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 937, Thu Sep 20 21:20:28 2001 UTC revision 1116, Tue Mar 5 23:17:18 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: Lal George
17    Date: 2002/03/05 17:24:48 EST
18    Tag: george-20020305-linkage-cluster
19    
20    In order to support the block placement optimization, the first
21    cluster that is generated (called the linkage cluster) contains a jump
22    to the entry point for the compilation unit. The linkage cluster
23    contains only one function, so block placement will have no effect on
24    the linkage cluster itself, but all the other clusters have full
25    freedom in the manner in which they reorder blocks or functions.
26    
27    On the x86 the typical linkage code that is generated is:
28       ----------------------
29            .align 2
30       L0:
31            addl    $L1-L0, 72(%esp)
32            jmp     L0
33    
34    
35            .align  2
36       L1:
37       ----------------------
38    
39    72(%esp) is the memory location for the stdlink register. This
40    must contain the address of the CPS function being called. In the
41    above example, it contains the address of memory for  L0; before
42    calling L1 (the real entry point for the compilation unit), it
43    must contain the address for L1, and hence
44    
45            addl $L1-L0, 72(%esp)
46    
47    I have tested this on all architectures except the hppa.
48    
49    ----------------------------------------------------------------------
50    Name: Allen Leung
51    Date: 2002/03/03 13:20:00 EST
52    Tag: leunga-20020303-mlrisc-tools
53    
54      Added #[ ... ] expressions to mlrisc tools
55    
56    ----------------------------------------------------------------------
57    Name: Matthias Blume
58    Date: 2002/02/27 12:29:00 EST
59    Tag: blume-20020227-cdebug
60    Description:
61    
62    - made types in structure C and C_Debug to be equal
63    - got rid of code duplication (c-int.sml vs. c-int-debug.sml)
64    - there no longer is a C_Int_Debug (C_Debug is directly derived from C)
65    
66    ----------------------------------------------------------------------
67    Name: Matthias Blume
68    Date: 2002/02/26 12:00:00 EST
69    Tag: blume-20020226-ffi
70    Description:
71    
72    1. Fixed a minor bug in CM's "noweb" tool:
73       If numbering is turned off, then truly don't number (i.e., do not
74       supply the -L option to noweb).  The previous behavior was to supply
75       -L'' -- which caused noweb to use the "default" line numbering scheme.
76       Thanks to Chris Richards for pointing this out (and supplying the fix).
77    
78    2. Once again, I reworked some aspects of the FFI:
79    
80       A. The incomplete/complete type business:
81    
82       - Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
83         gone!
84       - ML types representing an incomplete type are now *equal* to
85         ML types representing their corresponding complete types (just like
86         in C).  This is still safe because ml-nlffigen will not generate
87         RTTI for incomplete types, nor will it generate functions that
88         require access to such RTTI.   But when ML code generated from both
89         incomplete and complete versions of the C type meet, the ML types
90         are trivially interoperable.
91    
92         NOTE:  These changes restore the full generality of the translation
93         (which was previously lost when I eliminated functorization)!
94    
95       B. Enum types:
96    
97       - Structure C now has a type constructor "enum" that is similar to
98         how the "su" constructor works.  However, "enum" is not a phantom
99         type because each "T enum" has values (and is isomorphic to
100         MLRep.Signed.int).
101       - There are generic access operations for enum objects (using
102         MLRep.Signed.int).
103       - ml-nlffigen will generate a structure E_foo for each "enum foo".
104         * The structure contains the definition of type "mlrep" (the ML-side
105         representation type of the enum).  Normally, mlrep is the same
106         as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
107         then mlrep will be defined as a datatype -- thus facilitating
108         pattern matching on mlrep values.
109         ("-ec" will be suppressed if there are duplicate values in an
110          enumeration.)
111         * Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
112         will be generated for each C enum constant xxx.
113         * Conversion functions m2i and i2m convert between mlrep and
114         MLRep.Signed.int.  (Without "-ec", these functions are identities.)
115         * Coversion functions c and ml convert between mlrep and "tag enum".
116         * Access functions (get/set) fetch and store mlrep values.
117       - By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
118         enumerations are merged into one single enumeration represented by
119         structure E_'.
120    
121    ----------------------------------------------------------------------
122    Name: Allen Leung
123    Date: 2002/02/25 04:45:00 EST
124    Tag: leunga-20020225-cps-spill
125    
126    This is a new implementation of the CPS spill phase.
127    The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
128    In case of problems, replace it with the old file spill.sml
129    
130    The current compiler runs into some serious performance problems when
131    constructing a large record.  This can happen when we try to compile a
132    structure with many items.  Even a very simple structure like the following
133    makes the compiler slow down.
134    
135        structure Foo = struct
136           val x_1 = 0w1 : Word32.int
137           val x_2 = 0w2 : Word32.int
138           val x_3 = 0w3 : Word32.int
139           ...
