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 958, Fri Oct 12 03:54:43 2001 UTC revision 1094, Mon Feb 25 09:58:56 2002 UTC
# Line 8  Line 8 
8  The form of an entry should be:  The form of an entry should be:
9    
10  Name:  Name:
11  Date:  Date: yyyy/mm/dd
12  Tag: <post-commit CVS tag>  Tag: <post-commit CVS tag>
13  Description:  Description:
14    
15  ----------------------------------------------------------------------  ----------------------------------------------------------------------
16    Name: Allen Leung
17    Date: 2002/02/25 04:45:00 EST
18    Tag: leunga-20020225-cps-spill
19    
20    This is a new implementation of the CPS spill phase.
21    The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
22    In case of problems, replace it with the old file spill.sml
23    
24    The current compiler runs into some serious performance problems when
25    constructing a large record.  This can happen when we try to compile a
26    structure with many items.  Even a very simple structure like the following
27    makes the compiler slow down.
28    
29        structure Foo = struct
30           val x_1 = 0w1 : Word32.int
31           val x_2 = 0w2 : Word32.int
32           val x_3 = 0w3 : Word32.int
33           ...
34           val x_N = 0wN : Word32.int
35        end
36    
37    The following table shows the compile time, from N=1000 to N=4000,
38    with the old compiler:
39    
40    N
41    1000   CPS 100 spill                           0.04u  0.00s  0.00g
42           MLRISC ra                               0.06u  0.00s  0.05g
43              (spills = 0 reloads = 0)
44           TOTAL                                   0.63u  0.07s  0.21g
45    
46    1100   CPS 100 spill                           8.25u  0.32s  0.64g
47           MLRISC ra                               5.68u  0.59s  3.93g
48              (spills = 0 reloads = 0)
49           TOTAL                                   14.71u  0.99s  4.81g
50    
51    1500   CPS 100 spill                           58.55u  2.34s  1.74g
52           MLRISC ra                               5.54u  0.65s  3.91g
53              (spills = 543 reloads = 1082)
54           TOTAL                                   65.40u  3.13s  6.00g
55    
56    2000   CPS 100 spill                           126.69u  4.84s  3.08g
57           MLRISC ra                               0.80u  0.10s  0.55g
58              (spills = 42 reloads = 84)
59           TOTAL                                   129.42u  5.10s  4.13g
60    
61    3000   CPS 100 spill                           675.59u  19.03s  11.64g
62           MLRISC ra                               2.69u  0.27s  1.38g
63              (spills = 62 reloads = 124)
64           TOTAL                                   682.48u  19.61s  13.99g
65    
66    4000   CPS 100 spill                           2362.82u  56.28s  43.60g
67           MLRISC ra                               4.96u  0.27s  2.72g
68              (spills = 85 reloads = 170)
69           TOTAL                                   2375.26u  57.21s  48.00g
70    
71    As you can see the old cps spill module suffers from some serious
72    performance problem but since I cannot decipher the old code fully,
73    innstead of patching the problems up, I'm reimplementing it
74    with a different algorithm.  The new code is more modular,
75    smaller when compiled, and substantially faster
76    (O(n log n) time and O(n) space).  Timing of the new spill module:
77    
78    4000  CPS 100 spill                           0.02u  0.00s  0.00g
79          MLRISC ra                               0.25u  0.02s  0.15g
80             (spills=1 reloads=3)
81          TOTAL                                   7.74u  0.34s  1.62g
82    
83    Implementation details:
84    
85    As far as I can tell, the purpose of the CPS spill module is to make sure the
86    number of live variables at any program point (the bandwidth)
87    does not exceed a certain limit, which is determined by the
88    size of the spill area.
89    
90    When the bandwidth is too large, we decrease the register pressure by
91    packing live variables into spill records.  How we achieve this is
92    completely different than what we did in the old code.
93    
94    First, there is something about the MLRiscGen code generator
95    that we should be aware of:
96    
97    o MLRiscGen performs code motion!
98    
99       In particular, it will move floating point computations and
100       address computations involving only the heap pointer to
101       their use sites (if there is only a single use).
102       What this means is that if we have a CPS record construction
103       statement
104    
105           RECORD(k,vl,w,e)
106    
107       we should never count the new record address w as live if w
108       has only one use (which is often the case).
109    
110       We should do something similar to floating point, but the transformation
111       there is much more complex, so I won't deal with that.
112    
113    Secondly, there are now two new cps primops at our disposal:
114    
115     1. rawrecord of record_kind option
116        This pure operator allocates some uninitialized storage from the heap.
117        There are two forms:
118    
119         rawrecord NONE [INT n]  allocates a tagless record of length n
120         rawrecord (SOME rk) [INT n] allocates a tagged record of length n
121                                     and initializes the tag.
122    
123     2. rawupdate of cty
124          rawupdate cty (v,i,x)
125          Assigns to x to the ith component of record v.
126          The storelist is not updated.
127    
128    We use these new primops for both spilling and increment record construction.
129    
130     1. Spilling.
