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 975, Wed Oct 31 20:22:44 2001 UTC revision 1098, Wed Feb 27 17:29:47 2002 UTC
# Line 8  Line 8 
8  The form of an entry should be:  The form of an entry should be:
9    
10  Name:  Name:
11  Date:  Date: yyyy/mm/dd
12  Tag: <post-commit CVS tag>  Tag: <post-commit CVS tag>
13  Description:  Description:
14    
15  ----------------------------------------------------------------------  ----------------------------------------------------------------------
16  Name: Matthias Blume  Name: Matthias Blume
17    Date: 2002/02/27 12:29:00 EST
18    Tag: blume-20020227-cdebug
19    Description:
20    
21    - made types in structure C and C_Debug to be equal
22    - got rid of code duplication (c-int.sml vs. c-int-debug.sml)
23    - there no longer is a C_Int_Debug (C_Debug is directly derived from C)
24    
25    ----------------------------------------------------------------------
26    Name: Matthias Blume
27    Date: 2002/02/26 12:00:00 EST
28    Tag: blume-20020226-ffi
29    Description:
30    
31    1. Fixed a minor bug in CM's "noweb" tool:
32       If numbering is turned off, then truly don't number (i.e., do not
33       supply the -L option to noweb).  The previous behavior was to supply
34       -L'' -- which caused noweb to use the "default" line numbering scheme.
35       Thanks to Chris Richards for pointing this out (and supplying the fix).
36    
37    2. Once again, I reworked some aspects of the FFI:
38    
39       A. The incomplete/complete type business:
40    
41       - Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
42         gone!
43       - ML types representing an incomplete type are now *equal* to
44         ML types representing their corresponding complete types (just like
45         in C).  This is still safe because ml-nlffigen will not generate
46         RTTI for incomplete types, nor will it generate functions that
47         require access to such RTTI.   But when ML code generated from both
48         incomplete and complete versions of the C type meet, the ML types
49         are trivially interoperable.
50    
51         NOTE:  These changes restore the full generality of the translation
52         (which was previously lost when I eliminated functorization)!
53    
54       B. Enum types:
55    
56       - Structure C now has a type constructor "enum" that is similar to
57         how the "su" constructor works.  However, "enum" is not a phantom
58         type because each "T enum" has values (and is isomorphic to
59         MLRep.Signed.int).
60       - There are generic access operations for enum objects (using
61         MLRep.Signed.int).
62       - ml-nlffigen will generate a structure E_foo for each "enum foo".
63         * The structure contains the definition of type "mlrep" (the ML-side
64         representation type of the enum).  Normally, mlrep is the same
65         as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
66         then mlrep will be defined as a datatype -- thus facilitating
67         pattern matching on mlrep values.
68         ("-ec" will be suppressed if there are duplicate values in an
69          enumeration.)
70         * Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
71         will be generated for each C enum constant xxx.
72         * Conversion functions m2i and i2m convert between mlrep and
73         MLRep.Signed.int.  (Without "-ec", these functions are identities.)
74         * Coversion functions c and ml convert between mlrep and "tag enum".
75         * Access functions (get/set) fetch and store mlrep values.
76       - By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
77         enumerations are merged into one single enumeration represented by
78         structure E_'.
79    
80    ----------------------------------------------------------------------
81    Name: Allen Leung
82    Date: 2002/02/25 04:45:00 EST
83    Tag: leunga-20020225-cps-spill
84    
85    This is a new implementation of the CPS spill phase.
86    The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
87    In case of problems, replace it with the old file spill.sml
88    
89    The current compiler runs into some serious performance problems when
90    constructing a large record.  This can happen when we try to compile a
91    structure with many items.  Even a very simple structure like the following
92    makes the compiler slow down.
93    
94        structure Foo = struct
95           val x_1 = 0w1 : Word32.int
96           val x_2 = 0w2 : Word32.int
97           val x_3 = 0w3 : Word32.int
98           ...
