w2 : Word32.int
1055           val x_3 = 0w3 : Word32.int
1056           ...
1057           val x_N = 0wN : Word32.int
1058        end
1059    
1060    The following table shows the compile time, from N=1000 to N=4000,
1061    with the old compiler:
1062    
1063    N
1064    1000   CPS 100 spill                           0.04u  0.00s  0.00g
1065           MLRISC ra                               0.06u  0.00s  0.05g
1066              (spills = 0 reloads = 0)
1067           TOTAL                                   0.63u  0.07s  0.21g
1068    
1069    1100   CPS 100 spill                           8.25u  0.32s  0.64g
1070           MLRISC ra                               5.68u  0.59s  3.93g
1071              (spills = 0 reloads = 0)
1072           TOTAL                                   14.71u  0.99s  4.81g
1073    
1074    1500   CPS 100 spill                           58.55u  2.34s  1.74g
1075           MLRISC ra                               5.54u  0.65s  3.91g
1076              (spills = 543 reloads = 1082)
1077           TOTAL                                   65.40u  3.13s  6.00g
1078    
1079    2000   CPS 100 spill                           126.69u  4.84s  3.08g
1080           MLRISC ra                               0.80u  0.10s  0.55g
1081              (spills = 42 reloads = 84)
1082           TOTAL                                   129.42u  5.10s  4.13g
1083    
1084    3000   CPS 100 spill                           675.59u  19.03s  11.64g
1085           MLRISC ra                               2.69u  0.27s  1.38g
1086              (spills = 62 reloads = 124)
1087           TOTAL                                   682.48u  19.61s  13.99g
1088    
1089    4000   CPS 100 spill                           2362.82u  56.28s  43.60g
1090           MLRISC ra                               4.96u  0.27s  2.72g
1091              (spills = 85 reloads = 170)
1092           TOTAL                                   2375.26u  57.21s  48.00g
1093    
1094    As you can see the old cps spill module suffers from some serious
1095    performance problem.  But since I cannot decipher the old code fully,
1096    instead of patching the problems up, I'm reimplementing it
1097    with a different algorithm.  The new code is more modular,
1098    smaller when compiled, and substantially faster
1099    (O(n log n) time and O(n) space).  Timing of the new spill module:
1100    
1101    4000  CPS 100 spill                           0.02u  0.00s  0.00g
1102          MLRISC ra                               0.25u  0.02s  0.15g
1103             (spills=1 reloads=3)
1104          TOTAL                                   7.74u  0.34s  1.62g
1105    
1106    Implementation details:
1107    
1108    As far as I can tell, the purpose of the CPS spill module is to make sure the
1109    number of live variables at any program point (the bandwidth)
1110    does not exceed a certain limit, which is determined by the
1111    size of the spill area.
1112    
1113    When the bandwidth is too large, we decrease the register pressure by
1114    packing live variables into spill records.  How we achieve this is
1115    completely different than what we did in the old code.
1116    
1117    First, there is something about the MLRiscGen code generator
1118    that we should be aware of:
1119    
1120    o MLRiscGen performs code motion!
1121    
1122       In particular, it will move floating point computations and
1123       address computations involving only the heap pointer to
1124       their use sites (if there is only a single use).
1125       What this means is that if we have a CPS record construction
1126       statement
1127    
1128           RECORD(k,vl,w,e)
1129    
1130       we should never count the new record address w as live if w
1131       has only one use (which is often the case).
1132    
1133       We should do something similar to floating point, but the transformation
1134       there is much more complex, so I won't deal with that.
1135    
1136    Secondly, there are now two new cps primops at our disposal:
1137    
1138     1. rawrecord of record_kind option
1139        This pure operator allocates some uninitialized storage from the heap.
1140        There are two forms:
1141    
1142         rawrecord NONE [INT n]  allocates a tagless record of length n
1143         rawrecord (SOME rk) [INT n] allocates a tagged record of length n
1144                                     and initializes the tag.
1145    
1146     2. rawupdate of cty
1147          rawupdate cty (v,i,x)
1148          Assigns to x to the ith component of record v.
1149          The storelist is not updated.
1150    
1151    We use these new primops for both spilling and increment record construction.
1152    
1153     1. Spilling.
1154    
1155        This is implemented with a linear scan algorithm (but generalized
1156        to trees).  The algorithm will create a single spill record at the
1157        beginning of the cps function and use rawupdate to spill to it,
1158        and SELECT or SELp to reload from it.  So both spills and reloads
1159        are fine-grain operations.  In contrast, in the old algorithm
1160        "spills" have to be bundled together in records.
