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 1094, Mon Feb 25 09:58:56 2002 UTC revision 1204, Mon May 20 14:58:33 2002 UTC
# Line 13  Line 13 
13  Description:  Description:
14    
15  ----------------------------------------------------------------------  ----------------------------------------------------------------------
16    Name: Matthias Blume
17    Date: 2002/05/20 10:59:00 EDT
18    Tag: blume-20020520-preloads
19    Description:
20    
21    Added a lot of commented-out suggestions for things to be included
22    in config/preloads.
23    
24    ----------------------------------------------------------------------
25    Name: Allen Leung
26    Date: 2002/05/18 14:20:00 EDT
27    Tag: leunga-20020518-mdl
28    Description:
29    
30    o Made the mdl tool stuff compile and run again.
31    o I've disabled all the stuff that depends on RTL specifications; they
32      are all badly broken anyway.
33    
34    ----------------------------------------------------------------------
35    Name: Matthias Blume
36    Date: 2002/05/17 16:49:00 EDT
37    Tag: blume-20020517-controls
38    Description:
39    
40    0. John Reppy made several modifications to the SML/NJ library.
41       In particular, there is a shiny new controls-lib.cm.
42    
43    1. Pushed new controls interface through compiler so that everything
44       compiles again.
45    
46    2. Added FormatComb and FORMAT_COMB to the CML version of the
47       SML/NJ library (so that CML compiles again).
48    
49    3. Modified init scripts because XXX_DEFAULT environment variables
50       are no longer with us.  (Boot-time initialization is now done
51       using the same environment variables that are also used for
52       startup-time initialization of controls.)
53    
54    ----------------------------------------------------------------------
55    Name: Lal George
56    Date: 2002/05/15 09:20:10 EDT
57    Tag: george-20020515-pseudo-op-decls
58    Description:
59    
60            All pseudo-ops emitted before the first segment declaration
61    such as TEXT, DATA, and BSS directives are assumed to be global
62    declarations and are emitted first in the assembly file. This is
63    useful in a number of situations where one has pseudo-ops that are not
64    specific to any segment, and also works around the constraint that one
65    cannot have client pseudo-ops in the TEXT segment.
66    
67            Because no segment is associated with these declarations it is
68    an error to allocate any space or objects before the first segment
69    directive and an exception will be raised. However, we cannot make
70    this check for client pseudo-ops.
71    
72            These top level declarations are a field in the CFG graph_info.
73    In theory you can continue to add to this field after the CFG has been
74    built -- provided you know what you are doing;-)
75    
76    ----------------------------------------------------------------------
77    Name: Matthias Blume
78    Date: 2002/05/13 16:40:00 EDT
79    Tag: blume-20020513-pp-etc
80    Description:
81    
82    A few minor bugfixes:
83    
84      - Stopgap measure for bug recently reported by Elsa Gunter (ppDec).
85        (Bogus printouts for redefined bindings still occur.  Compiler
86        bug should no longer occur now.  We need to redo the prettyprinter
87        from scratch.)
88    
89      - CM pathname printer now also adds escape sequences for ( and )
90    
91      - commend and docu fixes for ml-nlffi
92    
93    ----------------------------------------------------------------------
94    Name: Matthias Blume
95    Date: 2002/05/10 16:40:00 EDT
96    Tag: blume-20020510-erg-textio
97    Description:
98    
99    Applied the following bugfix provided by Emden Gansner:
100    
101        Output is corrupted when outputSubstr is used rather than output.
102    
103        The problem occurs when a substring
104    
105            ss = (s, dataStart, dataLen)
106    
107        where dataStart > 0, fills a stream buffer with avail bytes left.
108        avail bytes of s, starting at index dataStart, are copied into the
109        buffer, the buffer is flushed, and then the remaining dataLen-avail
110        bytes of ss are copied into the beginning of the buffer. Instead of
111        starting this copy at index dataStart+avail in s, the current code
112        starts the copy at index avail.
113    
114        Fix:
115        In text-io-fn.sml, change line 695 from
116             val needsFlush = copyVec(v, avail, dataLen-avail, buf, 0)
117        to
118             val needsFlush = copyVec(v, dataStart+avail, dataLen-avail, buf, 0)
119    
120    ----------------------------------------------------------------------
121    Name: Matthias Blume
122    Date: 2002/04/12 13:55:00 EDT
123    Tag: blume-20020412-assyntax
124    Description:
125    
126    1. Grabbed newer assyntax.h from the XFree86 project.