99           val x_N = 0wN : Word32.int
100        end
101    
102    The following table shows the compile time, from N=1000 to N=4000,
103    with the old compiler:
104    
105    N
106    1000   CPS 100 spill                           0.04u  0.00s  0.00g
107           MLRISC ra                               0.06u  0.00s  0.05g
108              (spills = 0 reloads = 0)
109           TOTAL                                   0.63u  0.07s  0.21g
110    
111    1100   CPS 100 spill                           8.25u  0.32s  0.64g
112           MLRISC ra                               5.68u  0.59s  3.93g
113              (spills = 0 reloads = 0)
114           TOTAL                                   14.71u  0.99s  4.81g
115    
116    1500   CPS 100 spill                           58.55u  2.34s  1.74g
117           MLRISC ra                               5.54u  0.65s  3.91g
118              (spills = 543 reloads = 1082)
119           TOTAL                                   65.40u  3.13s  6.00g
120    
121    2000   CPS 100 spill                           126.69u  4.84s  3.08g
122           MLRISC ra                               0.80u  0.10s  0.55g
123              (spills = 42 reloads = 84)
124           TOTAL                                   129.42u  5.10s  4.13g
125    
126    3000   CPS 100 spill                           675.59u  19.03s  11.64g
127           MLRISC ra                               2.69u  0.27s  1.38g
128              (spills = 62 reloads = 124)
129           TOTAL                                   682.48u  19.61s  13.99g
130    
131    4000   CPS 100 spill                           2362.82u  56.28s  43.60g
132           MLRISC ra                               4.96u  0.27s  2.72g
133              (spills = 85 reloads = 170)
134           TOTAL                                   2375.26u  57.21s  48.00g
135    
136    As you can see the old cps spill module suffers from some serious
137    performance problem but since I cannot decipher the old code fully,
138    innstead of patching the problems up, I'm reimplementing it
139    with a different algorithm.  The new code is more modular,
140    smaller when compiled, and substantially faster
141    (O(n log n) time and O(n) space).  Timing of the new spill module:
142    
143    4000  CPS 100 spill                           0.02u  0.00s  0.00g
144          MLRISC ra                               0.25u  0.02s  0.15g
145             (spills=1 reloads=3)
146          TOTAL                                   7.74u  0.34s  1.62g
147    
148    Implementation details:
149    
150    As far as I can tell, the purpose of the CPS spill module is to make sure the
151    number of live variables at any program point (the bandwidth)
152    does not exceed a certain limit, which is determined by the
153    size of the spill area.
154    
155    When the bandwidth is too large, we decrease the register pressure by
156    packing live variables into spill records.  How we achieve this is
157    completely different than what we did in the old code.
158    
159    First, there is something about the MLRiscGen code generator
160    that we should be aware of:
161    
162    o MLRiscGen performs code motion!
163    
164       In particular, it will move floating point computations and
165       address computations involving only the heap pointer to
166       their use sites (if there is only a single use).
167       What this means is that if we have a CPS record construction
168       statement
169    
170           RECORD(k,vl,w,e)
171    
172       we should never count the new record address w as live if w
173       has only one use (which is often the case).
174    
175       We should do something similar to floating point, but the transformation
176       there is much more complex, so I won't deal with that.
177    
178    Secondly, there are now two new cps primops at our disposal:
179    
180     1. rawrecord of record_kind option
181        This pure operator allocates some uninitialized storage from the heap.
182        There are two forms:
183    
184         rawrecord NONE [INT n]  allocates a tagless record of length n
185         rawrecord (SOME rk) [INT n] allocates a tagged record of length n
186                                     and initializes the tag.
187    
188     2. rawupdate of cty
189          rawupdate cty (v,i,x)
190          Assigns to x to the ith component of record v.
191          The storelist is not updated.
192    
193    We use these new primops for both spilling and increment record construction.
194    
195     1. Spilling.
196    
197        This is implemented with a linear scan algorithm (but generalized
198        to trees).  The algorithm will create a single spill record at the
199        beginning of the cps function and use rawupdate to spill to it,
200        and SELECT or SELp to reload from it.  So both spills and reloads
201        are fine-grain operations.  In contrast, in the old algorithm
202        "spills" have to be bundled together in records.
203    
204        Ideally, we should sink the spill record construction to where
205        it is needed.  We can even split the spill record into multiple ones
206        at the places where they are needed.  But CPS is not a good
207        representation for global code motion, so I'll keep it simple and
208        am not attempting this.
209    
210     2. Incremental record construction (aka record splitting).
211    
212        Long records with many component values which are simulatenously live
213        (recall that single use record addresses are not considered to
214         be live) are constructed with rawrecord and rawupdate.
215        We allocate space on the heap with rawrecord first, then gradually
216        fill it in with rawupdate.  This is the technique suggested to me
217        by Matthias.
218    
219        Some restrictions on when this is applicable:
220        1. It is not a VECTOR record.  The code generator currently does not handle
221           this case. VECTOR record uses double indirection like arrays.
222        2. All the record component values are defined in the same "basic block"
223           as the record constructor.  This is to prevent speculative
224           record construction.
225    
226    ----------------------------------------------------------------------
227    Name: Allen Leung
228    Date: 2002/02/22 01:02:00 EST
229    Tag: leunga-20020222-mlrisc-tools
230    
231    Minor bug fixes in the parser and rewriter
232    
233    ----------------------------------------------------------------------
234    Name: Allen Leung
235    Date: 2002/02/21 20:20:00 EST
236    Tag: leunga-20020221-peephole
237    
238    Regenerated the peephole files.  Some contained typos in the specification
239    and some didn't compile because of pretty printing bugs in the old version
240    of 'nowhere'.