1161    
1162        Ideally, we should sink the spill record construction to where
1163        it is needed.  We can even split the spill record into multiple ones
1164        at the places where they are needed.  But CPS is not a good
1165        representation for global code motion, so I'll keep it simple and
1166        am not attempting this.
1167    
1168     2. Incremental record construction (aka record splitting).
1169    
1170        Long records with many component values which are simulatenously live
1171        (recall that single use record addresses are not considered to
1172         be live) are constructed with rawrecord and rawupdate.
1173        We allocate space on the heap with rawrecord first, then gradually
1174        fill it in with rawupdate.  This is the technique suggested to me
1175        by Matthias.
1176    
1177        Some restrictions on when this is applicable:
1178        1. It is not a VECTOR record.  The code generator currently does not handle
1179           this case. VECTOR record uses double indirection like arrays.
1180        2. All the record component values are defined in the same "basic block"
1181           as the record constructor.  This is to prevent speculative
1182           record construction.
1183    
1184    ----------------------------------------------------------------------
1185    Name: Allen Leung
1186    Date: 2002/02/22 01:02:00 EST
1187    Tag: leunga-20020222-mlrisc-tools
1188    
1189    Minor bug fixes in the parser and rewriter
1190    
1191    ----------------------------------------------------------------------
1192    Name: Allen Leung
1193    Date: 2002/02/21 20:20:00 EST
1194    Tag: leunga-20020221-peephole
1195    
1196    Regenerated the peephole files.  Some contained typos in the specification
1197    and some didn't compile because of pretty printing bugs in the old version
1198    of 'nowhere'.
1199    
1200    ----------------------------------------------------------------------
1201    Name: Allen Leung
1202    Date: 2002/02/19 20:20:00 EST
1203    Tag: leunga-20020219-mlrisc-tools
1204    Description:
1205    
1206       Minor bug fixes to the mlrisc-tools library:
1207    
1208       1.  Fixed up parsing colon suffixed keywords
1209       2.  Added the ability to shut the error messages up
1210       3.  Reimplemented the pretty printer and fixed up/improved
1211           the pretty printing of handle and -> types.
1212       4.  Fixed up generation of literal symbols in the nowhere tool.
1213       5.  Added some SML keywords to to sml.sty
1214    
1215    ----------------------------------------------------------------------
1216    Name: Matthias Blume
1217    Date: 2002/02/19 16:20:00 EST
1218    Tag: blume-20020219-cmffi
1219    Description:
1220    
1221    A wild mix of changes, some minor, some major:
1222    
1223    * All C FFI-related libraries are now anchored under $c:
1224        $/c.cm      --> $c/c.cm
1225        $/c-int.cm  --> $c/internals/c-int.cm
1226        $/memory.cm --> $c/memory/memory.cm
1227    
1228    * "make" tool (in CM) now treats its argument pathname slightly
1229      differently:
1230        1. If the native expansion is an absolute name, then before invoking
1231           the "make" command on it, CM will apply OS.Path.mkRelative
1232           (with relativeTo = OS.FileSys.getDir()) to it.
1233        2. The argument will be passed through to subsequent phases of CM
1234           processing without "going native".  In particular, if the argument
1235           was an anchored path, then "make" will not lose track of that anchor.
1236    
1237    * Compiler backends now "know" their respective C calling conventions
1238      instead of having to be told about it by ml-nlffigen.  This relieves
1239      ml-nlffigen from one of its burdens.
1240    
1241    * The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
1242    
1243    * Export C_DEBUG and C_Debug from $c/c.cm.
1244    
1245    * C type encoding in ml-nlffi-lib has been improved to model the conceptual
1246      subtyping relationship between incomplete pointers and their complete
1247      counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
1248      with the convention of instantiating 'o with ('t, 'c) obj whenever
1249      the pointer target type is complete.  In the incomplete case, 'o
1250      will be instantiated with some "'c iobj" -- a type obtained by
1251      using one of the functors PointerToIncompleteType or PointerToCompleteType.
1252    
1253      Operations that work on both incomplete and complete pointer types are
1254      typed as taking an 'o ptr while operations that require the target to
1255      be known are typed as taking some ('t, 'c) obj ptr.
1256    
1257      voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
1258      where void is an eqtype without any values.  This makes it possible
1259      to work on voidptr values using functions meant to operate on light
1260      incomplete pointers.
1261    
1262    * As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
1263      been vastly simplified.
