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 1078, Tue Feb 19 21:26:48 2002 UTC revision 1215, Tue May 21 20:05:42 2002 UTC
# Line 14  Line 14 
14    
15  ----------------------------------------------------------------------  ----------------------------------------------------------------------
16  Name: Matthias Blume  Name: Matthias Blume
17    Date: 2002/05/21 16:00:00 EDT
18    Tag: Release_110_40
19    Description:
20    
21    New working release (110.40).  New bootfiles.
22    
23    [Also: John Reppy fixed GC bug 1131.]
24    
25    ----------------------------------------------------------------------
26    Name: Matthias Blume
27    Date: 2002/05/21 12:35:00 EDT
28    Tag: blume-20020521-cmdoc
29    Description:
30    
31    CM documentation update.
32    
33    ----------------------------------------------------------------------
34    Name: Matthias Blume
35    Date: 2002/05/21 10:55:00 EDT
36    Tag: blume-20020521-misc
37    Description:
38    
39    - John tweaked runtime to be silent on heap export (except when
40      GC messages are on).
41    - I added a few more things (cross-compiling versions of CMB) to
42      config/preloads (as suggestions).
43    
44    ----------------------------------------------------------------------
45    Name: Matthias Blume
46    Date: 2002/05/20 22:25:00 EDT
47    Tag: blume-20020520-controls
48    Description:
49    
50    - Added ControlUtil structure to control-lib.cm.
51    - Use it throughout.
52    - Used Controls facility to define MLRISC controls (as opposed to
53      registering MLRISC control ref cells with Controls after the fact)
54    - Fixed messed-up controls priorities.
55    
56    * Removed again all the stuff from config/preloads that one wouldn't
57      be able to preload at the time the initial heap image is built.
58      (Many libraries, e.g., CML, do not exist yet at this time.  The
59       only libraries that can be preloaded via config/preloads are those
60       that come bundled with the bootfiles.)
61    
62    ----------------------------------------------------------------------
63    Name: Matthias Blume
64    Date: 2002/05/20 10:59:00 EDT
65    Tag: blume-20020520-preloads
66    Description:
67    
68    Added a lot of commented-out suggestions for things to be included
69    in config/preloads.
70    
71    ----------------------------------------------------------------------
72    Name: Allen Leung
73    Date: 2002/05/18 14:20:00 EDT
74    Tag: leunga-20020518-mdl
75    Description:
76    
77    o Made the mdl tool stuff compile and run again.
78    o I've disabled all the stuff that depends on RTL specifications; they
79      are all badly broken anyway.
80    
81    ----------------------------------------------------------------------
82    Name: Matthias Blume
83    Date: 2002/05/17 16:49:00 EDT
84    Tag: blume-20020517-controls
85    Description:
86    
87    0. John Reppy made several modifications to the SML/NJ library.
88       In particular, there is a shiny new controls-lib.cm.
89    
90    1. Pushed new controls interface through compiler so that everything
91       compiles again.
92    
93    2. Added FormatComb and FORMAT_COMB to the CML version of the
94       SML/NJ library (so that CML compiles again).
95    
96    3. Modified init scripts because XXX_DEFAULT environment variables
97       are no longer with us.  (Boot-time initialization is now done
98       using the same environment variables that are also used for
99       startup-time initialization of controls.)
100    
101    ----------------------------------------------------------------------
102    Name: Lal George
103    Date: 2002/05/15 09:20:10 EDT
104    Tag: george-20020515-pseudo-op-decls
105    Description:
106    
107            All pseudo-ops emitted before the first segment declaration
108    such as TEXT, DATA, and BSS directives are assumed to be global
109    declarations and are emitted first in the assembly file. This is
110    useful in a number of situations where one has pseudo-ops that are not
111    specific to any segment, and also works around the constraint that one
112    cannot have client pseudo-ops in the TEXT segment.
113    
114            Because no segment is associated with these declarations it is
115    an error to allocate any space or objects before the first segment
116    directive and an exception will be raised. However, we cannot make
117    this check for client pseudo-ops.
118    
119            These top level declarations are a field in the CFG graph_info.
120    In theory you can continue to add to this field after the CFG has been
121    built -- provided you know what you are doing;-)
122    
123    ----------------------------------------------------------------------
124    Name: Matthias Blume
125    Date: 2002/05/13 16:40:00 EDT
126    Tag: blume-20020513-pp-etc
127    Description:
128    
129    A few minor bugfixes:
130    
131      - Stopgap measure for bug recently reported by Elsa Gunter (ppDec).
132        (Bogus printouts for redefined bindings still occur.  Compiler
133        bug should no longer occur now.  We need to redo the prettyprinter
134        from scratch.)
135    
136      - CM pathname printer now also adds escape sequences for ( and )
137    
138      - commend and docu fixes for ml-nlffi
139    
140    ----------------------------------------------------------------------
141    Name: Matthias Blume
142    Date: 2002/05/10 16:40:00 EDT
143    Tag: blume-20020510-erg-textio
144    Description:
145    
146    Applied the following bugfix provided by Emden Gansner:
147    
148        Output is corrupted when outputSubstr is used rather than output.
149    
150        The problem occurs when a substring
151    
152            ss = (s, dataStart, dataLen)
153    
154        where dataStart > 0, fills a stream buffer with avail bytes left.
155        avail bytes of s, starting at index dataStart, are copied into the
156        buffer, the buffer is flushed, and then the remaining dataLen-avail
157        bytes of ss are copied into the beginning of the buffer. Instead of
158        starting this copy at index dataStart+avail in s, the current code
159        starts the copy at index avail.