241    
242    ----------------------------------------------------------------------
243    Name: Allen Leung
244    Date: 2002/02/19 20:20:00 EST
245    Tag: leunga-20020219-mlrisc-tools
246    Description:
247    
248       Minor bug fixes to the mlrisc-tools library:
249    
250       1.  Fixed up parsing colon suffixed keywords
251       2.  Added the ability to shut the error messages up
252       3.  Reimplemented the pretty printer and fixed up/improved
253           the pretty printing of handle and -> types.
254       4.  Fixed up generation of literal symbols in the nowhere tool.
255       5.  Added some SML keywords to to sml.sty
256    
257    ----------------------------------------------------------------------
258    Name: Matthias Blume
259    Date: 2002/02/19 16:20:00 EST
260    Tag: blume-20020219-cmffi
261    Description:
262    
263    A wild mix of changes, some minor, some major:
264    
265    * All C FFI-related libraries are now anchored under $c:
266        $/c.cm      --> $c/c.cm
267        $/c-int.cm  --> $c/internals/c-int.cm
268        $/memory.cm --> $c/memory/memory.cm
269    
270    * "make" tool (in CM) now treats its argument pathname slightly
271      differently:
272        1. If the native expansion is an absolute name, then before invoking
273           the "make" command on it, CM will apply OS.Path.mkRelative
274           (with relativeTo = OS.FileSys.getDir()) to it.
275        2. The argument will be passed through to subsequent phases of CM
276           processing without "going native".  In particular, if the argument
277           was an anchored path, then "make" will not lose track of that anchor.
278    
279    * Compiler backends now "know" their respective C calling conventions
280      instead of having to be told about it by ml-nlffigen.  This relieves
281      ml-nlffigen from one of its burdens.
282    
283    * The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
284    
285    * Export C_DEBUG and C_Debug from $c/c.cm.
286    
287    * C type encoding in ml-nlffi-lib has been improved to model the conceptual
288      subtyping relationship between incomplete pointers and their complete
289      counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
290      with the convention of instantiating 'o with ('t, 'c) obj whenever
291      the pointer target type is complete.  In the incomplete case, 'o
292      will be instantiated with some "'c iobj" -- a type obtained by
293      using one of the functors PointerToIncompleteType or PointerToCompleteType.
294    
295      Operations that work on both incomplete and complete pointer types are
296      typed as taking an 'o ptr while operations that require the target to
297      be known are typed as taking some ('t, 'c) obj ptr.
298    
299      voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
300      where void is an eqtype without any values.  This makes it possible
301      to work on voidptr values using functions meant to operate on light
302      incomplete pointers.
303    
304    * As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
305      been vastly simplified.
306    
307    ----------------------------------------------------------------------
308    Name: Matthias Blume
309    Date: 2002/02/19 10:48:00 EST
310    Tag: blume-20020219-pqfix
311    Description:
312    
313    Applied Chris Okasaki's bug fix for priority queues.
314    
315    ----------------------------------------------------------------------
316    Name: Matthias Blume
317    Date: 2002/02/15 17:05:00
318    Tag: Release_110_39
319    Description:
320    
321    Last-minute retagging is becoming a tradition... :-(
322    
323    This is the working release 110.39.
324    
325    ----------------------------------------------------------------------
326    Name: Matthias Blume
327    Date: 2002/02/15 16:00:00 EST
328    Tag: Release_110_39-orig
329    Description:
330    
331    Working release 110.39.  New bootfiles.
332    
333    (Update: There was a small bug in the installer so it wouldn't work
334    with all shells.  So I retagged. -Matthias)
335    
336    ----------------------------------------------------------------------
337    Name: Matthias Blume
338    Date: 2002/02/15 14:17:00 EST
339    Tag: blume-20020215-showbindings
340    Description:
341    
342    Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
343    the latter can be useful for exploring what bindings are available at
344    the interactive prompt.  (The first function returns only the list
345    of symbols that are really bound, the second prints those but also the
346    ones that CM's autoloading mechanism knows about.)
347    
348    ----------------------------------------------------------------------
349    Name: Matthias Blume
350    Date: 2002/02/15 12:08:00 EST
351    Tag: blume-20020215-iptrs
352    Description:
353    
354    Two improvements to ml-nlffigen:
355    
356      1. Write files only if they do not exist or if their current contents
357         do not coincide with what's being written.  (That is, avoid messing
358         with the time stamps unless absolutely necessary.)
359    
360      2. Implement a "repository" mechanism for generated files related
361         to "incomplete pointer types".   See the README file for details.
362    
363    ----------------------------------------------------------------------
364    Name: Matthias Blume
365    Date: 2002/02/14 11:50:00 EST
366    Tag: blume-20020214-quote
367    Description:
368    
369    Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
370    because of the new and improved tag generation scheme.  (Thanks to Allen
371    Leung for pointing it out.)
372    
373    ----------------------------------------------------------------------
374    Name: Lal George
375    Date: 2002/02/14 09:55:27 EST 2002
376    Tag: george-20020214-isabelle-bug
377    Description:
378    
379    Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
380    of Isabelle on the x86.