1264    
1265    ----------------------------------------------------------------------
1266    Name: Matthias Blume
1267    Date: 2002/02/19 10:48:00 EST
1268    Tag: blume-20020219-pqfix
1269    Description:
1270    
1271    Applied Chris Okasaki's bug fix for priority queues.
1272    
1273    ----------------------------------------------------------------------
1274    Name: Matthias Blume
1275    Date: 2002/02/15 17:05:00
1276    Tag: Release_110_39
1277    Description:
1278    
1279    Last-minute retagging is becoming a tradition... :-(
1280    
1281    This is the working release 110.39.
1282    
1283    ----------------------------------------------------------------------
1284    Name: Matthias Blume
1285    Date: 2002/02/15 16:00:00 EST
1286    Tag: Release_110_39-orig
1287    Description:
1288    
1289    Working release 110.39.  New bootfiles.
1290    
1291    (Update: There was a small bug in the installer so it wouldn't work
1292    with all shells.  So I retagged. -Matthias)
1293    
1294    ----------------------------------------------------------------------
1295    Name: Matthias Blume
1296    Date: 2002/02/15 14:17:00 EST
1297    Tag: blume-20020215-showbindings
1298    Description:
1299    
1300    Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
1301    the latter can be useful for exploring what bindings are available at
1302    the interactive prompt.  (The first function returns only the list
1303    of symbols that are really bound, the second prints those but also the
1304    ones that CM's autoloading mechanism knows about.)
1305    
1306    ----------------------------------------------------------------------
1307    Name: Matthias Blume
1308    Date: 2002/02/15 12:08:00 EST
1309    Tag: blume-20020215-iptrs
1310    Description:
1311    
1312    Two improvements to ml-nlffigen:
1313    
1314      1. Write files only if they do not exist or if their current contents
1315         do not coincide with what's being written.  (That is, avoid messing
1316         with the time stamps unless absolutely necessary.)
1317    
1318      2. Implement a "repository" mechanism for generated files related
1319         to "incomplete pointer types".   See the README file for details.
1320    
1321    ----------------------------------------------------------------------
1322    Name: Matthias Blume
1323    Date: 2002/02/14 11:50:00 EST
1324    Tag: blume-20020214-quote
1325    Description:
1326    
1327    Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
1328    because of the new and improved tag generation scheme.  (Thanks to Allen
1329    Leung for pointing it out.)
1330    
1331    ----------------------------------------------------------------------
1332    Name: Lal George
1333    Date: 2002/02/14 09:55:27 EST 2002
1334    Tag: george-20020214-isabelle-bug
1335    Description:
1336    
1337    Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
1338    of Isabelle on the x86.
1339    
1340    From Allen:
1341    -----------
1342     I've found the problem:
1343    
1344         in ra-core.sml, I use the counter "blocked" to keep track of the
1345         true number of elements in the freeze queue.  When the counter goes
1346         to zero, I skip examining the queue.  But I've messed up the
1347         bookkeeping in combine():
1348    
1349             else ();
1350             case !ucol of
1351               PSEUDO => (if !cntv > 0 then
1352                     (if !cntu > 0 then blocked := !blocked - 1 else ();
1353                                        ^^^^^^^^^^^^^^^^^^^^^^^
1354                      moveu := mergeMoveList(!movev, !moveu)
1355                     )
1356                  else ();
1357    
1358         combine() is called to coalesce two nodes u and v.
1359         I think I was thinking that if the move counts of u and v are both
1360         greater than zero then after they are coalesced then one node is
1361         removed from the freeze queue.  Apparently I was thinking that
1362         both u and v are of low degree, but that's clearly not necessarily true.
1363    
1364    
1365    02/12/2002:
1366        Here's the patch.  HOL now compiles.
1367    
1368        I don't know how this impact on performance (compile
1369        time or runtime).  This bug caused the RA (especially on the x86)
1370        to go thru the potential spill phase when there are still nodes on the
1371        freeze queue.
1372    
1373    
1374    
1375    
1376    ----------------------------------------------------------------------
1377    Name: Matthias Blume
1378    Date: 2002/02/13 22:40:00 EST
1379    Tag: blume-20020213-fptr-rtti
1380    Description:
1381    
1382    Fixed a bug in ml-nlffigen that was introduced with one of the previous
1383    updates.
1384    
1385    ----------------------------------------------------------------------
1386    Name: Matthias Blume
1387    Date: 2002/02/13 16:41:00 EST
1388    Tag: blume-20020213-cmlpq
1389    Description:
1390    
1391    Added new priority queue export symbols (which have just been added to
1392    smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
1393    not compile and the installer would choke.)
