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 1086, Fri Feb 22 05:56:29 2002 UTC revision 1261, Tue Jul 23 15:45:45 2002 UTC
# Line 13  Line 13 
13  Description:  Description:
14    
15  ----------------------------------------------------------------------  ----------------------------------------------------------------------
16    Name: Matthias Blume
17    Date: 2002/07/23 11:45:00
18    Tag: blume-20020723-smlnj-home
19    Description:
20    
21    Sml now senses the SMLNJ_HOME environment variable.  If this is set,
22    then the bin dir is assumed to be in $SMLNJ_HOME/bin and (unless
23    CM_PATHCONFIG is also set), the path configuration file is assumed
24    to be in $SMLNJ_HOME/lib/pathconfig.  This way one can easily move
25    the entire tree to some other place and everything will "just work".
26    
27    (Companion commands such as ml-build and ml-makedepend also sense this
28    variable.)
29    
30    ----------------------------------------------------------------------
31    Name: Matthias Blume
32    Date: 2002/07/12 21:19:00 EDT
33    Tag: blume-20020712-liveness
34    Description:
35    
36    Exported two useful "step" functions from liveness module (MLRISC).
37    
38    ----------------------------------------------------------------------
39    Name: Matthias Blume
40    Date: 2002/07/05 16:00 EDT
41    Tag: Release_110_41
42    Description:
43    
44    New working release.  New bootfiles.
45    
46    ----------------------------------------------------------------------
47    Name: Matthias Blume
48    Date: 2002/07/05 10:25:00 EDT
49    Tag: blume-20020705-btimp
50    Description:
51    
52    Exported structure BTImp from $smlnj/viscomp/debugprof.cm so that
53    other clients can set up backtracing support.
54    
55    ----------------------------------------------------------------------
56    Name: Matthias Blume
57    Date: 2002/06/25 17:23:00 EDT
58    Tag: blume-20020625-fpmax
59    Description:
60    
61    Fixed a bug in translation of INLMAX (and INLMIN) for the floating-point
62    case.  (The sense of the isNaN test was reversed -- which made min and
63    max always return their first argument.)
64    
65    ----------------------------------------------------------------------
66    Name: Matthias Blume
67    Date: 2002/06/11
68    Tag: blume-20020611-unixpath
69    Description:
70    
71    Back-ported OS.Path.{from,to}UnixPath from idlbasis-devel branch.
72    
73    ----------------------------------------------------------------------
74    Name: Matthias Blume
75    Date: 2002/06/10 16:35:00 EDT
76    Tag: blume-20020610-ieeereal
77    Description:
78    
79    I back-ported my implementation of IEEEReal.fromString from the
80    idlbasis-devel branch so that we can test it.
81    
82    Another small change is that ppDec tries to give more information
83    than just "<sig>" in the case of functors.  However, this code is
84    broken in some mysterious way if the functor's body's signature
85    has not been declared by ascription but gets inferred from the
86    implementation.  This needs fixing...
87    
88    ----------------------------------------------------------------------
89    Name: Matthias Blume
90    Date: 2002/05/31
91    Tag: blume-20020531-btrace-mode
92    Description:
93    
94    Resurrected SMLofNJ.Internals.BTrace.mode.  (It accidentally fell by
95    the wayside when I switched over to using Controls everywhere.)
96    
97    ----------------------------------------------------------------------
98    Name: Lal George
99    Date: 2002/05/23 12:21:40 EDT
100    Tag: george-20020523-visual-labels
101    Description:
102    
103    Labels are now displayed in the graphical output to make
104    the fall-through and target blocks obvious.
105    
106    ----------------------------------------------------------------------
107    Name: Matthias Blume
108    Date: 2002/05/22 11:03:00 EDT
109    Tag: blume-20020522-shrink
110    Description:
111    
112    John tweaked yesterday's fix for 1131 to handle an out-of-memory
113    situation that comes up when allocating huge arrays.
114    
115    ----------------------------------------------------------------------
116    Name: Matthias Blume
117    Date: 2002/05/21 16:00:00 EDT
118    Tag: Release_110_40
119    Description:
120    
121    New working release (110.40).  New bootfiles.
122    
123    [Also: John Reppy fixed GC bug 1131.]
124    
125    ----------------------------------------------------------------------
126    Name: Matthias Blume
127    Date: 2002/05/21 12:35:00 EDT
128    Tag: blume-20020521-cmdoc
129    Description:
130    
131    CM documentation update.
132    
133    ----------------------------------------------------------------------
134    Name: Matthias Blume
135    Date: 2002/05/21 10:55:00 EDT
136    Tag: blume-20020521-misc
137    Description:
138    
139    - John tweaked runtime to be silent on heap export (except when
140      GC messages are on).
141    - I added a few more things (cross-compiling versions of CMB) to
142      config/preloads (as suggestions).
143    
144    ----------------------------------------------------------------------
145    Name: Matthias Blume
146    Date: 2002/05/20 22:25:00 EDT
147    Tag: blume-20020520-controls
148    Description:
149    
150    - Added ControlUtil structure to control-lib.cm.
151    - Use it throughout.
152    - Used Controls facility to define MLRISC controls (as opposed to
153      registering MLRISC control ref cells with Controls after the fact)
154    - Fixed messed-up controls priorities.
155    
156    * Removed again all the stuff from config/preloads that one wouldn't
157      be able to preload at the time the initial heap image is built.
158      (Many libraries, e.g., CML, do not exist yet at this time.  The
159       only libraries that can be preloaded via config/preloads are those
160       that come bundled with the bootfiles.)
161    
162    ----------------------------------------------------------------------
163    Name: Matthias Blume
164    Date: 2002/05/20 10:59:00 EDT
165    Tag: blume-20020520-preloads
166    Description:
167    
168    Added a lot of commented-out suggestions for things to be included
169    in config/preloads.