381    
382    From Allen:
383    -----------
384     I've found the problem:
385    
386         in ra-core.sml, I use the counter "blocked" to keep track of the
387         true number of elements in the freeze queue.  When the counter goes
388         to zero, I skip examining the queue.  But I've messed up the
389         bookkeeping in combine():
390    
391             else ();
392             case !ucol of
393               PSEUDO => (if !cntv > 0 then
394                     (if !cntu > 0 then blocked := !blocked - 1 else ();
395                                        ^^^^^^^^^^^^^^^^^^^^^^^
396                      moveu := mergeMoveList(!movev, !moveu)
397                     )
398                  else ();
399    
400         combine() is called to coalesce two nodes u and v.
401         I think I was thinking that if the move counts of u and v are both
402         greater than zero then after they are coalesced then one node is
403         removed from the freeze queue.  Apparently I was thinking that
404         both u and v are of low degree, but that's clearly not necessarily true.
405    
406    
407    02/12/2002:
408        Here's the patch.  HOL now compiles.
409    
410        I don't know how this impact on performance (compile
411        time or runtime).  This bug caused the RA (especially on the x86)
412        to go thru the potential spill phase when there are still nodes on the
413        freeze queue.
414    
415    
416    
417    
418    ----------------------------------------------------------------------
419    Name: Matthias Blume
420    Date: 2002/02/13 22:40:00 EST
421    Tag: blume-20020213-fptr-rtti
422    Description:
423    
424    Fixed a bug in ml-nlffigen that was introduced with one of the previous
425    updates.
426    
427    ----------------------------------------------------------------------
428    Name: Matthias Blume
429    Date: 2002/02/13 16:41:00 EST
430    Tag: blume-20020213-cmlpq
431    Description:
432    
433    Added new priority queue export symbols (which have just been added to
434    smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
435    not compile and the installer would choke.)
436    
437    ----------------------------------------------------------------------
438    Name: Matthias Blume
439    Date: 2002/02/13 16:15:00 EST
440    Tag: blume-20020213-various
441    Description:
442    
443    1. More tweaks to ml-nlffigen:
444    
445       - better internal datastructures (resulting in slight speedup)
446       - "-match" option requires exact match
447       - "localized" gensym counters (untagged structs/unions nested within
448         other structs/unions or within typedefs get a fresh counter; their
449         tag will be prefixed by a concatenation of their parents' tags)
450       - bug fixes (related to calculation of transitive closure of types
451         to be included in the output)
452    
453    2. Minor Basis updates:
454    
455       - added implementations for List.collate and Option.app
456    
457    ----------------------------------------------------------------------
458    Name: Matthias Blume
459    Date: 2002/02/11 15:55:00 EST
460    Tag: blume-20020211-gensym
461    Description:
462    
463    Added a "-gensym" option to command line of ml-nlffigen.  This can be
464    used to specify a "stem" -- a string that is inserted in all "gensym'd"
465    names (ML structure names that correspond to unnamed C structs, unions,
466    and enums), so that separate runs of ml-nlffigen do not clash.
467    
468    ----------------------------------------------------------------------
469    Name: Matthias Blume
470    Date: 2002/02/11 12:05:00 EST
471    Tag: blume-20020211-gensml
472    Description:
473    
474    A quick fix for a problem with GenSML (in the pgraph-util library):
475    Make generation of toplevel "local" optional.  (Strictly speaking,
476    signature definitions within "local" are not legal SML.)
477    
478    Other than that: updates to INSTALL and cm/TODO.
479    
480    ----------------------------------------------------------------------
481    Name: Matthias Blume
482    Date: 2002/02/08 15:00:00 EST
483    Tag: blume-20020208-uniquepid
484    Description:
485    
486    0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
487    
488    1. The installer (config/install.sh) has gotten smarter:
489    
490         - Configuration options are a bit easier to specify now
491           (in config/targets).
492         - Bug in recognizing .tar.bz2 files fixed.
493         - Installer automatically resolves dependencies between
494           configuration options (e.g., if you ask for eXene, you will
495           also get cml -- regardless whether you asked for it or not).
496         - Installer can run in "quieter mode" by setting the environment
497           variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
498           "completely silent", though.
499         - Build HashCons library as part of smlnj-lib.
500    
501    2. A new scheme for assigning persistent identifiers to compilation
502       units (and, by extension, to types etc.) has been put into place.
503       This fixes a long-standing bug where types and even dynamic values
504       can get internally confused, thereby compromising type safety
505       (abstraction) and dynamic correctness.  See
506    
507         http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
508    
509       for an example of how things could go wrong until now.
510    
511       The downside of the new scheme is that pids are not quite as
512       persistent as they used to be: CM will generate a fresh pid
513       for every compilation unit that it thinks it sees for the first
514       time.  That means that if you compile starting from a clean, fresh
515       source tree at two different times, you end up with different
516       binaries.