140           val x_N = 0wN : Word32.int
141        end
142    
143    The following table shows the compile time, from N=1000 to N=4000,
144    with the old compiler:
145    
146    N
147    1000   CPS 100 spill                           0.04u  0.00s  0.00g
148           MLRISC ra                               0.06u  0.00s  0.05g
149              (spills = 0 reloads = 0)
150           TOTAL                                   0.63u  0.07s  0.21g
151    
152    1100   CPS 100 spill                           8.25u  0.32s  0.64g
153           MLRISC ra                               5.68u  0.59s  3.93g
154              (spills = 0 reloads = 0)
155           TOTAL                                   14.71u  0.99s  4.81g
156    
157    1500   CPS 100 spill                           58.55u  2.34s  1.74g
158           MLRISC ra                               5.54u  0.65s  3.91g
159              (spills = 543 reloads = 1082)
160           TOTAL                                   65.40u  3.13s  6.00g
161    
162    2000   CPS 100 spill                           126.69u  4.84s  3.08g
163           MLRISC ra                               0.80u  0.10s  0.55g
164              (spills = 42 reloads = 84)
165           TOTAL                                   129.42u  5.10s  4.13g
166    
167    3000   CPS 100 spill                           675.59u  19.03s  11.64g
168           MLRISC ra                               2.69u  0.27s  1.38g
169              (spills = 62 reloads = 124)
170           TOTAL                                   682.48u  19.61s  13.99g
171    
172    4000   CPS 100 spill                           2362.82u  56.28s  43.60g
173           MLRISC ra                               4.96u  0.27s  2.72g
174              (spills = 85 reloads = 170)
175           TOTAL                                   2375.26u  57.21s  48.00g
176    
177    As you can see the old cps spill module suffers from some serious
178    performance problem.  But since I cannot decipher the old code fully,
179    instead of patching the problems up, I'm reimplementing it
180    with a different algorithm.  The new code is more modular,
181    smaller when compiled, and substantially faster
182    (O(n log n) time and O(n) space).  Timing of the new spill module:
183    
184    4000  CPS 100 spill                           0.02u  0.00s  0.00g
185          MLRISC ra                               0.25u  0.02s  0.15g
186             (spills=1 reloads=3)
187          TOTAL                                   7.74u  0.34s  1.62g
188    
189    Implementation details:
190    
191    As far as I can tell, the purpose of the CPS spill module is to make sure the
192    number of live variables at any program point (the bandwidth)
193    does not exceed a certain limit, which is determined by the
194    size of the spill area.
195    
196    When the bandwidth is too large, we decrease the register pressure by
197    packing live variables into spill records.  How we achieve this is
198    completely different than what we did in the old code.
199    
200    First, there is something about the MLRiscGen code generator
201    that we should be aware of:
202    
203    o MLRiscGen performs code motion!
204    
205       In particular, it will move floating point computations and
206       address computations involving only the heap pointer to
207       their use sites (if there is only a single use).
208       What this means is that if we have a CPS record construction
209       statement
210    
211           RECORD(k,vl,w,e)
212    
213       we should never count the new record address w as live if w
214       has only one use (which is often the case).
215    
216       We should do something similar to floating point, but the transformation
217       there is much more complex, so I won't deal with that.
218    
219    Secondly, there are now two new cps primops at our disposal:
220    
221     1. rawrecord of record_kind option
222        This pure operator allocates some uninitialized storage from the heap.
223        There are two forms:
224    
225         rawrecord NONE [INT n]  allocates a tagless record of length n
226         rawrecord (SOME rk) [INT n] allocates a tagged record of length n
227                                     and initializes the tag.
228    
229     2. rawupdate of cty
230          rawupdate cty (v,i,x)
231          Assigns to x to the ith component of record v.
232          The storelist is not updated.
233    
234    We use these new primops for both spilling and increment record construction.
235    
236     1. Spilling.
237    
238        This is implemented with a linear scan algorithm (but generalized
239        to trees).  The algorithm will create a single spill record at the
240        beginning of the cps function and use rawupdate to spill to it,
241        and SELECT or SELp to reload from it.  So both spills and reloads
242        are fine-grain operations.  In contrast, in the old algorithm
243        "spills" have to be bundled together in records.
244    
245        Ideally, we should sink the spill record construction to where
246        it is needed.  We can even split the spill record into multiple ones
247        at the places where they are needed.  But CPS is not a good
248        representation for global code motion, so I'll keep it simple and
249        am not attempting this.
250    
251     2. Incremental record construction (aka record splitting).
252    
253        Long records with many component values which are simulatenously live
254        (recall that single use record addresses are not considered to
255         be live) are constructed with rawrecord and rawupdate.
256        We allocate space on the heap with rawrecord first, then gradually
257        fill it in with rawupdate.  This is the technique suggested to me
258        by Matthias.
259    
260        Some restrictions on when this is applicable:
261        1. It is not a VECTOR record.  The code generator currently does not handle
262           this case. VECTOR record uses double indirection like arrays.
263        2. All the record component values are defined in the same "basic block"
264           as the record constructor.  This is to prevent speculative
265           record construction.
266    
267    ----------------------------------------------------------------------
268    Name: Allen Leung
269    Date: 2002/02/22 01:02:00 EST
270    Tag: leunga-20020222-mlrisc-tools
271    
272    Minor bug fixes in the parser and rewriter
273    
274    ----------------------------------------------------------------------
275    Name: Allen Leung
276    Date: 2002/02/21 20:20:00 EST
277    Tag: leunga-20020221-peephole
278    
279    Regenerated the peephole files.  Some contained typos in the specification
280    and some didn't compile because of pretty printing bugs in the old version
281    of 'nowhere'.
282    
283    ----------------------------------------------------------------------
284    Name: Allen Leung
285    Date: 2002/02/19 20:20:00 EST
286    Tag: leunga-20020219-mlrisc-tools
287    Description:
288    
289       Minor bug fixes to the mlrisc-tools library:
290    
291       1.  Fixed up parsing colon suffixed keywords
292       2.  Added the ability to shut the error messages up
293       3.  Reimplemented the pretty printer and fixed up/improved
294           the pretty printing of handle and -> types.
295       4.  Fixed up generation of literal symbols in the nowhere tool.