127    2. Fiddled with how to compile X86.prim.asm without warnings.
128    3. (Very) Minor cleanup in CM.
129    
130    ----------------------------------------------------------------------
131    Name: Matthias Blume
132    Date: 2002/04/01 (no joke!) 17:07:00 EST
133    Tag: blume-20020401-x86div
134    Description:
135    
136    Added full support for div/mod/rem/quot on the x86, using the machine
137    instruction's two results (without clumsily recomputing the remainder)
138    directly where appropriate.
139    
140    Some more extensive power-of-two support was added to the x86 instruction
141    selector (avoiding expensive divs, mods, and muls where they can be
142    replaced with cheaper shifts and masks).  However, this sort of thing
143    ought to be done earlier, e.g., within the CPS optimizer so that
144    all architectures benefit from it.
145    
146    The compiler compiles to a fixed point, but changes might be somewhat
147    fragile nevertheless.  Please, report any strange things that you might
148    see wrt. div/mod/quot/rem...
149    
150    ----------------------------------------------------------------------
151    Name: Matthias Blume
152    Date: 2002/03/29 17:22:00
153    Tag: blume-20020329-div
154    Description:
155    
156    Fixed my broken div/mod logic.  Unfortunately, this means that the
157    inline code for div/mod now has one more comparison than before.
158    Fast paths (quotient > 0 or remainder = 0) are not affected, though.
159    The problem was with quotient = 0, because that alone does not tell
160    us which way the rounding went.  One then has to look at whether
161    remainder and divisor have the same sign...  :(
162    
163    Anyway, I replaced the bootfiles with fresh ones...
164    
165    ----------------------------------------------------------------------
166    Name: Matthias Blume
167    Date: 2002/03/29 14:10:00 EST
168    Tag: blume-20020329-inlprims
169    Description:
170    
171    NEW BOOTFILES!!!    Version number bumped to 110.39.3.
172    
173    Primops have changed. This means that the bin/boot-file formats have
174    changed as well.
175    
176    To make sure that there is no confusion, I made a new version.
177    
178    
179    CHANGES:
180    
181    * removed REMT from mltree (remainder should never overflow).
182    
183    * added primops to deal with divisions of all flavors to the frontend
184    
185    * handled these primops all the way through so they map to their respective
186      MLRISC support
187    
188    * used these primops in the implementation of Int, Int32, Word, Word32
189    
190    * removed INLDIV, INLMOD, and INLREM as they are no longer necessary
191    
192    * parameterized INLMIN, INLMAX, and INLABS by a numkind
193    
194    * translate.sml now deals with all flavors of INL{MIN,MAX,ABS}, including
195      floating point
196    
197    * used INL{MIN,MAX,ABS} in the implementation of Int, Int32, Word, Word32,
198      and Real (but Real.abs maps to a separate floating-point-only primop)
199    
200    
201    TODO items:
202    
203    * Hacked Alpha32 instruction selection, disabling the selection of REMx
204      instructions because the machine instruction encoder cannot handle
205      them.  (Hppa, PPC, and Sparc instruction selection did not handle
206      REM in the first place, and REM is supported by the x86 machine coder.)
207    
208    * Handle DIV and MOD with DIV_TO_NEGINF directly in the x86 instruction
209      selection phase.  (The two can be streamlined because the hardware
210      delivers both quotient and remainder at the same time anyway.)
211    
212    * Think about what to do with "valOf(Int32.minInt) div ~1" and friends.
213      (Currently the behavior is inconsistent both across architectures and
214      wrt. the draft Basis spec.)
215    
216    * Word8 should eventually be handled natively, too.
217    
218    * There seems to be one serious bug in mltree-gen.sml.  It appears, though,
219      as if there currently is no execution path that could trigger it in
220      SML/NJ.  (The assumptions underlying functions arith and promotable do not
221      hold for things like multiplication and division.)
222    
223    ----------------------------------------------------------------------
224    Name: Matthias Blume
225    Date: 2002/03/27 16:27:00 EST
226    Tag: blume-20020327-mlrisc-divisions
227    Description:
228    
229    Added support for all four division operations (ML's div, mod, quot,
230    and rem) to MLRISC.  In the course of doing so, I also rationalized
231    the naming (no more annoying switch-around of DIV and QUOT), by
232    parameterizing the operation by div_rounding_mode (which can be either
233    DIV_TO_ZERO or DIV_TO_NEGINF).
234    
235    The generic MLTreeGen functor takes care of compiling all four
236    operations down to only round-to-zero div.
