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 1212, Tue May 21 16:34:29 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: 2002/05/21 12:35:00 EDT
18    Tag: blume-20020521-cmdoc
19    Description:
20    
21    CM documentation update.
22    
23    ----------------------------------------------------------------------
24    Name: Matthias Blume
25    Date: 2002/05/21 10:55:00 EDT
26    Tag: blume-20020521-misc
27    Description:
28    
29    - John tweaked runtime to be silent on heap export (except when
30      GC messages are on).
31    - I added a few more things (cross-compiling versions of CMB) to
32      config/preloads (as suggestions).
33    
34    ----------------------------------------------------------------------
35    Name: Matthias Blume
36    Date: 2002/05/20 22:25:00 EDT
37    Tag: blume-20020520-controls
38    Description:
39    
40    - Added ControlUtil structure to control-lib.cm.
41    - Use it throughout.
42    - Used Controls facility to define MLRISC controls (as opposed to
43      registering MLRISC control ref cells with Controls after the fact)
44    - Fixed messed-up controls priorities.
45    
46    * Removed again all the stuff from config/preloads that one wouldn't
47      be able to preload at the time the initial heap image is built.
48      (Many libraries, e.g., CML, do not exist yet at this time.  The
49       only libraries that can be preloaded via config/preloads are those
50       that come bundled with the bootfiles.)
51    
52    ----------------------------------------------------------------------
53    Name: Matthias Blume
54    Date: 2002/05/20 10:59:00 EDT
55    Tag: blume-20020520-preloads
56    Description:
57    
58    Added a lot of commented-out suggestions for things to be included
59    in config/preloads.
60    
61    ----------------------------------------------------------------------
62    Name: Allen Leung
63    Date: 2002/05/18 14:20:00 EDT
64    Tag: leunga-20020518-mdl
65    Description:
66    
67    o Made the mdl tool stuff compile and run again.
68    o I've disabled all the stuff that depends on RTL specifications; they
69      are all badly broken anyway.
70    
71    ----------------------------------------------------------------------
72    Name: Matthias Blume
73    Date: 2002/05/17 16:49:00 EDT
74    Tag: blume-20020517-controls
75    Description:
76    
77    0. John Reppy made several modifications to the SML/NJ library.
78       In particular, there is a shiny new controls-lib.cm.
79    
80    1. Pushed new controls interface through compiler so that everything
81       compiles again.
82    
83    2. Added FormatComb and FORMAT_COMB to the CML version of the
84       SML/NJ library (so that CML compiles again).
85    
86    3. Modified init scripts because XXX_DEFAULT environment variables
87       are no longer with us.  (Boot-time initialization is now done
88       using the same environment variables that are also used for
89       startup-time initialization of controls.)
90    
91    ----------------------------------------------------------------------
92    Name: Lal George
93    Date: 2002/05/15 09:20:10 EDT
94    Tag: george-20020515-pseudo-op-decls
95    Description:
96    
97            All pseudo-ops emitted before the first segment declaration
98    such as TEXT, DATA, and BSS directives are assumed to be global
99    declarations and are emitted first in the assembly file. This is
100    useful in a number of situations where one has pseudo-ops that are not
101    specific to any segment, and also works around the constraint that one
102    cannot have client pseudo-ops in the TEXT segment.
103    
104            Because no segment is associated with these declarations it is
105    an error to allocate any space or objects before the first segment
106    directive and an exception will be raised. However, we cannot make
107    this check for client pseudo-ops.
108    
109            These top level declarations are a field in the CFG graph_info.
110    In theory you can continue to add to this field after the CFG has been
111    built -- provided you know what you are doing;-)
112    
113    ----------------------------------------------------------------------
114    Name: Matthias Blume
115    Date: 2002/05/13 16:40:00 EDT
116    Tag: blume-20020513-pp-etc
117    Description:
118    
119    A few minor bugfixes:
120    
121      - Stopgap measure for bug recently reported by Elsa Gunter (ppDec).
122        (Bogus printouts for redefined bindings still occur.  Compiler
123        bug should no longer occur now.  We need to redo the prettyprinter
124        from scratch.)
125    
126      - CM pathname printer now also adds escape sequences for ( and )
127    
128      - commend and docu fixes for ml-nlffi
129    
130    ----------------------------------------------------------------------
131    Name: Matthias Blume
132    Date: 2002/05/10 16:40:00 EDT
133    Tag: blume-20020510-erg-textio
134    Description:
135    
136    Applied the following bugfix provided by Emden Gansner:
137    
138        Output is corrupted when outputSubstr is used rather than output.
139    
140        The problem occurs when a substring
141    
142            ss = (s, dataStart, dataLen)
143    
144        where dataStart > 0, fills a stream buffer with avail bytes left.
145        avail bytes of s, starting at index dataStart, are copied into the
146        buffer, the buffer is flushed, and then the remaining dataLen-avail
147        bytes of ss are copied into the beginning of the buffer. Instead of
148        starting this copy at index dataStart+avail in s, the current code
149        starts the copy at index avail.
150    
151        Fix:
152        In text-io-fn.sml, change line 695 from
153             val needsFlush = copyVec(v, avail, dataLen-avail, buf, 0)
154        to
155             val needsFlush = copyVec(v, dataStart+avail, dataLen-avail, buf, 0)
156    
157    ----------------------------------------------------------------------
158    Name: Matthias Blume
159    Date: 2002/04/12 13:55:00 EDT
160    Tag: blume-20020412-assyntax
161    Description:
162    
163    1. Grabbed newer assyntax.h from the XFree86 project.
164    2. Fiddled with how to compile X86.prim.asm without warnings.
165    3. (Very) Minor cleanup in CM.
166    
167    ----------------------------------------------------------------------
168    Name: Matthias Blume
169    Date: 2002/04/01 (no joke!) 17:07:00 EST
170    Tag: blume-20020401-x86div
171    Description:
172    
173    Added full support for div/mod/rem/quot on the x86, using the machine
174    instruction's two results (without clumsily recomputing the remainder)
175    directly where appropriate.
176    
177    Some more extensive power-of-two support was added to the x86 instruction
178    selector (avoiding expensive divs, mods, and muls where they can be
179    replaced with cheaper shifts and masks).  However, this sort of thing
180    ought to be done earlier, e.g., within the CPS optimizer so that
181    all architectures benefit from it.
182    
183    The compiler compiles to a fixed point, but changes might be somewhat
184    fragile nevertheless.  Please, report any strange things that you might
185    see wrt. div/mod/quot/rem...
186    
187    ----------------------------------------------------------------------
188    Name: Matthias Blume
189    Date: 2002/03/29 17:22:00
190    Tag: blume-20020329-div
191    Description:
192    
193    Fixed my broken div/mod logic.  Unfortunately, this means that the
194    inline code for div/mod now has one more comparison than before.
195    Fast paths (quotient > 0 or remainder = 0) are not affected, though.
196    The problem was with quotient = 0, because that alone does not tell
197    us which way the rounding went.  One then has to look at whether
198    remainder and divisor have the same sign...  :(
199    
200    Anyway, I replaced the bootfiles with fresh ones...
201    
202    ----------------------------------------------------------------------
203    Name: Matthias Blume
204    Date: 2002/03/29 14:10:00 EST
205    Tag: blume-20020329-inlprims
206    Description:
207    
208    NEW BOOTFILES!!!    Version number bumped to 110.39.3.
209    
210    Primops have changed. This means that the bin/boot-file formats have
211    changed as well.
212    
213    To make sure that there is no confusion, I made a new version.
214    
215    
216    CHANGES:
217    
218    * removed REMT from mltree (remainder should never overflow).
219    
220    * added primops to deal with divisions of all flavors to the frontend
221    
222    * handled these primops all the way through so they map to their respective
223      MLRISC support
224    
225    * used these primops in the implementation of Int, Int32, Word, Word32
226    
227    * removed INLDIV, INLMOD, and INLREM as they are no longer necessary
228    
229    * parameterized INLMIN, INLMAX, and INLABS by a numkind
230    
231    * translate.sml now deals with all flavors of INL{MIN,MAX,ABS}, including
232      floating point
233    
234    * used INL{MIN,MAX,ABS} in the implementation of Int, Int32, Word, Word32,
235      and Real (but Real.abs maps to a separate floating-point-only primop)
236    
237    
238    TODO items:
239    
240    * Hacked Alpha32 instruction selection, disabling the selection of REMx
241      instructions because the machine instruction encoder cannot handle
242      them.  (Hppa, PPC, and Sparc instruction selection did not handle
243      REM in the first place, and REM is supported by the x86 machine coder.)
244    
245    * Handle DIV and MOD with DIV_TO_NEGINF directly in the x86 instruction
246      selection phase.  (The two can be streamlined because the hardware
247      delivers both quotient and remainder at the same time anyway.)
248    
249    * Think about what to do with "valOf(Int32.minInt) div ~1" and friends.