296       5.  Added some SML keywords to to sml.sty
297    
298    ----------------------------------------------------------------------
299    Name: Matthias Blume
300    Date: 2002/02/19 16:20:00 EST
301    Tag: blume-20020219-cmffi
302    Description:
303    
304    A wild mix of changes, some minor, some major:
305    
306    * All C FFI-related libraries are now anchored under $c:
307        $/c.cm      --> $c/c.cm
308        $/c-int.cm  --> $c/internals/c-int.cm
309        $/memory.cm --> $c/memory/memory.cm
310    
311    * "make" tool (in CM) now treats its argument pathname slightly
312      differently:
313        1. If the native expansion is an absolute name, then before invoking
314           the "make" command on it, CM will apply OS.Path.mkRelative
315           (with relativeTo = OS.FileSys.getDir()) to it.
316        2. The argument will be passed through to subsequent phases of CM
317           processing without "going native".  In particular, if the argument
318           was an anchored path, then "make" will not lose track of that anchor.
319    
320    * Compiler backends now "know" their respective C calling conventions
321      instead of having to be told about it by ml-nlffigen.  This relieves
322      ml-nlffigen from one of its burdens.
323    
324    * The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
325    
326    * Export C_DEBUG and C_Debug from $c/c.cm.
327    
328    * C type encoding in ml-nlffi-lib has been improved to model the conceptual
329      subtyping relationship between incomplete pointers and their complete
330      counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
331      with the convention of instantiating 'o with ('t, 'c) obj whenever
332      the pointer target type is complete.  In the incomplete case, 'o
333      will be instantiated with some "'c iobj" -- a type obtained by
334      using one of the functors PointerToIncompleteType or PointerToCompleteType.
335    
336      Operations that work on both incomplete and complete pointer types are
337      typed as taking an 'o ptr while operations that require the target to
338      be known are typed as taking some ('t, 'c) obj ptr.
339    
340      voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
341      where void is an eqtype without any values.  This makes it possible
342      to work on voidptr values using functions meant to operate on light
343      incomplete pointers.
344    
345    * As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
346      been vastly simplified.
347    
348    ----------------------------------------------------------------------
349    Name: Matthias Blume
350    Date: 2002/02/19 10:48:00 EST
351    Tag: blume-20020219-pqfix
352    Description:
353    
354    Applied Chris Okasaki's bug fix for priority queues.
355    
356    ----------------------------------------------------------------------
357    Name: Matthias Blume
358    Date: 2002/02/15 17:05:00
359    Tag: Release_110_39
360    Description:
361    
362    Last-minute retagging is becoming a tradition... :-(
363    
364    This is the working release 110.39.
365    
366    ----------------------------------------------------------------------
367    Name: Matthias Blume
368    Date: 2002/02/15 16:00:00 EST
369    Tag: Release_110_39-orig
370    Description:
371    
372    Working release 110.39.  New bootfiles.
373    
374    (Update: There was a small bug in the installer so it wouldn't work
375    with all shells.  So I retagged. -Matthias)
376    
377    ----------------------------------------------------------------------
378    Name: Matthias Blume
379    Date: 2002/02/15 14:17:00 EST
380    Tag: blume-20020215-showbindings
381    Description:
382    
383    Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
384    the latter can be useful for exploring what bindings are available at
385    the interactive prompt.  (The first function returns only the list
386    of symbols that are really bound, the second prints those but also the
387    ones that CM's autoloading mechanism knows about.)
388    
389    ----------------------------------------------------------------------
390    Name: Matthias Blume
391    Date: 2002/02/15 12:08:00 EST
392    Tag: blume-20020215-iptrs
393    Description:
394    
395    Two improvements to ml-nlffigen:
396    
397      1. Write files only if they do not exist or if their current contents
398         do not coincide with what's being written.  (That is, avoid messing
399         with the time stamps unless absolutely necessary.)
400    
401      2. Implement a "repository" mechanism for generated files related
402         to "incomplete pointer types".   See the README file for details.
403    
404    ----------------------------------------------------------------------
405    Name: Matthias Blume
406    Date: 2002/02/14 11:50:00 EST
407    Tag: blume-20020214-quote
408    Description:
409    
410    Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
411    because of the new and improved tag generation scheme.  (Thanks to Allen
412    Leung for pointing it out.)
413    
414    ----------------------------------------------------------------------
415    Name: Lal George
416    Date: 2002/02/14 09:55:27 EST 2002
417    Tag: george-20020214-isabelle-bug
418    Description:
419    
420    Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
421    of Isabelle on the x86.
422    
423    From Allen:
424    -----------
425     I've found the problem:
426    
427         in ra-core.sml, I use the counter "blocked" to keep track of the
428         true number of elements in the freeze queue.  When the counter goes
429         to zero, I skip examining the queue.  But I've messed up the
430         bookkeeping in combine():
431    
432             else ();
433             case !ucol of
434               PSEUDO => (if !cntv > 0 then
435                     (if !cntu > 0 then blocked := !blocked - 1 else ();
436                                        ^^^^^^^^^^^^^^^^^^^^^^^
437                      moveu := mergeMoveList(!movev, !moveu)
438                     )
439                  else ();
440    
441         combine() is called to coalesce two nodes u and v.
442         I think I was thinking that if the move counts of u and v are both
443         greater than zero then after they are coalesced then one node is
444         removed from the freeze queue.  Apparently I was thinking that
445         both u and v are of low degree, but that's clearly not necessarily true.
