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 1063, Wed Feb 13 21:41:53 2002 UTC revision 1184, Fri Mar 29 22:22:22 2002 UTC
# Line 14  Line 14 
14    
15  ----------------------------------------------------------------------  ----------------------------------------------------------------------
16  Name: Matthias Blume  Name: Matthias Blume
17    Date: 2002/03/29 17:22:00
18    Tag: blume-20020329-div
19    Description:
20    
21    Fixed my broken div/mod logic.  Unfortunately, this means that the
22    inline code for div/mod now has one more comparison than before.
23    Fast paths (quotient > 0 or remainder = 0) are not affected, though.
24    The problem was with quotient = 0, because that alone does not tell
25    us which way the rounding went.  One then has to look at whether
26    remainder and divisor have the same sign...  :(
27    
28    Anyway, I replaced the bootfiles with fresh ones...
29    
30    ----------------------------------------------------------------------
31    Name: Matthias Blume
32    Date: 2002/03/29 14:10:00 EST
33    Tag: blume-20020329-inlprims
34    Description:
35    
36    NEW BOOTFILES!!!    Version number bumped to 110.39.3.
37    
38    Primops have changed. This means that the bin/boot-file formats have
39    changed as well.
40    
41    To make sure that there is no confusion, I made a new version.
42    
43    
44    CHANGES:
45    
46    * removed REMT from mltree (remainder should never overflow).
47    
48    * added primops to deal with divisions of all flavors to the frontend
49    
50    * handled these primops all the way through so they map to their respective
51      MLRISC support
52    
53    * used these primops in the implementation of Int, Int32, Word, Word32
54    
55    * removed INLDIV, INLMOD, and INLREM as they are no longer necessary
56    
57    * parameterized INLMIN, INLMAX, and INLABS by a numkind
58    
59    * translate.sml now deals with all flavors of INL{MIN,MAX,ABS}, including
60      floating point
61    
62    * used INL{MIN,MAX,ABS} in the implementation of Int, Int32, Word, Word32,
63      and Real (but Real.abs maps to a separate floating-point-only primop)
64    
65    
66    TODO items:
67    
68    * Hacked Alpha32 instruction selection, disabling the selection of REMx
69      instructions because the machine instruction encoder cannot handle
70      them.  (Hppa, PPC, and Sparc instruction selection did not handle
71      REM in the first place, and REM is supported by the x86 machine coder.)
72    
73    * Handle DIV and MOD with DIV_TO_NEGINF directly in the x86 instruction
74      selection phase.  (The two can be streamlined because the hardware
75      delivers both quotient and remainder at the same time anyway.)
76    
77    * Think about what to do with "valOf(Int32.minInt) div ~1" and friends.
78      (Currently the behavior is inconsistent both across architectures and
79      wrt. the draft Basis spec.)
80    
81    * Word8 should eventually be handled natively, too.
82    
83    * There seems to be one serious bug in mltree-gen.sml.  It appears, though,
84      as if there currently is no execution path that could trigger it in
85      SML/NJ.  (The assumptions underlying functions arith and promotable do not
86      hold for things like multiplication and division.)
87    
88    ----------------------------------------------------------------------
89    Name: Matthias Blume
90    Date: 2002/03/27 16:27:00 EST
91    Tag: blume-20020327-mlrisc-divisions
92    Description:
93    
94    Added support for all four division operations (ML's div, mod, quot,
95    and rem) to MLRISC.  In the course of doing so, I also rationalized
96    the naming (no more annoying switch-around of DIV and QUOT), by
97    parameterizing the operation by div_rounding_mode (which can be either
98    DIV_TO_ZERO or DIV_TO_NEGINF).
99    
100    The generic MLTreeGen functor takes care of compiling all four
101    operations down to only round-to-zero div.
102    
103    Missing pieces:
104    
105      * Doing something smarter than relying on MLTreeGen on architectures
106        like, e.g., the x86 where hardware division delivers both quotient and
107        remainder at the same time.  With this, the implementation of the
108        round-to-neginf operations could be further streamlined.
109    
110      * Remove inlining support for div/mod/rem from the frontend and replace it
111        with primops that get carried through to the backend.  Do this for all
112        int and word types.