131    
132        This is implemented with a linear scan algorithm (but generalized
133        to trees).  The algorithm will create a single spill record at the
134        beginning of the cps function and use rawupdate to spill to it,
135        and SELECT or SELp to reload from it.  So both spills and reloads
136        are fine-grain operations.  In contrast, in the old algorithm
137        "spills" have to be bundled together in records.
138    
139        Ideally, we should sink the spill record construction to where
140        it is needed.  We can even split the spill record into multiple ones
141        at the places where they are needed.  But CPS is not a good
142        representation for global code motion, so I'll keep it simple and
143        am not attempting this.
144    
145     2. Incremental record construction (aka record splitting).
146    
147        Long records with many component values which are simulatenously live
148        (recall that single use record addresses are not considered to
149         be live) are constructed with rawrecord and rawupdate.
150        We allocate space on the heap with rawrecord first, then gradually
151        fill it in with rawupdate.  This is the technique suggested to me
152        by Matthias.
153    
154        Some restrictions on when this is applicable:
155        1. It is not a VECTOR record.  The code generator currently does not handle
156           this case. VECTOR record uses double indirection like arrays.
157        2. All the record component values are defined in the same "basic block"
158           as the record constructor.  This is to prevent speculative
159           record construction.
160    
161    ----------------------------------------------------------------------
162    Name: Allen Leung
163    Date: 2002/02/22 01:02:00 EST
164    Tag: leunga-20020222-mlrisc-tools
165    
166    Minor bug fixes in the parser and rewriter
167    
168    ----------------------------------------------------------------------
169    Name: Allen Leung
170    Date: 2002/02/21 20:20:00 EST
171    Tag: leunga-20020221-peephole
172    
173    Regenerated the peephole files.  Some contained typos in the specification
174    and some didn't compile because of pretty printing bugs in the old version
175    of 'nowhere'.
176    
177    ----------------------------------------------------------------------
178    Name: Allen Leung
179    Date: 2002/02/19 20:20:00 EST
180    Tag: leunga-20020219-mlrisc-tools
181    Description:
182    
183       Minor bug fixes to the mlrisc-tools library:
184    
185       1.  Fixed up parsing colon suffixed keywords
186       2.  Added the ability to shut the error messages up
187       3.  Reimplemented the pretty printer and fixed up/improved
188           the pretty printing of handle and -> types.
189       4.  Fixed up generation of literal symbols in the nowhere tool.
190       5.  Added some SML keywords to to sml.sty
191    
192    ----------------------------------------------------------------------
193    Name: Matthias Blume
194    Date: 2002/02/19 16:20:00 EST
195    Tag: blume-20020219-cmffi
196    Description:
197    
198    A wild mix of changes, some minor, some major:
199    
200    * All C FFI-related libraries are now anchored under $c:
201        $/c.cm      --> $c/c.cm
202        $/c-int.cm  --> $c/internals/c-int.cm
203        $/memory.cm --> $c/memory/memory.cm
204    
205    * "make" tool (in CM) now treats its argument pathname slightly
206      differently:
207        1. If the native expansion is an absolute name, then before invoking
208           the "make" command on it, CM will apply OS.Path.mkRelative
209           (with relativeTo = OS.FileSys.getDir()) to it.
210        2. The argument will be passed through to subsequent phases of CM
211           processing without "going native".  In particular, if the argument
212           was an anchored path, then "make" will not lose track of that anchor.
213    
214    * Compiler backends now "know" their respective C calling conventions
215      instead of having to be told about it by ml-nlffigen.  This relieves
216      ml-nlffigen from one of its burdens.
217    
218    * The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
219    
220    * Export C_DEBUG and C_Debug from $c/c.cm.
221    
222    * C type encoding in ml-nlffi-lib has been improved to model the conceptual
223      subtyping relationship between incomplete pointers and their complete
224      counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
225      with the convention of instantiating 'o with ('t, 'c) obj whenever
226      the pointer target type is complete.  In the incomplete case, 'o
227      will be instantiated with some "'c iobj" -- a type obtained by
228      using one of the functors PointerToIncompleteType or PointerToCompleteType.
229    
230      Operations that work on both incomplete and complete pointer types are
231      typed as taking an 'o ptr while operations that require the target to
232      be known are typed as taking some ('t, 'c) obj ptr.
233    
234      voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
235      where void is an eqtype without any values.  This makes it possible
236      to work on voidptr values using functions meant to operate on light
237      incomplete pointers.
238    
239    * As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
240      been vastly simplified.