1394    
1395    ----------------------------------------------------------------------
1396    Name: Matthias Blume
1397    Date: 2002/02/13 16:15:00 EST
1398    Tag: blume-20020213-various
1399    Description:
1400    
1401    1. More tweaks to ml-nlffigen:
1402    
1403       - better internal datastructures (resulting in slight speedup)
1404       - "-match" option requires exact match
1405       - "localized" gensym counters (untagged structs/unions nested within
1406         other structs/unions or within typedefs get a fresh counter; their
1407         tag will be prefixed by a concatenation of their parents' tags)
1408       - bug fixes (related to calculation of transitive closure of types
1409         to be included in the output)
1410    
1411    2. Minor Basis updates:
1412    
1413       - added implementations for List.collate and Option.app
1414    
1415    ----------------------------------------------------------------------
1416    Name: Matthias Blume
1417    Date: 2002/02/11 15:55:00 EST
1418    Tag: blume-20020211-gensym
1419    Description:
1420    
1421    Added a "-gensym" option to command line of ml-nlffigen.  This can be
1422    used to specify a "stem" -- a string that is inserted in all "gensym'd"
1423    names (ML structure names that correspond to unnamed C structs, unions,
1424    and enums), so that separate runs of ml-nlffigen do not clash.
1425    
1426    ----------------------------------------------------------------------
1427    Name: Matthias Blume
1428    Date: 2002/02/11 12:05:00 EST
1429    Tag: blume-20020211-gensml
1430    Description:
1431    
1432    A quick fix for a problem with GenSML (in the pgraph-util library):
1433    Make generation of toplevel "local" optional.  (Strictly speaking,
1434    signature definitions within "local" are not legal SML.)
1435    
1436    Other than that: updates to INSTALL and cm/TODO.
1437    
1438    ----------------------------------------------------------------------
1439    Name: Matthias Blume
1440    Date: 2002/02/08 15:00:00 EST
1441    Tag: blume-20020208-uniquepid
1442    Description:
1443    
1444    0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
1445    
1446    1. The installer (config/install.sh) has gotten smarter:
1447    
1448         - Configuration options are a bit easier to specify now
1449           (in config/targets).
1450         - Bug in recognizing .tar.bz2 files fixed.
1451         - Installer automatically resolves dependencies between
1452           configuration options (e.g., if you ask for eXene, you will
1453           also get cml -- regardless whether you asked for it or not).
1454         - Installer can run in "quieter mode" by setting the environment
1455           variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
1456           "completely silent", though.
1457         - Build HashCons library as part of smlnj-lib.
1458    
1459    2. A new scheme for assigning persistent identifiers to compilation
1460       units (and, by extension, to types etc.) has been put into place.
1461       This fixes a long-standing bug where types and even dynamic values
1462       can get internally confused, thereby compromising type safety
1463       (abstraction) and dynamic correctness.  See
1464    
1465         http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
1466    
1467       for an example of how things could go wrong until now.
1468    
1469       The downside of the new scheme is that pids are not quite as
1470       persistent as they used to be: CM will generate a fresh pid
1471       for every compilation unit that it thinks it sees for the first
1472       time.  That means that if you compile starting from a clean, fresh
1473       source tree at two different times, you end up with different
1474       binaries.
1475    
1476       Cutoff recompilation, however, has not been compromised because
1477       CM keeps pid information in special caches between runs.
1478    
1479    ----------------------------------------------------------------------
1480    Name: Lal George
1481    Date: 2002/02/07 15:34:13 EST 2002
1482    Tag: <none>
1483    Description:
1484    
1485    Compilers that generate assembly code may produce  global labels
1486    whose value is resolved at link time. The various peephole optimization
1487    modules did not take this in account.
1488    
1489    TODO. The Labels.addrOf function should really return an option
1490    type so that clients are forced to deal with this issue, rather
1491    than an exception being raised.
1492    
1493    ----------------------------------------------------------------------
1494    Name: Lal George
1495    Date: 2002/02/06 13:55:02 EST
1496    Tag: george-20020206-ra-breakup
1497    Description:
1498    
1499    1. A bug fix from Allen.
1500    
1501        A typo causes extra fstp %st(0)'s to be generated at compensation
1502        edges, which might cause stack underflow traps at runtime.  This
1503        occurs in fft where there are extraneous fstps right before the 'into'
1504        trap instruction (in this case they are harmless since none of the
1505        integers overflow.)