113    
114    ----------------------------------------------------------------------
115    Name: Matthias Blume
116    Date: 2002/03/25 17:25:00 EST
117    Tag: blume-20020325-divmod
118    Description:
119    
120    I improved (hopefully without breaking them) the implementation of Int.div,
121    Int.mod, and Int.rem.   For this, the code in translate.sml now takes
122    advantage of the following observations:
123    
124      Let  q = x quot y      r = x rem y
125           d = x div  y      m = x mod y
126    
127    where "quot" is the round-to-zero version of integer division that
128    hardware usually provides.  Then we have:
129    
130         r = x - q * y        where neither the * nor the - will overflow
131         d = if q >= 0 orelse x = q * y then q else q - 1
132                              where neither the * nor the - will overflow
133         m = if q >= 0 orelse r = 0 then r else r + y
134                              where the + will not overflow
135    
136    This results in substantial simplification of the generated code.
137    The following table shows the number of CFG nodes and edges generated
138    for
139            fun f (x, y) = x OPER y
140            (* with OPER \in div, mod, quot, rem *)
141    
142    
143        OPER | nodes(old) | edges(old) | nodes(new) | edges(new)
144        --------------------------------------------------------
145         div |         24 |         39 |         12 |         16
146         mod |         41 |         71 |         12 |         16
147        quot |          8 |         10 |          8 |         10
148         rem |         10 |         14 |          8 |         10
149    
150    
151    ----------------------------------------------------------------------
152    Name: Matthias Blume
153    Date: 2002/03/25 22:06:00 EST
154    Tag: blume-20020325-cprotobug
155    Description:
156    
157    Fixed a bug in cproto (c prototype decoder).
158    
159    ----------------------------------------------------------------------
160    Name: Matthias Blume
161    Date: 2002/03/25 16:00:00 EST
162    Tag: blume-20020325-raw-primops
163    Description:
164    
165    I did some cleanup to Allen's new primop code and
166    replaced yesterday's bootfiles with new ones.
167    (But they are stored in the same place.)
168    
169    ----------------------------------------------------------------------
170    Name: Matthias Blume
171    Date: 2002/03/24 22:40:00 EST
172    Tag: blume-20020324-bootfiles
173    Description:
174    
175    Made the bootfiles that Allen asked for.
176    
177    ----------------------------------------------------------------------
178    Name: Allen Leung
179    Date: 2002/03/23 15:50:00 EST
180    Tag: leunga-20020323-flint-cps-rcc-primops
181    Description:
182    
183      1. Changes to FLINT primops:
184    
185        (* make a call to a C-function;
186         * The primop carries C function prototype information and specifies
187         * which of its (ML-) arguments are floating point. C prototype
188         * information is for use by the backend, ML information is for
189         * use by the CPS converter. *)
190      | RAW_CCALL of { c_proto: CTypes.c_proto,
191                       ml_args: ccall_type list,
192                       ml_res_opt: ccall_type option,
193                       reentrant : bool
194                     } option
195       (* Allocate uninitialized storage on the heap.
196        * The record is meant to hold short-lived C objects, i.e., they
197        * are not ML pointers.  With the tag, the representation is
198        * the same as RECORD with tag tag_raw32 (sz=4), or tag_fblock (sz=8)
199        *)
200      | RAW_RECORD of {tag:bool,sz:int}
201      and ccall_type = CCALL_INT32 | CCALL_REAL64 | CCALL_ML_PTR
202    
203      2.  These CPS primops are now overloaded:
204    
205           rawload of {kind:numkind}
206           rawstore of {kind:numkind}
207    
208          The one argument form is:
209    
210             rawload {kind} address
211    
212          The two argument form is:
213    
214             rawload {kind} [ml object, byte-offset]
215    
216      3. RAW_CCALL/RCC now takes two extra arguments:
217    
218         a. The first is whether the C call is reentrant, i.e., whether
219            ML state should be saved and restored.
220         b. The second argument is a string argument specifying the name of
221            library and the C function.