241    
242    ----------------------------------------------------------------------
243    Name: Matthias Blume
244    Date: 2002/02/19 10:48:00 EST
245    Tag: blume-20020219-pqfix
246    Description:
247    
248    Applied Chris Okasaki's bug fix for priority queues.
249    
250    ----------------------------------------------------------------------
251    Name: Matthias Blume
252    Date: 2002/02/15 17:05:00
253    Tag: Release_110_39
254    Description:
255    
256    Last-minute retagging is becoming a tradition... :-(
257    
258    This is the working release 110.39.
259    
260    ----------------------------------------------------------------------
261    Name: Matthias Blume
262    Date: 2002/02/15 16:00:00 EST
263    Tag: Release_110_39-orig
264    Description:
265    
266    Working release 110.39.  New bootfiles.
267    
268    (Update: There was a small bug in the installer so it wouldn't work
269    with all shells.  So I retagged. -Matthias)
270    
271    ----------------------------------------------------------------------
272    Name: Matthias Blume
273    Date: 2002/02/15 14:17:00 EST
274    Tag: blume-20020215-showbindings
275    Description:
276    
277    Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
278    the latter can be useful for exploring what bindings are available at
279    the interactive prompt.  (The first function returns only the list
280    of symbols that are really bound, the second prints those but also the
281    ones that CM's autoloading mechanism knows about.)
282    
283    ----------------------------------------------------------------------
284    Name: Matthias Blume
285    Date: 2002/02/15 12:08:00 EST
286    Tag: blume-20020215-iptrs
287    Description:
288    
289    Two improvements to ml-nlffigen:
290    
291      1. Write files only if they do not exist or if their current contents
292         do not coincide with what's being written.  (That is, avoid messing
293         with the time stamps unless absolutely necessary.)
294    
295      2. Implement a "repository" mechanism for generated files related
296         to "incomplete pointer types".   See the README file for details.
297    
298    ----------------------------------------------------------------------
299    Name: Matthias Blume
300    Date: 2002/02/14 11:50:00 EST
301    Tag: blume-20020214-quote
302    Description:
303    
304    Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
305    because of the new and improved tag generation scheme.  (Thanks to Allen
306    Leung for pointing it out.)
307    
308    ----------------------------------------------------------------------
309    Name: Lal George
310    Date: 2002/02/14 09:55:27 EST 2002
311    Tag: george-20020214-isabelle-bug
312    Description:
313    
314    Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
315    of Isabelle on the x86.
316    
317    From Allen:
318    -----------
319     I've found the problem:
320    
321         in ra-core.sml, I use the counter "blocked" to keep track of the
322         true number of elements in the freeze queue.  When the counter goes
323         to zero, I skip examining the queue.  But I've messed up the
324         bookkeeping in combine():
325    
326             else ();
327             case !ucol of
328               PSEUDO => (if !cntv > 0 then
329                     (if !cntu > 0 then blocked := !blocked - 1 else ();
330                                        ^^^^^^^^^^^^^^^^^^^^^^^
331                      moveu := mergeMoveList(!movev, !moveu)
332                     )
333                  else ();
334    
335         combine() is called to coalesce two nodes u and v.
336         I think I was thinking that if the move counts of u and v are both
337         greater than zero then after they are coalesced then one node is
338         removed from the freeze queue.  Apparently I was thinking that
339         both u and v are of low degree, but that's clearly not necessarily true.
340    
341    
342    02/12/2002:
343        Here's the patch.  HOL now compiles.
344    
345        I don't know how this impact on performance (compile
346        time or runtime).  This bug caused the RA (especially on the x86)
347        to go thru the potential spill phase when there are still nodes on the
348        freeze queue.
349    
350    
351    
352    
353    ----------------------------------------------------------------------
354    Name: Matthias Blume
355    Date: 2002/02/13 22:40:00 EST
356    Tag: blume-20020213-fptr-rtti
357    Description:
358    
359    Fixed a bug in ml-nlffigen that was introduced with one of the previous
360    updates.
361    
362    ----------------------------------------------------------------------
363    Name: Matthias Blume
364    Date: 2002/02/13 16:41:00 EST
365    Tag: blume-20020213-cmlpq
366    Description:
367    
368    Added new priority queue export symbols (which have just been added to
369    smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
370    not compile and the installer would choke.)
371    
372    ----------------------------------------------------------------------
373    Name: Matthias Blume
374    Date: 2002/02/13 16:15:00 EST
375    Tag: blume-20020213-various
376    Description:
377    
378    1. More tweaks to ml-nlffigen:
379    
380       - better internal datastructures (resulting in slight speedup)
381       - "-match" option requires exact match
382       - "localized" gensym counters (untagged structs/unions nested within
383         other structs/unions or within typedefs get a fresh counter; their
384         tag will be prefixed by a concatenation of their parents' tags)
385       - bug fixes (related to calculation of transitive closure of types
386         to be included in the output)
387    
388    2. Minor Basis updates:
389    
390       - added implementations for List.collate and Option.app
391    
392    ----------------------------------------------------------------------
393    Name: Matthias Blume
394    Date: 2002/02/11 15:55:00 EST
395    Tag: blume-20020211-gensym
396    Description:
397    
398    Added a "-gensym" option to command line of ml-nlffigen.  This can be
399    used to specify a "stem" -- a string that is inserted in all "gensym'd"
400    names (ML structure names that correspond to unnamed C structs, unions,
401    and enums), so that separate runs of ml-nlffigen do not clash.