170    
171    ----------------------------------------------------------------------
172    Name: Allen Leung
173    Date: 2002/05/18 14:20:00 EDT
174    Tag: leunga-20020518-mdl
175    Description:
176    
177    o Made the mdl tool stuff compile and run again.
178    o I've disabled all the stuff that depends on RTL specifications; they
179      are all badly broken anyway.
180    
181    ----------------------------------------------------------------------
182    Name: Matthias Blume
183    Date: 2002/05/17 16:49:00 EDT
184    Tag: blume-20020517-controls
185    Description:
186    
187    0. John Reppy made several modifications to the SML/NJ library.
188       In particular, there is a shiny new controls-lib.cm.
189    
190    1. Pushed new controls interface through compiler so that everything
191       compiles again.
192    
193    2. Added FormatComb and FORMAT_COMB to the CML version of the
194       SML/NJ library (so that CML compiles again).
195    
196    3. Modified init scripts because XXX_DEFAULT environment variables
197       are no longer with us.  (Boot-time initialization is now done
198       using the same environment variables that are also used for
199       startup-time initialization of controls.)
200    
201    ----------------------------------------------------------------------
202    Name: Lal George
203    Date: 2002/05/15 09:20:10 EDT
204    Tag: george-20020515-pseudo-op-decls
205    Description:
206    
207            All pseudo-ops emitted before the first segment declaration
208    such as TEXT, DATA, and BSS directives are assumed to be global
209    declarations and are emitted first in the assembly file. This is
210    useful in a number of situations where one has pseudo-ops that are not
211    specific to any segment, and also works around the constraint that one
212    cannot have client pseudo-ops in the TEXT segment.
213    
214            Because no segment is associated with these declarations it is
215    an error to allocate any space or objects before the first segment
216    directive and an exception will be raised. However, we cannot make
217    this check for client pseudo-ops.
218    
219            These top level declarations are a field in the CFG graph_info.
220    In theory you can continue to add to this field after the CFG has been
221    built -- provided you know what you are doing;-)
222    
223    ----------------------------------------------------------------------
224    Name: Matthias Blume
225    Date: 2002/05/13 16:40:00 EDT
226    Tag: blume-20020513-pp-etc
227    Description:
228    
229    A few minor bugfixes:
230    
231      - Stopgap measure for bug recently reported by Elsa Gunter (ppDec).
232        (Bogus printouts for redefined bindings still occur.  Compiler
233        bug should no longer occur now.  We need to redo the prettyprinter
234        from scratch.)
235    
236      - CM pathname printer now also adds escape sequences for ( and )
237    
238      - commend and docu fixes for ml-nlffi
239    
240    ----------------------------------------------------------------------
241    Name: Matthias Blume
242    Date: 2002/05/10 16:40:00 EDT
243    Tag: blume-20020510-erg-textio
244    Description:
245    
246    Applied the following bugfix provided by Emden Gansner:
247    
248        Output is corrupted when outputSubstr is used rather than output.
249    
250        The problem occurs when a substring
251    
252            ss = (s, dataStart, dataLen)
253    
254        where dataStart > 0, fills a stream buffer with avail bytes left.
255        avail bytes of s, starting at index dataStart, are copied into the
256        buffer, the buffer is flushed, and then the remaining dataLen-avail
257        bytes of ss are copied into the beginning of the buffer. Instead of
258        starting this copy at index dataStart+avail in s, the current code
259        starts the copy at index avail.
260    
261        Fix:
262        In text-io-fn.sml, change line 695 from
263             val needsFlush = copyVec(v, avail, dataLen-avail, buf, 0)
264        to
265             val needsFlush = copyVec(v, dataStart+avail, dataLen-avail, buf, 0)
266    
267    ----------------------------------------------------------------------
268    Name: Matthias Blume
269    Date: 2002/04/12 13:55:00 EDT
270    Tag: blume-20020412-assyntax
271    Description:
272    
273    1. Grabbed newer assyntax.h from the XFree86 project.
274    2. Fiddled with how to compile X86.prim.asm without warnings.
275    3. (Very) Minor cleanup in CM.
276    
277    ----------------------------------------------------------------------
278    Name: Matthias Blume
279    Date: 2002/04/01 (no joke!) 17:07:00 EST
280    Tag: blume-20020401-x86div
281    Description:
282    
283    Added full support for div/mod/rem/quot on the x86, using the machine
284    instruction's two results (without clumsily recomputing the remainder)
285    directly where appropriate.
286    
287    Some more extensive power-of-two support was added to the x86 instruction
288    selector (avoiding expensive divs, mods, and muls where they can be
289    replaced with cheaper shifts and masks).  However, this sort of thing
290    ought to be done earlier, e.g., within the CPS optimizer so that
291    all architectures benefit from it.
292    
293    The compiler compiles to a fixed point, but changes might be somewhat
294    fragile nevertheless.  Please, report any strange things that you might
295    see wrt. div/mod/quot/rem...
296    
297    ----------------------------------------------------------------------
298    Name: Matthias Blume
299    Date: 2002/03/29 17:22:00
300    Tag: blume-20020329-div
301    Description:
302    
303    Fixed my broken div/mod logic.  Unfortunately, this means that the
304    inline code for div/mod now has one more comparison than before.
305    Fast paths (quotient > 0 or remainder = 0) are not affected, though.
306    The problem was with quotient = 0, because that alone does not tell
307    us which way the rounding went.  One then has to look at whether
308    remainder and divisor have the same sign...  :(
309    
310    Anyway, I replaced the bootfiles with fresh ones...
311    
312    ----------------------------------------------------------------------
313    Name: Matthias Blume
314    Date: 2002/03/29 14:10:00 EST
315    Tag: blume-20020329-inlprims
316    Description:
317    
318    NEW BOOTFILES!!!    Version number bumped to 110.39.3.
319    
320    Primops have changed. This means that the bin/boot-file formats have
321    changed as well.