160    
161        Fix:
162        In text-io-fn.sml, change line 695 from
163             val needsFlush = copyVec(v, avail, dataLen-avail, buf, 0)
164        to
165             val needsFlush = copyVec(v, dataStart+avail, dataLen-avail, buf, 0)
166    
167    ----------------------------------------------------------------------
168    Name: Matthias Blume
169    Date: 2002/04/12 13:55:00 EDT
170    Tag: blume-20020412-assyntax
171    Description:
172    
173    1. Grabbed newer assyntax.h from the XFree86 project.
174    2. Fiddled with how to compile X86.prim.asm without warnings.
175    3. (Very) Minor cleanup in CM.
176    
177    ----------------------------------------------------------------------
178    Name: Matthias Blume
179    Date: 2002/04/01 (no joke!) 17:07:00 EST
180    Tag: blume-20020401-x86div
181    Description:
182    
183    Added full support for div/mod/rem/quot on the x86, using the machine
184    instruction's two results (without clumsily recomputing the remainder)
185    directly where appropriate.
186    
187    Some more extensive power-of-two support was added to the x86 instruction
188    selector (avoiding expensive divs, mods, and muls where they can be
189    replaced with cheaper shifts and masks).  However, this sort of thing
190    ought to be done earlier, e.g., within the CPS optimizer so that
191    all architectures benefit from it.
192    
193    The compiler compiles to a fixed point, but changes might be somewhat
194    fragile nevertheless.  Please, report any strange things that you might
195    see wrt. div/mod/quot/rem...
196    
197    ----------------------------------------------------------------------
198    Name: Matthias Blume
199    Date: 2002/03/29 17:22:00
200    Tag: blume-20020329-div
201    Description:
202    
203    Fixed my broken div/mod logic.  Unfortunately, this means that the
204    inline code for div/mod now has one more comparison than before.
205    Fast paths (quotient > 0 or remainder = 0) are not affected, though.
206    The problem was with quotient = 0, because that alone does not tell
207    us which way the rounding went.  One then has to look at whether
208    remainder and divisor have the same sign...  :(
209    
210    Anyway, I replaced the bootfiles with fresh ones...
211    
212    ----------------------------------------------------------------------
213    Name: Matthias Blume
214    Date: 2002/03/29 14:10:00 EST
215    Tag: blume-20020329-inlprims
216    Description:
217    
218    NEW BOOTFILES!!!    Version number bumped to 110.39.3.
219    
220    Primops have changed. This means that the bin/boot-file formats have
221    changed as well.
222    
223    To make sure that there is no confusion, I made a new version.
224    
225    
226    CHANGES:
227    
228    * removed REMT from mltree (remainder should never overflow).
229    
230    * added primops to deal with divisions of all flavors to the frontend
231    
232    * handled these primops all the way through so they map to their respective
233      MLRISC support
234    
235    * used these primops in the implementation of Int, Int32, Word, Word32
236    
237    * removed INLDIV, INLMOD, and INLREM as they are no longer necessary
238    
239    * parameterized INLMIN, INLMAX, and INLABS by a numkind
240    
241    * translate.sml now deals with all flavors of INL{MIN,MAX,ABS}, including
242      floating point
243    
244    * used INL{MIN,MAX,ABS} in the implementation of Int, Int32, Word, Word32,
245      and Real (but Real.abs maps to a separate floating-point-only primop)
246    
247    
248    TODO items:
249    
250    * Hacked Alpha32 instruction selection, disabling the selection of REMx
251      instructions because the machine instruction encoder cannot handle
252      them.  (Hppa, PPC, and Sparc instruction selection did not handle
253      REM in the first place, and REM is supported by the x86 machine coder.)
254    
255    * Handle DIV and MOD with DIV_TO_NEGINF directly in the x86 instruction
256      selection phase.  (The two can be streamlined because the hardware
257      delivers both quotient and remainder at the same time anyway.)
258    
259    * Think about what to do with "valOf(Int32.minInt) div ~1" and friends.
260      (Currently the behavior is inconsistent both across architectures and
261      wrt. the draft Basis spec.)
262    
263    * Word8 should eventually be handled natively, too.
264    
265    * There seems to be one serious bug in mltree-gen.sml.  It appears, though,
266      as if there currently is no execution path that could trigger it in
267      SML/NJ.  (The assumptions underlying functions arith and promotable do not
268      hold for things like multiplication and division.)
269    
270    ----------------------------------------------------------------------
271    Name: Matthias Blume
272    Date: 2002/03/27 16:27:00 EST
273    Tag: blume-20020327-mlrisc-divisions
274    Description:
275    
276    Added support for all four division operations (ML's div, mod, quot,
277    and rem) to MLRISC.  In the course of doing so, I also rationalized
278    the naming (no more annoying switch-around of DIV and QUOT), by
279    parameterizing the operation by div_rounding_mode (which can be either
280    DIV_TO_ZERO or DIV_TO_NEGINF).
281    
282    The generic MLTreeGen functor takes care of compiling all four
283    operations down to only round-to-zero div.
284    
285    Missing pieces:
286    
287      * Doing something smarter than relying on MLTreeGen on architectures
288        like, e.g., the x86 where hardware division delivers both quotient and
289        remainder at the same time.  With this, the implementation of the
290        round-to-neginf operations could be further streamlined.