237    
238    Missing pieces:
239    
240      * Doing something smarter than relying on MLTreeGen on architectures
241        like, e.g., the x86 where hardware division delivers both quotient and
242        remainder at the same time.  With this, the implementation of the
243        round-to-neginf operations could be further streamlined.
244    
245      * Remove inlining support for div/mod/rem from the frontend and replace it
246        with primops that get carried through to the backend.  Do this for all
247        int and word types.
248    
249    ----------------------------------------------------------------------
250    Name: Matthias Blume
251    Date: 2002/03/25 17:25:00 EST
252    Tag: blume-20020325-divmod
253    Description:
254    
255    I improved (hopefully without breaking them) the implementation of Int.div,
256    Int.mod, and Int.rem.   For this, the code in translate.sml now takes
257    advantage of the following observations:
258    
259      Let  q = x quot y      r = x rem y
260           d = x div  y      m = x mod y
261    
262    where "quot" is the round-to-zero version of integer division that
263    hardware usually provides.  Then we have:
264    
265         r = x - q * y        where neither the * nor the - will overflow
266         d = if q >= 0 orelse x = q * y then q else q - 1
267                              where neither the * nor the - will overflow
268         m = if q >= 0 orelse r = 0 then r else r + y
269                              where the + will not overflow
270    
271    This results in substantial simplification of the generated code.
272    The following table shows the number of CFG nodes and edges generated
273    for
274            fun f (x, y) = x OPER y
275            (* with OPER \in div, mod, quot, rem *)
276    
277    
278        OPER | nodes(old) | edges(old) | nodes(new) | edges(new)
279        --------------------------------------------------------
280         div |         24 |         39 |         12 |         16
281         mod |         41 |         71 |         12 |         16
282        quot |          8 |         10 |          8 |         10
283         rem |         10 |         14 |          8 |         10
284    
285    
286    ----------------------------------------------------------------------
287    Name: Matthias Blume
288    Date: 2002/03/25 22:06:00 EST
289    Tag: blume-20020325-cprotobug
290    Description:
291    
292    Fixed a bug in cproto (c prototype decoder).
293    
294    ----------------------------------------------------------------------
295    Name: Matthias Blume
296    Date: 2002/03/25 16:00:00 EST
297    Tag: blume-20020325-raw-primops
298    Description:
299    
300    I did some cleanup to Allen's new primop code and
301    replaced yesterday's bootfiles with new ones.
302    (But they are stored in the same place.)
303    
304    ----------------------------------------------------------------------
305    Name: Matthias Blume
306    Date: 2002/03/24 22:40:00 EST
307    Tag: blume-20020324-bootfiles
308    Description:
309    
310    Made the bootfiles that Allen asked for.
311    
312    ----------------------------------------------------------------------
313    Name: Allen Leung
314    Date: 2002/03/23 15:50:00 EST
315    Tag: leunga-20020323-flint-cps-rcc-primops
316    Description:
317    
318      1. Changes to FLINT primops:
319    
320        (* make a call to a C-function;
321         * The primop carries C function prototype information and specifies
322         * which of its (ML-) arguments are floating point. C prototype
323         * information is for use by the backend, ML information is for
324         * use by the CPS converter. *)
325      | RAW_CCALL of { c_proto: CTypes.c_proto,
326                       ml_args: ccall_type list,
327                       ml_res_opt: ccall_type option,
328                       reentrant : bool
329                     } option
330       (* Allocate uninitialized storage on the heap.
331        * The record is meant to hold short-lived C objects, i.e., they
332        * are not ML pointers.  With the tag, the representation is
333        * the same as RECORD with tag tag_raw32 (sz=4), or tag_fblock (sz=8)
334        *)
335      | RAW_RECORD of {tag:bool,sz:int}
336      and ccall_type = CCALL_INT32 | CCALL_REAL64 | CCALL_ML_PTR
337    
338      2.  These CPS primops are now overloaded:
339    
340           rawload of {kind:numkind}
341           rawstore of {kind:numkind}
342    
343          The one argument form is:
344    
345             rawload {kind} address
346    
347          The two argument form is:
348    
349             rawload {kind} [ml object, byte-offset]
350    
351      3. RAW_CCALL/RCC now takes two extra arguments:
352    
353         a. The first is whether the C call is reentrant, i.e., whether
354            ML state should be saved and restored.
355         b. The second argument is a string argument specifying the name of
356            library and the C function.