322    
323    To make sure that there is no confusion, I made a new version.
324    
325    
326    CHANGES:
327    
328    * removed REMT from mltree (remainder should never overflow).
329    
330    * added primops to deal with divisions of all flavors to the frontend
331    
332    * handled these primops all the way through so they map to their respective
333      MLRISC support
334    
335    * used these primops in the implementation of Int, Int32, Word, Word32
336    
337    * removed INLDIV, INLMOD, and INLREM as they are no longer necessary
338    
339    * parameterized INLMIN, INLMAX, and INLABS by a numkind
340    
341    * translate.sml now deals with all flavors of INL{MIN,MAX,ABS}, including
342      floating point
343    
344    * used INL{MIN,MAX,ABS} in the implementation of Int, Int32, Word, Word32,
345      and Real (but Real.abs maps to a separate floating-point-only primop)
346    
347    
348    TODO items:
349    
350    * Hacked Alpha32 instruction selection, disabling the selection of REMx
351      instructions because the machine instruction encoder cannot handle
352      them.  (Hppa, PPC, and Sparc instruction selection did not handle
353      REM in the first place, and REM is supported by the x86 machine coder.)
354    
355    * Handle DIV and MOD with DIV_TO_NEGINF directly in the x86 instruction
356      selection phase.  (The two can be streamlined because the hardware
357      delivers both quotient and remainder at the same time anyway.)
358    
359    * Think about what to do with "valOf(Int32.minInt) div ~1" and friends.
360      (Currently the behavior is inconsistent both across architectures and
361      wrt. the draft Basis spec.)
362    
363    * Word8 should eventually be handled natively, too.
364    
365    * There seems to be one serious bug in mltree-gen.sml.  It appears, though,
366      as if there currently is no execution path that could trigger it in
367      SML/NJ.  (The assumptions underlying functions arith and promotable do not
368      hold for things like multiplication and division.)
369    
370    ----------------------------------------------------------------------
371    Name: Matthias Blume
372    Date: 2002/03/27 16:27:00 EST
373    Tag: blume-20020327-mlrisc-divisions
374    Description:
375    
376    Added support for all four division operations (ML's div, mod, quot,
377    and rem) to MLRISC.  In the course of doing so, I also rationalized
378    the naming (no more annoying switch-around of DIV and QUOT), by
379    parameterizing the operation by div_rounding_mode (which can be either
380    DIV_TO_ZERO or DIV_TO_NEGINF).
381    
382    The generic MLTreeGen functor takes care of compiling all four
383    operations down to only round-to-zero div.
384    
385    Missing pieces:
386    
387      * Doing something smarter than relying on MLTreeGen on architectures
388        like, e.g., the x86 where hardware division delivers both quotient and
389        remainder at the same time.  With this, the implementation of the
390        round-to-neginf operations could be further streamlined.
391    
392      * Remove inlining support for div/mod/rem from the frontend and replace it
393        with primops that get carried through to the backend.  Do this for all
394        int and word types.
395    
396    ----------------------------------------------------------------------
397    Name: Matthias Blume
398    Date: 2002/03/25 17:25:00 EST
399    Tag: blume-20020325-divmod
400    Description:
401    
402    I improved (hopefully without breaking them) the implementation of Int.div,
403    Int.mod, and Int.rem.   For this, the code in translate.sml now takes
404    advantage of the following observations:
405    
406      Let  q = x quot y      r = x rem y
407           d = x div  y      m = x mod y
408    
409    where "quot" is the round-to-zero version of integer division that
410    hardware usually provides.  Then we have:
411    
412         r = x - q * y        where neither the * nor the - will overflow
413         d = if q >= 0 orelse x = q * y then q else q - 1
414                              where neither the * nor the - will overflow
415         m = if q >= 0 orelse r = 0 then r else r + y
416                              where the + will not overflow
417    
418    This results in substantial simplification of the generated code.
419    The following table shows the number of CFG nodes and edges generated
420    for
421            fun f (x, y) = x OPER y
422            (* with OPER \in div, mod, quot, rem *)
423    
424    
425        OPER | nodes(old) | edges(old) | nodes(new) | edges(new)
426        --------------------------------------------------------
427         div |         24 |         39 |         12 |         16
428         mod |         41 |         71 |         12 |         16
429        quot |          8 |         10 |          8 |         10
430         rem |         10 |         14 |          8 |         10
431    
432    
433    ----------------------------------------------------------------------
434    Name: Matthias Blume
435    Date: 2002/03/25 22:06:00 EST
436    Tag: blume-20020325-cprotobug
437    Description:
438    
439    Fixed a bug in cproto (c prototype decoder).
440    
441    ----------------------------------------------------------------------
442    Name: Matthias Blume
443    Date: 2002/03/25 16:00:00 EST
444    Tag: blume-20020325-raw-primops
445    Description:
446    
447    I did some cleanup to Allen's new primop code and
448    replaced yesterday's bootfiles with new ones.
449    (But they are stored in the same place.)
450    
451    ----------------------------------------------------------------------
452    Name: Matthias Blume
453    Date: 2002/03/24 22:40:00 EST
454    Tag: blume-20020324-bootfiles
455    Description:
456    
457    Made the bootfiles that Allen asked for.
458    
459    ----------------------------------------------------------------------
460    Name: Allen Leung
461    Date: 2002/03/23 15:50:00 EST
462    Tag: leunga-20020323-flint-cps-rcc-primops
463    Description:
464    
465      1. Changes to FLINT primops:
466    
467        (* make a call to a C-function;
468         * The primop carries C function prototype information and specifies
469         * which of its (ML-) arguments are floating point. C prototype
470         * information is for use by the backend, ML information is for
471         * use by the CPS converter. *)
472      | RAW_CCALL of { c_proto: CTypes.c_proto,
473                       ml_args: ccall_type list,
474                       ml_res_opt: ccall_type option,
475                       reentrant : bool
476                     } option
477       (* Allocate uninitialized storage on the heap.