1506    
1507    2. Pulled out various utility modules that were embedded in the modules
1508       of the register allocator. I need these modules for other purposes, but
1509       they are not complete enough to put into a library (just yet).
1510    ----------------------------------------------------------------------
1511    Name: Matthias Blume
1512    Date: 2002/01/31 16:05:00 EST
1513    Tag: blume-20020131-sparc-ccalls
1514    Description:
1515    
1516    1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
1517       of extra stack space by mistake.  Fixed.
1518    
1519    2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
1520    
1521    ----------------------------------------------------------------------
1522    Name: Allen Leung
1523    Date: 2002/01/30
1524    Tag: leunga-20020130-nowhere-bug-fix
1525    Description:
1526    
1527       MLRISC bug fixes:
1528       1. Fixed a bindings computation bug in the 'nowhere' program generator tool.
1529       2. MachineInt.fromString was negating its value.
1530    
1531    ----------------------------------------------------------------------
1532    Name: Matthias Blume
1533    Date: 2002/01/29
1534    Tag: blume-20020129-INSTALL
1535    Description:
1536    
1537    - Added somewhat detailed installation instructions (file INSTALL).
1538    - Fixed curl-detection bug in config/install.sh.
1539    - It is now possible to select the URL getter using the URLGETTER
1540      environment variable:
1541    
1542          not set / "unknown"      --> automatic detection (script tries wget,
1543                                       curl, and lynx)
1544          "wget" / "curl" / "lynx" --> use the specified program (script "knows"
1545                                       how to properly invoke them)
1546          other                    --> use $URLGETTER directly, it must take
1547                                       precisely two command-line arguments
1548                                       (source URL and destination file name)
1549    
1550    ----------------------------------------------------------------------
1551    Name: Matthias Blume
1552    Date: 2002/01/28
1553    Tag: blume-20020128-sparc-ccalls
1554    Description:
1555    
1556    - Fixed problem with calculation of "used" registers in sparc-c-calls.
1557    - Make use of the allocParam argument in sparc-c-calls.
1558    
1559    ----------------------------------------------------------------------
1560    Name: Matthias Blume
1561    Date: 2002/01/28
1562    Tag: blume-20020128-allocParam
1563    Description:
1564    
1565    John Reppy:  Changes c-calls API to accept client-callback for
1566    allocating extra stack space.
1567    me: Corresponding changes to mlriscGen (using a dummy argument that
1568        does not change the current behavior).
1569    
1570    ----------------------------------------------------------------------
1571    Name: Matthias Blume
1572    Date: 2002/01/28 12:00:00
1573    Tag: Release_110_38
1574    Description:
1575    
1576    This time for real!!!
1577    
1578    ----------------------------------------------------------------------
1579    Name: Matthias Blume
1580    Date: 2002/01/28 10:56:00 EST
1581    Tag: blume-20020128-retraction
1582    Description:
1583    
1584    0. Retracted earlier 110.38.  (The Release_110_38 tag has been replaced
1585       with blume-Release_110_38-retracted.)
1586    
1587    1. Fixed a problem with incorrect rounding modes in real64.sml.
1588       (Thanks to Andrew Mccreight <andrew.mccreight@yale.edu>.)
1589    
1590    2. A bug in ml-nlffigen related to the handling of unnamed structs, unions,
1591       and enums fixed.  The naming of corresponding ML identifiers should
1592       now be consistent again.
1593    
1594    ----------------------------------------------------------------------
1595    Name: Allen Leung
1596    Date: 2002/01/27
1597    Tag: leunga-20020127-nowhere
1598    Description:
1599    
1600       Added a target called nowhere in the configuration scripts.
1601       Enabling this will build the MLRISC 'nowhere' tool (for translating
1602       programs with where-clauses into legal SML code) during installation.
1603    
1604    ----------------------------------------------------------------------
1605    Name: Matthias Blume
1606    Date: 2002/01/25 21:27:00 EST
1607    Tag: blume-Release_110_38-retracted
1608    Description:
1609    
1610    Call it a (working) release!  Version is 110.38. Bootfiles are ready.
1611    
1612    README will be added later.
1613    
1614    !!! NOTE:  Re-tagged as blume-Release_110_38-retracted. Original tag
1615    (Release_110_38) removed.  Reason: Last-minute bug fixes.