357    
358         These things are currently not handled in the code generator, yet.
359    
360      4. In CProto,
361    
362         An encoding type of "bool" means "ml object" and is mapped into
363         C prototype of PTR.  Note that "bool" is different than "string",
364         even though "string" is also mapped into PTR, because "bool"
365         is assigned an CPS type of BOGt, while "string" is assigned INT32t.
366    
367      5. Pickler/unpicker
368    
369         Changed to handle RAW_RECORD and newest RAW_CCALL
370    
371      6. MLRiscGen,
372    
373         1. Changed to handle the new rawload/rawstore/rawrecord operators.
374         2. Code for handling C Calls has been moved to a new module CPSCCalls,
375            in the file CodeGen/cpscompile/cps-c-calls.sml
376    
377      7. Added the conditional move operator
378    
379             condmove of branch
380    
381         to cps.  Generation of this is still buggy so it is currently
382         disabled.
383    
384    ----------------------------------------------------------------------
385    Name: Lal George
386    Date: 2002/03/22 14:18:25 EST
387    Tag: george-20020322-cps-branch-prob
388    Description:
389    
390    Implemented the Ball-Larus branch prediction-heuristics, and
391    incorporated graphical viewers for control flow graphs.
392    
393    Ball-Larus Heuristics:
394    ---------------------
395    See the file compiler/CodeGen/cpscompile/cpsBranchProb.sml.
396    
397    By design it uses the Dempster-Shafer theory for combining
398    probabilities.  For example, in the function:
399    
400        fun f(n,acc) = if n = 0 then acc else f(n-1, n*acc)
401    
402    the ball-larus heuristics predicts that the n=0 is unlikely
403    (OH-heuristic), and the 'then' branch is unlikely because of the
404    RH-heuristic -- giving the 'then' branch an even lower combined
405    probability using the Dempster-Shafer theory.
406    
407    Finally, John Reppy's loop analysis in MLRISC, further lowers the
408    probability of the 'then' branch because of the loop in the else
409    branch.
410    
411    
412    Graphical Viewing:
413    ------------------
414    I merely plugged in Allen's graphical viewers into the compiler. The
415    additional code is not much. At the top level, saying:
416    
417            Control.MLRISC.getFlag "cfg-graphical-view" := true;
418    
419    will display the graphical view of the control flow graph just before
420    back-patching.  daVinci must be in your path for this to work. If
421    daVinci is not available, then the default viewer can be changed
422    using:
423    
424            Control.MLRISC.getString "viewer"
425    
426    which can be set to "dot" or "vcg" for the corresponding viewers. Of
427    course, these viewers must be in your path.
428    
429    The above will display the compilation unit at the level of clusters,
430    many of which are small, boring, and un-interesting. Also setting:
431    
432            Control.MLRISC.getInt "cfg-graphical-view_size"
433    
434    will display clusters that are larger than the value set by the above.
435    
436    
437    ----------------------------------------------------------------------
438    Name: Matthias Blume
439    Date: 2002/03/21 22:20:00 EST
440    Tag: blume-20020321-kmp-bugfix
441    Description:
442    
443    Changed the interface to the KMP routine in PreString and fixed
444    a minor bug in one place where it was used.
445    
446    ----------------------------------------------------------------------
447    Name: Allen Leung
448    Date: 2002/03/21 20:30:00 EST
449    Tag: leunga-20020321-cfg
450    Description:
451    
452      Fixed a potential problem in cfg edge splitting.
453    
454    ----------------------------------------------------------------------
455    Name: Allen Leung
456    Date: 2002/03/21 17:15:00 EST
457    Tag: leunga-20020321-x86-fp-cfg
458    Description:
459    
460      1. Recoded the buggy parts of x86-fp.
461    
462         a. All the block reordering code has been removed.
463            We now depend on the block placement phases to do this work.
464    
465         b. Critical edge splitting code has been simplified and moved into the
466            CFG modules, as where they belong.
467    
468         Both of these were quite buggy and complex.  The code is now much, much
469         simpler.
470    
471      2. X86 backend.
472    
473         a. Added instructions for 64-bit support.  Instruction selection for
474            64-bit has not been committed, however, since that
475            requires changes to MLTREE which haven't been approved by
476            Lal and John.
477    
478         b. Added support for FUCOMI and FUCOMIP when generating code for
479            PentiumPro and above.  We only generate these instructions in
480            the fast-fp mode.
481    
482         c. Added cases for JP and JNP in X86FreqProps.