478        * The record is meant to hold short-lived C objects, i.e., they
479        * are not ML pointers.  With the tag, the representation is
480        * the same as RECORD with tag tag_raw32 (sz=4), or tag_fblock (sz=8)
481        *)
482      | RAW_RECORD of {tag:bool,sz:int}
483      and ccall_type = CCALL_INT32 | CCALL_REAL64 | CCALL_ML_PTR
484    
485      2.  These CPS primops are now overloaded:
486    
487           rawload of {kind:numkind}
488           rawstore of {kind:numkind}
489    
490          The one argument form is:
491    
492             rawload {kind} address
493    
494          The two argument form is:
495    
496             rawload {kind} [ml object, byte-offset]
497    
498      3. RAW_CCALL/RCC now takes two extra arguments:
499    
500         a. The first is whether the C call is reentrant, i.e., whether
501            ML state should be saved and restored.
502         b. The second argument is a string argument specifying the name of
503            library and the C function.
504    
505         These things are currently not handled in the code generator, yet.
506    
507      4. In CProto,
508    
509         An encoding type of "bool" means "ml object" and is mapped into
510         C prototype of PTR.  Note that "bool" is different than "string",
511         even though "string" is also mapped into PTR, because "bool"
512         is assigned an CPS type of BOGt, while "string" is assigned INT32t.
513    
514      5. Pickler/unpicker
515    
516         Changed to handle RAW_RECORD and newest RAW_CCALL
517    
518      6. MLRiscGen,
519    
520         1. Changed to handle the new rawload/rawstore/rawrecord operators.
521         2. Code for handling C Calls has been moved to a new module CPSCCalls,
522            in the file CodeGen/cpscompile/cps-c-calls.sml
523    
524      7. Added the conditional move operator
525    
526             condmove of branch
527    
528         to cps.  Generation of this is still buggy so it is currently
529         disabled.
530    
531    ----------------------------------------------------------------------
532    Name: Lal George
533    Date: 2002/03/22 14:18:25 EST
534    Tag: george-20020322-cps-branch-prob
535    Description:
536    
537    Implemented the Ball-Larus branch prediction-heuristics, and
538    incorporated graphical viewers for control flow graphs.
539    
540    Ball-Larus Heuristics:
541    ---------------------
542    See the file compiler/CodeGen/cpscompile/cpsBranchProb.sml.
543    
544    By design it uses the Dempster-Shafer theory for combining
545    probabilities.  For example, in the function:
546    
547        fun f(n,acc) = if n = 0 then acc else f(n-1, n*acc)
548    
549    the ball-larus heuristics predicts that the n=0 is unlikely
550    (OH-heuristic), and the 'then' branch is unlikely because of the
551    RH-heuristic -- giving the 'then' branch an even lower combined
552    probability using the Dempster-Shafer theory.
553    
554    Finally, John Reppy's loop analysis in MLRISC, further lowers the
555    probability of the 'then' branch because of the loop in the else
556    branch.
557    
558    
559    Graphical Viewing:
560    ------------------
561    I merely plugged in Allen's graphical viewers into the compiler. The
562    additional code is not much. At the top level, saying:
563    
564            Control.MLRISC.getFlag "cfg-graphical-view" := true;
565    
566    will display the graphical view of the control flow graph just before
567    back-patching.  daVinci must be in your path for this to work. If
568    daVinci is not available, then the default viewer can be changed
569    using:
570    
571            Control.MLRISC.getString "viewer"
572    
573    which can be set to "dot" or "vcg" for the corresponding viewers. Of
574    course, these viewers must be in your path.
575    
576    The above will display the compilation unit at the level of clusters,
577    many of which are small, boring, and un-interesting. Also setting:
578    
579            Control.MLRISC.getInt "cfg-graphical-view_size"
580    
581    will display clusters that are larger than the value set by the above.
582    
583    
584    ----------------------------------------------------------------------
585    Name: Matthias Blume
586    Date: 2002/03/21 22:20:00 EST
587    Tag: blume-20020321-kmp-bugfix
588    Description:
589    
590    Changed the interface to the KMP routine in PreString and fixed
591    a minor bug in one place where it was used.
592    
593    ----------------------------------------------------------------------
594    Name: Allen Leung
595    Date: 2002/03/21 20:30:00 EST
596    Tag: leunga-20020321-cfg
597    Description:
598    
599      Fixed a potential problem in cfg edge splitting.
600    
601    ----------------------------------------------------------------------
602    Name: Allen Leung
603    Date: 2002/03/21 17:15:00 EST
604    Tag: leunga-20020321-x86-fp-cfg
605    Description:
606    
607      1. Recoded the buggy parts of x86-fp.
608    
609         a. All the block reordering code has been removed.
610            We now depend on the block placement phases to do this work.
611    
612         b. Critical edge splitting code has been simplified and moved into the
613            CFG modules, as where they belong.
614    
615         Both of these were quite buggy and complex.  The code is now much, much
616         simpler.
617    
618      2. X86 backend.
619    
620         a. Added instructions for 64-bit support.  Instruction selection for
621            64-bit has not been committed, however, since that
622            requires changes to MLTREE which haven't been approved by
623            Lal and John.
624    
625         b. Added support for FUCOMI and FUCOMIP when generating code for
626            PentiumPro and above.  We only generate these instructions in
627            the fast-fp mode.
628    
629         c. Added cases for JP and JNP in X86FreqProps.
630    
631      3. CFG
632    
633         CFG now has a bunch of methods for edge splitting and merging.
634    
635      4. Machine description.