517    
518       Cutoff recompilation, however, has not been compromised because
519       CM keeps pid information in special caches between runs.
520    
521    ----------------------------------------------------------------------
522    Name: Lal George
523    Date: 2002/02/07 15:34:13 EST 2002
524    Tag: <none>
525    Description:
526    
527    Compilers that generate assembly code may produce  global labels
528    whose value is resolved at link time. The various peephole optimization
529    modules did not take this in account.
530    
531    TODO. The Labels.addrOf function should really return an option
532    type so that clients are forced to deal with this issue, rather
533    than an exception being raised.
534    
535    ----------------------------------------------------------------------
536    Name: Lal George
537    Date: 2002/02/06 13:55:02 EST
538    Tag: george-20020206-ra-breakup
539    Description:
540    
541    1. A bug fix from Allen.
542    
543        A typo causes extra fstp %st(0)'s to be generated at compensation
544        edges, which might cause stack underflow traps at runtime.  This
545        occurs in fft where there are extraneous fstps right before the 'into'
546        trap instruction (in this case they are harmless since none of the
547        integers overflow.)
548    
549    2. Pulled out various utility modules that were embedded in the modules
550       of the register allocator. I need these modules for other purposes, but
551       they are not complete enough to put into a library (just yet).
552    ----------------------------------------------------------------------
553    Name: Matthias Blume
554    Date: 2002/01/31 16:05:00 EST
555    Tag: blume-20020131-sparc-ccalls
556    Description:
557    
558    1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
559       of extra stack space by mistake.  Fixed.
560    
561    2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
562    
563    ----------------------------------------------------------------------
564    Name: Allen Leung
565    Date: 2002/01/30
566    Tag: leunga-20020130-nowhere-bug-fix
567    Description:
568    
569       MLRISC bug fixes:
570       1. Fixed a bindings computation bug in the 'nowhere' program generator tool.
571       2. MachineInt.fromString was negating its value.
572    
573    ----------------------------------------------------------------------
574    Name: Matthias Blume
575    Date: 2002/01/29
576    Tag: blume-20020129-INSTALL
577    Description:
578    
579    - Added somewhat detailed installation instructions (file INSTALL).
580    - Fixed curl-detection bug in config/install.sh.
581    - It is now possible to select the URL getter using the URLGETTER
582      environment variable:
583    
584          not set / "unknown"      --> automatic detection (script tries wget,
585                                       curl, and lynx)
586          "wget" / "curl" / "lynx" --> use the specified program (script "knows"
587                                       how to properly invoke them)
588          other                    --> use $URLGETTER directly, it must take
589                                       precisely two command-line arguments
590                                       (source URL and destination file name)
591    
592    ----------------------------------------------------------------------
593    Name: Matthias Blume
594    Date: 2002/01/28
595    Tag: blume-20020128-sparc-ccalls
596    Description:
597    
598    - Fixed problem with calculation of "used" registers in sparc-c-calls.
599    - Make use of the allocParam argument in sparc-c-calls.
600    
601    ----------------------------------------------------------------------
602    Name: Matthias Blume
603    Date: 2002/01/28
604    Tag: blume-20020128-allocParam
605    Description:
606    
607    John Reppy:  Changes c-calls API to accept client-callback for
608    allocating extra stack space.
609    me: Corresponding changes to mlriscGen (using a dummy argument that
610        does not change the current behavior).
611    
612    ----------------------------------------------------------------------
613    Name: Matthias Blume
614    Date: 2002/01/28 12:00:00
615    Tag: Release_110_38
616    Description:
617    
618    This time for real!!!
619    
620    ----------------------------------------------------------------------
621    Name: Matthias Blume
622    Date: 2002/01/28 10:56:00 EST
623    Tag: blume-20020128-retraction
624    Description:
625    
626    0. Retracted earlier 110.38.  (The Release_110_38 tag has been replaced
627       with blume-Release_110_38-retracted.)
628    
629    1. Fixed a problem with incorrect rounding modes in real64.sml.
630       (Thanks to Andrew Mccreight <andrew.mccreight@yale.edu>.)
631    
632    2. A bug in ml-nlffigen related to the handling of unnamed structs, unions,
633       and enums fixed.  The naming of corresponding ML identifiers should
634       now be consistent again.
635    
636    ----------------------------------------------------------------------
637    Name: Allen Leung
638    Date: 2002/01/27
639    Tag: leunga-20020127-nowhere
640    Description:
641    
642       Added a target called nowhere in the configuration scripts.
643       Enabling this will build the MLRISC 'nowhere' tool (for translating
644       programs with where-clauses into legal SML code) during installation.
645    
646    ----------------------------------------------------------------------
647    Name: Matthias Blume
648    Date: 2002/01/25 21:27:00 EST
649    Tag: blume-Release_110_38-retracted
650    Description:
651    
652    Call it a (working) release!  Version is 110.38. Bootfiles are ready.
653    
654    README will be added later.
655    
656    !!! NOTE:  Re-tagged as blume-Release_110_38-retracted. Original tag
657    (Release_110_38) removed.  Reason: Last-minute bug fixes.