402    
403    ----------------------------------------------------------------------
404    Name: Matthias Blume
405    Date: 2002/02/11 12:05:00 EST
406    Tag: blume-20020211-gensml
407    Description:
408    
409    A quick fix for a problem with GenSML (in the pgraph-util library):
410    Make generation of toplevel "local" optional.  (Strictly speaking,
411    signature definitions within "local" are not legal SML.)
412    
413    Other than that: updates to INSTALL and cm/TODO.
414    
415    ----------------------------------------------------------------------
416    Name: Matthias Blume
417    Date: 2002/02/08 15:00:00 EST
418    Tag: blume-20020208-uniquepid
419    Description:
420    
421    0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
422    
423    1. The installer (config/install.sh) has gotten smarter:
424    
425         - Configuration options are a bit easier to specify now
426           (in config/targets).
427         - Bug in recognizing .tar.bz2 files fixed.
428         - Installer automatically resolves dependencies between
429           configuration options (e.g., if you ask for eXene, you will
430           also get cml -- regardless whether you asked for it or not).
431         - Installer can run in "quieter mode" by setting the environment
432           variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
433           "completely silent", though.
434         - Build HashCons library as part of smlnj-lib.
435    
436    2. A new scheme for assigning persistent identifiers to compilation
437       units (and, by extension, to types etc.) has been put into place.
438       This fixes a long-standing bug where types and even dynamic values
439       can get internally confused, thereby compromising type safety
440       (abstraction) and dynamic correctness.  See
441    
442         http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
443    
444       for an example of how things could go wrong until now.
445    
446       The downside of the new scheme is that pids are not quite as
447       persistent as they used to be: CM will generate a fresh pid
448       for every compilation unit that it thinks it sees for the first
449       time.  That means that if you compile starting from a clean, fresh
450       source tree at two different times, you end up with different
451       binaries.
452    
453       Cutoff recompilation, however, has not been compromised because
454       CM keeps pid information in special caches between runs.
455    
456    ----------------------------------------------------------------------
457    Name: Lal George
458    Date: 2002/02/07 15:34:13 EST 2002
459    Tag: <none>
460    Description:
461    
462    Compilers that generate assembly code may produce  global labels
463    whose value is resolved at link time. The various peephole optimization
464    modules did not take this in account.
465    
466    TODO. The Labels.addrOf function should really return an option
467    type so that clients are forced to deal with this issue, rather
468    than an exception being raised.
469    
470    ----------------------------------------------------------------------
471    Name: Lal George
472    Date: 2002/02/06 13:55:02 EST
473    Tag: george-20020206-ra-breakup
474    Description:
475    
476    1. A bug fix from Allen.
477    
478        A typo causes extra fstp %st(0)'s to be generated at compensation
479        edges, which might cause stack underflow traps at runtime.  This
480        occurs in fft where there are extraneous fstps right before the 'into'
481        trap instruction (in this case they are harmless since none of the
482        integers overflow.)
483    
484    2. Pulled out various utility modules that were embedded in the modules
485       of the register allocator. I need these modules for other purposes, but
486       they are not complete enough to put into a library (just yet).
487    ----------------------------------------------------------------------
488    Name: Matthias Blume
489    Date: 2002/01/31 16:05:00 EST
490    Tag: blume-20020131-sparc-ccalls
491    Description:
492    
493    1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
494       of extra stack space by mistake.  Fixed.
495    
496    2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
497    
498    ----------------------------------------------------------------------
499    Name: Allen Leung
500    Date: 2002/01/30
501    Tag: leunga-20020130-nowhere-bug-fix
502    Description:
503    
504       MLRISC bug fixes:
505       1. Fixed a bindings computation bug in the 'nowhere' program generator tool.
506       2. MachineInt.fromString was negating its value.
507    
508    ----------------------------------------------------------------------
509    Name: Matthias Blume
510    Date: 2002/01/29
511    Tag: blume-20020129-INSTALL
512    Description:
513    
514    - Added somewhat detailed installation instructions (file INSTALL).
515    - Fixed curl-detection bug in config/install.sh.
516    - It is now possible to select the URL getter using the URLGETTER
517      environment variable:
518    
519          not set / "unknown"      --> automatic detection (script tries wget,
520                                       curl, and lynx)
521          "wget" / "curl" / "lynx" --> use the specified program (script "knows"
522                                       how to properly invoke them)
523          other                    --> use $URLGETTER directly, it must take
524                                       precisely two command-line arguments
525                                       (source URL and destination file name)
526    
527    ----------------------------------------------------------------------
528    Name: Matthias Blume
529    Date: 2002/01/28
530    Tag: blume-20020128-sparc-ccalls
531    Description:
532    
533    - Fixed problem with calculation of "used" registers in sparc-c-calls.