446    
447    
448    02/12/2002:
449        Here's the patch.  HOL now compiles.
450    
451        I don't know how this impact on performance (compile
452        time or runtime).  This bug caused the RA (especially on the x86)
453        to go thru the potential spill phase when there are still nodes on the
454        freeze queue.
455    
456    
457    
458    
459    ----------------------------------------------------------------------
460    Name: Matthias Blume
461    Date: 2002/02/13 22:40:00 EST
462    Tag: blume-20020213-fptr-rtti
463    Description:
464    
465    Fixed a bug in ml-nlffigen that was introduced with one of the previous
466    updates.
467    
468    ----------------------------------------------------------------------
469    Name: Matthias Blume
470    Date: 2002/02/13 16:41:00 EST
471    Tag: blume-20020213-cmlpq
472    Description:
473    
474    Added new priority queue export symbols (which have just been added to
475    smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
476    not compile and the installer would choke.)
477    
478    ----------------------------------------------------------------------
479    Name: Matthias Blume
480    Date: 2002/02/13 16:15:00 EST
481    Tag: blume-20020213-various
482    Description:
483    
484    1. More tweaks to ml-nlffigen:
485    
486       - better internal datastructures (resulting in slight speedup)
487       - "-match" option requires exact match
488       - "localized" gensym counters (untagged structs/unions nested within
489         other structs/unions or within typedefs get a fresh counter; their
490         tag will be prefixed by a concatenation of their parents' tags)
491       - bug fixes (related to calculation of transitive closure of types
492         to be included in the output)
493    
494    2. Minor Basis updates:
495    
496       - added implementations for List.collate and Option.app
497    
498    ----------------------------------------------------------------------
499    Name: Matthias Blume
500    Date: 2002/02/11 15:55:00 EST
501    Tag: blume-20020211-gensym
502    Description:
503    
504    Added a "-gensym" option to command line of ml-nlffigen.  This can be
505    used to specify a "stem" -- a string that is inserted in all "gensym'd"
506    names (ML structure names that correspond to unnamed C structs, unions,
507    and enums), so that separate runs of ml-nlffigen do not clash.
508    
509    ----------------------------------------------------------------------
510    Name: Matthias Blume
511    Date: 2002/02/11 12:05:00 EST
512    Tag: blume-20020211-gensml
513    Description:
514    
515    A quick fix for a problem with GenSML (in the pgraph-util library):
516    Make generation of toplevel "local" optional.  (Strictly speaking,
517    signature definitions within "local" are not legal SML.)
518    
519    Other than that: updates to INSTALL and cm/TODO.
520    
521    ----------------------------------------------------------------------
522    Name: Matthias Blume
523    Date: 2002/02/08 15:00:00 EST
524    Tag: blume-20020208-uniquepid
525    Description:
526    
527    0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
528    
529    1. The installer (config/install.sh) has gotten smarter:
530    
531         - Configuration options are a bit easier to specify now
532           (in config/targets).
533         - Bug in recognizing .tar.bz2 files fixed.
534         - Installer automatically resolves dependencies between
535           configuration options (e.g., if you ask for eXene, you will
536           also get cml -- regardless whether you asked for it or not).
537         - Installer can run in "quieter mode" by setting the environment
538           variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
539           "completely silent", though.
540         - Build HashCons library as part of smlnj-lib.
541    
542    2. A new scheme for assigning persistent identifiers to compilation
543       units (and, by extension, to types etc.) has been put into place.
544       This fixes a long-standing bug where types and even dynamic values
545       can get internally confused, thereby compromising type safety
546       (abstraction) and dynamic correctness.  See
547    
548         http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
549    
550       for an example of how things could go wrong until now.
551    
552       The downside of the new scheme is that pids are not quite as
553       persistent as they used to be: CM will generate a fresh pid
554       for every compilation unit that it thinks it sees for the first
555       time.  That means that if you compile starting from a clean, fresh
556       source tree at two different times, you end up with different
557       binaries.
558    
559       Cutoff recompilation, however, has not been compromised because
560       CM keeps pid information in special caches between runs.
561    
562    ----------------------------------------------------------------------
563    Name: Lal George
564    Date: 2002/02/07 15:34:13 EST 2002
565    Tag: <none>
566    Description:
567    
568    Compilers that generate assembly code may produce  global labels
569    whose value is resolved at link time. The various peephole optimization
570    modules did not take this in account.
571    
572    TODO. The Labels.addrOf function should really return an option
573    type so that clients are forced to deal with this issue, rather
574    than an exception being raised.
575    
576    ----------------------------------------------------------------------
577    Name: Lal George
578    Date: 2002/02/06 13:55:02 EST
579    Tag: george-20020206-ra-breakup
580    Description:
581    
582    1. A bug fix from Allen.
583    
584        A typo causes extra fstp %st(0)'s to be generated at compensation
585        edges, which might cause stack underflow traps at runtime.  This
586        occurs in fft where there are extraneous fstps right before the 'into'
587        trap instruction (in this case they are harmless since none of the
588        integers overflow.)
589    
590    2. Pulled out various utility modules that were embedded in the modules
591       of the register allocator. I need these modules for other purposes, but
592       they are not complete enough to put into a library (just yet).
593    ----------------------------------------------------------------------
594    Name: Matthias Blume
595    Date: 2002/01/31 16:05:00 EST
596    Tag: blume-20020131-sparc-ccalls
597    Description:
598    
599    1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
600       of extra stack space by mistake.  Fixed.
601    
602    2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
603    
604    ----------------------------------------------------------------------
605    Name: Allen Leung
606    Date: 2002/01/30
607    Tag: leunga-20020130-nowhere-bug-fix
608    Description:
609    
610       MLRISC bug fixes:
611       1. Fixed a bindings computation bug in the 'nowhere' program generator tool.
