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 938, Thu Sep 20 21:27:33 2001 UTC revision 1124, Thu Mar 7 19:49:22 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: 2002/03/07 14:44:24 EST 2002
18    Tag: george-20020307-weighted-block-placement
19    
20    Tested the weighted block placement optimization on all architectures
21    (except the hppa) using AMPL to generate the block and edge frequencies.
22    Changes were required in the machine properties to correctly
23    categorize trap instructions. There is an MLRISC flag
24    "weighted-block-placement" that can be used to enable weighted block
25    placement, but this will be ineffective without block/edge
26    frequencies (coming soon).
27    
28    
29    ----------------------------------------------------------------------
30    Name: Lal George
31    Date: 2002/03/05 17:24:48 EST
32    Tag: george-20020305-linkage-cluster
33    
34    In order to support the block placement optimization, a new cluster
35    is generated as the very first cluster (called the linkage cluster).
36    It contains a single jump to the 'real' entry point for the compilation
37    unit. Block placement has no effect on the linkage cluster itself, but
38    all the other clusters  have full freedom in the manner in which they
39    reorder blocks or functions.
40    
41    On the x86 the typical linkage code that is generated is:
42       ----------------------
43            .align 2
44       L0:
45            addl    $L1-L0, 72(%esp)
46            jmp     L1
47    
48    
49            .align  2
50       L1:
51       ----------------------
52    
53    72(%esp) is the memory location for the stdlink register. This
54    must contain the address of the CPS function being called. In the
55    above example, it contains the address of  L0; before
56    calling L1 (the real entry point for the compilation unit), it
57    must contain the address for L1, and hence
58    
59            addl $L1-L0, 72(%esp)
60    
61    I have tested this on all architectures except the hppa.The increase
62    in code size is of course negligible
63    
64    ----------------------------------------------------------------------
65    Name: Allen Leung
66    Date: 2002/03/03 13:20:00 EST
67    Tag: leunga-20020303-mlrisc-tools
68    
69      Added #[ ... ] expressions to mlrisc tools
70    
71    ----------------------------------------------------------------------
72    Name: Matthias Blume
73    Date: 2002/02/27 12:29:00 EST
74    Tag: blume-20020227-cdebug
75    Description:
76    
77    - made types in structure C and C_Debug to be equal
78    - got rid of code duplication (c-int.sml vs. c-int-debug.sml)
79    - there no longer is a C_Int_Debug (C_Debug is directly derived from C)
80    
81    ----------------------------------------------------------------------
82    Name: Matthias Blume
83    Date: 2002/02/26 12:00:00 EST
84    Tag: blume-20020226-ffi
85    Description:
86    
87    1. Fixed a minor bug in CM's "noweb" tool:
88       If numbering is turned off, then truly don't number (i.e., do not
89       supply the -L option to noweb).  The previous behavior was to supply
90       -L'' -- which caused noweb to use the "default" line numbering scheme.
91       Thanks to Chris Richards for pointing this out (and supplying the fix).
92    
93    2. Once again, I reworked some aspects of the FFI:
94    
95       A. The incomplete/complete type business:
96    
97       - Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
98         gone!
99       - ML types representing an incomplete type are now *equal* to
100         ML types representing their corresponding complete types (just like
101         in C).  This is still safe because ml-nlffigen will not generate
102         RTTI for incomplete types, nor will it generate functions that
103         require access to such RTTI.   But when ML code generated from both
104         incomplete and complete versions of the C type meet, the ML types
105         are trivially interoperable.
106    
107         NOTE:  These changes restore the full generality of the translation
108         (which was previously lost when I eliminated functorization)!
109    
110       B. Enum types:
111    
112       - Structure C now has a type constructor "enum" that is similar to
113         how the "su" constructor works.  However, "enum" is not a phantom
114         type because each "T enum" has values (and is isomorphic to
115         MLRep.Signed.int).
116       - There are generic access operations for enum objects (using
117         MLRep.Signed.int).
118       - ml-nlffigen will generate a structure E_foo for each "enum foo".
119         * The structure contains the definition of type "mlrep" (the ML-side
120         representation type of the enum).  Normally, mlrep is the same
121         as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
122         then mlrep will be defined as a datatype -- thus facilitating
123         pattern matching on mlrep values.
124         ("-ec" will be suppressed if there are duplicate values in an
125          enumeration.)
126         * Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
127         will be generated for each C enum constant xxx.
128         * Conversion functions m2i and i2m convert between mlrep and
129         MLRep.Signed.int.  (Without "-ec", these functions are identities.)
130         * Coversion functions c and ml convert between mlrep and "tag enum".
131         * Access functions (get/set) fetch and store mlrep values.
132       - By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
133         enumerations are merged into one single enumeration represented by
134         structure E_'.
135    
136    ----------------------------------------------------------------------
137    Name: Allen Leung
138    Date: 2002/02/25 04:45:00 EST
139    Tag: leunga-20020225-cps-spill
140    
141    This is a new implementation of the CPS spill phase.
142    The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
143    In case of problems, replace it with the old file spill.sml
144    
145    The current compiler runs into some serious performance problems when
146    constructing a large record.  This can happen when we try to compile a
147    structure with many items.  Even a very simple structure like the following
148    makes the compiler slow down.
149    
150        structure Foo = struct
151           val x_1 = 0w1 : Word32.int
152           val x_2 = 0w2 : Word32.int
153           val x_3 = 0w3 : Word32.int
154           ...