250      (Currently the behavior is inconsistent both across architectures and
251      wrt. the draft Basis spec.)
252    
253    * Word8 should eventually be handled natively, too.
254    
255    * There seems to be one serious bug in mltree-gen.sml.  It appears, though,
256      as if there currently is no execution path that could trigger it in
257      SML/NJ.  (The assumptions underlying functions arith and promotable do not
258      hold for things like multiplication and division.)
259    
260    ----------------------------------------------------------------------
261    Name: Matthias Blume
262    Date: 2002/03/27 16:27:00 EST
263    Tag: blume-20020327-mlrisc-divisions
264    Description:
265    
266    Added support for all four division operations (ML's div, mod, quot,
267    and rem) to MLRISC.  In the course of doing so, I also rationalized
268    the naming (no more annoying switch-around of DIV and QUOT), by
269    parameterizing the operation by div_rounding_mode (which can be either
270    DIV_TO_ZERO or DIV_TO_NEGINF).
271    
272    The generic MLTreeGen functor takes care of compiling all four
273    operations down to only round-to-zero div.
274    
275    Missing pieces:
276    
277      * Doing something smarter than relying on MLTreeGen on architectures
278        like, e.g., the x86 where hardware division delivers both quotient and
279        remainder at the same time.  With this, the implementation of the
280        round-to-neginf operations could be further streamlined.
281    
282      * Remove inlining support for div/mod/rem from the frontend and replace it
283        with primops that get carried through to the backend.  Do this for all
284        int and word types.
285    
286    ----------------------------------------------------------------------
287    Name: Matthias Blume
288    Date: 2002/03/25 17:25:00 EST
289    Tag: blume-20020325-divmod
290    Description:
291    
292    I improved (hopefully without breaking them) the implementation of Int.div,
293    Int.mod, and Int.rem.   For this, the code in translate.sml now takes
294    advantage of the following observations:
295    
296      Let  q = x quot y      r = x rem y
297           d = x div  y      m = x mod y
298    
299    where "quot" is the round-to-zero version of integer division that
300    hardware usually provides.  Then we have:
301    
302         r = x - q * y        where neither the * nor the - will overflow
303         d = if q >= 0 orelse x = q * y then q else q - 1
304                              where neither the * nor the - will overflow
305         m = if q >= 0 orelse r = 0 then r else r + y
306                              where the + will not overflow
307    
308    This results in substantial simplification of the generated code.
309    The following table shows the number of CFG nodes and edges generated
310    for
311            fun f (x, y) = x OPER y
312            (* with OPER \in div, mod, quot, rem *)
313    
314    
315        OPER | nodes(old) | edges(old) | nodes(new) | edges(new)
316        --------------------------------------------------------
317         div |         24 |         39 |         12 |         16
318         mod |         41 |         71 |         12 |         16
319        quot |          8 |         10 |          8 |         10
320         rem |         10 |         14 |          8 |         10
321    
322    
323    ----------------------------------------------------------------------
324    Name: Matthias Blume
325    Date: 2002/03/25 22:06:00 EST
326    Tag: blume-20020325-cprotobug
327    Description:
328    
329    Fixed a bug in cproto (c prototype decoder).
330    
331    ----------------------------------------------------------------------
332    Name: Matthias Blume
333    Date: 2002/03/25 16:00:00 EST
334    Tag: blume-20020325-raw-primops
335    Description:
336    
337    I did some cleanup to Allen's new primop code and
338    replaced yesterday's bootfiles with new ones.
339    (But they are stored in the same place.)
340    
341    ----------------------------------------------------------------------
342    Name: Matthias Blume
343    Date: 2002/03/24 22:40:00 EST
344    Tag: blume-20020324-bootfiles
345    Description:
346    
347    Made the bootfiles that Allen asked for.
348    
349    ----------------------------------------------------------------------
350    Name: Allen Leung
351    Date: 2002/03/23 15:50:00 EST
352    Tag: leunga-20020323-flint-cps-rcc-primops
353    Description:
354    
355      1. Changes to FLINT primops:
356    
357        (* make a call to a C-function;
358         * The primop carries C function prototype information and specifies
359         * which of its (ML-) arguments are floating point. C prototype
360         * information is for use by the backend, ML information is for
361         * use by the CPS converter. *)
362      | RAW_CCALL of { c_proto: CTypes.c_proto,
363                       ml_args: ccall_type list,
364                       ml_res_opt: ccall_type option,
365                       reentrant : bool
366                     } option
367       (* Allocate uninitialized storage on the heap.
368        * The record is meant to hold short-lived C objects, i.e., they
369        * are not ML pointers.  With the tag, the representation is
370        * the same as RECORD with tag tag_raw32 (sz=4), or tag_fblock (sz=8)
371        *)
372      | RAW_RECORD of {tag:bool,sz:int}
373      and ccall_type = CCALL_INT32 | CCALL_REAL64 | CCALL_ML_PTR
374    
375      2.  These CPS primops are now overloaded:
376    
377           rawload of {kind:numkind}
378           rawstore of {kind:numkind}
379    
380          The one argument form is:
381    
382             rawload {kind} address
383    
384          The two argument form is:
385    
386             rawload {kind} [ml object, byte-offset]
387    
388      3. RAW_CCALL/RCC now takes two extra arguments:
389    
390         a. The first is whether the C call is reentrant, i.e., whether
391            ML state should be saved and restored.
392         b. The second argument is a string argument specifying the name of
393            library and the C function.
394    
395         These things are currently not handled in the code generator, yet.
396    
397      4. In CProto,
398    
399         An encoding type of "bool" means "ml object" and is mapped into
400         C prototype of PTR.  Note that "bool" is different than "string",
401         even though "string" is also mapped into PTR, because "bool"
402         is assigned an CPS type of BOGt, while "string" is assigned INT32t.
403    
404      5. Pickler/unpicker
405    
406         Changed to handle RAW_RECORD and newest RAW_CCALL
407    
408      6. MLRiscGen,
409    
410         1. Changed to handle the new rawload/rawstore/rawrecord operators.
411         2. Code for handling C Calls has been moved to a new module CPSCCalls,
412            in the file CodeGen/cpscompile/cps-c-calls.sml
413    
414      7. Added the conditional move operator
415    
416             condmove of branch
417    
418         to cps.  Generation of this is still buggy so it is currently
419         disabled.
420    
421    ----------------------------------------------------------------------
422    Name: Lal George
423    Date: 2002/03/22 14:18:25 EST
424    Tag: george-20020322-cps-branch-prob
425    Description:
426    
427    Implemented the Ball-Larus branch prediction-heuristics, and
428    incorporated graphical viewers for control flow graphs.
429    
430    Ball-Larus Heuristics:
431    ---------------------
432    See the file compiler/CodeGen/cpscompile/cpsBranchProb.sml.
433    
434    By design it uses the Dempster-Shafer theory for combining
435    probabilities.  For example, in the function:
436    
437        fun f(n,acc) = if n = 0 then acc else f(n-1, n*acc)
438    
439    the ball-larus heuristics predicts that the n=0 is unlikely
440    (OH-heuristic), and the 'then' branch is unlikely because of the
441    RH-heuristic -- giving the 'then' branch an even lower combined
442    probability using the Dempster-Shafer theory.
443    
444    Finally, John Reppy's loop analysis in MLRISC, further lowers the
445    probability of the 'then' branch because of the loop in the else
446    branch.
447    
448    
449    Graphical Viewing:
450    ------------------
451    I merely plugged in Allen's graphical viewers into the compiler. The
452    additional code is not much. At the top level, saying:
453    
454            Control.MLRISC.getFlag "cfg-graphical-view" := true;
455    
456    will display the graphical view of the control flow graph just before
457    back-patching.  daVinci must be in your path for this to work. If
458    daVinci is not available, then the default viewer can be changed
459    using:
460    
461            Control.MLRISC.getString "viewer"
462    
463    which can be set to "dot" or "vcg" for the corresponding viewers. Of
464    course, these viewers must be in your path.
465    
466    The above will display the compilation unit at the level of clusters,
467    many of which are small, boring, and un-interesting. Also setting:
468    
469            Control.MLRISC.getInt "cfg-graphical-view_size"
470    
471    will display clusters that are larger than the value set by the above.
472    
473    
474    ----------------------------------------------------------------------
475    Name: Matthias Blume
476    Date: 2002/03/21 22:20:00 EST
477    Tag: blume-20020321-kmp-bugfix
478    Description:
479    
480    Changed the interface to the KMP routine in PreString and fixed
481    a minor bug in one place where it was used.
482    
483    ----------------------------------------------------------------------
484    Name: Allen Leung
485    Date: 2002/03/21 20:30:00 EST
486    Tag: leunga-20020321-cfg
487    Description:
488    
489      Fixed a potential problem in cfg edge splitting.
490    
491    ----------------------------------------------------------------------
492    Name: Allen Leung
493    Date: 2002/03/21 17:15:00 EST
494    Tag: leunga-20020321-x86-fp-cfg
495    Description:
496    
497      1. Recoded the buggy parts of x86-fp.