1616    
1617    ----------------------------------------------------------------------
1618    Name: Matthias Blume
1619    Date: 2002/01/25
1620    Tag: blume-20020125-ffi
1621    Description:
1622    
1623    A large number of tweaks and improvements to ml-nlffi-lib and
1624    ml-nlffigen:
1625    
1626       - ML represenation types have been streamlined
1627       - getter and setter functions work with concrete values, not abstract
1628         ones where possible
1629       - ml-nlffigen command line more flexible (see README file there)
1630       - some bugs have been fixed (hopefully)
1631    
1632    ----------------------------------------------------------------------
1633    Name: Lal George
1634    Date: 2002/01/24
1635    Tag: george-20020124-risc-ra-interface
1636    Description:
1637    
1638       There is a dramatic simplification in the interface to the
1639       register allocator for RISC architectures as a result of making
1640       parallel copy instructions explicit.
1641    
1642    ----------------------------------------------------------------------
1643    Name: Matthias Blume
1644    Date: 2002/01/22
1645    Tag: blume-20020122-x86-ccalls
1646    Description:
1647    
1648    Bug fix for c-calls on x86 (having to do with how char- and
1649    short-arguments are being handled).
1650    
1651    ----------------------------------------------------------------------
1652    Name: Matthias Blume
1653    Date: 2002/01/21
1654    Tag: blume-20020121-ff
1655    Description:
1656    
1657    Another day of fiddling with the FFI...
1658    
1659    1. Bug fix/workaround:  CKIT does not complain about negative array
1660       dimensions, so ml-nlffigen has to guard itself against this possibility.
1661       (Otherwise a negative dimension would send it into an infinite loop.)
1662    
1663    2. Some of the abstract types (light objects, light pointers, most "base"
1664       types) in structure C are now eqtypes.
1665    
1666    3. Added constructors and test functions for NULL function pointers.
1667    
1668    ----------------------------------------------------------------------
1669    Name: Matthias Blume
1670    Date: 2002/01/18
1671    Tag: blume-20020118-ready-for-new-release
1672    Description:
1673    
1674    Made config/srcarchiveurl point to a new place.  (Will provide boot
1675    files shortly.)
1676    
1677    Maybe we christen this to be 110.38?
1678    
1679    ----------------------------------------------------------------------
1680    Name: Matthias Blume
1681    Date: 2002/01/18
1682    Tag: blume-20020118-more-ffifiddle
1683    Description:
1684    
1685    Today's FFI fiddling:
1686    
1687      - Provided a structure CGetSet with "convenient" versions of C.Get.* and
1688        C.Set.* that use concrete (MLRep.*) arguments and results instead
1689        of abstract ones.
1690    
1691      - Provided word-style bit operations etc. for "int" representation
1692        types in MLRep.S<Foo>Bitops where <Foo> ranges over Char, Int, Short,
1693        and Long.
1694    
1695    ----------------------------------------------------------------------
1696    Name: Matthias Blume
1697    Date: 2002/01/18
1698    Tag: blume-20020118-use-x86-fp
1699    Description:
1700    
1701    Now that x86-fast-fp seems to be working, I turned it back on again
1702    by default.  (Seems to work fine now, even with the FFI.)
1703    
1704    Other than that, I added some documentation about the FFI to
1705    src/ml-nlffigen/README and updated the FFI test examples in
1706    src/ml-nlffi-lib/Tests/*.
1707    
1708    ----------------------------------------------------------------------
1709    Name: Allen Leung
1710    Date: 2002/01/17
1711    Tag: leunga-20020117-x86-fast-fp-call
1712    Description:
1713    
1714       1. Fixed a problem with handling return fp values when x86's fast fp
1715          mode is turned on.
1716    
1717       2. Minor pretty printing fix for cellset.  Print %st(0) as %st(0) instead
1718          of %f32.
1719    
1720       3. Added a constructor INT32lit to the ast of MLRISC tools.
1721    
1722    ----------------------------------------------------------------------
1723    Name: Matthias Blume
1724    Date: 2002/01/16
1725    Tag: blume-20020116-ffifiddle
1726    Description:
1727    
1728    More fiddling with the FFI interface:
1729    
1730     - Make constness 'c instead of rw wherever possible.  This eliminates
1731       the need for certain explicit coercions.  (However, due to ML's
1732       value polymorphism, there will still be many cases where explicit
1733       coercions are necessary.  Phantom types are not the whole answer
1734       to modeling a subtyping relationship in ML.)
1735    
1736     - ro/rw coersions for pointers added.  (Avoids the detour through */&.)
1737    
1738     - "printf" test example added to src/ml-nlffi-lib/Tests.  (Demonstrates
1739       clumsy workaround for varargs problem.)