291    
292      * Remove inlining support for div/mod/rem from the frontend and replace it
293        with primops that get carried through to the backend.  Do this for all
294        int and word types.
295    
296    ----------------------------------------------------------------------
297    Name: Matthias Blume
298    Date: 2002/03/25 17:25:00 EST
299    Tag: blume-20020325-divmod
300    Description:
301    
302    I improved (hopefully without breaking them) the implementation of Int.div,
303    Int.mod, and Int.rem.   For this, the code in translate.sml now takes
304    advantage of the following observations:
305    
306      Let  q = x quot y      r = x rem y
307           d = x div  y      m = x mod y
308    
309    where "quot" is the round-to-zero version of integer division that
310    hardware usually provides.  Then we have:
311    
312         r = x - q * y        where neither the * nor the - will overflow
313         d = if q >= 0 orelse x = q * y then q else q - 1
314                              where neither the * nor the - will overflow
315         m = if q >= 0 orelse r = 0 then r else r + y
316                              where the + will not overflow
317    
318    This results in substantial simplification of the generated code.
319    The following table shows the number of CFG nodes and edges generated
320    for
321            fun f (x, y) = x OPER y
322            (* with OPER \in div, mod, quot, rem *)
323    
324    
325        OPER | nodes(old) | edges(old) | nodes(new) | edges(new)
326        --------------------------------------------------------
327         div |         24 |         39 |         12 |         16
328         mod |         41 |         71 |         12 |         16
329        quot |          8 |         10 |          8 |         10
330         rem |         10 |         14 |          8 |         10
331    
332    
333    ----------------------------------------------------------------------
334    Name: Matthias Blume
335    Date: 2002/03/25 22:06:00 EST
336    Tag: blume-20020325-cprotobug
337    Description:
338    
339    Fixed a bug in cproto (c prototype decoder).
340    
341    ----------------------------------------------------------------------
342    Name: Matthias Blume
343    Date: 2002/03/25 16:00:00 EST
344    Tag: blume-20020325-raw-primops
345    Description:
346    
347    I did some cleanup to Allen's new primop code and
348    replaced yesterday's bootfiles with new ones.
349    (But they are stored in the same place.)
350    
351    ----------------------------------------------------------------------
352    Name: Matthias Blume
353    Date: 2002/03/24 22:40:00 EST
354    Tag: blume-20020324-bootfiles
355    Description:
356    
357    Made the bootfiles that Allen asked for.
358    
359    ----------------------------------------------------------------------
360    Name: Allen Leung
361    Date: 2002/03/23 15:50:00 EST
362    Tag: leunga-20020323-flint-cps-rcc-primops
363    Description:
364    
365      1. Changes to FLINT primops:
366    
367        (* make a call to a C-function;
368         * The primop carries C function prototype information and specifies
369         * which of its (ML-) arguments are floating point. C prototype
370         * information is for use by the backend, ML information is for
371         * use by the CPS converter. *)
372      | RAW_CCALL of { c_proto: CTypes.c_proto,
373                       ml_args: ccall_type list,
374                       ml_res_opt: ccall_type option,
375                       reentrant : bool
376                     } option
377       (* Allocate uninitialized storage on the heap.
378        * The record is meant to hold short-lived C objects, i.e., they
379        * are not ML pointers.  With the tag, the representation is
380        * the same as RECORD with tag tag_raw32 (sz=4), or tag_fblock (sz=8)
381        *)
382      | RAW_RECORD of {tag:bool,sz:int}
383      and ccall_type = CCALL_INT32 | CCALL_REAL64 | CCALL_ML_PTR
384    
385      2.  These CPS primops are now overloaded:
386    
387           rawload of {kind:numkind}
388           rawstore of {kind:numkind}
389    
390          The one argument form is:
391    
392             rawload {kind} address
393    
394          The two argument form is:
395    
396             rawload {kind} [ml object, byte-offset]
397    
398      3. RAW_CCALL/RCC now takes two extra arguments:
399    
400         a. The first is whether the C call is reentrant, i.e., whether
401            ML state should be saved and restored.
402         b. The second argument is a string argument specifying the name of
403            library and the C function.
404    
405         These things are currently not handled in the code generator, yet.
406    
407      4. In CProto,
408    
409         An encoding type of "bool" means "ml object" and is mapped into
410         C prototype of PTR.  Note that "bool" is different than "string",
411         even though "string" is also mapped into PTR, because "bool"
412         is assigned an CPS type of BOGt, while "string" is assigned INT32t.
413    
414      5. Pickler/unpicker
415    
416         Changed to handle RAW_RECORD and newest RAW_CCALL
417    
418      6. MLRiscGen,
419    
420         1. Changed to handle the new rawload/rawstore/rawrecord operators.
421         2. Code for handling C Calls has been moved to a new module CPSCCalls,
422            in the file CodeGen/cpscompile/cps-c-calls.sml
423    
424      7. Added the conditional move operator
425    
426             condmove of branch
427    
428         to cps.  Generation of this is still buggy so it is currently
429         disabled.
430    
431    ----------------------------------------------------------------------
432    Name: Lal George
433    Date: 2002/03/22 14:18:25 EST
434    Tag: george-20020322-cps-branch-prob
435    Description:
436    
437    Implemented the Ball-Larus branch prediction-heuristics, and
438    incorporated graphical viewers for control flow graphs.