498    
499         a. All the block reordering code has been removed.
500            We now depend on the block placement phases to do this work.
501    
502         b. Critical edge splitting code has been simplified and moved into the
503            CFG modules, as where they belong.
504    
505         Both of these were quite buggy and complex.  The code is now much, much
506         simpler.
507    
508      2. X86 backend.
509    
510         a. Added instructions for 64-bit support.  Instruction selection for
511            64-bit has not been committed, however, since that
512            requires changes to MLTREE which haven't been approved by
513            Lal and John.
514    
515         b. Added support for FUCOMI and FUCOMIP when generating code for
516            PentiumPro and above.  We only generate these instructions in
517            the fast-fp mode.
518    
519         c. Added cases for JP and JNP in X86FreqProps.
520    
521      3. CFG
522    
523         CFG now has a bunch of methods for edge splitting and merging.
524    
525      4. Machine description.
526    
527         John's simplification of MLTREE_BASIS.fcond broke a few machine
528         description things:
529    
530         rtl-build.{sig,sml} and hppa.mdl fixed.
531    
532         NOTE: the machine description stuff in the repository is still broken.
533               Again, I can't put my fixes in because that involves
534               changes to MLTREE.
535    
536    ----------------------------------------------------------------------
537    Name: Matthias Blume
538    Date: 2002/03/20 15:55:00 EST
539    Tag: blume-20020320-kmp
540    Description:
541    
542    Implemented Knuth-Morris-Pratt string matching in PreString and used
543    it for String.isSubstring, Substring.isSubstring, and
544    Substring.position.
545    
546    (Might need some stress-testing.  Simple examples worked fine.)
547    
548    ----------------------------------------------------------------------
549    Name: Matthias Blume
550    Date: 2002/03/19 16:37:00 EST
551    Tag: blume-20020319-witnesses
552    Description:
553    
554    Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
555    
556    This implements a generic mechanism for changing constness qualifiers
557    anywhere within big C types without resorting to outright "casts".
558    (So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
559    modify the constness at the outermost level.)
560    The implementation of "convert" is based on the idea of "witness"
561    values -- values that are not used by the operation but whose types
562    "testify" to their applicability.  On the implementation side, "convert"
563    is simply a projection (returning its second curried argument).  With
564    cross-module inlining, it should not result in any machine code being
565    generated.
566    
567    ----------------------------------------------------------------------
568    Name: Matthias Blume
569    Date: 2002/03/15 16:40:00 EST
570    Tag: blume-20020315-basis
571    Description:
572    
573    Provided (preliminary?) implementations for
574    
575      {String,Substring}.{concatWith,isSuffix,isSubstring}
576    
577    and
578    
579      Substring.full
580    
581    Those are in the Basis spec but they were missing in SML/NJ.
582    
583    ----------------------------------------------------------------------
584    Name: Matthias Blume
585    Date: 2002/03/14 21:30:00 EST
586    Tag: blume-20020314-controls
587    Description:
588    
589    Controls:
590    ---------
591    
592    1. Factored out the recently-added Controls : CONTROLS stuff and put
593       it into its own library $/controls-lib.cm.  The source tree for
594       this is under src/smlnj-lib/Controls.
595    
596    2. Changed the names of types and functions in this interface, so they
597       make a bit more "sense":
598    
599          module -> registry
600          'a registry -> 'a group
601    
602    3. The interface now deals in ref cells only.  The getter/setter interface
603       is (mostly) gone.
604    
605    4. Added a function that lets one register an already-existing ref cell.
606    
607    5. Made the corresponding modifications to the rest of the code so that
608       everything compiles again.
609    
610    6. Changed the implementation of Controls.MLRISC back to something closer
611       to the original.  In particular, this module (and therefore MLRISC)
612       does not depend on Controls.  There now is some link-time code in
613       int-sys.sml that registers the MLRISC controls with the Controls
614       module.
615    
616    CM:
617    ---
618    
619      * One can now specify the lambda-split aggressiveness in init.cmi.
620    
621    ----------------------------------------------------------------------
622    Name: Allen Leung
623    Date: 2002/03/13 17:30:00 EST
624    Tag: leunga-20020313-x86-fp-unary
625    Description:
626    
627    Bug fix for:
628    
629    > leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
630    > Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
631    > - fun f(x,(y,z)) = Real.~ y;
632    > [autoloading]
633    > [autoloading done]
634    >       fchsl   (%eax), 184(%esp)
635    > Error: MLRisc bug: X86MCEmitter.emitInstr
636    >
637    > uncaught exception Error
638    >   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
639    
640    The problem was that the code generator did not generate any fp registers
641    in this case, and the ra didn't know that it needed to run the X86FP phase to
642    translate the pseudo fp instruction.   This only happened with unary fp
643    operators in certain situations.
644    
645    ----------------------------------------------------------------------
646    Name: Matthias Blume
647    Date: 2002/03/13 14:00:00 EST
648    Tag: blume-20020313-overload-etc
649    Description:
650    
651    1. Added _overload as a synonym for overload for backward compatibility.
652       (Control.overloadKW must be true for either version to be accepted.)
653    
654    2. Fixed bug in install script that caused more things to be installed
655       than what was requested in config/targets.
656    
657    3. Made CM aware of the (_)overload construct so that autoloading
658       works.
659    
660    ----------------------------------------------------------------------
661    Name: Matthias Blume
662    Date: 2002/03/12 22:03:00 EST
663    Tag: blume-20020312-url
664    Description:
665    
666    Forgot to update BOOT and srcarchiveurl.
667    
668    ----------------------------------------------------------------------
669    Name: Matthias Blume
670    Date: 2002/03/12 17:30:00 EST
671    Tag: blume-20020312-version110392
672    Description:
673    
674    Yet another version number bump (because of small changes to the
675    binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
676    
677    Changes:
678    
679      The new pid generation scheme described a few weeks ago was overly
680      complicated.  I implemented a new mechanism that is simpler and
681      provides a bit more "stability":  Once CM has seen a compilation
682      unit, it keeps its identity constant (as long as you do not delete
683      those crucial CM/GUID/* files).  This means that when you change
684      an interface, compile, then go back to the old interface, and
685      compile again, you arrive at the original pid.
686    
687      There now also is a mechanism that instructs CM to use the plain
688      environment hash as a module's pid (effectively making its GUID
689      the empty string).  For this, "noguid" must be specified as an
690      option to the .sml file in question within its .cm file.
691      This is most useful for code that is being generated by tools such
692      as ml-nlffigen (because during development programmers tend to
693      erase the tool's entire output directory tree including CM's cached
694      GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
695      revert to the old, broken behavior of SML/NJ, but in specific cases
696      where there is no danger of interface confusion, its use is ok
697      (I think).
698    
699      ml-nlffigen by default generates "noguid" annotations.  They can be
700      turned off by specifying -guid in its command line.
701    
702    ----------------------------------------------------------------------
703    Name: Lal George
704    Date: 2002/03/12 12 14:42:36 EST
705    Tag: george-20020312-frequency-computation
706    Description:
707    
708    Integrated jump chaining and static block frequency into the
709    compiler. More details and numbers later.
710    
711    ----------------------------------------------------------------------
712    Name: Lal George
713    Date: 2002/03/11 11 22:38:53 EST
714    Tag: george-20020311-jump-chain-elim
715    Description:
716    
717    Tested the jump chain elimination on all architectures (except the
718    hppa).  This is on by default right now and is profitable for the
719    alpha and x86, however, it may not be profitable for the sparc and ppc
720    when compiling the compiler.
721    
722    The gc test will typically jump to a label at the end of the cluster,
723    where there is another jump to an external cluster containing the actual
724    code to invoke gc. This is to allow factoring of common gc invocation
725    sequences. That is to say, we generate:
726    
727            f:
728               testgc
729               ja   L1      % jump if above to L1
730    
731            L1:
732               jmp L2
733    
734    
735    After jump chain elimination the 'ja L1' instructions is converted to
736    'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
737    up being implemented in their long form (if L2 is far away) using:
738    
739            jbe     L3      % jump if below or equal to L3
740            jmp     L2
741         L3:
742            ...
743    
744    
745    For large compilation units L2  may be far away.
746    
747    
748    ----------------------------------------------------------------------
749    Name: Matthias Blume
750    Date: 2002/03/11 13:30:00 EST
751    Tag: blume-20020311-mltreeeval
752    Description:
753    
754    A functor parameter was missing.
755    
756    ----------------------------------------------------------------------
757    Name: Allen Leung
758    Date: 2002/03/11 10:30:00 EST
759    Tag: leunga-20020311-runtime-string0
760    Description:
761    
762       The representation of the empty string now points to a
763    legal null terminated C string instead of unit.  It is now possible
764    to convert an ML string into C string with InlineT.CharVector.getData.