658    
659    ----------------------------------------------------------------------
660    Name: Matthias Blume
661    Date: 2002/01/25
662    Tag: blume-20020125-ffi
663    Description:
664    
665    A large number of tweaks and improvements to ml-nlffi-lib and
666    ml-nlffigen:
667    
668       - ML represenation types have been streamlined
669       - getter and setter functions work with concrete values, not abstract
670         ones where possible
671       - ml-nlffigen command line more flexible (see README file there)
672       - some bugs have been fixed (hopefully)
673    
674    ----------------------------------------------------------------------
675    Name: Lal George
676    Date: 2002/01/24
677    Tag: george-20020124-risc-ra-interface
678    Description:
679    
680       There is a dramatic simplification in the interface to the
681       register allocator for RISC architectures as a result of making
682       parallel copy instructions explicit.
683    
684    ----------------------------------------------------------------------
685    Name: Matthias Blume
686    Date: 2002/01/22
687    Tag: blume-20020122-x86-ccalls
688    Description:
689    
690    Bug fix for c-calls on x86 (having to do with how char- and
691    short-arguments are being handled).
692    
693    ----------------------------------------------------------------------
694    Name: Matthias Blume
695    Date: 2002/01/21
696    Tag: blume-20020121-ff
697    Description:
698    
699    Another day of fiddling with the FFI...
700    
701    1. Bug fix/workaround:  CKIT does not complain about negative array
702       dimensions, so ml-nlffigen has to guard itself against this possibility.
703       (Otherwise a negative dimension would send it into an infinite loop.)
704    
705    2. Some of the abstract types (light objects, light pointers, most "base"
706       types) in structure C are now eqtypes.
707    
708    3. Added constructors and test functions for NULL function pointers.
709    
710    ----------------------------------------------------------------------
711    Name: Matthias Blume
712    Date: 2002/01/18
713    Tag: blume-20020118-ready-for-new-release
714    Description:
715    
716    Made config/srcarchiveurl point to a new place.  (Will provide boot
717    files shortly.)
718    
719    Maybe we christen this to be 110.38?
720    
721    ----------------------------------------------------------------------
722    Name: Matthias Blume
723    Date: 2002/01/18
724    Tag: blume-20020118-more-ffifiddle
725    Description:
726    
727    Today's FFI fiddling:
728    
729      - Provided a structure CGetSet with "convenient" versions of C.Get.* and
730        C.Set.* that use concrete (MLRep.*) arguments and results instead
731        of abstract ones.
732    
733      - Provided word-style bit operations etc. for "int" representation
734        types in MLRep.S<Foo>Bitops where <Foo> ranges over Char, Int, Short,
735        and Long.
736    
737    ----------------------------------------------------------------------
738    Name: Matthias Blume
739    Date: 2002/01/18
740    Tag: blume-20020118-use-x86-fp
741    Description:
742    
743    Now that x86-fast-fp seems to be working, I turned it back on again
744    by default.  (Seems to work fine now, even with the FFI.)
745    
746    Other than that, I added some documentation about the FFI to
747    src/ml-nlffigen/README and updated the FFI test examples in
748    src/ml-nlffi-lib/Tests/*.
749    
750    ----------------------------------------------------------------------
751    Name: Allen Leung
752    Date: 2002/01/17
753    Tag: leunga-20020117-x86-fast-fp-call
754    Description:
755    
756       1. Fixed a problem with handling return fp values when x86's fast fp
757          mode is turned on.
758    
759       2. Minor pretty printing fix for cellset.  Print %st(0) as %st(0) instead
760          of %f32.
761    
762       3. Added a constructor INT32lit to the ast of MLRISC tools.
763    
764    ----------------------------------------------------------------------
765    Name: Matthias Blume
766    Date: 2002/01/16
767    Tag: blume-20020116-ffifiddle
768    Description:
769    
770    More fiddling with the FFI interface:
771    
772     - Make constness 'c instead of rw wherever possible.  This eliminates
773       the need for certain explicit coercions.  (However, due to ML's
774       value polymorphism, there will still be many cases where explicit
775       coercions are necessary.  Phantom types are not the whole answer
776       to modeling a subtyping relationship in ML.)
777    
778     - ro/rw coersions for pointers added.  (Avoids the detour through */&.)
779    
780     - "printf" test example added to src/ml-nlffi-lib/Tests.  (Demonstrates
781       clumsy workaround for varargs problem.)
782    
783    ----------------------------------------------------------------------
784    Name: Lal George
785    Date: 2002/01/15
786    Tag: <none>
787    Description:
788    
789    1. Since COPY instructions are no longer native to the architecture,
790       a generic functor can be used to implement the expandCopies function.