155           val x_N = 0wN : Word32.int
156        end
157    
158    The following table shows the compile time, from N=1000 to N=4000,
159    with the old compiler:
160    
161    N
162    1000   CPS 100 spill                           0.04u  0.00s  0.00g
163           MLRISC ra                               0.06u  0.00s  0.05g
164              (spills = 0 reloads = 0)
165           TOTAL                                   0.63u  0.07s  0.21g
166    
167    1100   CPS 100 spill                           8.25u  0.32s  0.64g
168           MLRISC ra                               5.68u  0.59s  3.93g
169              (spills = 0 reloads = 0)
170           TOTAL                                   14.71u  0.99s  4.81g
171    
172    1500   CPS 100 spill                           58.55u  2.34s  1.74g
173           MLRISC ra                               5.54u  0.65s  3.91g
174              (spills = 543 reloads = 1082)
175           TOTAL                                   65.40u  3.13s  6.00g
176    
177    2000   CPS 100 spill                           126.69u  4.84s  3.08g
178           MLRISC ra                               0.80u  0.10s  0.55g
179              (spills = 42 reloads = 84)
180           TOTAL                                   129.42u  5.10s  4.13g
181    
182    3000   CPS 100 spill                           675.59u  19.03s  11.64g
183           MLRISC ra                               2.69u  0.27s  1.38g
184              (spills = 62 reloads = 124)
185           TOTAL                                   682.48u  19.61s  13.99g
186    
187    4000   CPS 100 spill                           2362.82u  56.28s  43.60g
188           MLRISC ra                               4.96u  0.27s  2.72g
189              (spills = 85 reloads = 170)
190           TOTAL                                   2375.26u  57.21s  48.00g
191    
192    As you can see the old cps spill module suffers from some serious
193    performance problem.  But since I cannot decipher the old code fully,
194    instead of patching the problems up, I'm reimplementing it
195    with a different algorithm.  The new code is more modular,
196    smaller when compiled, and substantially faster
197    (O(n log n) time and O(n) space).  Timing of the new spill module:
198    
199    4000  CPS 100 spill                           0.02u  0.00s  0.00g
200          MLRISC ra                               0.25u  0.02s  0.15g
201             (spills=1 reloads=3)
202          TOTAL                                   7.74u  0.34s  1.62g
203    
204    Implementation details:
205    
206    As far as I can tell, the purpose of the CPS spill module is to make sure the
207    number of live variables at any program point (the bandwidth)
208    does not exceed a certain limit, which is determined by the
209    size of the spill area.
210    
211    When the bandwidth is too large, we decrease the register pressure by
212    packing live variables into spill records.  How we achieve this is
213    completely different than what we did in the old code.
214    
215    First, there is something about the MLRiscGen code generator
216    that we should be aware of:
217    
218    o MLRiscGen performs code motion!
219    
220       In particular, it will move floating point computations and
221       address computations involving only the heap pointer to
222       their use sites (if there is only a single use).
223       What this means is that if we have a CPS record construction
224       statement
225    
226           RECORD(k,vl,w,e)
227    
228       we should never count the new record address w as live if w
229       has only one use (which is often the case).
230    
231       We should do something similar to floating point, but the transformation
232       there is much more complex, so I won't deal with that.
233    
234    Secondly, there are now two new cps primops at our disposal:
235    
236     1. rawrecord of record_kind option
237        This pure operator allocates some uninitialized storage from the heap.
238        There are two forms:
239    
240         rawrecord NONE [INT n]  allocates a tagless record of length n
241         rawrecord (SOME rk) [INT n] allocates a tagged record of length n
242                                     and initializes the tag.
243    
244     2. rawupdate of cty
245          rawupdate cty (v,i,x)
246          Assigns to x to the ith component of record v.
247          The storelist is not updated.
248    
249    We use these new primops for both spilling and increment record construction.
250    
251     1. Spilling.
252    
253        This is implemented with a linear scan algorithm (but generalized
254        to trees).  The algorithm will create a single spill record at the
255        beginning of the cps function and use rawupdate to spill to it,
256        and SELECT or SELp to reload from it.  So both spills and reloads
257        are fine-grain operations.  In contrast, in the old algorithm
258        "spills" have to be bundled together in records.
259    
260        Ideally, we should sink the spill record construction to where
261        it is needed.  We can even split the spill record into multiple ones
262        at the places where they are needed.  But CPS is not a good
263        representation for global code motion, so I'll keep it simple and
264        am not attempting this.
265    
266     2. Incremental record construction (aka record splitting).
267    
268        Long records with many component values which are simulatenously live
269        (recall that single use record addresses are not considered to
270         be live) are constructed with rawrecord and rawupdate.
271        We allocate space on the heap with rawrecord first, then gradually
272        fill it in with rawupdate.  This is the technique suggested to me
273        by Matthias.
274    
275        Some restrictions on when this is applicable:
276        1. It is not a VECTOR record.  The code generator currently does not handle
277           this case. VECTOR record uses double indirection like arrays.
278        2. All the record component values are defined in the same "basic block"
279           as the record constructor.  This is to prevent speculative
280           record construction.
281    
282    ----------------------------------------------------------------------
283    Name: Allen Leung
284    Date: 2002/02/22 01:02:00 EST
285    Tag: leunga-20020222-mlrisc-tools
286    
287    Minor bug fixes in the parser and rewriter
288    
289    ----------------------------------------------------------------------
290    Name: Allen Leung
291    Date: 2002/02/21 20:20:00 EST
292    Tag: leunga-20020221-peephole
293    
294    Regenerated the peephole files.  Some contained typos in the specification
295    and some didn't compile because of pretty printing bugs in the old version
296    of 'nowhere'.