439    
440    Ball-Larus Heuristics:
441    ---------------------
442    See the file compiler/CodeGen/cpscompile/cpsBranchProb.sml.
443    
444    By design it uses the Dempster-Shafer theory for combining
445    probabilities.  For example, in the function:
446    
447        fun f(n,acc) = if n = 0 then acc else f(n-1, n*acc)
448    
449    the ball-larus heuristics predicts that the n=0 is unlikely
450    (OH-heuristic), and the 'then' branch is unlikely because of the
451    RH-heuristic -- giving the 'then' branch an even lower combined
452    probability using the Dempster-Shafer theory.
453    
454    Finally, John Reppy's loop analysis in MLRISC, further lowers the
455    probability of the 'then' branch because of the loop in the else
456    branch.
457    
458    
459    Graphical Viewing:
460    ------------------
461    I merely plugged in Allen's graphical viewers into the compiler. The
462    additional code is not much. At the top level, saying:
463    
464            Control.MLRISC.getFlag "cfg-graphical-view" := true;
465    
466    will display the graphical view of the control flow graph just before
467    back-patching.  daVinci must be in your path for this to work. If
468    daVinci is not available, then the default viewer can be changed
469    using:
470    
471            Control.MLRISC.getString "viewer"
472    
473    which can be set to "dot" or "vcg" for the corresponding viewers. Of
474    course, these viewers must be in your path.
475    
476    The above will display the compilation unit at the level of clusters,
477    many of which are small, boring, and un-interesting. Also setting:
478    
479            Control.MLRISC.getInt "cfg-graphical-view_size"
480    
481    will display clusters that are larger than the value set by the above.
482    
483    
484    ----------------------------------------------------------------------
485    Name: Matthias Blume
486    Date: 2002/03/21 22:20:00 EST
487    Tag: blume-20020321-kmp-bugfix
488    Description:
489    
490    Changed the interface to the KMP routine in PreString and fixed
491    a minor bug in one place where it was used.
492    
493    ----------------------------------------------------------------------
494    Name: Allen Leung
495    Date: 2002/03/21 20:30:00 EST
496    Tag: leunga-20020321-cfg
497    Description:
498    
499      Fixed a potential problem in cfg edge splitting.
500    
501    ----------------------------------------------------------------------
502    Name: Allen Leung
503    Date: 2002/03/21 17:15:00 EST
504    Tag: leunga-20020321-x86-fp-cfg
505    Description:
506    
507      1. Recoded the buggy parts of x86-fp.
508    
509         a. All the block reordering code has been removed.
510            We now depend on the block placement phases to do this work.
511    
512         b. Critical edge splitting code has been simplified and moved into the
513            CFG modules, as where they belong.
514    
515         Both of these were quite buggy and complex.  The code is now much, much
516         simpler.
517    
518      2. X86 backend.
519    
520         a. Added instructions for 64-bit support.  Instruction selection for
521            64-bit has not been committed, however, since that
522            requires changes to MLTREE which haven't been approved by
523            Lal and John.
524    
525         b. Added support for FUCOMI and FUCOMIP when generating code for
526            PentiumPro and above.  We only generate these instructions in
527            the fast-fp mode.
528    
529         c. Added cases for JP and JNP in X86FreqProps.
530    
531      3. CFG
532    
533         CFG now has a bunch of methods for edge splitting and merging.
534    
535      4. Machine description.
536    
537         John's simplification of MLTREE_BASIS.fcond broke a few machine
538         description things:
539    
540         rtl-build.{sig,sml} and hppa.mdl fixed.
541    
542         NOTE: the machine description stuff in the repository is still broken.
543               Again, I can't put my fixes in because that involves
544               changes to MLTREE.
545    
546    ----------------------------------------------------------------------
547    Name: Matthias Blume
548    Date: 2002/03/20 15:55:00 EST
549    Tag: blume-20020320-kmp
550    Description:
551    
552    Implemented Knuth-Morris-Pratt string matching in PreString and used
553    it for String.isSubstring, Substring.isSubstring, and
554    Substring.position.
555    
556    (Might need some stress-testing.  Simple examples worked fine.)
557    
558    ----------------------------------------------------------------------
559    Name: Matthias Blume
560    Date: 2002/03/19 16:37:00 EST
561    Tag: blume-20020319-witnesses
562    Description:
563    
564    Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
565    
566    This implements a generic mechanism for changing constness qualifiers
567    anywhere within big C types without resorting to outright "casts".
568    (So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
569    modify the constness at the outermost level.)
570    The implementation of "convert" is based on the idea of "witness"
571    values -- values that are not used by the operation but whose types
572    "testify" to their applicability.  On the implementation side, "convert"
573    is simply a projection (returning its second curried argument).  With
574    cross-module inlining, it should not result in any machine code being
575    generated.
576    
577    ----------------------------------------------------------------------
578    Name: Matthias Blume
579    Date: 2002/03/15 16:40:00 EST
580    Tag: blume-20020315-basis
581    Description:
582    
583    Provided (preliminary?) implementations for
584    
585      {String,Substring}.{concatWith,isSuffix,isSubstring}
586    
587    and
588    
589      Substring.full
590    
591    Those are in the Basis spec but they were missing in SML/NJ.
592    
593    ----------------------------------------------------------------------
594    Name: Matthias Blume
595    Date: 2002/03/14 21:30:00 EST
596    Tag: blume-20020314-controls
597    Description:
598    
599    Controls:
600    ---------
601    
602    1. Factored out the recently-added Controls : CONTROLS stuff and put
603       it into its own library $/controls-lib.cm.  The source tree for
604       this is under src/smlnj-lib/Controls.