791    
792    2. Allowed EXPORT and IMPORT pseudo-op declarations to appear inside a
793       TEXT segment.
794    
795    ----------------------------------------------------------------------
796    Name: Matthias Blume
797    Date: 2002/01/15
798    Tag: blume-20020115-ffiupdates
799    Description:
800    
801    1. Fix for bug resulting in single-precision float values being returned
802       incorrectly from FFI calls.
803    
804    2. Small modifications to C FFI API:
805    
806        - memory-allocation routines return straight objects (no options)
807          and raise an exception in out-of-memory situations
808        - unsafe extensions to cast between function pointers and pointers
809          from/to ints
810        - added structure C_Debug as an alternative to structure C where
811          pointer-dereferencing (|*| and |*!) always check for null-pointers
812        - added open_lib' to DynLinkage;  open_lib' works like open_lib
813          but also takes a (possibly empty) list of existing library handles
814          that the current library depends on
815    
816    ----------------------------------------------------------------------
817    Name: Matthias Blume
818    Date: 2002/01/10
819    Tag: blume-20020110-newffigen
820    Description:
821    
822    1. Updates to portable graph code.
823    
824    2. Major update to ml-nlffigen and ml-nlffi-lib.  Things are much
825       more scalable now so that even huge interfaces such as the one
826       for GTK compile in finite time and space. :-)
827       See src/ml-nlffigen/README for details on what's new.
828    
829    ----------------------------------------------------------------------
830    Name: Lal George
831    Date: 2001/01/09 14:31:35 EST 2002
832    Tag: george-20011206-rm-native-copy
833    Description:
834    
835            Removed the native COPY and FCOPY instructions
836            from all the architectures and replaced it with the
837            explicit COPY instruction from the previous commit.
838    
839            It is now possible to simplify many of the optimizations
840            modules that manipulate copies. This has not been
841            done in this change.
842    
843    ----------------------------------------------------------------------
844    Name: Lal George
845    Date: 2001/12/06 16:50:13 EST 2001
846    Tag: george-20011206-mlrisc-instruction
847    Description:
848    
849    Changed the representation of instructions from being fully abstract
850    to being partially concrete. That is to say:
851    
852      from
853            type instruction
854    
855      to
856            type instr                              (* machine instruction *)
857    
858            datatype instruction =
859                LIVE of {regs: C.cellset, spilled: C.cellset}
860              | KILL of {regs: C.cellset, spilled: C.cellset}
861              | COPYXXX of {k: CB.cellkind, dst: CB.cell list, src: CB.cell list}
862              | ANNOTATION of {i: instruction, a: Annotations.annotation}
863              | INSTR of instr
864    
865    This makes the handling of certain special instructions that appear on
866    all architectures easier and uniform.
867    
868    LIVE and KILL say that a list of registers are live or killed at the
869    program point where they appear. No spill code is generated when an
870    element of the 'regs' field is spilled, but the register is moved to
871    the 'spilled' (which is present, more for debugging than anything else).
872    
873    LIVE replaces the (now deprecated) DEFFREG instruction on the alpha.
874    We used to generate:
875    
876            DEFFREG f1
877            f1 := f2 + f3
878            trapb
879    
880    but now generate:
881    
882            f1 := f2 + f3
883            trapb
884            LIVE {regs=[f1,f2,f3], spilled=[]}
885    
886    Furthermore, the DEFFREG (hack) required that all floating point instruction
887    use all registers mentioned in the instruction. Therefore f1 := f2 + f3,
888    defines f1 and uses [f1,f2,f3]! This hack is no longer required resulting
889    in a cleaner alpha implementation. (Hopefully, intel will not get rid of
890    this architecture).
891    
892    COPYXXX is intended to replace the parallel COPY and FCOPY  available on
893    all the architectures. This will result in further simplification of the
894    register allocator that must be aware of them for coalescing purposes, and
895    will also simplify certain aspects of the machine description that provides
896    callbacks related to parallel copies.
897    
898    ANNOTATION should be obvious, and now INSTR represents the honest to God
899    machine instruction set!
900    
901    The <arch>/instructions/<arch>Instr.sml files define certain utility
902    functions for making porting easier -- essentially converting upper case
903    to lower case. All machine instructions (of type instr) are in upper case,
904    and the lower case form generates an MLRISC instruction. For example on
905    the alpha we have:
906    
907      datatype instr =
908         LDA of {r:cell, b:cell, d:operand}
909       | ...
910    
911      val lda : {r:cell, b:cell, d:operand} -> instruction
912        ...
913    
914    where lda is just (INSTR o LDA), etc.
915    
916    ----------------------------------------------------------------------
917    Name: Matthias Blume
918    Date: 2001/11/22 21:40:00 EST
919    Tag: Release_110_37
920    Description:
921    
922    Release 110.37.  This time for real.
923    
924    ----------------------------------------------------------------------
925    Name: Matthias Blume
926    Date: 2001/11/21 16:35:00 EST
927    Tag: blume-20011121-foot-in-mouth
928    Description:
929    
930    Removed the "Release_110_37" tag because of a serious bug.
931    This will be re-tagged once the bug is fixed.
932    
933    ----------------------------------------------------------------------
934    Name: Matthias Blume
935    Date: 2001/11/21 16:14:00 EST
936    Tag: blume-20011121-forgottenfile
937    Description:
938    
939    Forgot to add a file.  (Just a .tex-file -- part of
940    the CM manual source.)