297    
298    ----------------------------------------------------------------------
299    Name: Allen Leung
300    Date: 2002/02/19 20:20:00 EST
301    Tag: leunga-20020219-mlrisc-tools
302    Description:
303    
304       Minor bug fixes to the mlrisc-tools library:
305    
306       1.  Fixed up parsing colon suffixed keywords
307       2.  Added the ability to shut the error messages up
308       3.  Reimplemented the pretty printer and fixed up/improved
309           the pretty printing of handle and -> types.
310       4.  Fixed up generation of literal symbols in the nowhere tool.
311       5.  Added some SML keywords to to sml.sty
312    
313    ----------------------------------------------------------------------
314    Name: Matthias Blume
315    Date: 2002/02/19 16:20:00 EST
316    Tag: blume-20020219-cmffi
317    Description:
318    
319    A wild mix of changes, some minor, some major:
320    
321    * All C FFI-related libraries are now anchored under $c:
322        $/c.cm      --> $c/c.cm
323        $/c-int.cm  --> $c/internals/c-int.cm
324        $/memory.cm --> $c/memory/memory.cm
325    
326    * "make" tool (in CM) now treats its argument pathname slightly
327      differently:
328        1. If the native expansion is an absolute name, then before invoking
329           the "make" command on it, CM will apply OS.Path.mkRelative
330           (with relativeTo = OS.FileSys.getDir()) to it.
331        2. The argument will be passed through to subsequent phases of CM
332           processing without "going native".  In particular, if the argument
333           was an anchored path, then "make" will not lose track of that anchor.
334    
335    * Compiler backends now "know" their respective C calling conventions
336      instead of having to be told about it by ml-nlffigen.  This relieves
337      ml-nlffigen from one of its burdens.
338    
339    * The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
340    
341    * Export C_DEBUG and C_Debug from $c/c.cm.
342    
343    * C type encoding in ml-nlffi-lib has been improved to model the conceptual
344      subtyping relationship between incomplete pointers and their complete
345      counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
346      with the convention of instantiating 'o with ('t, 'c) obj whenever
347      the pointer target type is complete.  In the incomplete case, 'o
348      will be instantiated with some "'c iobj" -- a type obtained by
349      using one of the functors PointerToIncompleteType or PointerToCompleteType.
350    
351      Operations that work on both incomplete and complete pointer types are
352      typed as taking an 'o ptr while operations that require the target to
353      be known are typed as taking some ('t, 'c) obj ptr.
354    
355      voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
356      where void is an eqtype without any values.  This makes it possible
357      to work on voidptr values using functions meant to operate on light
358      incomplete pointers.
359    
360    * As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
361      been vastly simplified.
362    
363    ----------------------------------------------------------------------
364    Name: Matthias Blume
365    Date: 2002/02/19 10:48:00 EST
366    Tag: blume-20020219-pqfix
367    Description:
368    
369    Applied Chris Okasaki's bug fix for priority queues.
370    
371    ----------------------------------------------------------------------
372    Name: Matthias Blume
373    Date: 2002/02/15 17:05:00
374    Tag: Release_110_39
375    Description:
376    
377    Last-minute retagging is becoming a tradition... :-(
378    
379    This is the working release 110.39.
380    
381    ----------------------------------------------------------------------
382    Name: Matthias Blume
383    Date: 2002/02/15 16:00:00 EST
384    Tag: Release_110_39-orig
385    Description:
386    
387    Working release 110.39.  New bootfiles.
388    
389    (Update: There was a small bug in the installer so it wouldn't work
390    with all shells.  So I retagged. -Matthias)
391    
392    ----------------------------------------------------------------------
393    Name: Matthias Blume
394    Date: 2002/02/15 14:17:00 EST
395    Tag: blume-20020215-showbindings
396    Description:
397    
398    Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
399    the latter can be useful for exploring what bindings are available at
400    the interactive prompt.  (The first function returns only the list
401    of symbols that are really bound, the second prints those but also the
402    ones that CM's autoloading mechanism knows about.)
403    
404    ----------------------------------------------------------------------
405    Name: Matthias Blume
406    Date: 2002/02/15 12:08:00 EST
407    Tag: blume-20020215-iptrs
408    Description:
409    
410    Two improvements to ml-nlffigen:
411    
412      1. Write files only if they do not exist or if their current contents
413         do not coincide with what's being written.  (That is, avoid messing
414         with the time stamps unless absolutely necessary.)
415    
416      2. Implement a "repository" mechanism for generated files related
417         to "incomplete pointer types".   See the README file for details.
418    
419    ----------------------------------------------------------------------
420    Name: Matthias Blume
421    Date: 2002/02/14 11:50:00 EST
422    Tag: blume-20020214-quote
423    Description:
424    
425    Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
426    because of the new and improved tag generation scheme.  (Thanks to Allen
427    Leung for pointing it out.)
428    
429    ----------------------------------------------------------------------
430    Name: Lal George
431    Date: 2002/02/14 09:55:27 EST 2002
432    Tag: george-20020214-isabelle-bug
433    Description:
434    
435    Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
436    of Isabelle on the x86.
437    
438    From Allen:
439    -----------
440     I've found the problem:
441    
442         in ra-core.sml, I use the counter "blocked" to keep track of the
443         true number of elements in the freeze queue.  When the counter goes
444         to zero, I skip examining the queue.  But I've messed up the
445         bookkeeping in combine():
446    
447             else ();
448             case !ucol of
449               PSEUDO => (if !cntv > 0 then
450                     (if !cntu > 0 then blocked := !blocked - 1 else ();
451                                        ^^^^^^^^^^^^^^^^^^^^^^^
452                      moveu := mergeMoveList(!movev, !moveu)
453                     )
454                  else ();
455    
456         combine() is called to coalesce two nodes u and v.