636    
637         John's simplification of MLTREE_BASIS.fcond broke a few machine
638         description things:
639    
640         rtl-build.{sig,sml} and hppa.mdl fixed.
641    
642         NOTE: the machine description stuff in the repository is still broken.
643               Again, I can't put my fixes in because that involves
644               changes to MLTREE.
645    
646    ----------------------------------------------------------------------
647    Name: Matthias Blume
648    Date: 2002/03/20 15:55:00 EST
649    Tag: blume-20020320-kmp
650    Description:
651    
652    Implemented Knuth-Morris-Pratt string matching in PreString and used
653    it for String.isSubstring, Substring.isSubstring, and
654    Substring.position.
655    
656    (Might need some stress-testing.  Simple examples worked fine.)
657    
658    ----------------------------------------------------------------------
659    Name: Matthias Blume
660    Date: 2002/03/19 16:37:00 EST
661    Tag: blume-20020319-witnesses
662    Description:
663    
664    Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
665    
666    This implements a generic mechanism for changing constness qualifiers
667    anywhere within big C types without resorting to outright "casts".
668    (So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
669    modify the constness at the outermost level.)
670    The implementation of "convert" is based on the idea of "witness"
671    values -- values that are not used by the operation but whose types
672    "testify" to their applicability.  On the implementation side, "convert"
673    is simply a projection (returning its second curried argument).  With
674    cross-module inlining, it should not result in any machine code being
675    generated.
676    
677    ----------------------------------------------------------------------
678    Name: Matthias Blume
679    Date: 2002/03/15 16:40:00 EST
680    Tag: blume-20020315-basis
681    Description:
682    
683    Provided (preliminary?) implementations for
684    
685      {String,Substring}.{concatWith,isSuffix,isSubstring}
686    
687    and
688    
689      Substring.full
690    
691    Those are in the Basis spec but they were missing in SML/NJ.
692    
693    ----------------------------------------------------------------------
694    Name: Matthias Blume
695    Date: 2002/03/14 21:30:00 EST
696    Tag: blume-20020314-controls
697    Description:
698    
699    Controls:
700    ---------
701    
702    1. Factored out the recently-added Controls : CONTROLS stuff and put
703       it into its own library $/controls-lib.cm.  The source tree for
704       this is under src/smlnj-lib/Controls.
705    
706    2. Changed the names of types and functions in this interface, so they
707       make a bit more "sense":
708    
709          module -> registry
710          'a registry -> 'a group
711    
712    3. The interface now deals in ref cells only.  The getter/setter interface
713       is (mostly) gone.
714    
715    4. Added a function that lets one register an already-existing ref cell.
716    
717    5. Made the corresponding modifications to the rest of the code so that
718       everything compiles again.
719    
720    6. Changed the implementation of Controls.MLRISC back to something closer
721       to the original.  In particular, this module (and therefore MLRISC)
722       does not depend on Controls.  There now is some link-time code in
723       int-sys.sml that registers the MLRISC controls with the Controls
724       module.
725    
726    CM:
727    ---
728    
729      * One can now specify the lambda-split aggressiveness in init.cmi.
730    
731    ----------------------------------------------------------------------
732    Name: Allen Leung
733    Date: 2002/03/13 17:30:00 EST
734    Tag: leunga-20020313-x86-fp-unary
735    Description:
736    
737    Bug fix for:
738    
739    > leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
740    > Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
741    > - fun f(x,(y,z)) = Real.~ y;
742    > [autoloading]
743    > [autoloading done]
744    >       fchsl   (%eax), 184(%esp)
745    > Error: MLRisc bug: X86MCEmitter.emitInstr
746    >
747    > uncaught exception Error
748    >   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
749    
750    The problem was that the code generator did not generate any fp registers
751    in this case, and the ra didn't know that it needed to run the X86FP phase to
752    translate the pseudo fp instruction.   This only happened with unary fp
753    operators in certain situations.
754    
755    ----------------------------------------------------------------------
756    Name: Matthias Blume
757    Date: 2002/03/13 14:00:00 EST
758    Tag: blume-20020313-overload-etc
759    Description:
760    
761    1. Added _overload as a synonym for overload for backward compatibility.
762       (Control.overloadKW must be true for either version to be accepted.)
763    
764    2. Fixed bug in install script that caused more things to be installed
765       than what was requested in config/targets.
766    
767    3. Made CM aware of the (_)overload construct so that autoloading
768       works.
769    
770    ----------------------------------------------------------------------
771    Name: Matthias Blume
772    Date: 2002/03/12 22:03:00 EST
773    Tag: blume-20020312-url
774    Description:
775    
776    Forgot to update BOOT and srcarchiveurl.
777    
778    ----------------------------------------------------------------------
779    Name: Matthias Blume
780    Date: 2002/03/12 17:30:00 EST
781    Tag: blume-20020312-version110392
782    Description:
783    
784    Yet another version number bump (because of small changes to the
785    binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
786    
787    Changes:
788    
789      The new pid generation scheme described a few weeks ago was overly
790      complicated.  I implemented a new mechanism that is simpler and
791      provides a bit more "stability":  Once CM has seen a compilation
792      unit, it keeps its identity constant (as long as you do not delete
793      those crucial CM/GUID/* files).  This means that when you change
794      an interface, compile, then go back to the old interface, and
795      compile again, you arrive at the original pid.
796    
797      There now also is a mechanism that instructs CM to use the plain
798      environment hash as a module's pid (effectively making its GUID
799      the empty string).  For this, "noguid" must be specified as an
800      option to the .sml file in question within its .cm file.
801      This is most useful for code that is being generated by tools such
802      as ml-nlffigen (because during development programmers tend to
803      erase the tool's entire output directory tree including CM's cached
804      GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
805      revert to the old, broken behavior of SML/NJ, but in specific cases
806      where there is no danger of interface confusion, its use is ok
807      (I think).