483    
484      3. CFG
485    
486         CFG now has a bunch of methods for edge splitting and merging.
487    
488      4. Machine description.
489    
490         John's simplification of MLTREE_BASIS.fcond broke a few machine
491         description things:
492    
493         rtl-build.{sig,sml} and hppa.mdl fixed.
494    
495         NOTE: the machine description stuff in the repository is still broken.
496               Again, I can't put my fixes in because that involves
497               changes to MLTREE.
498    
499    ----------------------------------------------------------------------
500    Name: Matthias Blume
501    Date: 2002/03/20 15:55:00 EST
502    Tag: blume-20020320-kmp
503    Description:
504    
505    Implemented Knuth-Morris-Pratt string matching in PreString and used
506    it for String.isSubstring, Substring.isSubstring, and
507    Substring.position.
508    
509    (Might need some stress-testing.  Simple examples worked fine.)
510    
511    ----------------------------------------------------------------------
512    Name: Matthias Blume
513    Date: 2002/03/19 16:37:00 EST
514    Tag: blume-20020319-witnesses
515    Description:
516    
517    Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
518    
519    This implements a generic mechanism for changing constness qualifiers
520    anywhere within big C types without resorting to outright "casts".
521    (So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
522    modify the constness at the outermost level.)
523    The implementation of "convert" is based on the idea of "witness"
524    values -- values that are not used by the operation but whose types
525    "testify" to their applicability.  On the implementation side, "convert"
526    is simply a projection (returning its second curried argument).  With
527    cross-module inlining, it should not result in any machine code being
528    generated.
529    
530    ----------------------------------------------------------------------
531    Name: Matthias Blume
532    Date: 2002/03/15 16:40:00 EST
533    Tag: blume-20020315-basis
534    Description:
535    
536    Provided (preliminary?) implementations for
537    
538      {String,Substring}.{concatWith,isSuffix,isSubstring}
539    
540    and
541    
542      Substring.full
543    
544    Those are in the Basis spec but they were missing in SML/NJ.
545    
546    ----------------------------------------------------------------------
547    Name: Matthias Blume
548    Date: 2002/03/14 21:30:00 EST
549    Tag: blume-20020314-controls
550    Description:
551    
552    Controls:
553    ---------
554    
555    1. Factored out the recently-added Controls : CONTROLS stuff and put
556       it into its own library $/controls-lib.cm.  The source tree for
557       this is under src/smlnj-lib/Controls.
558    
559    2. Changed the names of types and functions in this interface, so they
560       make a bit more "sense":
561    
562          module -> registry
563          'a registry -> 'a group
564    
565    3. The interface now deals in ref cells only.  The getter/setter interface
566       is (mostly) gone.
567    
568    4. Added a function that lets one register an already-existing ref cell.
569    
570    5. Made the corresponding modifications to the rest of the code so that
571       everything compiles again.
572    
573    6. Changed the implementation of Controls.MLRISC back to something closer
574       to the original.  In particular, this module (and therefore MLRISC)
575       does not depend on Controls.  There now is some link-time code in
576       int-sys.sml that registers the MLRISC controls with the Controls
577       module.
578    
579    CM:
580    ---
581    
582      * One can now specify the lambda-split aggressiveness in init.cmi.
583    
584    ----------------------------------------------------------------------
585    Name: Allen Leung
586    Date: 2002/03/13 17:30:00 EST
587    Tag: leunga-20020313-x86-fp-unary
588    Description:
589    
590    Bug fix for:
591    
592    > leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
593    > Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
594    > - fun f(x,(y,z)) = Real.~ y;
595    > [autoloading]
596    > [autoloading done]
597    >       fchsl   (%eax), 184(%esp)
598    > Error: MLRisc bug: X86MCEmitter.emitInstr
599    >
600    > uncaught exception Error
601    >   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
602    
603    The problem was that the code generator did not generate any fp registers
604    in this case, and the ra didn't know that it needed to run the X86FP phase to
605    translate the pseudo fp instruction.   This only happened with unary fp
606    operators in certain situations.
607    
608    ----------------------------------------------------------------------
609    Name: Matthias Blume
610    Date: 2002/03/13 14:00:00 EST
611    Tag: blume-20020313-overload-etc
612    Description:
613    
614    1. Added _overload as a synonym for overload for backward compatibility.
615       (Control.overloadKW must be true for either version to be accepted.)
616    
617    2. Fixed bug in install script that caused more things to be installed
618       than what was requested in config/targets.