1740    
1741    ----------------------------------------------------------------------
1742    Name: Lal George
1743    Date: 2002/01/15
1744    Tag: <none>
1745    Description:
1746    
1747    1. Since COPY instructions are no longer native to the architecture,
1748       a generic functor can be used to implement the expandCopies function.
1749    
1750    2. Allowed EXPORT and IMPORT pseudo-op declarations to appear inside a
1751       TEXT segment.
1752    
1753    ----------------------------------------------------------------------
1754    Name: Matthias Blume
1755    Date: 2002/01/15
1756    Tag: blume-20020115-ffiupdates
1757    Description:
1758    
1759    1. Fix for bug resulting in single-precision float values being returned
1760       incorrectly from FFI calls.
1761    
1762    2. Small modifications to C FFI API:
1763    
1764        - memory-allocation routines return straight objects (no options)
1765          and raise an exception in out-of-memory situations
1766        - unsafe extensions to cast between function pointers and pointers
1767          from/to ints
1768        - added structure C_Debug as an alternative to structure C where
1769          pointer-dereferencing (|*| and |*!) always check for null-pointers
1770        - added open_lib' to DynLinkage;  open_lib' works like open_lib
1771          but also takes a (possibly empty) list of existing library handles
1772          that the current library depends on
1773    
1774    ----------------------------------------------------------------------
1775    Name: Matthias Blume
1776    Date: 2002/01/10
1777    Tag: blume-20020110-newffigen
1778    Description:
1779    
1780    1. Updates to portable graph code.
1781    
1782    2. Major update to ml-nlffigen and ml-nlffi-lib.  Things are much
1783       more scalable now so that even huge interfaces such as the one
1784       for GTK compile in finite time and space. :-)
1785       See src/ml-nlffigen/README for details on what's new.
1786    
1787    ----------------------------------------------------------------------
1788    Name: Lal George
1789    Date: 2001/01/09 14:31:35 EST 2002
1790    Tag: george-20011206-rm-native-copy
1791    Description:
1792    
1793            Removed the native COPY and FCOPY instructions
1794            from all the architectures and replaced it with the
1795            explicit COPY instruction from the previous commit.
1796    
1797            It is now possible to simplify many of the optimizations
1798            modules that manipulate copies. This has not been
1799            done in this change.
1800    
1801    ----------------------------------------------------------------------
1802    Name: Lal George
1803    Date: 2001/12/06 16:50:13 EST 2001
1804    Tag: george-20011206-mlrisc-instruction
1805    Description:
1806    
1807    Changed the representation of instructions from being fully abstract
1808    to being partially concrete. That is to say:
1809    
1810      from
1811            type instruction
1812    
1813      to
1814            type instr                              (* machine instruction *)
1815    
1816            datatype instruction =
1817                LIVE of {regs: C.cellset, spilled: C.cellset}
1818              | KILL of {regs: C.cellset, spilled: C.cellset}
1819              | COPYXXX of {k: CB.cellkind, dst: CB.cell list, src: CB.cell list}
1820              | ANNOTATION of {i: instruction, a: Annotations.annotation}
1821              | INSTR of instr
1822    
1823    This makes the handling of certain special instructions that appear on
1824    all architectures easier and uniform.
1825    
1826    LIVE and KILL say that a list of registers are live or killed at the
1827    program point where they appear. No spill code is generated when an
1828    element of the 'regs' field is spilled, but the register is moved to
1829    the 'spilled' (which is present, more for debugging than anything else).
1830    
1831    LIVE replaces the (now deprecated) DEFFREG instruction on the alpha.
1832    We used to generate:
1833    
1834            DEFFREG f1
1835            f1 := f2 + f3
1836            trapb
1837    
1838    but now generate:
1839    
1840            f1 := f2 + f3
1841            trapb
1842            LIVE {regs=[f1,f2,f3], spilled=[]}
1843    
1844    Furthermore, the DEFFREG (hack) required that all floating point instruction
1845    use all registers mentioned in the instruction. Therefore f1 := f2 + f3,
1846    defines f1 and uses [f1,f2,f3]! This hack is no longer required resulting
1847    in a cleaner alpha implementation. (Hopefully, intel will not get rid of
1848    this architecture).
1849    
1850    COPYXXX is intended to replace the parallel COPY and FCOPY  available on
1851    all the architectures. This will result in further simplification of the
1852    register allocator that must be aware of them for coalescing purposes, and
1853    will also simplify certain aspects of the machine description that provides
1854    callbacks related to parallel copies.
1855    
1856    ANNOTATION should be obvious, and now INSTR represents the honest to God
1857    machine instruction set!