808    
809      ml-nlffigen by default generates "noguid" annotations.  They can be
810      turned off by specifying -guid in its command line.
811    
812    ----------------------------------------------------------------------
813    Name: Lal George
814    Date: 2002/03/12 12 14:42:36 EST
815    Tag: george-20020312-frequency-computation
816    Description:
817    
818    Integrated jump chaining and static block frequency into the
819    compiler. More details and numbers later.
820    
821    ----------------------------------------------------------------------
822    Name: Lal George
823    Date: 2002/03/11 11 22:38:53 EST
824    Tag: george-20020311-jump-chain-elim
825    Description:
826    
827    Tested the jump chain elimination on all architectures (except the
828    hppa).  This is on by default right now and is profitable for the
829    alpha and x86, however, it may not be profitable for the sparc and ppc
830    when compiling the compiler.
831    
832    The gc test will typically jump to a label at the end of the cluster,
833    where there is another jump to an external cluster containing the actual
834    code to invoke gc. This is to allow factoring of common gc invocation
835    sequences. That is to say, we generate:
836    
837            f:
838               testgc
839               ja   L1      % jump if above to L1
840    
841            L1:
842               jmp L2
843    
844    
845    After jump chain elimination the 'ja L1' instructions is converted to
846    'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
847    up being implemented in their long form (if L2 is far away) using:
848    
849            jbe     L3      % jump if below or equal to L3
850            jmp     L2
851         L3:
852            ...
853    
854    
855    For large compilation units L2  may be far away.
856    
857    
858    ----------------------------------------------------------------------
859    Name: Matthias Blume
860    Date: 2002/03/11 13:30:00 EST
861    Tag: blume-20020311-mltreeeval
862    Description:
863    
864    A functor parameter was missing.
865    
866    ----------------------------------------------------------------------
867    Name: Allen Leung
868    Date: 2002/03/11 10:30:00 EST
869    Tag: leunga-20020311-runtime-string0
870    Description:
871    
872       The representation of the empty string now points to a
873    legal null terminated C string instead of unit.  It is now possible
874    to convert an ML string into C string with InlineT.CharVector.getData.
875    This compiles into one single machine instruction.
876    
877    ----------------------------------------------------------------------
878    Name: Allen Leung
879    Date: 2002/03/10 23:55:00 EST
880    Tag: leunga-20020310-x86-call
881    Description:
882    
883       Added machine generation for CALL instruction (relative displacement mode)
884    
885    ----------------------------------------------------------------------
886    Name: Matthias Blume
887    Date: 2002/03/08 16:05:00
888    Tag: blume-20020308-entrypoints
889    Description:
890    
891    Version number bumped to 110.39.1.  NEW BOOTFILES!
892    
893    Entrypoints: non-zero offset into a code object where execution should begin.
894    
895    - Added the notion of an entrypoint to CodeObj.
896    - Added reading/writing of entrypoint info to Binfile.
897    - Made runtime system bootloader aware of entrypoints.
898    - Use the address of the label of the first function given to mlriscGen
899      as the entrypoint.  This address is currently always 0, but it will
900      not be 0 once we turn on block placement.
901    - Removed the linkage cluster code (which was The Other Way(tm) of dealing
902      with entry points) from mlriscGen.
903    
904    ----------------------------------------------------------------------
905    Name: Allen Leung
906    Date: 2002/03/07 20:45:00 EST
907    Tag: leunga-20020307-x86-cmov
908    Description:
909    
910       Bug fixes for CMOVcc on x86.
911    
912       1. Added machine code generation for CMOVcc
913       2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
914       3. CMOVcc cannot have an immediate operand as argument.
915    
916    ----------------------------------------------------------------------
917    Name: Matthias Blume
918    Date: 2002/03/07 16:15:00 EST
919    Tag: blume-20020307-controls
920    Description:
921    
922    This is a very large but mostly boring patch which makes (almost)
923    every tuneable compiler knob (i.e., pretty much everything under
924    Control.* plus a few other things) configurable via both the command
925    line and environment variables in the style CM did its configuration
926    until now.
927    
928    Try starting sml with '-h' (or, if you are brave, '-H')
929    
930    To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
931    implements the underlying generic mechanism.
932    
933    The interface to some of the existing such facilities has changed somewhat.
934    For example, the MLRiscControl module now provides mkFoo instead of getFoo.
935    (The getFoo interface is still there for backward-compatibility, but its
936    use is deprecated.)
937    
938    The ml-build script passes -Cxxx=yyy command-line arguments through so
939    that one can now twiddle the compiler settings when using this "batch"
940    compiler.
941    
942    TODO items:
943    
944    We should go through and throw out all controls that are no longer
945    connected to anything.  Moreover, we should go through and provide
946    meaningful (and correct!) documentation strings for those controls
947    that still are connected.
948    
949    Currently, multiple calls to Controls.new are accepted (only the first
950    has any effect).  Eventually we should make sure that every control
951    is being made (via Controls.new) exactly once.  Future access can then
952    be done using Controls.acc.
953    
954    Finally, it would probably be a good idea to use the getter-setter
955    interface to controls rather than ref cells.  For the time being, both
956    styles are provided by the Controls module, but getter-setter pairs are
957    better if thread-safety is of any concern because they can be wrapped.
958    
959    *****************************************
960    
961    One bug fix: The function blockPlacement in three of the MLRISC
962    backpatch files used to be hard-wired to one of two possibilities at
963    link time (according to the value of the placementFlag).  But (I
964    think) it should rather sense the flag every time.