765    This compiles into one single machine instruction.
766    
767    ----------------------------------------------------------------------
768    Name: Allen Leung
769    Date: 2002/03/10 23:55:00 EST
770    Tag: leunga-20020310-x86-call
771    Description:
772    
773       Added machine generation for CALL instruction (relative displacement mode)
774    
775    ----------------------------------------------------------------------
776    Name: Matthias Blume
777    Date: 2002/03/08 16:05:00
778    Tag: blume-20020308-entrypoints
779    Description:
780    
781    Version number bumped to 110.39.1.  NEW BOOTFILES!
782    
783    Entrypoints: non-zero offset into a code object where execution should begin.
784    
785    - Added the notion of an entrypoint to CodeObj.
786    - Added reading/writing of entrypoint info to Binfile.
787    - Made runtime system bootloader aware of entrypoints.
788    - Use the address of the label of the first function given to mlriscGen
789      as the entrypoint.  This address is currently always 0, but it will
790      not be 0 once we turn on block placement.
791    - Removed the linkage cluster code (which was The Other Way(tm) of dealing
792      with entry points) from mlriscGen.
793    
794    ----------------------------------------------------------------------
795    Name: Allen Leung
796    Date: 2002/03/07 20:45:00 EST
797    Tag: leunga-20020307-x86-cmov
798    Description:
799    
800       Bug fixes for CMOVcc on x86.
801    
802       1. Added machine code generation for CMOVcc
803       2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
804       3. CMOVcc cannot have an immediate operand as argument.
805    
806    ----------------------------------------------------------------------
807    Name: Matthias Blume
808    Date: 2002/03/07 16:15:00 EST
809    Tag: blume-20020307-controls
810    Description:
811    
812    This is a very large but mostly boring patch which makes (almost)
813    every tuneable compiler knob (i.e., pretty much everything under
814    Control.* plus a few other things) configurable via both the command
815    line and environment variables in the style CM did its configuration
816    until now.
817    
818    Try starting sml with '-h' (or, if you are brave, '-H')
819    
820    To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
821    implements the underlying generic mechanism.
822    
823    The interface to some of the existing such facilities has changed somewhat.
824    For example, the MLRiscControl module now provides mkFoo instead of getFoo.
825    (The getFoo interface is still there for backward-compatibility, but its
826    use is deprecated.)
827    
828    The ml-build script passes -Cxxx=yyy command-line arguments through so
829    that one can now twiddle the compiler settings when using this "batch"
830    compiler.
831    
832    TODO items:
833    
834    We should go through and throw out all controls that are no longer
835    connected to anything.  Moreover, we should go through and provide
836    meaningful (and correct!) documentation strings for those controls
837    that still are connected.
838    
839    Currently, multiple calls to Controls.new are accepted (only the first
840    has any effect).  Eventually we should make sure that every control
841    is being made (via Controls.new) exactly once.  Future access can then
842    be done using Controls.acc.
843    
844    Finally, it would probably be a good idea to use the getter-setter
845    interface to controls rather than ref cells.  For the time being, both
846    styles are provided by the Controls module, but getter-setter pairs are
847    better if thread-safety is of any concern because they can be wrapped.
848    
849    *****************************************
850    
851    One bug fix: The function blockPlacement in three of the MLRISC
852    backpatch files used to be hard-wired to one of two possibilities at
853    link time (according to the value of the placementFlag).  But (I
854    think) it should rather sense the flag every time.
855    
856    *****************************************
857    
858    Other assorted changes (by other people who did not supply a HISTORY entry):
859    
860    1. the cross-module inliner now works much better (Monnier)
861    2. representation of weights, frequencies, and probabilities in MLRISC
862       changed in preparation of using those for weighted block placement
863       (Reppy, George)
864    
865    ----------------------------------------------------------------------
866    Name: Lal George
867    Date: 2002/03/07 14:44:24 EST 2002
868    Tag: george-20020307-weighted-block-placement
869    
870    Tested the weighted block placement optimization on all architectures
871    (except the hppa) using AMPL to generate the block and edge frequencies.
872    Changes were required in the machine properties to correctly
873    categorize trap instructions. There is an MLRISC flag
874    "weighted-block-placement" that can be used to enable weighted block
875    placement, but this will be ineffective without block/edge
876    frequencies (coming soon).
877    
878    
879    ----------------------------------------------------------------------
880    Name: Lal George
881    Date: 2002/03/05 17:24:48 EST
882    Tag: george-20020305-linkage-cluster
883    
884    In order to support the block placement optimization, a new cluster
885    is generated as the very first cluster (called the linkage cluster).
886    It contains a single jump to the 'real' entry point for the compilation
887    unit. Block placement has no effect on the linkage cluster itself, but
888    all the other clusters  have full freedom in the manner in which they
889    reorder blocks or functions.
890    
891    On the x86 the typical linkage code that is generated is:
892       ----------------------
893            .align 2
894       L0:
895            addl    $L1-L0, 72(%esp)
896            jmp     L1
897    
898    
899            .align  2
900       L1:
901       ----------------------
902    
903    72(%esp) is the memory location for the stdlink register. This
904    must contain the address of the CPS function being called. In the
905    above example, it contains the address of  L0; before
906    calling L1 (the real entry point for the compilation unit), it
907    must contain the address for L1, and hence
908    
909            addl $L1-L0, 72(%esp)
910    
911    I have tested this on all architectures except the hppa.The increase
912    in code size is of course negligible
913    
914    ----------------------------------------------------------------------
915    Name: Allen Leung
916    Date: 2002/03/03 13:20:00 EST
917    Tag: leunga-20020303-mlrisc-tools
918    
919      Added #[ ... ] expressions to mlrisc tools
920    
921    ----------------------------------------------------------------------
922    Name: Matthias Blume
923    Date: 2002/02/27 12:29:00 EST
924    Tag: blume-20020227-cdebug
925    Description:
926    
927    - made types in structure C and C_Debug to be equal
928    - got rid of code duplication (c-int.sml vs. c-int-debug.sml)
929    - there no longer is a C_Int_Debug (C_Debug is directly derived from C)
930    
931    ----------------------------------------------------------------------
932    Name: Matthias Blume
933    Date: 2002/02/26 12:00:00 EST
934    Tag: blume-20020226-ffi
935    Description:
936    
937    1. Fixed a minor bug in CM's "noweb" tool:
938       If numbering is turned off, then truly don't number (i.e., do not
939       supply the -L option to noweb).  The previous behavior was to supply
940       -L'' -- which caused noweb to use the "default" line numbering scheme.
941       Thanks to Chris Richards for pointing this out (and supplying the fix).
942    
943    2. Once again, I reworked some aspects of the FFI:
944    
945       A. The incomplete/complete type business:
946    
947       - Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
948         gone!
949       - ML types representing an incomplete type are now *equal* to
950         ML types representing their corresponding complete types (just like
951         in C).  This is still safe because ml-nlffigen will not generate
952         RTTI for incomplete types, nor will it generate functions that
953         require access to such RTTI.   But when ML code generated from both
954         incomplete and complete versions of the C type meet, the ML types
955         are trivially interoperable.
956    
957         NOTE:  These changes restore the full generality of the translation
958         (which was previously lost when I eliminated functorization)!
959    
960       B. Enum types:
961    
962       - Structure C now has a type constructor "enum" that is similar to
963         how the "su" constructor works.  However, "enum" is not a phantom
964         type because each "T enum" has values (and is isomorphic to
965         MLRep.Signed.int).
966       - There are generic access operations for enum objects (using
967         MLRep.Signed.int).
968       - ml-nlffigen will generate a structure E_foo for each "enum foo".
969         * The structure contains the definition of type "mlrep" (the ML-side
970         representation type of the enum).  Normally, mlrep is the same
971         as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
972         then mlrep will be defined as a datatype -- thus facilitating
973         pattern matching on mlrep values.
974         ("-ec" will be suppressed if there are duplicate values in an
975          enumeration.)
976         * Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
977         will be generated for each C enum constant xxx.
978         * Conversion functions m2i and i2m convert between mlrep and
979         MLRep.Signed.int.  (Without "-ec", these functions are identities.)
980         * Coversion functions c and ml convert between mlrep and "tag enum".
981         * Access functions (get/set) fetch and store mlrep values.
982       - By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
983         enumerations are merged into one single enumeration represented by
984         structure E_'.
985    
986    ----------------------------------------------------------------------
987    Name: Allen Leung
988    Date: 2002/02/25 04:45:00 EST
989    Tag: leunga-20020225-cps-spill
990    
991    This is a new implementation of the CPS spill phase.
992    The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
993    In case of problems, replace it with the old file spill.sml
994    
995    The current compiler runs into some serious performance problems when
996    constructing a large record.  This can happen when we try to compile a
997    structure with many items.  Even a very simple structure like the following
998    makes the compiler slow down.