534    - Make use of the allocParam argument in sparc-c-calls.
535    
536    ----------------------------------------------------------------------
537    Name: Matthias Blume
538    Date: 2002/01/28
539    Tag: blume-20020128-allocParam
540    Description:
541    
542    John Reppy:  Changes c-calls API to accept client-callback for
543    allocating extra stack space.
544    me: Corresponding changes to mlriscGen (using a dummy argument that
545        does not change the current behavior).
546    
547    ----------------------------------------------------------------------
548    Name: Matthias Blume
549    Date: 2002/01/28 12:00:00
550    Tag: Release_110_38
551    Description:
552    
553    This time for real!!!
554    
555    ----------------------------------------------------------------------
556    Name: Matthias Blume
557    Date: 2002/01/28 10:56:00 EST
558    Tag: blume-20020128-retraction
559    Description:
560    
561    0. Retracted earlier 110.38.  (The Release_110_38 tag has been replaced
562       with blume-Release_110_38-retracted.)
563    
564    1. Fixed a problem with incorrect rounding modes in real64.sml.
565       (Thanks to Andrew Mccreight <andrew.mccreight@yale.edu>.)
566    
567    2. A bug in ml-nlffigen related to the handling of unnamed structs, unions,
568       and enums fixed.  The naming of corresponding ML identifiers should
569       now be consistent again.
570    
571    ----------------------------------------------------------------------
572    Name: Allen Leung
573    Date: 2002/01/27
574    Tag: leunga-20020127-nowhere
575    Description:
576    
577       Added a target called nowhere in the configuration scripts.
578       Enabling this will build the MLRISC 'nowhere' tool (for translating
579       programs with where-clauses into legal SML code) during installation.
580    
581    ----------------------------------------------------------------------
582    Name: Matthias Blume
583    Date: 2002/01/25 21:27:00 EST
584    Tag: blume-Release_110_38-retracted
585    Description:
586    
587    Call it a (working) release!  Version is 110.38. Bootfiles are ready.
588    
589    README will be added later.
590    
591    !!! NOTE:  Re-tagged as blume-Release_110_38-retracted. Original tag
592    (Release_110_38) removed.  Reason: Last-minute bug fixes.
593    
594    ----------------------------------------------------------------------
595    Name: Matthias Blume
596    Date: 2002/01/25
597    Tag: blume-20020125-ffi
598    Description:
599    
600    A large number of tweaks and improvements to ml-nlffi-lib and
601    ml-nlffigen:
602    
603       - ML represenation types have been streamlined
604       - getter and setter functions work with concrete values, not abstract
605         ones where possible
606       - ml-nlffigen command line more flexible (see README file there)
607       - some bugs have been fixed (hopefully)
608    
609    ----------------------------------------------------------------------
610    Name: Lal George
611    Date: 2002/01/24
612    Tag: george-20020124-risc-ra-interface
613    Description:
614    
615       There is a dramatic simplification in the interface to the
616       register allocator for RISC architectures as a result of making
617       parallel copy instructions explicit.
618    
619    ----------------------------------------------------------------------
620    Name: Matthias Blume
621    Date: 2002/01/22
622    Tag: blume-20020122-x86-ccalls
623    Description:
624    
625    Bug fix for c-calls on x86 (having to do with how char- and
626    short-arguments are being handled).
627    
628    ----------------------------------------------------------------------
629    Name: Matthias Blume
630    Date: 2002/01/21
631    Tag: blume-20020121-ff
632    Description:
633    
634    Another day of fiddling with the FFI...
635    
636    1. Bug fix/workaround:  CKIT does not complain about negative array
637       dimensions, so ml-nlffigen has to guard itself against this possibility.
638       (Otherwise a negative dimension would send it into an infinite loop.)
639    
640    2. Some of the abstract types (light objects, light pointers, most "base"
641       types) in structure C are now eqtypes.
642    
643    3. Added constructors and test functions for NULL function pointers.
644    
645    ----------------------------------------------------------------------
646    Name: Matthias Blume
647    Date: 2002/01/18
648    Tag: blume-20020118-ready-for-new-release
649    Description:
650    
651    Made config/srcarchiveurl point to a new place.  (Will provide boot
652    files shortly.)
653    
654    Maybe we christen this to be 110.38?
655    
656    ----------------------------------------------------------------------
657    Name: Matthias Blume
658    Date: 2002/01/18
659    Tag: blume-20020118-more-ffifiddle
660    Description:
661    
662    Today's FFI fiddling:
663    
664      - Provided a structure CGetSet with "convenient" versions of C.Get.* and
665        C.Set.* that use concrete (MLRep.*) arguments and results instead
666        of abstract ones.