619    
620    3. Made CM aware of the (_)overload construct so that autoloading
621       works.
622    
623    ----------------------------------------------------------------------
624    Name: Matthias Blume
625    Date: 2002/03/12 22:03:00 EST
626    Tag: blume-20020312-url
627    Description:
628    
629    Forgot to update BOOT and srcarchiveurl.
630    
631    ----------------------------------------------------------------------
632    Name: Matthias Blume
633    Date: 2002/03/12 17:30:00 EST
634    Tag: blume-20020312-version110392
635    Description:
636    
637    Yet another version number bump (because of small changes to the
638    binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
639    
640    Changes:
641    
642      The new pid generation scheme described a few weeks ago was overly
643      complicated.  I implemented a new mechanism that is simpler and
644      provides a bit more "stability":  Once CM has seen a compilation
645      unit, it keeps its identity constant (as long as you do not delete
646      those crucial CM/GUID/* files).  This means that when you change
647      an interface, compile, then go back to the old interface, and
648      compile again, you arrive at the original pid.
649    
650      There now also is a mechanism that instructs CM to use the plain
651      environment hash as a module's pid (effectively making its GUID
652      the empty string).  For this, "noguid" must be specified as an
653      option to the .sml file in question within its .cm file.
654      This is most useful for code that is being generated by tools such
655      as ml-nlffigen (because during development programmers tend to
656      erase the tool's entire output directory tree including CM's cached
657      GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
658      revert to the old, broken behavior of SML/NJ, but in specific cases
659      where there is no danger of interface confusion, its use is ok
660      (I think).
661    
662      ml-nlffigen by default generates "noguid" annotations.  They can be
663      turned off by specifying -guid in its command line.
664    
665    ----------------------------------------------------------------------
666    Name: Lal George
667    Date: 2002/03/12 12 14:42:36 EST
668    Tag: george-20020312-frequency-computation
669    Description:
670    
671    Integrated jump chaining and static block frequency into the
672    compiler. More details and numbers later.
673    
674    ----------------------------------------------------------------------
675    Name: Lal George
676    Date: 2002/03/11 11 22:38:53 EST
677    Tag: george-20020311-jump-chain-elim
678    Description:
679    
680    Tested the jump chain elimination on all architectures (except the
681    hppa).  This is on by default right now and is profitable for the
682    alpha and x86, however, it may not be profitable for the sparc and ppc
683    when compiling the compiler.
684    
685    The gc test will typically jump to a label at the end of the cluster,
686    where there is another jump to an external cluster containing the actual
687    code to invoke gc. This is to allow factoring of common gc invocation
688    sequences. That is to say, we generate:
689    
690            f:
691               testgc
692               ja   L1      % jump if above to L1
693    
694            L1:
695               jmp L2
696    
697    
698    After jump chain elimination the 'ja L1' instructions is converted to
699    'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
700    up being implemented in their long form (if L2 is far away) using:
701    
702            jbe     L3      % jump if below or equal to L3
703            jmp     L2
704         L3:
705            ...
706    
707    
708    For large compilation units L2  may be far away.
709    
710    
711    ----------------------------------------------------------------------
712    Name: Matthias Blume
713    Date: 2002/03/11 13:30:00 EST
714    Tag: blume-20020311-mltreeeval
715    Description:
716    
717    A functor parameter was missing.
718    
719    ----------------------------------------------------------------------
720    Name: Allen Leung
721    Date: 2002/03/11 10:30:00 EST
722    Tag: leunga-20020311-runtime-string0
723    Description:
724    
725       The representation of the empty string now points to a
726    legal null terminated C string instead of unit.  It is now possible
727    to convert an ML string into C string with InlineT.CharVector.getData.
728    This compiles into one single machine instruction.
729    
730    ----------------------------------------------------------------------
731    Name: Allen Leung
732    Date: 2002/03/10 23:55:00 EST
733    Tag: leunga-20020310-x86-call
734    Description:
735    
736       Added machine generation for CALL instruction (relative displacement mode)
737    
738    ----------------------------------------------------------------------
739    Name: Matthias Blume
740    Date: 2002/03/08 16:05:00
741    Tag: blume-20020308-entrypoints
742    Description:
743    
744    Version number bumped to 110.39.1.  NEW BOOTFILES!
745    
746    Entrypoints: non-zero offset into a code object where execution should begin.
747    
748    - Added the notion of an entrypoint to CodeObj.