605    
606    2. Changed the names of types and functions in this interface, so they
607       make a bit more "sense":
608    
609          module -> registry
610          'a registry -> 'a group
611    
612    3. The interface now deals in ref cells only.  The getter/setter interface
613       is (mostly) gone.
614    
615    4. Added a function that lets one register an already-existing ref cell.
616    
617    5. Made the corresponding modifications to the rest of the code so that
618       everything compiles again.
619    
620    6. Changed the implementation of Controls.MLRISC back to something closer
621       to the original.  In particular, this module (and therefore MLRISC)
622       does not depend on Controls.  There now is some link-time code in
623       int-sys.sml that registers the MLRISC controls with the Controls
624       module.
625    
626    CM:
627    ---
628    
629      * One can now specify the lambda-split aggressiveness in init.cmi.
630    
631    ----------------------------------------------------------------------
632    Name: Allen Leung
633    Date: 2002/03/13 17:30:00 EST
634    Tag: leunga-20020313-x86-fp-unary
635    Description:
636    
637    Bug fix for:
638    
639    > leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
640    > Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
641    > - fun f(x,(y,z)) = Real.~ y;
642    > [autoloading]
643    > [autoloading done]
644    >       fchsl   (%eax), 184(%esp)
645    > Error: MLRisc bug: X86MCEmitter.emitInstr
646    >
647    > uncaught exception Error
648    >   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
649    
650    The problem was that the code generator did not generate any fp registers
651    in this case, and the ra didn't know that it needed to run the X86FP phase to
652    translate the pseudo fp instruction.   This only happened with unary fp
653    operators in certain situations.
654    
655    ----------------------------------------------------------------------
656    Name: Matthias Blume
657    Date: 2002/03/13 14:00:00 EST
658    Tag: blume-20020313-overload-etc
659    Description:
660    
661    1. Added _overload as a synonym for overload for backward compatibility.
662       (Control.overloadKW must be true for either version to be accepted.)
663    
664    2. Fixed bug in install script that caused more things to be installed
665       than what was requested in config/targets.
666    
667    3. Made CM aware of the (_)overload construct so that autoloading
668       works.
669    
670    ----------------------------------------------------------------------
671    Name: Matthias Blume
672    Date: 2002/03/12 22:03:00 EST
673    Tag: blume-20020312-url
674    Description:
675    
676    Forgot to update BOOT and srcarchiveurl.
677    
678    ----------------------------------------------------------------------
679    Name: Matthias Blume
680    Date: 2002/03/12 17:30:00 EST
681    Tag: blume-20020312-version110392
682    Description:
683    
684    Yet another version number bump (because of small changes to the
685    binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
686    
687    Changes:
688    
689      The new pid generation scheme described a few weeks ago was overly
690      complicated.  I implemented a new mechanism that is simpler and
691      provides a bit more "stability":  Once CM has seen a compilation
692      unit, it keeps its identity constant (as long as you do not delete
693      those crucial CM/GUID/* files).  This means that when you change
694      an interface, compile, then go back to the old interface, and
695      compile again, you arrive at the original pid.
696    
697      There now also is a mechanism that instructs CM to use the plain
698      environment hash as a module's pid (effectively making its GUID
699      the empty string).  For this, "noguid" must be specified as an
700      option to the .sml file in question within its .cm file.
701      This is most useful for code that is being generated by tools such
702      as ml-nlffigen (because during development programmers tend to
703      erase the tool's entire output directory tree including CM's cached
704      GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
705      revert to the old, broken behavior of SML/NJ, but in specific cases
706      where there is no danger of interface confusion, its use is ok
707      (I think).
708    
709      ml-nlffigen by default generates "noguid" annotations.  They can be
710      turned off by specifying -guid in its command line.
711    
712    ----------------------------------------------------------------------
713    Name: Lal George
714    Date: 2002/03/12 12 14:42:36 EST
715    Tag: george-20020312-frequency-computation
716    Description:
717    
718    Integrated jump chaining and static block frequency into the
719    compiler. More details and numbers later.
720    
721    ----------------------------------------------------------------------
722    Name: Lal George
723    Date: 2002/03/11 11 22:38:53 EST
724    Tag: george-20020311-jump-chain-elim
725    Description:
726    
727    Tested the jump chain elimination on all architectures (except the
728    hppa).  This is on by default right now and is profitable for the
729    alpha and x86, however, it may not be profitable for the sparc and ppc
730    when compiling the compiler.
731    
732    The gc test will typically jump to a label at the end of the cluster,
733    where there is another jump to an external cluster containing the actual
734    code to invoke gc. This is to allow factoring of common gc invocation
735    sequences. That is to say, we generate:
736    
737            f:
738               testgc
739               ja   L1      % jump if above to L1
740    
741            L1:
742               jmp L2
743    
744    
745    After jump chain elimination the 'ja L1' instructions is converted to
746    'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
747    up being implemented in their long form (if L2 is far away) using:
748    
749            jbe     L3      % jump if below or equal to L3
750            jmp     L2
751         L3:
752            ...
753    
754    
755    For large compilation units L2  may be far away.
756    
757    
758    ----------------------------------------------------------------------
759    Name: Matthias Blume
760    Date: 2002/03/11 13:30:00 EST
761    Tag: blume-20020311-mltreeeval
762    Description:
763    
764    A functor parameter was missing.
765    
766    ----------------------------------------------------------------------
767    Name: Allen Leung
768    Date: 2002/03/11 10:30:00 EST
769    Tag: leunga-20020311-runtime-string0
770    Description:
771    
772       The representation of the empty string now points to a
773    legal null terminated C string instead of unit.  It is now possible
774    to convert an ML string into C string with InlineT.CharVector.getData.