457         I think I was thinking that if the move counts of u and v are both
458         greater than zero then after they are coalesced then one node is
459         removed from the freeze queue.  Apparently I was thinking that
460         both u and v are of low degree, but that's clearly not necessarily true.
461    
462    
463    02/12/2002:
464        Here's the patch.  HOL now compiles.
465    
466        I don't know how this impact on performance (compile
467        time or runtime).  This bug caused the RA (especially on the x86)
468        to go thru the potential spill phase when there are still nodes on the
469        freeze queue.
470    
471    
472    
473    
474    ----------------------------------------------------------------------
475    Name: Matthias Blume
476    Date: 2002/02/13 22:40:00 EST
477    Tag: blume-20020213-fptr-rtti
478    Description:
479    
480    Fixed a bug in ml-nlffigen that was introduced with one of the previous
481    updates.
482    
483    ----------------------------------------------------------------------
484    Name: Matthias Blume
485    Date: 2002/02/13 16:41:00 EST
486    Tag: blume-20020213-cmlpq
487    Description:
488    
489    Added new priority queue export symbols (which have just been added to
490    smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
491    not compile and the installer would choke.)
492    
493    ----------------------------------------------------------------------
494    Name: Matthias Blume
495    Date: 2002/02/13 16:15:00 EST
496    Tag: blume-20020213-various
497    Description:
498    
499    1. More tweaks to ml-nlffigen:
500    
501       - better internal datastructures (resulting in slight speedup)
502       - "-match" option requires exact match
503       - "localized" gensym counters (untagged structs/unions nested within
504         other structs/unions or within typedefs get a fresh counter; their
505         tag will be prefixed by a concatenation of their parents' tags)
506       - bug fixes (related to calculation of transitive closure of types
507         to be included in the output)
508    
509    2. Minor Basis updates:
510    
511       - added implementations for List.collate and Option.app
512    
513    ----------------------------------------------------------------------
514    Name: Matthias Blume
515    Date: 2002/02/11 15:55:00 EST
516    Tag: blume-20020211-gensym
517    Description:
518    
519    Added a "-gensym" option to command line of ml-nlffigen.  This can be
520    used to specify a "stem" -- a string that is inserted in all "gensym'd"
521    names (ML structure names that correspond to unnamed C structs, unions,
522    and enums), so that separate runs of ml-nlffigen do not clash.
523    
524    ----------------------------------------------------------------------
525    Name: Matthias Blume
526    Date: 2002/02/11 12:05:00 EST
527    Tag: blume-20020211-gensml
528    Description:
529    
530    A quick fix for a problem with GenSML (in the pgraph-util library):
531    Make generation of toplevel "local" optional.  (Strictly speaking,
532    signature definitions within "local" are not legal SML.)
533    
534    Other than that: updates to INSTALL and cm/TODO.
535    
536    ----------------------------------------------------------------------
537    Name: Matthias Blume
538    Date: 2002/02/08 15:00:00 EST
539    Tag: blume-20020208-uniquepid
540    Description:
541    
542    0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
543    
544    1. The installer (config/install.sh) has gotten smarter:
545    
546         - Configuration options are a bit easier to specify now
547           (in config/targets).
548         - Bug in recognizing .tar.bz2 files fixed.
549         - Installer automatically resolves dependencies between
550           configuration options (e.g., if you ask for eXene, you will
551           also get cml -- regardless whether you asked for it or not).
552         - Installer can run in "quieter mode" by setting the environment
553           variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
554           "completely silent", though.
555         - Build HashCons library as part of smlnj-lib.
556    
557    2. A new scheme for assigning persistent identifiers to compilation
558       units (and, by extension, to types etc.) has been put into place.
559       This fixes a long-standing bug where types and even dynamic values
560       can get internally confused, thereby compromising type safety
561       (abstraction) and dynamic correctness.  See
562    
563         http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
564    
565       for an example of how things could go wrong until now.
566    
567       The downside of the new scheme is that pids are not quite as
568       persistent as they used to be: CM will generate a fresh pid
569       for every compilation unit that it thinks it sees for the first
570       time.  That means that if you compile starting from a clean, fresh
571       source tree at two different times, you end up with different
572       binaries.
573    
574       Cutoff recompilation, however, has not been compromised because
575       CM keeps pid information in special caches between runs.
576    
577    ----------------------------------------------------------------------
578    Name: Lal George
579    Date: 2002/02/07 15:34:13 EST 2002
580    Tag: <none>
581    Description:
582    
583    Compilers that generate assembly code may produce  global labels
584    whose value is resolved at link time. The various peephole optimization
585    modules did not take this in account.
586    
587    TODO. The Labels.addrOf function should really return an option
588    type so that clients are forced to deal with this issue, rather
589    than an exception being raised.
590    
591    ----------------------------------------------------------------------
592    Name: Lal George
593    Date: 2002/02/06 13:55:02 EST
594    Tag: george-20020206-ra-breakup
595    Description:
596    
597    1. A bug fix from Allen.
598    
599        A typo causes extra fstp %st(0)'s to be generated at compensation
600        edges, which might cause stack underflow traps at runtime.  This
601        occurs in fft where there are extraneous fstps right before the 'into'
602        trap instruction (in this case they are harmless since none of the
603        integers overflow.)
604    
605    2. Pulled out various utility modules that were embedded in the modules
606       of the register allocator. I need these modules for other purposes, but
607       they are not complete enough to put into a library (just yet).