612       2. MachineInt.fromString was negating its value.
613    
614    ----------------------------------------------------------------------
615    Name: Matthias Blume
616    Date: 2002/01/29
617    Tag: blume-20020129-INSTALL
618    Description:
619    
620    - Added somewhat detailed installation instructions (file INSTALL).
621    - Fixed curl-detection bug in config/install.sh.
622    - It is now possible to select the URL getter using the URLGETTER
623      environment variable:
624    
625          not set / "unknown"      --> automatic detection (script tries wget,
626                                       curl, and lynx)
627          "wget" / "curl" / "lynx" --> use the specified program (script "knows"
628                                       how to properly invoke them)
629          other                    --> use $URLGETTER directly, it must take
630                                       precisely two command-line arguments
631                                       (source URL and destination file name)
632    
633    ----------------------------------------------------------------------
634    Name: Matthias Blume
635    Date: 2002/01/28
636    Tag: blume-20020128-sparc-ccalls
637    Description:
638    
639    - Fixed problem with calculation of "used" registers in sparc-c-calls.
640    - Make use of the allocParam argument in sparc-c-calls.
641    
642    ----------------------------------------------------------------------
643    Name: Matthias Blume
644    Date: 2002/01/28
645    Tag: blume-20020128-allocParam
646    Description:
647    
648    John Reppy:  Changes c-calls API to accept client-callback for
649    allocating extra stack space.
650    me: Corresponding changes to mlriscGen (using a dummy argument that
651        does not change the current behavior).
652    
653    ----------------------------------------------------------------------
654    Name: Matthias Blume
655    Date: 2002/01/28 12:00:00
656    Tag: Release_110_38
657    Description:
658    
659    This time for real!!!
660    
661    ----------------------------------------------------------------------
662    Name: Matthias Blume
663    Date: 2002/01/28 10:56:00 EST
664    Tag: blume-20020128-retraction
665    Description:
666    
667    0. Retracted earlier 110.38.  (The Release_110_38 tag has been replaced
668       with blume-Release_110_38-retracted.)
669    
670    1. Fixed a problem with incorrect rounding modes in real64.sml.
671       (Thanks to Andrew Mccreight <andrew.mccreight@yale.edu>.)
672    
673    2. A bug in ml-nlffigen related to the handling of unnamed structs, unions,
674       and enums fixed.  The naming of corresponding ML identifiers should
675       now be consistent again.
676    
677    ----------------------------------------------------------------------
678    Name: Allen Leung
679    Date: 2002/01/27
680    Tag: leunga-20020127-nowhere
681    Description:
682    
683       Added a target called nowhere in the configuration scripts.
684       Enabling this will build the MLRISC 'nowhere' tool (for translating
685       programs with where-clauses into legal SML code) during installation.
686    
687    ----------------------------------------------------------------------
688    Name: Matthias Blume
689    Date: 2002/01/25 21:27:00 EST
690    Tag: blume-Release_110_38-retracted
691    Description:
692    
693    Call it a (working) release!  Version is 110.38. Bootfiles are ready.
694    
695    README will be added later.
696    
697    !!! NOTE:  Re-tagged as blume-Release_110_38-retracted. Original tag
698    (Release_110_38) removed.  Reason: Last-minute bug fixes.
699    
700    ----------------------------------------------------------------------
701    Name: Matthias Blume
702    Date: 2002/01/25
703    Tag: blume-20020125-ffi
704    Description:
705    
706    A large number of tweaks and improvements to ml-nlffi-lib and
707    ml-nlffigen:
708    
709       - ML represenation types have been streamlined
710       - getter and setter functions work with concrete values, not abstract
711         ones where possible
712       - ml-nlffigen command line more flexible (see README file there)
713       - some bugs have been fixed (hopefully)
714    
715    ----------------------------------------------------------------------
716    Name: Lal George
717    Date: 2002/01/24
718    Tag: george-20020124-risc-ra-interface
719    Description:
720    
721       There is a dramatic simplification in the interface to the
722       register allocator for RISC architectures as a result of making
723       parallel copy instructions explicit.
724    
725    ----------------------------------------------------------------------
726    Name: Matthias Blume
727    Date: 2002/01/22
728    Tag: blume-20020122-x86-ccalls
729    Description:
730    
731    Bug fix for c-calls on x86 (having to do with how char- and
732    short-arguments are being handled).
733    
734    ----------------------------------------------------------------------
735    Name: Matthias Blume
736    Date: 2002/01/21
737    Tag: blume-20020121-ff
738    Description:
739    
740    Another day of fiddling with the FFI...
741    
742    1. Bug fix/workaround:  CKIT does not complain about negative array
743       dimensions, so ml-nlffigen has to guard itself against this possibility.
744       (Otherwise a negative dimension would send it into an infinite loop.)
745    
746    2. Some of the abstract types (light objects, light pointers, most "base"
747       types) in structure C are now eqtypes.
748    
749    3. Added constructors and test functions for NULL function pointers.
750    
751    ----------------------------------------------------------------------
752    Name: Matthias Blume
753    Date: 2002/01/18
754    Tag: blume-20020118-ready-for-new-release
755    Description:
756    
757    Made config/srcarchiveurl point to a new place.  (Will provide boot
758    files shortly.)