749    - Added reading/writing of entrypoint info to Binfile.
750    - Made runtime system bootloader aware of entrypoints.
751    - Use the address of the label of the first function given to mlriscGen
752      as the entrypoint.  This address is currently always 0, but it will
753      not be 0 once we turn on block placement.
754    - Removed the linkage cluster code (which was The Other Way(tm) of dealing
755      with entry points) from mlriscGen.
756    
757    ----------------------------------------------------------------------
758    Name: Allen Leung
759    Date: 2002/03/07 20:45:00 EST
760    Tag: leunga-20020307-x86-cmov
761    Description:
762    
763       Bug fixes for CMOVcc on x86.
764    
765       1. Added machine code generation for CMOVcc
766       2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
767       3. CMOVcc cannot have an immediate operand as argument.
768    
769    ----------------------------------------------------------------------
770    Name: Matthias Blume
771    Date: 2002/03/07 16:15:00 EST
772    Tag: blume-20020307-controls
773    Description:
774    
775    This is a very large but mostly boring patch which makes (almost)
776    every tuneable compiler knob (i.e., pretty much everything under
777    Control.* plus a few other things) configurable via both the command
778    line and environment variables in the style CM did its configuration
779    until now.
780    
781    Try starting sml with '-h' (or, if you are brave, '-H')
782    
783    To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
784    implements the underlying generic mechanism.
785    
786    The interface to some of the existing such facilities has changed somewhat.
787    For example, the MLRiscControl module now provides mkFoo instead of getFoo.
788    (The getFoo interface is still there for backward-compatibility, but its
789    use is deprecated.)
790    
791    The ml-build script passes -Cxxx=yyy command-line arguments through so
792    that one can now twiddle the compiler settings when using this "batch"
793    compiler.
794    
795    TODO items:
796    
797    We should go through and throw out all controls that are no longer
798    connected to anything.  Moreover, we should go through and provide
799    meaningful (and correct!) documentation strings for those controls
800    that still are connected.
801    
802    Currently, multiple calls to Controls.new are accepted (only the first
803    has any effect).  Eventually we should make sure that every control
804    is being made (via Controls.new) exactly once.  Future access can then
805    be done using Controls.acc.
806    
807    Finally, it would probably be a good idea to use the getter-setter
808    interface to controls rather than ref cells.  For the time being, both
809    styles are provided by the Controls module, but getter-setter pairs are
810    better if thread-safety is of any concern because they can be wrapped.
811    
812    *****************************************
813    
814    One bug fix: The function blockPlacement in three of the MLRISC
815    backpatch files used to be hard-wired to one of two possibilities at
816    link time (according to the value of the placementFlag).  But (I
817    think) it should rather sense the flag every time.
818    
819    *****************************************
820    
821    Other assorted changes (by other people who did not supply a HISTORY entry):
822    
823    1. the cross-module inliner now works much better (Monnier)
824    2. representation of weights, frequencies, and probabilities in MLRISC
825       changed in preparation of using those for weighted block placement
826       (Reppy, George)
827    
828    ----------------------------------------------------------------------
829    Name: Lal George
830    Date: 2002/03/07 14:44:24 EST 2002
831    Tag: george-20020307-weighted-block-placement
832    
833    Tested the weighted block placement optimization on all architectures
834    (except the hppa) using AMPL to generate the block and edge frequencies.
835    Changes were required in the machine properties to correctly
836    categorize trap instructions. There is an MLRISC flag
837    "weighted-block-placement" that can be used to enable weighted block
838    placement, but this will be ineffective without block/edge
839    frequencies (coming soon).
840    
841    
842    ----------------------------------------------------------------------
843    Name: Lal George
844    Date: 2002/03/05 17:24:48 EST
845    Tag: george-20020305-linkage-cluster
846    
847    In order to support the block placement optimization, a new cluster
848    is generated as the very first cluster (called the linkage cluster).
849    It contains a single jump to the 'real' entry point for the compilation
850    unit. Block placement has no effect on the linkage cluster itself, but
851    all the other clusters  have full freedom in the manner in which they
852    reorder blocks or functions.