667    
668      - Provided word-style bit operations etc. for "int" representation
669        types in MLRep.S<Foo>Bitops where <Foo> ranges over Char, Int, Short,
670        and Long.
671    
672    ----------------------------------------------------------------------
673    Name: Matthias Blume
674    Date: 2002/01/18
675    Tag: blume-20020118-use-x86-fp
676    Description:
677    
678    Now that x86-fast-fp seems to be working, I turned it back on again
679    by default.  (Seems to work fine now, even with the FFI.)
680    
681    Other than that, I added some documentation about the FFI to
682    src/ml-nlffigen/README and updated the FFI test examples in
683    src/ml-nlffi-lib/Tests/*.
684    
685    ----------------------------------------------------------------------
686    Name: Allen Leung
687    Date: 2002/01/17
688    Tag: leunga-20020117-x86-fast-fp-call
689    Description:
690    
691       1. Fixed a problem with handling return fp values when x86's fast fp
692          mode is turned on.
693    
694       2. Minor pretty printing fix for cellset.  Print %st(0) as %st(0) instead
695          of %f32.
696    
697       3. Added a constructor INT32lit to the ast of MLRISC tools.
698    
699    ----------------------------------------------------------------------
700    Name: Matthias Blume
701    Date: 2002/01/16
702    Tag: blume-20020116-ffifiddle
703    Description:
704    
705    More fiddling with the FFI interface:
706    
707     - Make constness 'c instead of rw wherever possible.  This eliminates
708       the need for certain explicit coercions.  (However, due to ML's
709       value polymorphism, there will still be many cases where explicit
710       coercions are necessary.  Phantom types are not the whole answer
711       to modeling a subtyping relationship in ML.)
712    
713     - ro/rw coersions for pointers added.  (Avoids the detour through */&.)
714    
715     - "printf" test example added to src/ml-nlffi-lib/Tests.  (Demonstrates
716       clumsy workaround for varargs problem.)
717    
718    ----------------------------------------------------------------------
719    Name: Lal George
720    Date: 2002/01/15
721    Tag: <none>
722    Description:
723    
724    1. Since COPY instructions are no longer native to the architecture,
725       a generic functor can be used to implement the expandCopies function.
726    
727    2. Allowed EXPORT and IMPORT pseudo-op declarations to appear inside a
728       TEXT segment.
729    
730    ----------------------------------------------------------------------
731    Name: Matthias Blume
732    Date: 2002/01/15
733    Tag: blume-20020115-ffiupdates
734    Description:
735    
736    1. Fix for bug resulting in single-precision float values being returned
737       incorrectly from FFI calls.
738    
739    2. Small modifications to C FFI API:
740    
741        - memory-allocation routines return straight objects (no options)
742          and raise an exception in out-of-memory situations
743        - unsafe extensions to cast between function pointers and pointers
744          from/to ints
745        - added structure C_Debug as an alternative to structure C where
746          pointer-dereferencing (|*| and |*!) always check for null-pointers
747        - added open_lib' to DynLinkage;  open_lib' works like open_lib
748          but also takes a (possibly empty) list of existing library handles
749          that the current library depends on
750    
751    ----------------------------------------------------------------------
752    Name: Matthias Blume
753    Date: 2002/01/10
754    Tag: blume-20020110-newffigen
755    Description:
756    
757    1. Updates to portable graph code.
758    
759    2. Major update to ml-nlffigen and ml-nlffi-lib.  Things are much
760       more scalable now so that even huge interfaces such as the one
761       for GTK compile in finite time and space. :-)
762       See src/ml-nlffigen/README for details on what's new.
763    
764    ----------------------------------------------------------------------
765    Name: Lal George
766    Date: 2001/01/09 14:31:35 EST 2002
767    Tag: george-20011206-rm-native-copy
768    Description:
769    
770            Removed the native COPY and FCOPY instructions
771            from all the architectures and replaced it with the
772            explicit COPY instruction from the previous commit.
773    
774            It is now possible to simplify many of the optimizations
775            modules that manipulate copies. This has not been
776            done in this change.
777    
778    ----------------------------------------------------------------------
779    Name: Lal George
780    Date: 2001/12/06 16:50:13 EST 2001
781    Tag: george-20011206-mlrisc-instruction
782    Description:
783    
784    Changed the representation of instructions from being fully abstract
785    to being partially concrete. That is to say:
786    
787      from
788            type instruction
789    
790      to
791            type instr                              (* machine instruction *)
792    
793            datatype instruction =
794                LIVE of {regs: C.cellset, spilled: C.cellset}
795              | KILL of {regs: C.cellset, spilled: C.cellset}
796              | COPYXXX of {k: CB.cellkind, dst: CB.cell list, src: CB.cell list}
797              | ANNOTATION of {i: instruction, a: Annotations.annotation}
798              | INSTR of instr
799    
800    This makes the handling of certain special instructions that appear on
801    all architectures easier and uniform.