941    
942    ----------------------------------------------------------------------
943    Name: Matthias Blume
944    Date: 2001/11/21 16:10:00 EST
945    Tag: blume-20011121-invalid_110_37
946    Description:
947    
948    Note: I removed the original tag "Release_110_37" from this commit
949          because we found a serious bug in all non-x86 backends.
950          - Matthias
951    
952    1. Modifications to the SML/NJ code generator and to the runtime system
953       so that code object name strings are directly inserted into code
954       objects at code generation time.  The only business the runtime system
955       has with this is now to read the name strings on occasions.
956       (The encoding of the name string has also changed somewhat.)
957    
958    2. CM now implements a simple "set calculus" for specifying export lists.
959       In particular, it is now possible to refer to the export lists of
960       other libraries/groups/sources and form unions as well as differences.
961       See the latest CM manual for details.
962    
963    3. An separate notion of "proxy" libraries has again be eliminated from
964       CM's model.  (Proxy libraries are now simply a special case of using
965       the export list calculus.)
966    
967    4. Some of the existing libraries now take advantage of the new set
968       calculus.
969       (Notice that not all libraries have been converted because some
970       of the existing .cm-files are supposed to be backward compatible
971       with 110.0.x.)
972    
973    5. Some cleanup in stand-alone programs.  (Don't use "exnMessage" -- use
974       "General.exnMessage"!  The former relies on a certain hook to be
975       initialized, and that often does not happen in the stand-alone case.)
976    
977    ----------------------------------------------------------------------
978    Name: Lal George
979    Date: 2001/11/21  13:56:18 EST
980    Tag: george-2001121-pseudo-ops
981    Description:
982    
983      Implemented a complete redesign of MLRISC pseudo-ops. Now there
984      ought to never be any question of incompatabilities with
985      pseudo-op syntax expected by host assemblers.
986    
987      For now, only modules supporting GAS syntax are implemented
988      but more should follow, such as MASM, and vendor assembler
989      syntax, e.g. IBM as, Sun as, etc.
990    
991    ----------------------------------------------------------------------
992    Name: Matthias Blume
993    Date: 2001/11/14 11:52:00 EST
994    Tag: blume-20011114-srcname
995    Description:
996    
997    1. Routed the name of the current source file to mlriscgen where it
998       should be directly emitted into the code object.  (This last part
999       is yet to be done.)
1000    
1001    2. Some cleanup of the pgraph code to make it match the proposal that
1002       I put out the other day.  (The proposal notwithstanding, things are
1003       still in flux here.)
1004    
1005    ----------------------------------------------------------------------
1006    Name: Lal George
1007    Date: 2001/11/14 09:44:04 EST
1008    Tag:
1009    Description:
1010    
1011      Fix for a backpatching bug reported by Allen.
1012    
1013      Because the boundary between short and long span-dependent
1014      instructions is +/- 128, there are an astounding number of
1015      span-dependent instructions whose size is over estimated.
1016    
1017      Allen came up with the idea of letting the size of span
1018      dependent instructions be non-monotonic, for a maxIter
1019      number of times, after which the size must be monotonically
1020      increasing.
1021    
1022      This table shows the number of span-dependent instructions
1023      whose size was over-estimated as a function of maxIter, for the
1024      file Parse/parse/ml.grm.sml:
1025    
1026         maxIter            # of instructions:
1027            10                      687
1028            20                      438
1029            30                      198
1030            40                        0
1031    
1032      In compiling the compiler, there is no significant difference in
1033      compilation speed between maxIter=10 and maxIter=40. Actually,
1034      my measurements showed that maxIter=40 was a tad faster than
1035      maxIter=10! Also 96% of the  files in the compiler reach a fix
1036      point within 13 iterations, so fixing maxIter at 40, while high,
1037      is okay.
1038    
1039    ----------------------------------------------------------------------
1040    Name: Matthias Blume
1041  Date: 2001/10/31 15:25:00 EST  Date: 2001/10/31 15:25:00 EST
1042  Tag: blume-20011031-pgraph  Tag: blume-20011031-pgraph
1043  Description:  Description:
# Line 146  Line 1170 
1170    
1171  ----------------------------------------------------------------------  ----------------------------------------------------------------------
1172  Name: Matthias Blume  Name: Matthias Blume
 >>>>>>> 1.169  
1173  Date: 2001/09/18 15:35:00 EDT  Date: 2001/09/18 15:35:00 EDT
1174  Tag: blume-20010918-readme11036  Tag: blume-20010918-readme11036
1175  Description:  Description:

Legend:
Removed from v.975  
changed lines
  Added in v.1098

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