608    ----------------------------------------------------------------------
609    Name: Matthias Blume
610    Date: 2002/01/31 16:05:00 EST
611    Tag: blume-20020131-sparc-ccalls
612    Description:
613    
614    1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
615       of extra stack space by mistake.  Fixed.
616    
617    2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
618    
619    ----------------------------------------------------------------------
620    Name: Allen Leung
621    Date: 2002/01/30
622    Tag: leunga-20020130-nowhere-bug-fix
623    Description:
624    
625       MLRISC bug fixes:
626       1. Fixed a bindings computation bug in the 'nowhere' program generator tool.
627       2. MachineInt.fromString was negating its value.
628    
629    ----------------------------------------------------------------------
630    Name: Matthias Blume
631    Date: 2002/01/29
632    Tag: blume-20020129-INSTALL
633    Description:
634    
635    - Added somewhat detailed installation instructions (file INSTALL).
636    - Fixed curl-detection bug in config/install.sh.
637    - It is now possible to select the URL getter using the URLGETTER
638      environment variable:
639    
640          not set / "unknown"      --> automatic detection (script tries wget,
641                                       curl, and lynx)
642          "wget" / "curl" / "lynx" --> use the specified program (script "knows"
643                                       how to properly invoke them)
644          other                    --> use $URLGETTER directly, it must take
645                                       precisely two command-line arguments
646                                       (source URL and destination file name)
647    
648    ----------------------------------------------------------------------
649    Name: Matthias Blume
650    Date: 2002/01/28
651    Tag: blume-20020128-sparc-ccalls
652    Description:
653    
654    - Fixed problem with calculation of "used" registers in sparc-c-calls.
655    - Make use of the allocParam argument in sparc-c-calls.
656    
657    ----------------------------------------------------------------------
658    Name: Matthias Blume
659    Date: 2002/01/28
660    Tag: blume-20020128-allocParam
661    Description:
662    
663    John Reppy:  Changes c-calls API to accept client-callback for
664    allocating extra stack space.
665    me: Corresponding changes to mlriscGen (using a dummy argument that
666        does not change the current behavior).
667    
668    ----------------------------------------------------------------------
669    Name: Matthias Blume
670    Date: 2002/01/28 12:00:00
671    Tag: Release_110_38
672    Description:
673    
674    This time for real!!!
675    
676    ----------------------------------------------------------------------
677    Name: Matthias Blume
678    Date: 2002/01/28 10:56:00 EST
679    Tag: blume-20020128-retraction
680    Description:
681    
682    0. Retracted earlier 110.38.  (The Release_110_38 tag has been replaced
683       with blume-Release_110_38-retracted.)
684    
685    1. Fixed a problem with incorrect rounding modes in real64.sml.
686       (Thanks to Andrew Mccreight <andrew.mccreight@yale.edu>.)
687    
688    2. A bug in ml-nlffigen related to the handling of unnamed structs, unions,
689       and enums fixed.  The naming of corresponding ML identifiers should
690       now be consistent again.
691    
692    ----------------------------------------------------------------------
693    Name: Allen Leung
694    Date: 2002/01/27
695    Tag: leunga-20020127-nowhere
696    Description:
697    
698       Added a target called nowhere in the configuration scripts.
699       Enabling this will build the MLRISC 'nowhere' tool (for translating
700       programs with where-clauses into legal SML code) during installation.
701    
702    ----------------------------------------------------------------------
703    Name: Matthias Blume
704    Date: 2002/01/25 21:27:00 EST
705    Tag: blume-Release_110_38-retracted
706    Description:
707    
708    Call it a (working) release!  Version is 110.38. Bootfiles are ready.
709    
710    README will be added later.
711    
712    !!! NOTE:  Re-tagged as blume-Release_110_38-retracted. Original tag
713    (Release_110_38) removed.  Reason: Last-minute bug fixes.
714    
715    ----------------------------------------------------------------------
716    Name: Matthias Blume
717    Date: 2002/01/25
718    Tag: blume-20020125-ffi
719    Description:
720    
721    A large number of tweaks and improvements to ml-nlffi-lib and
722    ml-nlffigen:
723    
724       - ML represenation types have been streamlined
725       - getter and setter functions work with concrete values, not abstract
726         ones where possible
727       - ml-nlffigen command line more flexible (see README file there)
728       - some bugs have been fixed (hopefully)
729    
730    ----------------------------------------------------------------------
731    Name: Lal George
732    Date: 2002/01/24
733    Tag: george-20020124-risc-ra-interface
734    Description:
735    
736       There is a dramatic simplification in the interface to the
737       register allocator for RISC architectures as a result of making
738       parallel copy instructions explicit.
739    
740    ----------------------------------------------------------------------
741    Name: Matthias Blume
742    Date: 2002/01/22
743    Tag: blume-20020122-x86-ccalls
744    Description:
745    
746    Bug fix for c-calls on x86 (having to do with how char- and
747    short-arguments are being handled).
748    
749    ----------------------------------------------------------------------
750    Name: Matthias Blume
751    Date: 2002/01/21
752    Tag: blume-20020121-ff
753    Description:
754    
755    Another day of fiddling with the FFI...
756    
757    1. Bug fix/workaround:  CKIT does not complain about negative array
758       dimensions, so ml-nlffigen has to guard itself against this possibility.
759       (Otherwise a negative dimension would send it into an infinite loop.)
760    
761    2. Some of the abstract types (light objects, light pointers, most "base"
762       types) in structure C are now eqtypes.
763    
764    3. Added constructors and test functions for NULL function pointers.