1858    
1859    The <arch>/instructions/<arch>Instr.sml files define certain utility
1860    functions for making porting easier -- essentially converting upper case
1861    to lower case. All machine instructions (of type instr) are in upper case,
1862    and the lower case form generates an MLRISC instruction. For example on
1863    the alpha we have:
1864    
1865      datatype instr =
1866         LDA of {r:cell, b:cell, d:operand}
1867       | ...
1868    
1869      val lda : {r:cell, b:cell, d:operand} -> instruction
1870        ...
1871    
1872    where lda is just (INSTR o LDA), etc.
1873    
1874    ----------------------------------------------------------------------
1875    Name: Matthias Blume
1876    Date: 2001/11/22 21:40:00 EST
1877    Tag: Release_110_37
1878    Description:
1879    
1880    Release 110.37.  This time for real.
1881    
1882    ----------------------------------------------------------------------
1883    Name: Matthias Blume
1884    Date: 2001/11/21 16:35:00 EST
1885    Tag: blume-20011121-foot-in-mouth
1886    Description:
1887    
1888    Removed the "Release_110_37" tag because of a serious bug.
1889    This will be re-tagged once the bug is fixed.
1890    
1891    ----------------------------------------------------------------------
1892    Name: Matthias Blume
1893    Date: 2001/11/21 16:14:00 EST
1894    Tag: blume-20011121-forgottenfile
1895    Description:
1896    
1897    Forgot to add a file.  (Just a .tex-file -- part of
1898    the CM manual source.)
1899    
1900    ----------------------------------------------------------------------
1901    Name: Matthias Blume
1902    Date: 2001/11/21 16:10:00 EST
1903    Tag: blume-20011121-invalid_110_37
1904    Description:
1905    
1906    Note: I removed the original tag "Release_110_37" from this commit
1907          because we found a serious bug in all non-x86 backends.
1908          - Matthias
1909    
1910    1. Modifications to the SML/NJ code generator and to the runtime system
1911       so that code object name strings are directly inserted into code
1912       objects at code generation time.  The only business the runtime system
1913       has with this is now to read the name strings on occasions.
1914       (The encoding of the name string has also changed somewhat.)
1915    
1916    2. CM now implements a simple "set calculus" for specifying export lists.
1917       In particular, it is now possible to refer to the export lists of
1918       other libraries/groups/sources and form unions as well as differences.
1919       See the latest CM manual for details.
1920    
1921    3. An separate notion of "proxy" libraries has again be eliminated from
1922       CM's model.  (Proxy libraries are now simply a special case of using
1923       the export list calculus.)
1924    
1925    4. Some of the existing libraries now take advantage of the new set
1926       calculus.
1927       (Notice that not all libraries have been converted because some
1928       of the existing .cm-files are supposed to be backward compatible
1929       with 110.0.x.)
1930    
1931    5. Some cleanup in stand-alone programs.  (Don't use "exnMessage" -- use
1932       "General.exnMessage"!  The former relies on a certain hook to be
1933       initialized, and that often does not happen in the stand-alone case.)
1934    
1935    ----------------------------------------------------------------------
1936    Name: Lal George
1937    Date: 2001/11/21  13:56:18 EST
1938    Tag: george-2001121-pseudo-ops
1939    Description:
1940    
1941      Implemented a complete redesign of MLRISC pseudo-ops. Now there
1942      ought to never be any question of incompatabilities with
1943      pseudo-op syntax expected by host assemblers.
1944    
1945      For now, only modules supporting GAS syntax are implemented
1946      but more should follow, such as MASM, and vendor assembler
1947      syntax, e.g. IBM as, Sun as, etc.
1948    
1949    ----------------------------------------------------------------------
1950    Name: Matthias Blume
1951  Date: 2001/11/14 11:52:00 EST  Date: 2001/11/14 11:52:00 EST
1952  Tag: blume-20011114-srcname  Tag: blume-20011114-srcname
1953  Description:  Description:
# Line 194  Line 2128 
2128    
2129  ----------------------------------------------------------------------  ----------------------------------------------------------------------
2130  Name: Matthias Blume  Name: Matthias Blume
 >>>>>>> 1.169  
2131  Date: 2001/09/18 15:35:00 EDT  Date: 2001/09/18 15:35:00 EDT
2132  Tag: blume-20010918-readme11036  Tag: blume-20010918-readme11036
2133  Description:  Description:

Legend:
Removed from v.977  
changed lines
  Added in v.1238

root@smlnj-gforge.cs.uchicago.edu
ViewVC Help
Powered by ViewVC 1.0.0