759    
760    Maybe we christen this to be 110.38?
761    
762    ----------------------------------------------------------------------
763    Name: Matthias Blume
764    Date: 2002/01/18
765    Tag: blume-20020118-more-ffifiddle
766    Description:
767    
768    Today's FFI fiddling:
769    
770      - Provided a structure CGetSet with "convenient" versions of C.Get.* and
771        C.Set.* that use concrete (MLRep.*) arguments and results instead
772        of abstract ones.
773    
774      - Provided word-style bit operations etc. for "int" representation
775        types in MLRep.S<Foo>Bitops where <Foo> ranges over Char, Int, Short,
776        and Long.
777    
778    ----------------------------------------------------------------------
779    Name: Matthias Blume
780    Date: 2002/01/18
781    Tag: blume-20020118-use-x86-fp
782    Description:
783    
784    Now that x86-fast-fp seems to be working, I turned it back on again
785    by default.  (Seems to work fine now, even with the FFI.)
786    
787    Other than that, I added some documentation about the FFI to
788    src/ml-nlffigen/README and updated the FFI test examples in
789    src/ml-nlffi-lib/Tests/*.
790    
791    ----------------------------------------------------------------------
792    Name: Allen Leung
793    Date: 2002/01/17
794    Tag: leunga-20020117-x86-fast-fp-call
795    Description:
796    
797       1. Fixed a problem with handling return fp values when x86's fast fp
798          mode is turned on.
799    
800       2. Minor pretty printing fix for cellset.  Print %st(0) as %st(0) instead
801          of %f32.
802    
803       3. Added a constructor INT32lit to the ast of MLRISC tools.
804    
805    ----------------------------------------------------------------------
806    Name: Matthias Blume
807    Date: 2002/01/16
808    Tag: blume-20020116-ffifiddle
809    Description:
810    
811    More fiddling with the FFI interface:
812    
813     - Make constness 'c instead of rw wherever possible.  This eliminates
814       the need for certain explicit coercions.  (However, due to ML's
815       value polymorphism, there will still be many cases where explicit
816       coercions are necessary.  Phantom types are not the whole answer
817       to modeling a subtyping relationship in ML.)
818    
819     - ro/rw coersions for pointers added.  (Avoids the detour through */&.)
820    
821     - "printf" test example added to src/ml-nlffi-lib/Tests.  (Demonstrates
822       clumsy workaround for varargs problem.)
823    
824    ----------------------------------------------------------------------
825    Name: Lal George
826    Date: 2002/01/15
827    Tag: <none>
828    Description:
829    
830    1. Since COPY instructions are no longer native to the architecture,
831       a generic functor can be used to implement the expandCopies function.
832    
833    2. Allowed EXPORT and IMPORT pseudo-op declarations to appear inside a
834       TEXT segment.
835    
836    ----------------------------------------------------------------------
837    Name: Matthias Blume
838    Date: 2002/01/15
839    Tag: blume-20020115-ffiupdates
840    Description:
841    
842    1. Fix for bug resulting in single-precision float values being returned
843       incorrectly from FFI calls.
844    
845    2. Small modifications to C FFI API:
846    
847        - memory-allocation routines return straight objects (no options)
848          and raise an exception in out-of-memory situations
849        - unsafe extensions to cast between function pointers and pointers
850          from/to ints
851        - added structure C_Debug as an alternative to structure C where
852          pointer-dereferencing (|*| and |*!) always check for null-pointers
853        - added open_lib' to DynLinkage;  open_lib' works like open_lib
854          but also takes a (possibly empty) list of existing library handles
855          that the current library depends on
856    
857    ----------------------------------------------------------------------
858    Name: Matthias Blume
859    Date: 2002/01/10
860    Tag: blume-20020110-newffigen
861    Description:
862    
863    1. Updates to portable graph code.
864    
865    2. Major update to ml-nlffigen and ml-nlffi-lib.  Things are much
866       more scalable now so that even huge interfaces such as the one
867       for GTK compile in finite time and space. :-)
868       See src/ml-nlffigen/README for details on what's new.
869    
870    ----------------------------------------------------------------------
871    Name: Lal George
872    Date: 2001/01/09 14:31:35 EST 2002
873    Tag: george-20011206-rm-native-copy
874    Description:
875    
876            Removed the native COPY and FCOPY instructions
877            from all the architectures and replaced it with the
878            explicit COPY instruction from the previous commit.
879    
880            It is now possible to simplify many of the optimizations
881            modules that manipulate copies. This has not been
882            done in this change.
883    
884    ----------------------------------------------------------------------
885    Name: Lal George
886    Date: 2001/12/06 16:50:13 EST 2001
887    Tag: george-20011206-mlrisc-instruction
888    Description:
889    
890    Changed the representation of instructions from being fully abstract
891    to being partially concrete. That is to say:
892    
893      from
894            type instruction
895    
896      to
897            type instr                              (* machine instruction *)
898    
899            datatype instruction =
900                LIVE of {regs: C.cellset, spilled: C.cellset}
901              | KILL of {regs: C.cellset, spilled: C.cellset}
902              | COPYXXX of {k: CB.cellkind, dst: CB.cell list, src: CB.cell list}
903              | ANNOTATION of {i: instruction, a: Annotations.annotation}
904              | INSTR of instr
905    
906    This makes the handling of certain special instructions that appear on
907    all architectures easier and uniform.
908    
909    LIVE and KILL say that a list of registers are live or killed at the
910    program point where they appear. No spill code is generated when an
911    element of the 'regs' field is spilled, but the register is moved to
912    the 'spilled' (which is present, more for debugging than anything else).