765    
766    ----------------------------------------------------------------------
767    Name: Matthias Blume
768    Date: 2002/01/18
769    Tag: blume-20020118-ready-for-new-release
770    Description:
771    
772    Made config/srcarchiveurl point to a new place.  (Will provide boot
773    files shortly.)
774    
775    Maybe we christen this to be 110.38?
776    
777    ----------------------------------------------------------------------
778    Name: Matthias Blume
779    Date: 2002/01/18
780    Tag: blume-20020118-more-ffifiddle
781    Description:
782    
783    Today's FFI fiddling:
784    
785      - Provided a structure CGetSet with "convenient" versions of C.Get.* and
786        C.Set.* that use concrete (MLRep.*) arguments and results instead
787        of abstract ones.
788    
789      - Provided word-style bit operations etc. for "int" representation
790        types in MLRep.S<Foo>Bitops where <Foo> ranges over Char, Int, Short,
791        and Long.
792    
793    ----------------------------------------------------------------------
794    Name: Matthias Blume
795    Date: 2002/01/18
796    Tag: blume-20020118-use-x86-fp
797    Description:
798    
799    Now that x86-fast-fp seems to be working, I turned it back on again
800    by default.  (Seems to work fine now, even with the FFI.)
801    
802    Other than that, I added some documentation about the FFI to
803    src/ml-nlffigen/README and updated the FFI test examples in
804    src/ml-nlffi-lib/Tests/*.
805    
806    ----------------------------------------------------------------------
807    Name: Allen Leung
808    Date: 2002/01/17
809    Tag: leunga-20020117-x86-fast-fp-call
810    Description:
811    
812       1. Fixed a problem with handling return fp values when x86's fast fp
813          mode is turned on.
814    
815       2. Minor pretty printing fix for cellset.  Print %st(0) as %st(0) instead
816          of %f32.
817    
818       3. Added a constructor INT32lit to the ast of MLRISC tools.
819    
820    ----------------------------------------------------------------------
821    Name: Matthias Blume
822    Date: 2002/01/16
823    Tag: blume-20020116-ffifiddle
824    Description:
825    
826    More fiddling with the FFI interface:
827    
828     - Make constness 'c instead of rw wherever possible.  This eliminates
829       the need for certain explicit coercions.  (However, due to ML's
830       value polymorphism, there will still be many cases where explicit
831       coercions are necessary.  Phantom types are not the whole answer
832       to modeling a subtyping relationship in ML.)
833    
834     - ro/rw coersions for pointers added.  (Avoids the detour through */&.)
835    
836     - "printf" test example added to src/ml-nlffi-lib/Tests.  (Demonstrates
837       clumsy workaround for varargs problem.)
838    
839    ----------------------------------------------------------------------
840    Name: Lal George
841    Date: 2002/01/15
842    Tag: <none>
843    Description:
844    
845    1. Since COPY instructions are no longer native to the architecture,
846       a generic functor can be used to implement the expandCopies function.
847    
848    2. Allowed EXPORT and IMPORT pseudo-op declarations to appear inside a
849       TEXT segment.
850    
851    ----------------------------------------------------------------------
852    Name: Matthias Blume
853    Date: 2002/01/15
854    Tag: blume-20020115-ffiupdates
855    Description:
856    
857    1. Fix for bug resulting in single-precision float values being returned
858       incorrectly from FFI calls.
859    
860    2. Small modifications to C FFI API:
861    
862        - memory-allocation routines return straight objects (no options)
863          and raise an exception in out-of-memory situations
864        - unsafe extensions to cast between function pointers and pointers
865          from/to ints
866        - added structure C_Debug as an alternative to structure C where
867          pointer-dereferencing (|*| and |*!) always check for null-pointers
868        - added open_lib' to DynLinkage;  open_lib' works like open_lib
869          but also takes a (possibly empty) list of existing library handles
870          that the current library depends on
871    
872    ----------------------------------------------------------------------
873    Name: Matthias Blume
874    Date: 2002/01/10
875    Tag: blume-20020110-newffigen
876    Description:
877    
878    1. Updates to portable graph code.
879    
880    2. Major update to ml-nlffigen and ml-nlffi-lib.  Things are much
881       more scalable now so that even huge interfaces such as the one
882       for GTK compile in finite time and space. :-)
883       See src/ml-nlffigen/README for details on what's new.
884    
885    ----------------------------------------------------------------------
886    Name: Lal George
887    Date: 2001/01/09 14:31:35 EST 2002
888    Tag: george-20011206-rm-native-copy
889    Description:
890    
891            Removed the native COPY and FCOPY instructions
892            from all the architectures and replaced it with the
893            explicit COPY instruction from the previous commit.
894    
895            It is now possible to simplify many of the optimizations
896            modules that manipulate copies. This has not been
897            done in this change.
898    
899    ----------------------------------------------------------------------
900    Name: Lal George
901    Date: 2001/12/06 16:50:13 EST 2001
902    Tag: george-20011206-mlrisc-instruction
903    Description:
904    
905    Changed the representation of instructions from being fully abstract
906    to being partially concrete. That is to say:
907    
908      from
909            type instruction
910    
911      to
912            type instr                              (* machine instruction *)
913    
914            datatype instruction =
915                LIVE of {regs: C.cellset, spilled: C.cellset}
916              | KILL of {regs: C.cellset, spilled: C.cellset}
917              | COPYXXX of {k: CB.cellkind, dst: CB.cell list, src: CB.cell list}
918              | ANNOTATION of {i: instruction, a: Annotations.annotation}
919              | INSTR of instr
920    
921    This makes the handling of certain special instructions that appear on
922    all architectures easier and uniform.