802    
803    LIVE and KILL say that a list of registers are live or killed at the
804    program point where they appear. No spill code is generated when an
805    element of the 'regs' field is spilled, but the register is moved to
806    the 'spilled' (which is present, more for debugging than anything else).
807    
808    LIVE replaces the (now deprecated) DEFFREG instruction on the alpha.
809    We used to generate:
810    
811            DEFFREG f1
812            f1 := f2 + f3
813            trapb
814    
815    but now generate:
816    
817            f1 := f2 + f3
818            trapb
819            LIVE {regs=[f1,f2,f3], spilled=[]}
820    
821    Furthermore, the DEFFREG (hack) required that all floating point instruction
822    use all registers mentioned in the instruction. Therefore f1 := f2 + f3,
823    defines f1 and uses [f1,f2,f3]! This hack is no longer required resulting
824    in a cleaner alpha implementation. (Hopefully, intel will not get rid of
825    this architecture).
826    
827    COPYXXX is intended to replace the parallel COPY and FCOPY  available on
828    all the architectures. This will result in further simplification of the
829    register allocator that must be aware of them for coalescing purposes, and
830    will also simplify certain aspects of the machine description that provides
831    callbacks related to parallel copies.
832    
833    ANNOTATION should be obvious, and now INSTR represents the honest to God
834    machine instruction set!
835    
836    The <arch>/instructions/<arch>Instr.sml files define certain utility
837    functions for making porting easier -- essentially converting upper case
838    to lower case. All machine instructions (of type instr) are in upper case,
839    and the lower case form generates an MLRISC instruction. For example on
840    the alpha we have:
841    
842      datatype instr =
843         LDA of {r:cell, b:cell, d:operand}
844       | ...
845    
846      val lda : {r:cell, b:cell, d:operand} -> instruction
847        ...
848    
849    where lda is just (INSTR o LDA), etc.
850    
851    ----------------------------------------------------------------------
852    Name: Matthias Blume
853    Date: 2001/11/22 21:40:00 EST
854    Tag: Release_110_37
855    Description:
856    
857    Release 110.37.  This time for real.
858    
859    ----------------------------------------------------------------------
860    Name: Matthias Blume
861    Date: 2001/11/21 16:35:00 EST
862    Tag: blume-20011121-foot-in-mouth
863    Description:
864    
865    Removed the "Release_110_37" tag because of a serious bug.
866    This will be re-tagged once the bug is fixed.
867    
868    ----------------------------------------------------------------------
869    Name: Matthias Blume
870    Date: 2001/11/21 16:14:00 EST
871    Tag: blume-20011121-forgottenfile
872    Description:
873    
874    Forgot to add a file.  (Just a .tex-file -- part of
875    the CM manual source.)
876    
877    ----------------------------------------------------------------------
878    Name: Matthias Blume
879    Date: 2001/11/21 16:10:00 EST
880    Tag: blume-20011121-invalid_110_37
881    Description:
882    
883    Note: I removed the original tag "Release_110_37" from this commit
884          because we found a serious bug in all non-x86 backends.
885          - Matthias
886    
887    1. Modifications to the SML/NJ code generator and to the runtime system
888       so that code object name strings are directly inserted into code
889       objects at code generation time.  The only business the runtime system
890       has with this is now to read the name strings on occasions.
891       (The encoding of the name string has also changed somewhat.)
892    
893    2. CM now implements a simple "set calculus" for specifying export lists.
894       In particular, it is now possible to refer to the export lists of
895       other libraries/groups/sources and form unions as well as differences.
896       See the latest CM manual for details.
897    
898    3. An separate notion of "proxy" libraries has again be eliminated from
899       CM's model.  (Proxy libraries are now simply a special case of using
900       the export list calculus.)
901    
902    4. Some of the existing libraries now take advantage of the new set
903       calculus.
904       (Notice that not all libraries have been converted because some
905       of the existing .cm-files are supposed to be backward compatible
906       with 110.0.x.)
907    
908    5. Some cleanup in stand-alone programs.  (Don't use "exnMessage" -- use
909       "General.exnMessage"!  The former relies on a certain hook to be
910       initialized, and that often does not happen in the stand-alone case.)
911    
912    ----------------------------------------------------------------------
913    Name: Lal George
914    Date: 2001/11/21  13:56:18 EST
915    Tag: george-2001121-pseudo-ops
916    Description:
917    
918      Implemented a complete redesign of MLRISC pseudo-ops. Now there
919      ought to never be any question of incompatabilities with
920      pseudo-op syntax expected by host assemblers.
921    
922      For now, only modules supporting GAS syntax are implemented
923      but more should follow, such as MASM, and vendor assembler
924      syntax, e.g. IBM as, Sun as, etc.
925    
926    ----------------------------------------------------------------------
927    Name: Matthias Blume
928    Date: 2001/11/14 11:52:00 EST
929    Tag: blume-20011114-srcname
930    Description:
931    
932    1. Routed the name of the current source file to mlriscgen where it
933       should be directly emitted into the code object.  (This last part
934       is yet to be done.)