913    
914    LIVE replaces the (now deprecated) DEFFREG instruction on the alpha.
915    We used to generate:
916    
917            DEFFREG f1
918            f1 := f2 + f3
919            trapb
920    
921    but now generate:
922    
923            f1 := f2 + f3
924            trapb
925            LIVE {regs=[f1,f2,f3], spilled=[]}
926    
927    Furthermore, the DEFFREG (hack) required that all floating point instruction
928    use all registers mentioned in the instruction. Therefore f1 := f2 + f3,
929    defines f1 and uses [f1,f2,f3]! This hack is no longer required resulting
930    in a cleaner alpha implementation. (Hopefully, intel will not get rid of
931    this architecture).
932    
933    COPYXXX is intended to replace the parallel COPY and FCOPY  available on
934    all the architectures. This will result in further simplification of the
935    register allocator that must be aware of them for coalescing purposes, and
936    will also simplify certain aspects of the machine description that provides
937    callbacks related to parallel copies.
938    
939    ANNOTATION should be obvious, and now INSTR represents the honest to God
940    machine instruction set!
941    
942    The <arch>/instructions/<arch>Instr.sml files define certain utility
943    functions for making porting easier -- essentially converting upper case
944    to lower case. All machine instructions (of type instr) are in upper case,
945    and the lower case form generates an MLRISC instruction. For example on
946    the alpha we have:
947    
948      datatype instr =
949         LDA of {r:cell, b:cell, d:operand}
950       | ...
951    
952      val lda : {r:cell, b:cell, d:operand} -> instruction
953        ...
954    
955    where lda is just (INSTR o LDA), etc.
956    
957    ----------------------------------------------------------------------
958    Name: Matthias Blume
959    Date: 2001/11/22 21:40:00 EST
960    Tag: Release_110_37
961    Description:
962    
963    Release 110.37.  This time for real.
964    
965    ----------------------------------------------------------------------
966    Name: Matthias Blume
967    Date: 2001/11/21 16:35:00 EST
968    Tag: blume-20011121-foot-in-mouth
969    Description:
970    
971    Removed the "Release_110_37" tag because of a serious bug.
972    This will be re-tagged once the bug is fixed.
973    
974    ----------------------------------------------------------------------
975    Name: Matthias Blume
976    Date: 2001/11/21 16:14:00 EST
977    Tag: blume-20011121-forgottenfile
978    Description:
979    
980    Forgot to add a file.  (Just a .tex-file -- part of
981    the CM manual source.)
982    
983    ----------------------------------------------------------------------
984    Name: Matthias Blume
985    Date: 2001/11/21 16:10:00 EST
986    Tag: blume-20011121-invalid_110_37
987    Description:
988    
989    Note: I removed the original tag "Release_110_37" from this commit
990          because we found a serious bug in all non-x86 backends.
991          - Matthias
992    
993    1. Modifications to the SML/NJ code generator and to the runtime system
994       so that code object name strings are directly inserted into code
995       objects at code generation time.  The only business the runtime system
996       has with this is now to read the name strings on occasions.
997       (The encoding of the name string has also changed somewhat.)
998    
999    2. CM now implements a simple "set calculus" for specifying export lists.
1000       In particular, it is now possible to refer to the export lists of
1001       other libraries/groups/sources and form unions as well as differences.
1002       See the latest CM manual for details.
1003    
1004    3. An separate notion of "proxy" libraries has again be eliminated from
1005       CM's model.  (Proxy libraries are now simply a special case of using
1006       the export list calculus.)
1007    
1008    4. Some of the existing libraries now take advantage of the new set
1009       calculus.
1010       (Notice that not all libraries have been converted because some
1011       of the existing .cm-files are supposed to be backward compatible
1012       with 110.0.x.)
1013    
1014    5. Some cleanup in stand-alone programs.  (Don't use "exnMessage" -- use
1015       "General.exnMessage"!  The former relies on a certain hook to be
1016       initialized, and that often does not happen in the stand-alone case.)
1017    
1018    ----------------------------------------------------------------------
1019    Name: Lal George
1020    Date: 2001/11/21  13:56:18 EST
1021    Tag: george-2001121-pseudo-ops
1022    Description:
1023    
1024      Implemented a complete redesign of MLRISC pseudo-ops. Now there
1025      ought to never be any question of incompatabilities with
1026      pseudo-op syntax expected by host assemblers.
1027    
1028      For now, only modules supporting GAS syntax are implemented
1029      but more should follow, such as MASM, and vendor assembler
1030      syntax, e.g. IBM as, Sun as, etc.
1031    
1032    ----------------------------------------------------------------------
1033    Name: Matthias Blume
1034    Date: 2001/11/14 11:52:00 EST
1035    Tag: blume-20011114-srcname
1036    Description:
1037    
1038    1. Routed the name of the current source file to mlriscgen where it
1039       should be directly emitted into the code object.  (This last part
1040       is yet to be done.)
1041    
1042    2. Some cleanup of the pgraph code to make it match the proposal that
1043       I put out the other day.  (The proposal notwithstanding, things are
1044       still in flux here.)
1045    
1046    ----------------------------------------------------------------------
1047    Name: Lal George
1048    Date: 2001/11/14 09:44:04 EST
1049    Tag:
1050    Description:
1051    
1052      Fix for a backpatching bug reported by Allen.
1053    
1054      Because the boundary between short and long span-dependent
1055      instructions is +/- 128, there are an astounding number of
1056      span-dependent instructions whose size is over estimated.
1057    
1058      Allen came up with the idea of letting the size of span
1059      dependent instructions be non-monotonic, for a maxIter
1060      number of times, after which the size must be monotonically
1061      increasing.