999    
1000        structure Foo = struct
1001           val x_1 = 0w1 : Word32.int
1002           val x_2 = 0w2 : Word32.int
1003           val x_3 = 0w3 : Word32.int
1004           ...
1005           val x_N = 0wN : Word32.int
1006        end
1007    
1008    The following table shows the compile time, from N=1000 to N=4000,
1009    with the old compiler:
1010    
1011    N
1012    1000   CPS 100 spill                           0.04u  0.00s  0.00g
1013           MLRISC ra                               0.06u  0.00s  0.05g
1014              (spills = 0 reloads = 0)
1015           TOTAL                                   0.63u  0.07s  0.21g
1016    
1017    1100   CPS 100 spill                           8.25u  0.32s  0.64g
1018           MLRISC ra                               5.68u  0.59s  3.93g
1019              (spills = 0 reloads = 0)
1020           TOTAL                                   14.71u  0.99s  4.81g
1021    
1022    1500   CPS 100 spill                           58.55u  2.34s  1.74g
1023           MLRISC ra                               5.54u  0.65s  3.91g
1024              (spills = 543 reloads = 1082)
1025           TOTAL                                   65.40u  3.13s  6.00g
1026    
1027    2000   CPS 100 spill                           126.69u  4.84s  3.08g
1028           MLRISC ra                               0.80u  0.10s  0.55g
1029              (spills = 42 reloads = 84)
1030           TOTAL                                   129.42u  5.10s  4.13g
1031    
1032    3000   CPS 100 spill                           675.59u  19.03s  11.64g
1033           MLRISC ra                               2.69u  0.27s  1.38g
1034              (spills = 62 reloads = 124)
1035           TOTAL                                   682.48u  19.61s  13.99g
1036    
1037    4000   CPS 100 spill                           2362.82u  56.28s  43.60g
1038           MLRISC ra                               4.96u  0.27s  2.72g
1039              (spills = 85 reloads = 170)
1040           TOTAL                                   2375.26u  57.21s  48.00g
1041    
1042    As you can see the old cps spill module suffers from some serious
1043    performance problem.  But since I cannot decipher the old code fully,
1044    instead of patching the problems up, I'm reimplementing it
1045    with a different algorithm.  The new code is more modular,
1046    smaller when compiled, and substantially faster
1047    (O(n log n) time and O(n) space).  Timing of the new spill module:
1048    
1049    4000  CPS 100 spill                           0.02u  0.00s  0.00g
1050          MLRISC ra                               0.25u  0.02s  0.15g
1051             (spills=1 reloads=3)
1052          TOTAL                                   7.74u  0.34s  1.62g
1053    
1054    Implementation details:
1055    
1056    As far as I can tell, the purpose of the CPS spill module is to make sure the
1057    number of live variables at any program point (the bandwidth)
1058    does not exceed a certain limit, which is determined by the
1059    size of the spill area.
1060    
1061    When the bandwidth is too large, we decrease the register pressure by
1062    packing live variables into spill records.  How we achieve this is
1063    completely different than what we did in the old code.
1064    
1065    First, there is something about the MLRiscGen code generator
1066    that we should be aware of:
1067    
1068    o MLRiscGen performs code motion!
1069    
1070       In particular, it will move floating point computations and
1071       address computations involving only the heap pointer to
1072       their use sites (if there is only a single use).
1073       What this means is that if we have a CPS record construction
1074       statement
1075    
1076           RECORD(k,vl,w,e)
1077    
1078       we should never count the new record address w as live if w
1079       has only one use (which is often the case).
1080    
1081       We should do something similar to floating point, but the transformation
1082       there is much more complex, so I won't deal with that.
1083    
1084    Secondly, there are now two new cps primops at our disposal:
1085    
1086     1. rawrecord of record_kind option
1087        This pure operator allocates some uninitialized storage from the heap.
1088        There are two forms:
1089    
1090         rawrecord NONE [INT n]  allocates a tagless record of length n
1091         rawrecord (SOME rk) [INT n] allocates a tagged record of length n
1092                                     and initializes the tag.
1093    
1094     2. rawupdate of cty
1095          rawupdate cty (v,i,x)
1096          Assigns to x to the ith component of record v.
1097          The storelist is not updated.
1098    
1099    We use these new primops for both spilling and increment record construction.
1100    
1101     1. Spilling.
1102    
1103        This is implemented with a linear scan algorithm (but generalized
1104        to trees).  The algorithm will create a single spill record at the
1105        beginning of the cps function and use rawupdate to spill to it,
1106        and SELECT or SELp to reload from it.  So both spills and reloads
1107        are fine-grain operations.  In contrast, in the old algorithm
1108        "spills" have to be bundled together in records.
1109    
1110        Ideally, we should sink the spill record construction to where
1111        it is needed.  We can even split the spill record into multiple ones
1112        at the places where they are needed.  But CPS is not a good
1113        representation for global code motion, so I'll keep it simple and
1114        am not attempting this.
1115    
1116     2. Incremental record construction (aka record splitting).
1117    
1118        Long records with many component values which are simulatenously live
1119        (recall that single use record addresses are not considered to
1120         be live) are constructed with rawrecord and rawupdate.
1121        We allocate space on the heap with rawrecord first, then gradually
1122        fill it in with rawupdate.  This is the technique suggested to me
1123        by Matthias.
1124    
1125        Some restrictions on when this is applicable:
1126        1. It is not a VECTOR record.  The code generator currently does not handle
1127           this case. VECTOR record uses double indirection like arrays.
1128        2. All the record component values are defined in the same "basic block"
1129           as the record constructor.  This is to prevent speculative
1130           record construction.
1131    
1132    ----------------------------------------------------------------------
1133    Name: Allen Leung
1134    Date: 2002/02/22 01:02:00 EST
1135    Tag: leunga-20020222-mlrisc-tools
1136    
1137    Minor bug fixes in the parser and rewriter
1138    
1139    ----------------------------------------------------------------------
1140    Name: Allen Leung
1141    Date: 2002/02/21 20:20:00 EST
1142    Tag: leunga-20020221-peephole
1143    
1144    Regenerated the peephole files.  Some contained typos in the specification
1145    and some didn't compile because of pretty printing bugs in the old version
1146    of 'nowhere'.
1147    
1148    ----------------------------------------------------------------------
1149    Name: Allen Leung
1150    Date: 2002/02/19 20:20:00 EST
1151    Tag: leunga-20020219-mlrisc-tools
1152    Description:
1153    
1154       Minor bug fixes to the mlrisc-tools library:
1155    
1156       1.  Fixed up parsing colon suffixed keywords
1157       2.  Added the ability to shut the error messages up
1158       3.  Reimplemented the pretty printer and fixed up/improved
1159           the pretty printing of handle and -> types.
1160       4.  Fixed up generation of literal symbols in the nowhere tool.
1161       5.  Added some SML keywords to to sml.sty
1162    
1163    ----------------------------------------------------------------------
1164    Name: Matthias Blume
1165    Date: 2002/02/19 16:20:00 EST
1166    Tag: blume-20020219-cmffi
1167    Description:
1168    
1169    A wild mix of changes, some minor, some major:
1170    
1171    * All C FFI-related libraries are now anchored under $c:
1172        $/c.cm      --> $c/c.cm
1173        $/c-int.cm  --> $c/internals/c-int.cm
1174        $/memory.cm --> $c/memory/memory.cm
1175    
1176    * "make" tool (in CM) now treats its argument pathname slightly
1177      differently:
1178        1. If the native expansion is an absolute name, then before invoking
1179           the "make" command on it, CM will apply OS.Path.mkRelative
1180           (with relativeTo = OS.FileSys.getDir()) to it.
1181        2. The argument will be passed through to subsequent phases of CM
1182           processing without "going native".  In particular, if the argument
1183           was an anchored path, then "make" will not lose track of that anchor.
1184    
1185    * Compiler backends now "know" their respective C calling conventions
1186      instead of having to be told about it by ml-nlffigen.  This relieves
1187      ml-nlffigen from one of its burdens.
1188    
1189    * The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
1190    
1191    * Export C_DEBUG and C_Debug from $c/c.cm.
1192    
1193    * C type encoding in ml-nlffi-lib has been improved to model the conceptual
1194      subtyping relationship between incomplete pointers and their complete
1195      counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
1196      with the convention of instantiating 'o with ('t, 'c) obj whenever
1197      the pointer target type is complete.  In the incomplete case, 'o
1198      will be instantiated with some "'c iobj" -- a type obtained by
1199      using one of the functors PointerToIncompleteType or PointerToCompleteType.
1200    
1201      Operations that work on both incomplete and complete pointer types are
1202      typed as taking an 'o ptr while operations that require the target to
1203      be known are typed as taking some ('t, 'c) obj ptr.
1204    
1205      voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
1206      where void is an eqtype without any values.  This makes it possible
1207      to work on voidptr values using functions meant to operate on light
1208      incomplete pointers.
1209    
1210    * As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
1211      been vastly simplified.