935    
936    2. Some cleanup of the pgraph code to make it match the proposal that
937       I put out the other day.  (The proposal notwithstanding, things are
938       still in flux here.)
939    
940    ----------------------------------------------------------------------
941    Name: Lal George
942    Date: 2001/11/14 09:44:04 EST
943    Tag:
944    Description:
945    
946      Fix for a backpatching bug reported by Allen.
947    
948      Because the boundary between short and long span-dependent
949      instructions is +/- 128, there are an astounding number of
950      span-dependent instructions whose size is over estimated.
951    
952      Allen came up with the idea of letting the size of span
953      dependent instructions be non-monotonic, for a maxIter
954      number of times, after which the size must be monotonically
955      increasing.
956    
957      This table shows the number of span-dependent instructions
958      whose size was over-estimated as a function of maxIter, for the
959      file Parse/parse/ml.grm.sml:
960    
961         maxIter            # of instructions:
962            10                      687
963            20                      438
964            30                      198
965            40                        0
966    
967      In compiling the compiler, there is no significant difference in
968      compilation speed between maxIter=10 and maxIter=40. Actually,
969      my measurements showed that maxIter=40 was a tad faster than
970      maxIter=10! Also 96% of the  files in the compiler reach a fix
971      point within 13 iterations, so fixing maxIter at 40, while high,
972      is okay.
973    
974    ----------------------------------------------------------------------
975    Name: Matthias Blume
976    Date: 2001/10/31 15:25:00 EST
977    Tag: blume-20011031-pgraph
978    Description:
979    
980    CKIT:
981    * Changed the "Function" constructor of type Ast.ctype to carry optional
982      argument identifiers.
983    * Changed the return type of TypeUtil.getFunction accordingly.
984    * Type equality ignores the argument names.
985    * TypeUtil.composite tries to preserve argument names but gives up quickly
986      if there is a mismatch.
987    
988    installation script:
989    * attempts to use "curl" if available (unless "wget" is available as well)
990    
991    CM:
992    * has an experimental implementation of "portable graphs" which I will
993      soon propose as an implementation-independent library format
994    * there are also new libraries $/pgraph.cm and $/pgraph-util.cm
995    
996    NLFFI-LIB:
997    * some cleanup (all cosmetic)
998    
999    NLFFIGEN:
1000    * temporarily disabled the mechanism that suppresses ML output for
1001      C definitions whose identifiers start with an underscore character
1002    * generate val bindings for enum constants
1003    * user can request that only one style (light or heavy) is being used;
1004      default is to use both (command-line arguments: -heavy and -light)
1005    * fixed bug in handling of function types involving incomplete pointers
1006    * generate ML entry points that take record arguments (i.e., using
1007      named arguments) for C functions that have a prototype with named
1008      arguments
1009      (see changes to CKIT)
1010    
1011    ----------------------------------------------------------------------
1012    Name: Allen Leung
1013    Date: 2001/10/27 20:34:00 EDT
1014    Tag: leunga-20011027-x86-fast-fp-call
1015    Description:
1016    
1017       Fixed the bug described in blume-20010920-slowfp.
1018    
1019       The fix involves
1020          1. generating FCOPYs in FSTP in ia32-svid
1021          2. marking a CALL with the appropriate annotation
1022    
1023    ----------------------------------------------------------------------
1024    Name: Matthias Blume
1025    Date: 2001/10/16 11:32:00 EDT
1026    Tag: blume-20011016-netbsd
1027    Description:
1028    
1029    Underscore patch from Chris Richards (fixing problem with compiling
1030    runtime system under recent NetBSD).
1031    
1032    ----------------------------------------------------------------------
1033    Name: Allen Leung
1034    Date: 2001/10/12 17:18:32 EDT 2001
1035    Tag: leung-20011012-x86-printflowgraph
1036    Description:
1037    
1038    X86RA now uses a valid (instead of dummy) PrintFlowgraph module.
1039    
1040    ----------------------------------------------------------------------
1041  Name: Lal George  Name: Lal George
1042  Date: 2001/10/11 23:51:34 EDT  Date: 2001/10/11 23:51:34 EDT
1043  Tag: george-20011011-too-many-instrs  Tag: george-20011011-too-many-instrs
# Line 80  Line 1105 
1105    
1106  ----------------------------------------------------------------------  ----------------------------------------------------------------------
1107  Name: Matthias Blume  Name: Matthias Blume
 >>>>>>> 1.169  
1108  Date: 2001/09/18 15:35:00 EDT  Date: 2001/09/18 15:35:00 EDT
1109  Tag: blume-20010918-readme11036  Tag: blume-20010918-readme11036
1110  Description:  Description:

Legend:
Removed from v.958  
changed lines
  Added in v.1094

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