965    
966    *****************************************
967    
968    Other assorted changes (by other people who did not supply a HISTORY entry):
969    
970    1. the cross-module inliner now works much better (Monnier)
971    2. representation of weights, frequencies, and probabilities in MLRISC
972       changed in preparation of using those for weighted block placement
973       (Reppy, George)
974    
975    ----------------------------------------------------------------------
976    Name: Lal George
977    Date: 2002/03/07 14:44:24 EST 2002
978    Tag: george-20020307-weighted-block-placement
979    
980    Tested the weighted block placement optimization on all architectures
981    (except the hppa) using AMPL to generate the block and edge frequencies.
982    Changes were required in the machine properties to correctly
983    categorize trap instructions. There is an MLRISC flag
984    "weighted-block-placement" that can be used to enable weighted block
985    placement, but this will be ineffective without block/edge
986    frequencies (coming soon).
987    
988    
989    ----------------------------------------------------------------------
990    Name: Lal George
991    Date: 2002/03/05 17:24:48 EST
992    Tag: george-20020305-linkage-cluster
993    
994    In order to support the block placement optimization, a new cluster
995    is generated as the very first cluster (called the linkage cluster).
996    It contains a single jump to the 'real' entry point for the compilation
997    unit. Block placement has no effect on the linkage cluster itself, but
998    all the other clusters  have full freedom in the manner in which they
999    reorder blocks or functions.
1000    
1001    On the x86 the typical linkage code that is generated is:
1002       ----------------------
1003            .align 2
1004       L0:
1005            addl    $L1-L0, 72(%esp)
1006            jmp     L1
1007    
1008    
1009            .align  2
1010       L1:
1011       ----------------------
1012    
1013    72(%esp) is the memory location for the stdlink register. This
1014    must contain the address of the CPS function being called. In the
1015    above example, it contains the address of  L0; before
1016    calling L1 (the real entry point for the compilation unit), it
1017    must contain the address for L1, and hence
1018    
1019            addl $L1-L0, 72(%esp)
1020    
1021    I have tested this on all architectures except the hppa.The increase
1022    in code size is of course negligible
1023    
1024    ----------------------------------------------------------------------
1025    Name: Allen Leung
1026    Date: 2002/03/03 13:20:00 EST
1027    Tag: leunga-20020303-mlrisc-tools
1028    
1029      Added #[ ... ] expressions to mlrisc tools
1030    
1031    ----------------------------------------------------------------------
1032    Name: Matthias Blume
1033    Date: 2002/02/27 12:29:00 EST
1034    Tag: blume-20020227-cdebug
1035    Description:
1036    
1037    - made types in structure C and C_Debug to be equal
1038    - got rid of code duplication (c-int.sml vs. c-int-debug.sml)
1039    - there no longer is a C_Int_Debug (C_Debug is directly derived from C)
1040    
1041    ----------------------------------------------------------------------
1042    Name: Matthias Blume
1043    Date: 2002/02/26 12:00:00 EST
1044    Tag: blume-20020226-ffi
1045    Description:
1046    
1047    1. Fixed a minor bug in CM's "noweb" tool:
1048       If numbering is turned off, then truly don't number (i.e., do not
1049       supply the -L option to noweb).  The previous behavior was to supply
1050       -L'' -- which caused noweb to use the "default" line numbering scheme.
1051       Thanks to Chris Richards for pointing this out (and supplying the fix).
1052    
1053    2. Once again, I reworked some aspects of the FFI:
1054    
1055       A. The incomplete/complete type business:
1056    
1057       - Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
1058         gone!
1059       - ML types representing an incomplete type are now *equal* to
1060         ML types representing their corresponding complete types (just like
1061         in C).  This is still safe because ml-nlffigen will not generate
1062         RTTI for incomplete types, nor will it generate functions that
1063         require access to such RTTI.   But when ML code generated from both
1064         incomplete and complete versions of the C type meet, the ML types
1065         are trivially interoperable.
1066    
1067         NOTE:  These changes restore the full generality of the translation
1068         (which was previously lost when I eliminated functorization)!
1069    
1070       B. Enum types:
1071    
1072       - Structure C now has a type constructor "enum" that is similar to
1073         how the "su" constructor works.  However, "enum" is not a phantom
1074         type because each "T enum" has values (and is isomorphic to
1075         MLRep.Signed.int).
1076       - There are generic access operations for enum objects (using
1077         MLRep.Signed.int).
1078       - ml-nlffigen will generate a structure E_foo for each "enum foo".
1079         * The structure contains the definition of type "mlrep" (the ML-side
1080         representation type of the enum).  Normally, mlrep is the same
1081         as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
1082         then mlrep will be defined as a datatype -- thus facilitating
1083         pattern matching on mlrep values.
1084         ("-ec" will be suppressed if there are duplicate values in an
1085          enumeration.)
1086         * Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
1087         will be generated for each C enum constant xxx.
1088         * Conversion functions m2i and i2m convert between mlrep and
1089         MLRep.Signed.int.  (Without "-ec", these functions are identities.)
1090         * Coversion functions c and ml convert between mlrep and "tag enum".
1091         * Access functions (get/set) fetch and store mlrep values.
1092       - By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
1093         enumerations are merged into one single enumeration represented by
1094         structure E_'.
1095    
1096    ----------------------------------------------------------------------
1097    Name: Allen Leung
1098    Date: 2002/02/25 04:45:00 EST
1099    Tag: leunga-20020225-cps-spill
1100    
1101    This is a new implementation of the CPS spill phase.
1102    The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
1103    In case of problems, replace it with the old file spill.sml
1104    
1105    The current compiler runs into some serious performance problems when
1106    constructing a large record.  This can happen when we try to compile a
1107    structure with many items.  Even a very simple structure like the following
1108    makes the compiler slow down.
1109    
1110        structure Foo = struct
1111           val x_1 = 0w1 : Word32.int
1112           val x_2 = 0w2 : Word32.int
1113           val x_3 = 0w3 : Word32.int
1114           ...