1212    
1213    ----------------------------------------------------------------------
1214    Name: Matthias Blume
1215    Date: 2002/02/19 10:48:00 EST
1216    Tag: blume-20020219-pqfix
1217    Description:
1218    
1219    Applied Chris Okasaki's bug fix for priority queues.
1220    
1221    ----------------------------------------------------------------------
1222    Name: Matthias Blume
1223    Date: 2002/02/15 17:05:00
1224    Tag: Release_110_39
1225    Description:
1226    
1227    Last-minute retagging is becoming a tradition... :-(
1228    
1229    This is the working release 110.39.
1230    
1231    ----------------------------------------------------------------------
1232    Name: Matthias Blume
1233    Date: 2002/02/15 16:00:00 EST
1234    Tag: Release_110_39-orig
1235    Description:
1236    
1237    Working release 110.39.  New bootfiles.
1238    
1239    (Update: There was a small bug in the installer so it wouldn't work
1240    with all shells.  So I retagged. -Matthias)
1241    
1242    ----------------------------------------------------------------------
1243    Name: Matthias Blume
1244    Date: 2002/02/15 14:17:00 EST
1245    Tag: blume-20020215-showbindings
1246    Description:
1247    
1248    Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
1249    the latter can be useful for exploring what bindings are available at
1250    the interactive prompt.  (The first function returns only the list
1251    of symbols that are really bound, the second prints those but also the
1252    ones that CM's autoloading mechanism knows about.)
1253    
1254    ----------------------------------------------------------------------
1255    Name: Matthias Blume
1256    Date: 2002/02/15 12:08:00 EST
1257    Tag: blume-20020215-iptrs
1258    Description:
1259    
1260    Two improvements to ml-nlffigen:
1261    
1262      1. Write files only if they do not exist or if their current contents
1263         do not coincide with what's being written.  (That is, avoid messing
1264         with the time stamps unless absolutely necessary.)
1265    
1266      2. Implement a "repository" mechanism for generated files related
1267         to "incomplete pointer types".   See the README file for details.
1268    
1269    ----------------------------------------------------------------------
1270    Name: Matthias Blume
1271    Date: 2002/02/14 11:50:00 EST
1272    Tag: blume-20020214-quote
1273    Description:
1274    
1275    Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
1276    because of the new and improved tag generation scheme.  (Thanks to Allen
1277    Leung for pointing it out.)
1278    
1279    ----------------------------------------------------------------------
1280    Name: Lal George
1281    Date: 2002/02/14 09:55:27 EST 2002
1282    Tag: george-20020214-isabelle-bug
1283    Description:
1284    
1285    Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
1286    of Isabelle on the x86.
1287    
1288    From Allen:
1289    -----------
1290     I've found the problem:
1291    
1292         in ra-core.sml, I use the counter "blocked" to keep track of the
1293         true number of elements in the freeze queue.  When the counter goes
1294         to zero, I skip examining the queue.  But I've messed up the
1295         bookkeeping in combine():
1296    
1297             else ();
1298             case !ucol of
1299               PSEUDO => (if !cntv > 0 then
1300                     (if !cntu > 0 then blocked := !blocked - 1 else ();
1301                                        ^^^^^^^^^^^^^^^^^^^^^^^
1302                      moveu := mergeMoveList(!movev, !moveu)
1303                     )
1304                  else ();
1305    
1306         combine() is called to coalesce two nodes u and v.
1307         I think I was thinking that if the move counts of u and v are both
1308         greater than zero then after they are coalesced then one node is
1309         removed from the freeze queue.  Apparently I was thinking that
1310         both u and v are of low degree, but that's clearly not necessarily true.
1311    
1312    
1313    02/12/2002:
1314        Here's the patch.  HOL now compiles.
1315    
1316        I don't know how this impact on performance (compile
1317        time or runtime).  This bug caused the RA (especially on the x86)
1318        to go thru the potential spill phase when there are still nodes on the
1319        freeze queue.
1320    
1321    
1322    
1323    
1324    ----------------------------------------------------------------------
1325    Name: Matthias Blume
1326    Date: 2002/02/13 22:40:00 EST
1327    Tag: blume-20020213-fptr-rtti
1328    Description:
1329    
1330    Fixed a bug in ml-nlffigen that was introduced with one of the previous
1331    updates.
1332    
1333    ----------------------------------------------------------------------
1334    Name: Matthias Blume
1335    Date: 2002/02/13 16:41:00 EST
1336    Tag: blume-20020213-cmlpq
1337    Description:
1338    
1339    Added new priority queue export symbols (which have just been added to
1340    smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
1341    not compile and the installer would choke.)
1342    
1343    ----------------------------------------------------------------------
1344    Name: Matthias Blume
1345    Date: 2002/02/13 16:15:00 EST
1346    Tag: blume-20020213-various
1347    Description:
1348    
1349    1. More tweaks to ml-nlffigen:
1350    
1351       - better internal datastructures (resulting in slight speedup)
1352       - "-match" option requires exact match
1353       - "localized" gensym counters (untagged structs/unions nested within
1354         other structs/unions or within typedefs get a fresh counter; their
1355         tag will be prefixed by a concatenation of their parents' tags)
1356       - bug fixes (related to calculation of transitive closure of types
1357         to be included in the output)
1358    
1359    2. Minor Basis updates:
1360    
1361       - added implementations for List.collate and Option.app
1362    
1363    ----------------------------------------------------------------------
1364    Name: Matthias Blume
1365    Date: 2002/02/11 15:55:00 EST
1366    Tag: blume-20020211-gensym
1367    Description:
1368    
1369    Added a "-gensym" option to command line of ml-nlffigen.  This can be
1370    used to specify a "stem" -- a string that is inserted in all "gensym'd"
1371    names (ML structure names that correspond to unnamed C structs, unions,
1372    and enums), so that separate runs of ml-nlffigen do not clash.
1373    
1374    ----------------------------------------------------------------------
1375    Name: Matthias Blume
1376    Date: 2002/02/11 12:05:00 EST
1377    Tag: blume-20020211-gensml
1378    Description:
1379    
1380    A quick fix for a problem with GenSML (in the pgraph-util library):
1381    Make generation of toplevel "local" optional.  (Strictly speaking,
1382    signature definitions within "local" are not legal SML.)
1383    
1384    Other than that: updates to INSTALL and cm/TODO.
1385    
1386    ----------------------------------------------------------------------
1387    Name: Matthias Blume
1388    Date: 2002/02/08 15:00:00 EST
1389    Tag: blume-20020208-uniquepid
1390    Description:
1391    
1392    0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
1393    
1394    1. The installer (config/install.sh) has gotten smarter:
1395    
1396         - Configuration options are a bit easier to specify now
1397           (in config/targets).
1398         - Bug in recognizing .tar.bz2 files fixed.
1399         - Installer automatically resolves dependencies between
1400           configuration options (e.g., if you ask for eXene, you will
1401           also get cml -- regardless whether you asked for it or not).
1402         - Installer can run in "quieter mode" by setting the environment
1403           variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
1404           "completely silent", though.
1405         - Build HashCons library as part of smlnj-lib.
1406    
1407    2. A new scheme for assigning persistent identifiers to compilation
1408       units (and, by extension, to types etc.) has been put into place.
1409       This fixes a long-standing bug where types and even dynamic values
1410       can get internally confused, thereby compromising type safety
1411       (abstraction) and dynamic correctness.  See
1412    
1413         http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
1414    
1415       for an example of how things could go wrong until now.
1416    
1417       The downside of the new scheme is that pids are not quite as
1418       persistent as they used to be: CM will generate a fresh pid
1419       for every compilation unit that it thinks it sees for the first
1420       time.  That means that if you compile starting from a clean, fresh
1421       source tree at two different times, you end up with different
1422       binaries.
1423    
1424       Cutoff recompilation, however, has not been compromised because
1425       CM keeps pid information in special caches between runs.
1426    
1427    ----------------------------------------------------------------------
1428    Name: Lal George
1429    Date: 2002/02/07 15:34:13 EST 2002
1430    Tag: <none>
1431    Description:
1432    
1433    Compilers that generate assembly code may produce  global labels
1434    whose value is resolved at link time. The various peephole optimization
1435    modules did not take this in account.
1436    
1437    TODO. The Labels.addrOf function should really return an option
1438    type so that clients are forced to deal with this issue, rather
1439    than an exception being raised.
1440    
1441    ----------------------------------------------------------------------
1442    Name: Lal George
1443    Date: 2002/02/06 13:55:02 EST
1444    Tag: george-20020206-ra-breakup
1445    Description:
1446    
1447    1. A bug fix from Allen.
1448    
1449        A typo causes extra fstp %st(0)'s to be generated at compensation
1450        edges, which might cause stack underflow traps at runtime.  This
1451        occurs in fft where there are extraneous fstps right before the 'into'
1452        trap instruction (in this case they are harmless since none of the
1453        integers overflow.)