853    
854    On the x86 the typical linkage code that is generated is:
855       ----------------------
856            .align 2
857       L0:
858            addl    $L1-L0, 72(%esp)
859            jmp     L1
860    
861    
862            .align  2
863       L1:
864       ----------------------
865    
866    72(%esp) is the memory location for the stdlink register. This
867    must contain the address of the CPS function being called. In the
868    above example, it contains the address of  L0; before
869    calling L1 (the real entry point for the compilation unit), it
870    must contain the address for L1, and hence
871    
872            addl $L1-L0, 72(%esp)
873    
874    I have tested this on all architectures except the hppa.The increase
875    in code size is of course negligible
876    
877    ----------------------------------------------------------------------
878    Name: Allen Leung
879    Date: 2002/03/03 13:20:00 EST
880    Tag: leunga-20020303-mlrisc-tools
881    
882      Added #[ ... ] expressions to mlrisc tools
883    
884    ----------------------------------------------------------------------
885    Name: Matthias Blume
886    Date: 2002/02/27 12:29:00 EST
887    Tag: blume-20020227-cdebug
888    Description:
889    
890    - made types in structure C and C_Debug to be equal
891    - got rid of code duplication (c-int.sml vs. c-int-debug.sml)
892    - there no longer is a C_Int_Debug (C_Debug is directly derived from C)
893    
894    ----------------------------------------------------------------------
895    Name: Matthias Blume
896    Date: 2002/02/26 12:00:00 EST
897    Tag: blume-20020226-ffi
898    Description:
899    
900    1. Fixed a minor bug in CM's "noweb" tool:
901       If numbering is turned off, then truly don't number (i.e., do not
902       supply the -L option to noweb).  The previous behavior was to supply
903       -L'' -- which caused noweb to use the "default" line numbering scheme.
904       Thanks to Chris Richards for pointing this out (and supplying the fix).
905    
906    2. Once again, I reworked some aspects of the FFI:
907    
908       A. The incomplete/complete type business:
909    
910       - Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
911         gone!
912       - ML types representing an incomplete type are now *equal* to
913         ML types representing their corresponding complete types (just like
914         in C).  This is still safe because ml-nlffigen will not generate
915         RTTI for incomplete types, nor will it generate functions that
916         require access to such RTTI.   But when ML code generated from both
917         incomplete and complete versions of the C type meet, the ML types
918         are trivially interoperable.
919    
920         NOTE:  These changes restore the full generality of the translation
921         (which was previously lost when I eliminated functorization)!
922    
923       B. Enum types:
924    
925       - Structure C now has a type constructor "enum" that is similar to
926         how the "su" constructor works.  However, "enum" is not a phantom
927         type because each "T enum" has values (and is isomorphic to
928         MLRep.Signed.int).
929       - There are generic access operations for enum objects (using
930         MLRep.Signed.int).
931       - ml-nlffigen will generate a structure E_foo for each "enum foo".
932         * The structure contains the definition of type "mlrep" (the ML-side
933         representation type of the enum).  Normally, mlrep is the same
934         as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
935         then mlrep will be defined as a datatype -- thus facilitating
936         pattern matching on mlrep values.
937         ("-ec" will be suppressed if there are duplicate values in an
938          enumeration.)
939         * Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
940         will be generated for each C enum constant xxx.
941         * Conversion functions m2i and i2m convert between mlrep and
942         MLRep.Signed.int.  (Without "-ec", these functions are identities.)
943         * Coversion functions c and ml convert between mlrep and "tag enum".
944         * Access functions (get/set) fetch and store mlrep values.
945       - By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
946         enumerations are merged into one single enumeration represented by
947         structure E_'.
948    
949    ----------------------------------------------------------------------
950  Name: Allen Leung  Name: Allen Leung
951  Date: 2002/02/25 04:45:00 EST  Date: 2002/02/25 04:45:00 EST
952  Tag: leunga-20020225-cps-spill  Tag: leunga-20020225-cps-spill
# Line 69  Line 1003 
1003         TOTAL                                   2375.26u  57.21s  48.00g         TOTAL                                   2375.26u  57.21s  48.00g
1004    
1005  As you can see the old cps spill module suffers from some serious  As you can see the old cps spill module suffers from some serious
1006  performance problem but since I cannot decipher the old code fully,  performance problem.  But since I cannot decipher the old code fully,
1007  innstead of patching the problems up, I'm reimplementing it  instead of patching the problems up, I'm reimplementing it
1008  with a different algorithm.  The new code is more modular,  with a different algorithm.  The new code is more modular,
1009  smaller when compiled, and substantially faster  smaller when compiled, and substantially faster
1010  (O(n log n) time and O(n) space).  Timing of the new spill module:  (O(n log n) time and O(n) space).  Timing of the new spill module:

Legend:
Removed from v.1094  
changed lines
  Added in v.1204

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