1115           val x_N = 0wN : Word32.int
1116        end
1117    
1118    The following table shows the compile time, from N=1000 to N=4000,
1119    with the old compiler:
1120    
1121    N
1122    1000   CPS 100 spill                           0.04u  0.00s  0.00g
1123           MLRISC ra                               0.06u  0.00s  0.05g
1124              (spills = 0 reloads = 0)
1125           TOTAL                                   0.63u  0.07s  0.21g
1126    
1127    1100   CPS 100 spill                           8.25u  0.32s  0.64g
1128           MLRISC ra                               5.68u  0.59s  3.93g
1129              (spills = 0 reloads = 0)
1130           TOTAL                                   14.71u  0.99s  4.81g
1131    
1132    1500   CPS 100 spill                           58.55u  2.34s  1.74g
1133           MLRISC ra                               5.54u  0.65s  3.91g
1134              (spills = 543 reloads = 1082)
1135           TOTAL                                   65.40u  3.13s  6.00g
1136    
1137    2000   CPS 100 spill                           126.69u  4.84s  3.08g
1138           MLRISC ra                               0.80u  0.10s  0.55g
1139              (spills = 42 reloads = 84)
1140           TOTAL                                   129.42u  5.10s  4.13g
1141    
1142    3000   CPS 100 spill                           675.59u  19.03s  11.64g
1143           MLRISC ra                               2.69u  0.27s  1.38g
1144              (spills = 62 reloads = 124)
1145           TOTAL                                   682.48u  19.61s  13.99g
1146    
1147    4000   CPS 100 spill                           2362.82u  56.28s  43.60g
1148           MLRISC ra                               4.96u  0.27s  2.72g
1149              (spills = 85 reloads = 170)
1150           TOTAL                                   2375.26u  57.21s  48.00g
1151    
1152    As you can see the old cps spill module suffers from some serious
1153    performance problem.  But since I cannot decipher the old code fully,
1154    instead of patching the problems up, I'm reimplementing it
1155    with a different algorithm.  The new code is more modular,
1156    smaller when compiled, and substantially faster
1157    (O(n log n) time and O(n) space).  Timing of the new spill module:
1158    
1159    4000  CPS 100 spill                           0.02u  0.00s  0.00g
1160          MLRISC ra                               0.25u  0.02s  0.15g
1161             (spills=1 reloads=3)
1162          TOTAL                                   7.74u  0.34s  1.62g
1163    
1164    Implementation details:
1165    
1166    As far as I can tell, the purpose of the CPS spill module is to make sure the
1167    number of live variables at any program point (the bandwidth)
1168    does not exceed a certain limit, which is determined by the
1169    size of the spill area.
1170    
1171    When the bandwidth is too large, we decrease the register pressure by
1172    packing live variables into spill records.  How we achieve this is
1173    completely different than what we did in the old code.
1174    
1175    First, there is something about the MLRiscGen code generator
1176    that we should be aware of:
1177    
1178    o MLRiscGen performs code motion!
1179    
1180       In particular, it will move floating point computations and
1181       address computations involving only the heap pointer to
1182       their use sites (if there is only a single use).
1183       What this means is that if we have a CPS record construction
1184       statement
1185    
1186           RECORD(k,vl,w,e)
1187    
1188       we should never count the new record address w as live if w
1189       has only one use (which is often the case).
1190    
1191       We should do something similar to floating point, but the transformation
1192       there is much more complex, so I won't deal with that.
1193    
1194    Secondly, there are now two new cps primops at our disposal:
1195    
1196     1. rawrecord of record_kind option
1197        This pure operator allocates some uninitialized storage from the heap.
1198        There are two forms:
1199    
1200         rawrecord NONE [INT n]  allocates a tagless record of length n
1201         rawrecord (SOME rk) [INT n] allocates a tagged record of length n
1202                                     and initializes the tag.
1203    
1204     2. rawupdate of cty
1205          rawupdate cty (v,i,x)
1206          Assigns to x to the ith component of record v.
1207          The storelist is not updated.
1208    
1209    We use these new primops for both spilling and increment record construction.
1210    
1211     1. Spilling.
1212    
1213        This is implemented with a linear scan algorithm (but generalized
1214        to trees).  The algorithm will create a single spill record at the
1215        beginning of the cps function and use rawupdate to spill to it,
1216        and SELECT or SELp to reload from it.  So both spills and reloads
1217        are fine-grain operations.  In contrast, in the old algorithm
1218        "spills" have to be bundled together in records.
1219    
1220        Ideally, we should sink the spill record construction to where
1221        it is needed.  We can even split the spill record into multiple ones
1222        at the places where they are needed.  But CPS is not a good
1223        representation for global code motion, so I'll keep it simple and
1224        am not attempting this.
1225    
1226     2. Incremental record construction (aka record splitting).
1227    
1228        Long records with many component values which are simulatenously live
1229        (recall that single use record addresses are not considered to
1230         be live) are constructed with rawrecord and rawupdate.
1231        We allocate space on the heap with rawrecord first, then gradually
1232        fill it in with rawupdate.  This is the technique suggested to me
1233        by Matthias.
1234    
1235        Some restrictions on when this is applicable:
1236        1. It is not a VECTOR record.  The code generator currently does not handle
1237           this case. VECTOR record uses double indirection like arrays.
1238        2. All the record component values are defined in the same "basic block"
1239           as the record constructor.  This is to prevent speculative
1240           record construction.
1241    
1242    ----------------------------------------------------------------------
1243  Name: Allen Leung  Name: Allen Leung
1244  Date: 2002/02/22 01:02:00 EST  Date: 2002/02/22 01:02:00 EST
1245  Tag: leunga-20020222-mlrisc-tools  Tag: leunga-20020222-mlrisc-tools

Legend:
Removed from v.1086  
changed lines
  Added in v.1261

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