1454    
1455    2. Pulled out various utility modules that were embedded in the modules
1456       of the register allocator. I need these modules for other purposes, but
1457       they are not complete enough to put into a library (just yet).
1458    ----------------------------------------------------------------------
1459    Name: Matthias Blume
1460    Date: 2002/01/31 16:05:00 EST
1461    Tag: blume-20020131-sparc-ccalls
1462    Description:
1463    
1464    1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
1465       of extra stack space by mistake.  Fixed.
1466    
1467    2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
1468    
1469    ----------------------------------------------------------------------
1470    Name: Allen Leung
1471    Date: 2002/01/30
1472    Tag: leunga-20020130-nowhere-bug-fix
1473    Description:
1474    
1475       MLRISC bug fixes:
1476       1. Fixed a bindings computation bug in the 'nowhere' program generator tool.
1477       2. MachineInt.fromString was negating its value.
1478    
1479    ----------------------------------------------------------------------
1480    Name: Matthias Blume
1481    Date: 2002/01/29
1482    Tag: blume-20020129-INSTALL
1483    Description:
1484    
1485    - Added somewhat detailed installation instructions (file INSTALL).
1486    - Fixed curl-detection bug in config/install.sh.
1487    - It is now possible to select the URL getter using the URLGETTER
1488      environment variable:
1489    
1490          not set / "unknown"      --> automatic detection (script tries wget,
1491                                       curl, and lynx)
1492          "wget" / "curl" / "lynx" --> use the specified program (script "knows"
1493                                       how to properly invoke them)
1494          other                    --> use $URLGETTER directly, it must take
1495                                       precisely two command-line arguments
1496                                       (source URL and destination file name)
1497    
1498    ----------------------------------------------------------------------
1499    Name: Matthias Blume
1500    Date: 2002/01/28
1501    Tag: blume-20020128-sparc-ccalls
1502    Description:
1503    
1504    - Fixed problem with calculation of "used" registers in sparc-c-calls.
1505    - Make use of the allocParam argument in sparc-c-calls.
1506    
1507    ----------------------------------------------------------------------
1508    Name: Matthias Blume
1509    Date: 2002/01/28
1510    Tag: blume-20020128-allocParam
1511    Description:
1512    
1513    John Reppy:  Changes c-calls API to accept client-callback for
1514    allocating extra stack space.
1515    me: Corresponding changes to mlriscGen (using a dummy argument that
1516        does not change the current behavior).
1517    
1518    ----------------------------------------------------------------------
1519    Name: Matthias Blume
1520    Date: 2002/01/28 12:00:00
1521    Tag: Release_110_38
1522    Description:
1523    
1524    This time for real!!!
1525    
1526    ----------------------------------------------------------------------
1527    Name: Matthias Blume
1528    Date: 2002/01/28 10:56:00 EST
1529    Tag: blume-20020128-retraction
1530    Description:
1531    
1532    0. Retracted earlier 110.38.  (The Release_110_38 tag has been replaced
1533       with blume-Release_110_38-retracted.)
1534    
1535    1. Fixed a problem with incorrect rounding modes in real64.sml.
1536       (Thanks to Andrew Mccreight <andrew.mccreight@yale.edu>.)
1537    
1538    2. A bug in ml-nlffigen related to the handling of unnamed structs, unions,
1539       and enums fixed.  The naming of corresponding ML identifiers should
1540       now be consistent again.
1541    
1542    ----------------------------------------------------------------------
1543    Name: Allen Leung
1544    Date: 2002/01/27
1545    Tag: leunga-20020127-nowhere
1546    Description:
1547    
1548       Added a target called nowhere in the configuration scripts.
1549       Enabling this will build the MLRISC 'nowhere' tool (for translating
1550       programs with where-clauses into legal SML code) during installation.
1551    
1552    ----------------------------------------------------------------------
1553    Name: Matthias Blume
1554    Date: 2002/01/25 21:27:00 EST
1555    Tag: blume-Release_110_38-retracted
1556    Description:
1557    
1558    Call it a (working) release!  Version is 110.38. Bootfiles are ready.
1559    
1560    README will be added later.
1561    
1562    !!! NOTE:  Re-tagged as blume-Release_110_38-retracted. Original tag
1563    (Release_110_38) removed.  Reason: Last-minute bug fixes.
1564    
1565    ----------------------------------------------------------------------
1566    Name: Matthias Blume
1567    Date: 2002/01/25
1568    Tag: blume-20020125-ffi
1569    Description:
1570    
1571    A large number of tweaks and improvements to ml-nlffi-lib and
1572    ml-nlffigen:
1573    
1574       - ML represenation types have been streamlined
1575       - getter and setter functions work with concrete values, not abstract
1576         ones where possible
1577       - ml-nlffigen command line more flexible (see README file there)
1578       - some bugs have been fixed (hopefully)
1579    
1580    ----------------------------------------------------------------------
1581    Name: Lal George
1582    Date: 2002/01/24
1583    Tag: george-20020124-risc-ra-interface
1584    Description:
1585    
1586       There is a dramatic simplification in the interface to the
1587       register allocator for RISC architectures as a result of making
1588       parallel copy instructions explicit.
1589    
1590    ----------------------------------------------------------------------
1591    Name: Matthias Blume
1592    Date: 2002/01/22
1593    Tag: blume-20020122-x86-ccalls
1594    Description:
1595    
1596    Bug fix for c-calls on x86 (having to do with how char- and
1597    short-arguments are being handled).
1598    
1599    ----------------------------------------------------------------------
1600    Name: Matthias Blume
1601    Date: 2002/01/21
1602    Tag: blume-20020121-ff
1603    Description:
1604    
1605    Another day of fiddling with the FFI...
1606    
1607    1. Bug fix/workaround:  CKIT does not complain about negative array
1608       dimensions, so ml-nlffigen has to guard itself against this possibility.
1609       (Otherwise a negative dimension would send it into an infinite loop.)
1610    
1611    2. Some of the abstract types (light objects, light pointers, most "base"
1612       types) in structure C are now eqtypes.
1613    
1614    3. Added constructors and test functions for NULL function pointers.
1615    
1616    ----------------------------------------------------------------------
1617    Name: Matthias Blume
1618    Date: 2002/01/18
1619    Tag: blume-20020118-ready-for-new-release
1620    Description:
1621    
1622    Made config/srcarchiveurl point to a new place.  (Will provide boot
1623    files shortly.)
1624    
1625    Maybe we christen this to be 110.38?
1626    
1627    ----------------------------------------------------------------------
1628    Name: Matthias Blume
1629    Date: 2002/01/18
1630    Tag: blume-20020118-more-ffifiddle
1631    Description:
1632    
1633    Today's FFI fiddling:
1634    
1635      - Provided a structure CGetSet with "convenient" versions of C.Get.* and
1636        C.Set.* that use concrete (MLRep.*) arguments and results instead
1637        of abstract ones.
1638    
1639      - Provided word-style bit operations etc. for "int" representation
1640        types in MLRep.S<Foo>Bitops where <Foo> ranges over Char, Int, Short,
1641        and Long.
1642    
1643    ----------------------------------------------------------------------
1644    Name: Matthias Blume
1645    Date: 2002/01/18
1646    Tag: blume-20020118-use-x86-fp
1647    Description:
1648    
1649    Now that x86-fast-fp seems to be working, I turned it back on again
1650    by default.  (Seems to work fine now, even with the FFI.)
1651    
1652    Other than that, I added some documentation about the FFI to
1653    src/ml-nlffigen/README and updated the FFI test examples in
1654    src/ml-nlffi-lib/Tests/*.
1655    
1656    ----------------------------------------------------------------------
1657    Name: Allen Leung
1658    Date: 2002/01/17
1659    Tag: leunga-20020117-x86-fast-fp-call
1660    Description:
1661    
1662       1. Fixed a problem with handling return fp values when x86's fast fp
1663          mode is turned on.
1664    
1665       2. Minor pretty printing fix for cellset.  Print %st(0) as %st(0) instead
1666          of %f32.
1667    
1668       3. Added a constructor INT32lit to the ast of MLRISC tools.
1669    
1670    ----------------------------------------------------------------------
1671    Name: Matthias Blume
1672    Date: 2002/01/16
1673    Tag: blume-20020116-ffifiddle
1674    Description:
1675    
1676    More fiddling with the FFI interface:
1677    
1678     - Make constness 'c instead of rw wherever possible.  This eliminates
1679       the need for certain explicit coercions.  (However, due to ML's
1680       value polymorphism, there will still be many cases where explicit
1681       coercions are necessary.  Phantom types are not the whole answer
1682       to modeling a subtyping relationship in ML.)
1683    
1684     - ro/rw coersions for pointers added.  (Avoids the detour through */&.)
1685    
1686     - "printf" test example added to src/ml-nlffi-lib/Tests.  (Demonstrates
1687       clumsy workaround for varargs problem.)