775    This compiles into one single machine instruction.
776    
777    ----------------------------------------------------------------------
778    Name: Allen Leung
779    Date: 2002/03/10 23:55:00 EST
780    Tag: leunga-20020310-x86-call
781    Description:
782    
783       Added machine generation for CALL instruction (relative displacement mode)
784    
785    ----------------------------------------------------------------------
786    Name: Matthias Blume
787    Date: 2002/03/08 16:05:00
788    Tag: blume-20020308-entrypoints
789    Description:
790    
791    Version number bumped to 110.39.1.  NEW BOOTFILES!
792    
793    Entrypoints: non-zero offset into a code object where execution should begin.
794    
795    - Added the notion of an entrypoint to CodeObj.
796    - Added reading/writing of entrypoint info to Binfile.
797    - Made runtime system bootloader aware of entrypoints.
798    - Use the address of the label of the first function given to mlriscGen
799      as the entrypoint.  This address is currently always 0, but it will
800      not be 0 once we turn on block placement.
801    - Removed the linkage cluster code (which was The Other Way(tm) of dealing
802      with entry points) from mlriscGen.
803    
804    ----------------------------------------------------------------------
805    Name: Allen Leung
806    Date: 2002/03/07 20:45:00 EST
807    Tag: leunga-20020307-x86-cmov
808    Description:
809    
810       Bug fixes for CMOVcc on x86.
811    
812       1. Added machine code generation for CMOVcc
813       2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
814       3. CMOVcc cannot have an immediate operand as argument.
815    
816    ----------------------------------------------------------------------
817    Name: Matthias Blume
818    Date: 2002/03/07 16:15:00 EST
819    Tag: blume-20020307-controls
820    Description:
821    
822    This is a very large but mostly boring patch which makes (almost)
823    every tuneable compiler knob (i.e., pretty much everything under
824    Control.* plus a few other things) configurable via both the command
825    line and environment variables in the style CM did its configuration
826    until now.
827    
828    Try starting sml with '-h' (or, if you are brave, '-H')
829    
830    To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
831    implements the underlying generic mechanism.
832    
833    The interface to some of the existing such facilities has changed somewhat.
834    For example, the MLRiscControl module now provides mkFoo instead of getFoo.
835    (The getFoo interface is still there for backward-compatibility, but its
836    use is deprecated.)
837    
838    The ml-build script passes -Cxxx=yyy command-line arguments through so
839    that one can now twiddle the compiler settings when using this "batch"
840    compiler.
841    
842    TODO items:
843    
844    We should go through and throw out all controls that are no longer
845    connected to anything.  Moreover, we should go through and provide
846    meaningful (and correct!) documentation strings for those controls
847    that still are connected.
848    
849    Currently, multiple calls to Controls.new are accepted (only the first
850    has any effect).  Eventually we should make sure that every control
851    is being made (via Controls.new) exactly once.  Future access can then
852    be done using Controls.acc.
853    
854    Finally, it would probably be a good idea to use the getter-setter
855    interface to controls rather than ref cells.  For the time being, both
856    styles are provided by the Controls module, but getter-setter pairs are
857    better if thread-safety is of any concern because they can be wrapped.
858    
859    *****************************************
860    
861    One bug fix: The function blockPlacement in three of the MLRISC
862    backpatch files used to be hard-wired to one of two possibilities at
863    link time (according to the value of the placementFlag).  But (I
864    think) it should rather sense the flag every time.
865    
866    *****************************************
867    
868    Other assorted changes (by other people who did not supply a HISTORY entry):
869    
870    1. the cross-module inliner now works much better (Monnier)
871    2. representation of weights, frequencies, and probabilities in MLRISC
872       changed in preparation of using those for weighted block placement
873       (Reppy, George)
874    
875    ----------------------------------------------------------------------
876    Name: Lal George
877    Date: 2002/03/07 14:44:24 EST 2002
878    Tag: george-20020307-weighted-block-placement
879    
880    Tested the weighted block placement optimization on all architectures
881    (except the hppa) using AMPL to generate the block and edge frequencies.
882    Changes were required in the machine properties to correctly
883    categorize trap instructions. There is an MLRISC flag
884    "weighted-block-placement" that can be used to enable weighted block
885    placement, but this will be ineffective without block/edge
886    frequencies (coming soon).
887    
888    
889    ----------------------------------------------------------------------
890    Name: Lal George
891    Date: 2002/03/05 17:24:48 EST
892    Tag: george-20020305-linkage-cluster
893    
894    In order to support the block placement optimization, a new cluster
895    is generated as the very first cluster (called the linkage cluster).
896    It contains a single jump to the 'real' entry point for the compilation
897    unit. Block placement has no effect on the linkage cluster itself, but
898    all the other clusters  have full freedom in the manner in which they
899    reorder blocks or functions.