1062    
1063      This table shows the number of span-dependent instructions
1064      whose size was over-estimated as a function of maxIter, for the
1065      file Parse/parse/ml.grm.sml:
1066    
1067         maxIter            # of instructions:
1068            10                      687
1069            20                      438
1070            30                      198
1071            40                        0
1072    
1073      In compiling the compiler, there is no significant difference in
1074      compilation speed between maxIter=10 and maxIter=40. Actually,
1075      my measurements showed that maxIter=40 was a tad faster than
1076      maxIter=10! Also 96% of the  files in the compiler reach a fix
1077      point within 13 iterations, so fixing maxIter at 40, while high,
1078      is okay.
1079    
1080    ----------------------------------------------------------------------
1081    Name: Matthias Blume
1082    Date: 2001/10/31 15:25:00 EST
1083    Tag: blume-20011031-pgraph
1084    Description:
1085    
1086    CKIT:
1087    * Changed the "Function" constructor of type Ast.ctype to carry optional
1088      argument identifiers.
1089    * Changed the return type of TypeUtil.getFunction accordingly.
1090    * Type equality ignores the argument names.
1091    * TypeUtil.composite tries to preserve argument names but gives up quickly
1092      if there is a mismatch.
1093    
1094    installation script:
1095    * attempts to use "curl" if available (unless "wget" is available as well)
1096    
1097    CM:
1098    * has an experimental implementation of "portable graphs" which I will
1099      soon propose as an implementation-independent library format
1100    * there are also new libraries $/pgraph.cm and $/pgraph-util.cm
1101    
1102    NLFFI-LIB:
1103    * some cleanup (all cosmetic)
1104    
1105    NLFFIGEN:
1106    * temporarily disabled the mechanism that suppresses ML output for
1107      C definitions whose identifiers start with an underscore character
1108    * generate val bindings for enum constants
1109    * user can request that only one style (light or heavy) is being used;
1110      default is to use both (command-line arguments: -heavy and -light)
1111    * fixed bug in handling of function types involving incomplete pointers
1112    * generate ML entry points that take record arguments (i.e., using
1113      named arguments) for C functions that have a prototype with named
1114      arguments
1115      (see changes to CKIT)
1116    
1117    ----------------------------------------------------------------------
1118    Name: Allen Leung
1119    Date: 2001/10/27 20:34:00 EDT
1120    Tag: leunga-20011027-x86-fast-fp-call
1121    Description:
1122    
1123       Fixed the bug described in blume-20010920-slowfp.
1124    
1125       The fix involves
1126          1. generating FCOPYs in FSTP in ia32-svid
1127          2. marking a CALL with the appropriate annotation
1128    
1129    ----------------------------------------------------------------------
1130    Name: Matthias Blume
1131    Date: 2001/10/16 11:32:00 EDT
1132    Tag: blume-20011016-netbsd
1133    Description:
1134    
1135    Underscore patch from Chris Richards (fixing problem with compiling
1136    runtime system under recent NetBSD).
1137    
1138    ----------------------------------------------------------------------
1139    Name: Allen Leung
1140    Date: 2001/10/12 17:18:32 EDT 2001
1141    Tag: leung-20011012-x86-printflowgraph
1142    Description:
1143    
1144    X86RA now uses a valid (instead of dummy) PrintFlowgraph module.
1145    
1146    ----------------------------------------------------------------------
1147    Name: Lal George
1148    Date: 2001/10/11 23:51:34 EDT
1149    Tag: george-20011011-too-many-instrs
1150    Description:
1151    
1152    The representation of a program point never expected to see more
1153    than 65536 instructions in a basic block!
1154    
1155    ----------------------------------------------------------------------
1156    Name: Lal George
1157    Date: 2001/10/09 09:41:37 EDT
1158    Tag: george-20011008-mlrisc-labels
1159    Description:
1160    
1161    Changed the machine description files to support printing of
1162    local and global labels in assembly code, based on host assembler
1163    conventions.
1164    
1165    ----------------------------------------------------------------------
1166    Name: Matthias Blume
1167    Date: 2001/09/25 15:25:00 EDT
1168    Tag: blume-20010925-exninfo
1169    Description:
1170    
1171    I provided a non-hook implementation of exnName (at the toplevel) and
1172    made the "dummy" implementation of exnMessage (at the toplevel) more
1173    useful: if nothing gets "hooked in", then at least you are going to
1174    see the exception name and a message indicating why you don't see more.
1175    
1176    [For the time being, programs that need exnMessage and want to use
1177    ml-build should either use General.exnMessage (strongly recommended) or
1178    refer to structure General at some other point so that CM sees a
1179    static dependency.]
1180    
1181    [Similar remarks go for "print" and "use":  If you want to use their
1182    functionality in stand-alone programs generated by ml-build, then use
1183    TextIO.output and Backend.Interact.useFile (from $smlnj/compiler.cm).]
1184    
1185    ----------------------------------------------------------------------
1186    Name: Matthias Blume
1187    Date: 2001/09/20 17:28:00 EDT
1188    Tag: blume-20010920-slowfp
1189    Description:
1190    
1191    Allen says that x86-fast-fp is not safe yet, so I turned it off again...
1192    
1193    ----------------------------------------------------------------------
1194  Name: Matthias Blume  Name: Matthias Blume
1195  Date: 2001/09/20 17:20:00 EDT  Date: 2001/09/20 17:20:00 EDT
1196  Tag: blume-20010920-canonicalpaths  Tag: blume-20010920-canonicalpaths

Legend:
Removed from v.937  
changed lines
  Added in v.1116

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