923    
924    LIVE and KILL say that a list of registers are live or killed at the
925    program point where they appear. No spill code is generated when an
926    element of the 'regs' field is spilled, but the register is moved to
927    the 'spilled' (which is present, more for debugging than anything else).
928    
929    LIVE replaces the (now deprecated) DEFFREG instruction on the alpha.
930    We used to generate:
931    
932            DEFFREG f1
933            f1 := f2 + f3
934            trapb
935    
936    but now generate:
937    
938            f1 := f2 + f3
939            trapb
940            LIVE {regs=[f1,f2,f3], spilled=[]}
941    
942    Furthermore, the DEFFREG (hack) required that all floating point instruction
943    use all registers mentioned in the instruction. Therefore f1 := f2 + f3,
944    defines f1 and uses [f1,f2,f3]! This hack is no longer required resulting
945    in a cleaner alpha implementation. (Hopefully, intel will not get rid of
946    this architecture).
947    
948    COPYXXX is intended to replace the parallel COPY and FCOPY  available on
949    all the architectures. This will result in further simplification of the
950    register allocator that must be aware of them for coalescing purposes, and
951    will also simplify certain aspects of the machine description that provides
952    callbacks related to parallel copies.
953    
954    ANNOTATION should be obvious, and now INSTR represents the honest to God
955    machine instruction set!
956    
957    The <arch>/instructions/<arch>Instr.sml files define certain utility
958    functions for making porting easier -- essentially converting upper case
959    to lower case. All machine instructions (of type instr) are in upper case,
960    and the lower case form generates an MLRISC instruction. For example on
961    the alpha we have:
962    
963      datatype instr =
964         LDA of {r:cell, b:cell, d:operand}
965       | ...
966    
967      val lda : {r:cell, b:cell, d:operand} -> instruction
968        ...
969    
970    where lda is just (INSTR o LDA), etc.
971    
972    ----------------------------------------------------------------------
973    Name: Matthias Blume
974    Date: 2001/11/22 21:40:00 EST
975    Tag: Release_110_37
976    Description:
977    
978    Release 110.37.  This time for real.
979    
980    ----------------------------------------------------------------------
981    Name: Matthias Blume
982    Date: 2001/11/21 16:35:00 EST
983    Tag: blume-20011121-foot-in-mouth
984    Description:
985    
986    Removed the "Release_110_37" tag because of a serious bug.
987    This will be re-tagged once the bug is fixed.
988    
989    ----------------------------------------------------------------------
990    Name: Matthias Blume
991    Date: 2001/11/21 16:14:00 EST
992    Tag: blume-20011121-forgottenfile
993    Description:
994    
995    Forgot to add a file.  (Just a .tex-file -- part of
996    the CM manual source.)
997    
998    ----------------------------------------------------------------------
999    Name: Matthias Blume
1000    Date: 2001/11/21 16:10:00 EST
1001    Tag: blume-20011121-invalid_110_37
1002    Description:
1003    
1004    Note: I removed the original tag "Release_110_37" from this commit
1005          because we found a serious bug in all non-x86 backends.
1006          - Matthias
1007    
1008    1. Modifications to the SML/NJ code generator and to the runtime system
1009       so that code object name strings are directly inserted into code
1010       objects at code generation time.  The only business the runtime system
1011       has with this is now to read the name strings on occasions.
1012       (The encoding of the name string has also changed somewhat.)
1013    
1014    2. CM now implements a simple "set calculus" for specifying export lists.
1015       In particular, it is now possible to refer to the export lists of
1016       other libraries/groups/sources and form unions as well as differences.
1017       See the latest CM manual for details.
1018    
1019    3. An separate notion of "proxy" libraries has again be eliminated from
1020       CM's model.  (Proxy libraries are now simply a special case of using
1021       the export list calculus.)
1022    
1023    4. Some of the existing libraries now take advantage of the new set
1024       calculus.
1025       (Notice that not all libraries have been converted because some
1026       of the existing .cm-files are supposed to be backward compatible
1027       with 110.0.x.)
1028    
1029    5. Some cleanup in stand-alone programs.  (Don't use "exnMessage" -- use
1030       "General.exnMessage"!  The former relies on a certain hook to be
1031       initialized, and that often does not happen in the stand-alone case.)
1032    
1033    ----------------------------------------------------------------------
1034    Name: Lal George
1035    Date: 2001/11/21  13:56:18 EST
1036    Tag: george-2001121-pseudo-ops
1037    Description:
1038    
1039      Implemented a complete redesign of MLRISC pseudo-ops. Now there
1040      ought to never be any question of incompatabilities with
1041      pseudo-op syntax expected by host assemblers.
1042    
1043      For now, only modules supporting GAS syntax are implemented
1044      but more should follow, such as MASM, and vendor assembler
1045      syntax, e.g. IBM as, Sun as, etc.
1046    
1047    ----------------------------------------------------------------------
1048    Name: Matthias Blume
1049    Date: 2001/11/14 11:52:00 EST
1050    Tag: blume-20011114-srcname
1051    Description:
1052    
1053    1. Routed the name of the current source file to mlriscgen where it
1054       should be directly emitted into the code object.  (This last part
1055       is yet to be done.)
1056    
1057    2. Some cleanup of the pgraph code to make it match the proposal that
1058       I put out the other day.  (The proposal notwithstanding, things are
1059       still in flux here.)
1060    
1061    ----------------------------------------------------------------------
1062    Name: Lal George
1063    Date: 2001/11/14 09:44:04 EST
1064    Tag:
1065    Description:
1066    
1067      Fix for a backpatching bug reported by Allen.
1068    
1069      Because the boundary between short and long span-dependent
1070      instructions is +/- 128, there are an astounding number of
1071      span-dependent instructions whose size is over estimated.