900    
901    On the x86 the typical linkage code that is generated is:
902       ----------------------
903            .align 2
904       L0:
905            addl    $L1-L0, 72(%esp)
906            jmp     L1
907    
908    
909            .align  2
910       L1:
911       ----------------------
912    
913    72(%esp) is the memory location for the stdlink register. This
914    must contain the address of the CPS function being called. In the
915    above example, it contains the address of  L0; before
916    calling L1 (the real entry point for the compilation unit), it
917    must contain the address for L1, and hence
918    
919            addl $L1-L0, 72(%esp)
920    
921    I have tested this on all architectures except the hppa.The increase
922    in code size is of course negligible
923    
924    ----------------------------------------------------------------------
925    Name: Allen Leung
926    Date: 2002/03/03 13:20:00 EST
927    Tag: leunga-20020303-mlrisc-tools
928    
929      Added #[ ... ] expressions to mlrisc tools
930    
931    ----------------------------------------------------------------------
932    Name: Matthias Blume
933    Date: 2002/02/27 12:29:00 EST
934    Tag: blume-20020227-cdebug
935    Description:
936    
937    - made types in structure C and C_Debug to be equal
938    - got rid of code duplication (c-int.sml vs. c-int-debug.sml)
939    - there no longer is a C_Int_Debug (C_Debug is directly derived from C)
940    
941    ----------------------------------------------------------------------
942    Name: Matthias Blume
943    Date: 2002/02/26 12:00:00 EST
944    Tag: blume-20020226-ffi
945    Description:
946    
947    1. Fixed a minor bug in CM's "noweb" tool:
948       If numbering is turned off, then truly don't number (i.e., do not
949       supply the -L option to noweb).  The previous behavior was to supply
950       -L'' -- which caused noweb to use the "default" line numbering scheme.
951       Thanks to Chris Richards for pointing this out (and supplying the fix).
952    
953    2. Once again, I reworked some aspects of the FFI:
954    
955       A. The incomplete/complete type business:
956    
957       - Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
958         gone!
959       - ML types representing an incomplete type are now *equal* to
960         ML types representing their corresponding complete types (just like
961         in C).  This is still safe because ml-nlffigen will not generate
962         RTTI for incomplete types, nor will it generate functions that
963         require access to such RTTI.   But when ML code generated from both
964         incomplete and complete versions of the C type meet, the ML types
965         are trivially interoperable.
966    
967         NOTE:  These changes restore the full generality of the translation
968         (which was previously lost when I eliminated functorization)!
969    
970       B. Enum types:
971    
972       - Structure C now has a type constructor "enum" that is similar to
973         how the "su" constructor works.  However, "enum" is not a phantom
974         type because each "T enum" has values (and is isomorphic to
975         MLRep.Signed.int).
976       - There are generic access operations for enum objects (using
977         MLRep.Signed.int).
978       - ml-nlffigen will generate a structure E_foo for each "enum foo".
979         * The structure contains the definition of type "mlrep" (the ML-side
980         representation type of the enum).  Normally, mlrep is the same
981         as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
982         then mlrep will be defined as a datatype -- thus facilitating
983         pattern matching on mlrep values.
984         ("-ec" will be suppressed if there are duplicate values in an
985          enumeration.)
986         * Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
987         will be generated for each C enum constant xxx.
988         * Conversion functions m2i and i2m convert between mlrep and
989         MLRep.Signed.int.  (Without "-ec", these functions are identities.)
990         * Coversion functions c and ml convert between mlrep and "tag enum".
991         * Access functions (get/set) fetch and store mlrep values.
992       - By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
993         enumerations are merged into one single enumeration represented by
994         structure E_'.
995    
996    ----------------------------------------------------------------------
997    Name: Allen Leung
998    Date: 2002/02/25 04:45:00 EST
999    Tag: leunga-20020225-cps-spill
1000    
1001    This is a new implementation of the CPS spill phase.
1002    The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
1003    In case of problems, replace it with the old file spill.sml
1004    
1005    The current compiler runs into some serious performance problems when
1006    constructing a large record.  This can happen when we try to compile a
1007    structure with many items.  Even a very simple structure like the following
1008    makes the compiler slow down.
1009    
1010        structure Foo = struct
1011           val x_1 = 0w1 : Word32.int
1012           val x_2 = 0w2 : Word32.int
1013           val x_3 = 0w3 : Word32.int
1014           ...
1015           val x_N = 0wN : Word32.int
1016        end
1017    
1018    The following table shows the compile time, from N=1000 to N=4000,
1019    with the old compiler:
1020    
1021    N
1022    1000   CPS 100 spill                           0.04u  0.00s  0.00g
1023           MLRISC ra                               0.06u  0.00s  0.05g
1024              (spills = 0 reloads = 0)
1025           TOTAL                                   0.63u  0.07s  0.21g
1026    
1027    1100   CPS 100 spill                           8.25u  0.32s  0.64g
1028           MLRISC ra                               5.68u  0.59s  3.93g
1029              (spills = 0 reloads = 0)
1030           TOTAL                                   14.71u  0.99s  4.81g
1031    
1032    1500   CPS 100 spill                           58.55u  2.34s  1.74g
1033           MLRISC ra                               5.54u  0.65s  3.91g
1034              (spills = 543 reloads = 1082)
1035           TOTAL                                   65.40u  3.13s  6.00g
1036    
1037    2000   CPS 100 spill                           126.69u  4.84s  3.08g
1038           MLRISC ra                               0.80u  0.10s  0.55g
1039              (spills = 42 reloads = 84)
1040           TOTAL                                   129.42u  5.10s  4.13g
1041    
1042    3000   CPS 100 spill                           675.59u  19.03s  11.64g
1043           MLRISC ra                               2.69u  0.27s  1.38g
1044              (spills = 62 reloads = 124)
1045           TOTAL                                   682.48u  19.61s  13.99g
1046    
1047    4000   CPS 100 spill                           2362.82u  56.28s  43.60g
1048           MLRISC ra                               4.96u  0.27s  2.72g
1049              (spills = 85 reloads = 170)
1050           TOTAL                                   2375.26u  57.21s  48.00g
1051    
1052    As you can see the old cps spill module suffers from some serious
1053    performance problem.  But since I cannot decipher the old code fully,
1054    instead of patching the problems up, I'm reimplementing it
1055    with a different algorithm.  The new code is more modular,
1056    smaller when compiled, and substantially faster
1057    (O(n log n) time and O(n) space).  Timing of the new spill module:
1058    
1059    4000  CPS 100 spill                           0.02u  0.00s  0.00g
1060          MLRISC ra                               0.25u  0.02s  0.15g
1061             (spills=1 reloads=3)
1062          TOTAL                                   7.74u  0.34s  1.62g
1063    
1064    Implementation details:
1065    
1066    As far as I can tell, the purpose of the CPS spill module is to make sure the
1067    number of live variables at any program point (the bandwidth)
1068    does not exceed a certain limit, which is determined by the
1069    size of the spill area.