1688    
1689    ----------------------------------------------------------------------
1690    Name: Lal George
1691    Date: 2002/01/15
1692    Tag: <none>
1693    Description:
1694    
1695    1. Since COPY instructions are no longer native to the architecture,
1696       a generic functor can be used to implement the expandCopies function.
1697    
1698    2. Allowed EXPORT and IMPORT pseudo-op declarations to appear inside a
1699       TEXT segment.
1700    
1701    ----------------------------------------------------------------------
1702    Name: Matthias Blume
1703    Date: 2002/01/15
1704    Tag: blume-20020115-ffiupdates
1705    Description:
1706    
1707    1. Fix for bug resulting in single-precision float values being returned
1708       incorrectly from FFI calls.
1709    
1710    2. Small modifications to C FFI API:
1711    
1712        - memory-allocation routines return straight objects (no options)
1713          and raise an exception in out-of-memory situations
1714        - unsafe extensions to cast between function pointers and pointers
1715          from/to ints
1716        - added structure C_Debug as an alternative to structure C where
1717          pointer-dereferencing (|*| and |*!) always check for null-pointers
1718        - added open_lib' to DynLinkage;  open_lib' works like open_lib
1719          but also takes a (possibly empty) list of existing library handles
1720          that the current library depends on
1721    
1722    ----------------------------------------------------------------------
1723    Name: Matthias Blume
1724    Date: 2002/01/10
1725    Tag: blume-20020110-newffigen
1726    Description:
1727    
1728    1. Updates to portable graph code.
1729    
1730    2. Major update to ml-nlffigen and ml-nlffi-lib.  Things are much
1731       more scalable now so that even huge interfaces such as the one
1732       for GTK compile in finite time and space. :-)
1733       See src/ml-nlffigen/README for details on what's new.
1734    
1735    ----------------------------------------------------------------------
1736    Name: Lal George
1737    Date: 2001/01/09 14:31:35 EST 2002
1738    Tag: george-20011206-rm-native-copy
1739    Description:
1740    
1741            Removed the native COPY and FCOPY instructions
1742            from all the architectures and replaced it with the
1743            explicit COPY instruction from the previous commit.
1744    
1745            It is now possible to simplify many of the optimizations
1746            modules that manipulate copies. This has not been
1747            done in this change.
1748    
1749    ----------------------------------------------------------------------
1750    Name: Lal George
1751    Date: 2001/12/06 16:50:13 EST 2001
1752    Tag: george-20011206-mlrisc-instruction
1753    Description:
1754    
1755    Changed the representation of instructions from being fully abstract
1756    to being partially concrete. That is to say:
1757    
1758      from
1759            type instruction
1760    
1761      to
1762            type instr                              (* machine instruction *)
1763    
1764            datatype instruction =
1765                LIVE of {regs: C.cellset, spilled: C.cellset}
1766              | KILL of {regs: C.cellset, spilled: C.cellset}
1767              | COPYXXX of {k: CB.cellkind, dst: CB.cell list, src: CB.cell list}
1768              | ANNOTATION of {i: instruction, a: Annotations.annotation}
1769              | INSTR of instr
1770    
1771    This makes the handling of certain special instructions that appear on
1772    all architectures easier and uniform.
1773    
1774    LIVE and KILL say that a list of registers are live or killed at the
1775    program point where they appear. No spill code is generated when an
1776    element of the 'regs' field is spilled, but the register is moved to
1777    the 'spilled' (which is present, more for debugging than anything else).
1778    
1779    LIVE replaces the (now deprecated) DEFFREG instruction on the alpha.
1780    We used to generate:
1781    
1782            DEFFREG f1
1783            f1 := f2 + f3
1784            trapb
1785    
1786    but now generate:
1787    
1788            f1 := f2 + f3
1789            trapb
1790            LIVE {regs=[f1,f2,f3], spilled=[]}
1791    
1792    Furthermore, the DEFFREG (hack) required that all floating point instruction
1793    use all registers mentioned in the instruction. Therefore f1 := f2 + f3,
1794    defines f1 and uses [f1,f2,f3]! This hack is no longer required resulting
1795    in a cleaner alpha implementation. (Hopefully, intel will not get rid of
1796    this architecture).
1797    
1798    COPYXXX is intended to replace the parallel COPY and FCOPY  available on
1799    all the architectures. This will result in further simplification of the
1800    register allocator that must be aware of them for coalescing purposes, and
1801    will also simplify certain aspects of the machine description that provides
1802    callbacks related to parallel copies.
1803    
1804    ANNOTATION should be obvious, and now INSTR represents the honest to God
1805    machine instruction set!
1806    
1807    The <arch>/instructions/<arch>Instr.sml files define certain utility
1808    functions for making porting easier -- essentially converting upper case
1809    to lower case. All machine instructions (of type instr) are in upper case,
1810    and the lower case form generates an MLRISC instruction. For example on
1811    the alpha we have:
1812    
1813      datatype instr =
1814         LDA of {r:cell, b:cell, d:operand}
1815       | ...
1816    
1817      val lda : {r:cell, b:cell, d:operand} -> instruction
1818        ...
1819    
1820    where lda is just (INSTR o LDA), etc.
1821    
1822    ----------------------------------------------------------------------
1823    Name: Matthias Blume
1824    Date: 2001/11/22 21:40:00 EST
1825    Tag: Release_110_37
1826    Description:
1827    
1828    Release 110.37.  This time for real.
1829    
1830    ----------------------------------------------------------------------
1831    Name: Matthias Blume
1832    Date: 2001/11/21 16:35:00 EST
1833    Tag: blume-20011121-foot-in-mouth
1834    Description:
1835    
1836    Removed the "Release_110_37" tag because of a serious bug.
1837    This will be re-tagged once the bug is fixed.
1838    
1839    ----------------------------------------------------------------------
1840    Name: Matthias Blume
1841    Date: 2001/11/21 16:14:00 EST
1842    Tag: blume-20011121-forgottenfile
1843    Description:
1844    
1845    Forgot to add a file.  (Just a .tex-file -- part of
1846    the CM manual source.)
1847    
1848    ----------------------------------------------------------------------
1849    Name: Matthias Blume
1850    Date: 2001/11/21 16:10:00 EST
1851    Tag: blume-20011121-invalid_110_37
1852    Description:
1853    
1854    Note: I removed the original tag "Release_110_37" from this commit
1855          because we found a serious bug in all non-x86 backends.
1856          - Matthias
1857    
1858    1. Modifications to the SML/NJ code generator and to the runtime system
1859       so that code object name strings are directly inserted into code
1860       objects at code generation time.  The only business the runtime system
1861       has with this is now to read the name strings on occasions.
1862       (The encoding of the name string has also changed somewhat.)
1863    
1864    2. CM now implements a simple "set calculus" for specifying export lists.
1865       In particular, it is now possible to refer to the export lists of
1866       other libraries/groups/sources and form unions as well as differences.
1867       See the latest CM manual for details.
1868    
1869    3. An separate notion of "proxy" libraries has again be eliminated from
1870       CM's model.  (Proxy libraries are now simply a special case of using
1871       the export list calculus.)
1872    
1873    4. Some of the existing libraries now take advantage of the new set
1874       calculus.
1875       (Notice that not all libraries have been converted because some
1876       of the existing .cm-files are supposed to be backward compatible
1877       with 110.0.x.)
1878    
1879    5. Some cleanup in stand-alone programs.  (Don't use "exnMessage" -- use
1880       "General.exnMessage"!  The former relies on a certain hook to be
1881       initialized, and that often does not happen in the stand-alone case.)
1882    
1883    ----------------------------------------------------------------------
1884    Name: Lal George
1885    Date: 2001/11/21  13:56:18 EST
1886    Tag: george-2001121-pseudo-ops
1887    Description:
1888    
1889      Implemented a complete redesign of MLRISC pseudo-ops. Now there
1890      ought to never be any question of incompatabilities with
1891      pseudo-op syntax expected by host assemblers.
1892    
1893      For now, only modules supporting GAS syntax are implemented
1894      but more should follow, such as MASM, and vendor assembler
1895      syntax, e.g. IBM as, Sun as, etc.
1896    
1897    ----------------------------------------------------------------------
1898    Name: Matthias Blume
1899  Date: 2001/11/14 11:52:00 EST  Date: 2001/11/14 11:52:00 EST
1900  Tag: blume-20011114-srcname  Tag: blume-20011114-srcname
1901  Description:  Description:
# Line 194  Line 2076 
2076    
2077  ----------------------------------------------------------------------  ----------------------------------------------------------------------
2078  Name: Matthias Blume  Name: Matthias Blume
 >>>>>>> 1.169  
2079  Date: 2001/09/18 15:35:00 EDT  Date: 2001/09/18 15:35:00 EDT
2080  Tag: blume-20010918-readme11036  Tag: blume-20010918-readme11036
2081  Description:  Description:

Legend:
Removed from v.977  
changed lines
  Added in v.1212

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