1072    
1073      Allen came up with the idea of letting the size of span
1074      dependent instructions be non-monotonic, for a maxIter
1075      number of times, after which the size must be monotonically
1076      increasing.
1077    
1078      This table shows the number of span-dependent instructions
1079      whose size was over-estimated as a function of maxIter, for the
1080      file Parse/parse/ml.grm.sml:
1081    
1082         maxIter            # of instructions:
1083            10                      687
1084            20                      438
1085            30                      198
1086            40                        0
1087    
1088      In compiling the compiler, there is no significant difference in
1089      compilation speed between maxIter=10 and maxIter=40. Actually,
1090      my measurements showed that maxIter=40 was a tad faster than
1091      maxIter=10! Also 96% of the  files in the compiler reach a fix
1092      point within 13 iterations, so fixing maxIter at 40, while high,
1093      is okay.
1094    
1095    ----------------------------------------------------------------------
1096    Name: Matthias Blume
1097    Date: 2001/10/31 15:25:00 EST
1098    Tag: blume-20011031-pgraph
1099    Description:
1100    
1101    CKIT:
1102    * Changed the "Function" constructor of type Ast.ctype to carry optional
1103      argument identifiers.
1104    * Changed the return type of TypeUtil.getFunction accordingly.
1105    * Type equality ignores the argument names.
1106    * TypeUtil.composite tries to preserve argument names but gives up quickly
1107      if there is a mismatch.
1108    
1109    installation script:
1110    * attempts to use "curl" if available (unless "wget" is available as well)
1111    
1112    CM:
1113    * has an experimental implementation of "portable graphs" which I will
1114      soon propose as an implementation-independent library format
1115    * there are also new libraries $/pgraph.cm and $/pgraph-util.cm
1116    
1117    NLFFI-LIB:
1118    * some cleanup (all cosmetic)
1119    
1120    NLFFIGEN:
1121    * temporarily disabled the mechanism that suppresses ML output for
1122      C definitions whose identifiers start with an underscore character
1123    * generate val bindings for enum constants
1124    * user can request that only one style (light or heavy) is being used;
1125      default is to use both (command-line arguments: -heavy and -light)
1126    * fixed bug in handling of function types involving incomplete pointers
1127    * generate ML entry points that take record arguments (i.e., using
1128      named arguments) for C functions that have a prototype with named
1129      arguments
1130      (see changes to CKIT)
1131    
1132    ----------------------------------------------------------------------
1133    Name: Allen Leung
1134    Date: 2001/10/27 20:34:00 EDT
1135    Tag: leunga-20011027-x86-fast-fp-call
1136    Description:
1137    
1138       Fixed the bug described in blume-20010920-slowfp.
1139    
1140       The fix involves
1141          1. generating FCOPYs in FSTP in ia32-svid
1142          2. marking a CALL with the appropriate annotation
1143    
1144    ----------------------------------------------------------------------
1145    Name: Matthias Blume
1146    Date: 2001/10/16 11:32:00 EDT
1147    Tag: blume-20011016-netbsd
1148    Description:
1149    
1150    Underscore patch from Chris Richards (fixing problem with compiling
1151    runtime system under recent NetBSD).
1152    
1153    ----------------------------------------------------------------------
1154    Name: Allen Leung
1155    Date: 2001/10/12 17:18:32 EDT 2001
1156    Tag: leung-20011012-x86-printflowgraph
1157    Description:
1158    
1159    X86RA now uses a valid (instead of dummy) PrintFlowgraph module.
1160    
1161    ----------------------------------------------------------------------
1162    Name: Lal George
1163    Date: 2001/10/11 23:51:34 EDT
1164    Tag: george-20011011-too-many-instrs
1165    Description:
1166    
1167    The representation of a program point never expected to see more
1168    than 65536 instructions in a basic block!
1169    
1170    ----------------------------------------------------------------------
1171    Name: Lal George
1172    Date: 2001/10/09 09:41:37 EDT
1173    Tag: george-20011008-mlrisc-labels
1174    Description:
1175    
1176    Changed the machine description files to support printing of
1177    local and global labels in assembly code, based on host assembler
1178    conventions.
1179    
1180    ----------------------------------------------------------------------
1181    Name: Matthias Blume
1182    Date: 2001/09/25 15:25:00 EDT
1183    Tag: blume-20010925-exninfo
1184    Description:
1185    
1186    I provided a non-hook implementation of exnName (at the toplevel) and
1187    made the "dummy" implementation of exnMessage (at the toplevel) more
1188    useful: if nothing gets "hooked in", then at least you are going to
1189    see the exception name and a message indicating why you don't see more.
1190    
1191    [For the time being, programs that need exnMessage and want to use
1192    ml-build should either use General.exnMessage (strongly recommended) or
1193    refer to structure General at some other point so that CM sees a
1194    static dependency.]
1195    
1196    [Similar remarks go for "print" and "use":  If you want to use their
1197    functionality in stand-alone programs generated by ml-build, then use
1198    TextIO.output and Backend.Interact.useFile (from $smlnj/compiler.cm).]
1199    
1200    ----------------------------------------------------------------------
1201  Name: Matthias Blume  Name: Matthias Blume
1202  Date: 2001/09/20 17:28:00 EDT  Date: 2001/09/20 17:28:00 EDT
1203  Tag: blume-20010920-slowfp  Tag: blume-20010920-slowfp

Legend:
Removed from v.938  
changed lines
  Added in v.1124

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