1070    
1071    When the bandwidth is too large, we decrease the register pressure by
1072    packing live variables into spill records.  How we achieve this is
1073    completely different than what we did in the old code.
1074    
1075    First, there is something about the MLRiscGen code generator
1076    that we should be aware of:
1077    
1078    o MLRiscGen performs code motion!
1079    
1080       In particular, it will move floating point computations and
1081       address computations involving only the heap pointer to
1082       their use sites (if there is only a single use).
1083       What this means is that if we have a CPS record construction
1084       statement
1085    
1086           RECORD(k,vl,w,e)
1087    
1088       we should never count the new record address w as live if w
1089       has only one use (which is often the case).
1090    
1091       We should do something similar to floating point, but the transformation
1092       there is much more complex, so I won't deal with that.
1093    
1094    Secondly, there are now two new cps primops at our disposal:
1095    
1096     1. rawrecord of record_kind option
1097        This pure operator allocates some uninitialized storage from the heap.
1098        There are two forms:
1099    
1100         rawrecord NONE [INT n]  allocates a tagless record of length n
1101         rawrecord (SOME rk) [INT n] allocates a tagged record of length n
1102                                     and initializes the tag.
1103    
1104     2. rawupdate of cty
1105          rawupdate cty (v,i,x)
1106          Assigns to x to the ith component of record v.
1107          The storelist is not updated.
1108    
1109    We use these new primops for both spilling and increment record construction.
1110    
1111     1. Spilling.
1112    
1113        This is implemented with a linear scan algorithm (but generalized
1114        to trees).  The algorithm will create a single spill record at the
1115        beginning of the cps function and use rawupdate to spill to it,
1116        and SELECT or SELp to reload from it.  So both spills and reloads
1117        are fine-grain operations.  In contrast, in the old algorithm
1118        "spills" have to be bundled together in records.
1119    
1120        Ideally, we should sink the spill record construction to where
1121        it is needed.  We can even split the spill record into multiple ones
1122        at the places where they are needed.  But CPS is not a good
1123        representation for global code motion, so I'll keep it simple and
1124        am not attempting this.
1125    
1126     2. Incremental record construction (aka record splitting).
1127    
1128        Long records with many component values which are simulatenously live
1129        (recall that single use record addresses are not considered to
1130         be live) are constructed with rawrecord and rawupdate.
1131        We allocate space on the heap with rawrecord first, then gradually
1132        fill it in with rawupdate.  This is the technique suggested to me
1133        by Matthias.
1134    
1135        Some restrictions on when this is applicable:
1136        1. It is not a VECTOR record.  The code generator currently does not handle
1137           this case. VECTOR record uses double indirection like arrays.
1138        2. All the record component values are defined in the same "basic block"
1139           as the record constructor.  This is to prevent speculative
1140           record construction.
1141    
1142    ----------------------------------------------------------------------
1143    Name: Allen Leung
1144    Date: 2002/02/22 01:02:00 EST
1145    Tag: leunga-20020222-mlrisc-tools
1146    
1147    Minor bug fixes in the parser and rewriter
1148    
1149    ----------------------------------------------------------------------
1150    Name: Allen Leung
1151    Date: 2002/02/21 20:20:00 EST
1152    Tag: leunga-20020221-peephole
1153    
1154    Regenerated the peephole files.  Some contained typos in the specification
1155    and some didn't compile because of pretty printing bugs in the old version
1156    of 'nowhere'.
1157    
1158    ----------------------------------------------------------------------
1159    Name: Allen Leung
1160    Date: 2002/02/19 20:20:00 EST
1161    Tag: leunga-20020219-mlrisc-tools
1162    Description:
1163    
1164       Minor bug fixes to the mlrisc-tools library:
1165    
1166       1.  Fixed up parsing colon suffixed keywords
1167       2.  Added the ability to shut the error messages up
1168       3.  Reimplemented the pretty printer and fixed up/improved
1169           the pretty printing of handle and -> types.
1170       4.  Fixed up generation of literal symbols in the nowhere tool.
1171       5.  Added some SML keywords to to sml.sty
1172    
1173    ----------------------------------------------------------------------
1174    Name: Matthias Blume
1175  Date: 2002/02/19 16:20:00 EST  Date: 2002/02/19 16:20:00 EST
1176  Tag: blume-20020219-cmffi  Tag: blume-20020219-cmffi
1177  Description:  Description:

Legend:
Removed from v.1078  
changed lines
  Added in v.1215

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