222    
223         These things are currently not handled in the code generator, yet.
224    
225      4. In CProto,
226    
227         An encoding type of "bool" means "ml object" and is mapped into
228         C prototype of PTR.  Note that "bool" is different than "string",
229         even though "string" is also mapped into PTR, because "bool"
230         is assigned an CPS type of BOGt, while "string" is assigned INT32t.
231    
232      5. Pickler/unpicker
233    
234         Changed to handle RAW_RECORD and newest RAW_CCALL
235    
236      6. MLRiscGen,
237    
238         1. Changed to handle the new rawload/rawstore/rawrecord operators.
239         2. Code for handling C Calls has been moved to a new module CPSCCalls,
240            in the file CodeGen/cpscompile/cps-c-calls.sml
241    
242      7. Added the conditional move operator
243    
244             condmove of branch
245    
246         to cps.  Generation of this is still buggy so it is currently
247         disabled.
248    
249    ----------------------------------------------------------------------
250    Name: Lal George
251    Date: 2002/03/22 14:18:25 EST
252    Tag: george-20020322-cps-branch-prob
253    Description:
254    
255    Implemented the Ball-Larus branch prediction-heuristics, and
256    incorporated graphical viewers for control flow graphs.
257    
258    Ball-Larus Heuristics:
259    ---------------------
260    See the file compiler/CodeGen/cpscompile/cpsBranchProb.sml.
261    
262    By design it uses the Dempster-Shafer theory for combining
263    probabilities.  For example, in the function:
264    
265        fun f(n,acc) = if n = 0 then acc else f(n-1, n*acc)
266    
267    the ball-larus heuristics predicts that the n=0 is unlikely
268    (OH-heuristic), and the 'then' branch is unlikely because of the
269    RH-heuristic -- giving the 'then' branch an even lower combined
270    probability using the Dempster-Shafer theory.
271    
272    Finally, John Reppy's loop analysis in MLRISC, further lowers the
273    probability of the 'then' branch because of the loop in the else
274    branch.
275    
276    
277    Graphical Viewing:
278    ------------------
279    I merely plugged in Allen's graphical viewers into the compiler. The
280    additional code is not much. At the top level, saying:
281    
282            Control.MLRISC.getFlag "cfg-graphical-view" := true;
283    
284    will display the graphical view of the control flow graph just before
285    back-patching.  daVinci must be in your path for this to work. If
286    daVinci is not available, then the default viewer can be changed
287    using:
288    
289            Control.MLRISC.getString "viewer"
290    
291    which can be set to "dot" or "vcg" for the corresponding viewers. Of
292    course, these viewers must be in your path.
293    
294    The above will display the compilation unit at the level of clusters,
295    many of which are small, boring, and un-interesting. Also setting:
296    
297            Control.MLRISC.getInt "cfg-graphical-view_size"
298    
299    will display clusters that are larger than the value set by the above.
300    
301    
302    ----------------------------------------------------------------------
303    Name: Matthias Blume
304    Date: 2002/03/21 22:20:00 EST
305    Tag: blume-20020321-kmp-bugfix
306    Description:
307    
308    Changed the interface to the KMP routine in PreString and fixed
309    a minor bug in one place where it was used.
310    
311    ----------------------------------------------------------------------
312    Name: Allen Leung
313    Date: 2002/03/21 20:30:00 EST
314    Tag: leunga-20020321-cfg
315    Description:
316    
317      Fixed a potential problem in cfg edge splitting.
318    
319    ----------------------------------------------------------------------
320    Name: Allen Leung
321    Date: 2002/03/21 17:15:00 EST
322    Tag: leunga-20020321-x86-fp-cfg
323    Description:
324    
325      1. Recoded the buggy parts of x86-fp.
326    
327         a. All the block reordering code has been removed.
328            We now depend on the block placement phases to do this work.
329    
330         b. Critical edge splitting code has been simplified and moved into the
331            CFG modules, as where they belong.
332    
333         Both of these were quite buggy and complex.  The code is now much, much
334         simpler.
335    
336      2. X86 backend.
337    
338         a. Added instructions for 64-bit support.  Instruction selection for
339            64-bit has not been committed, however, since that
340            requires changes to MLTREE which haven't been approved by
341            Lal and John.
342    
343         b. Added support for FUCOMI and FUCOMIP when generating code for
344            PentiumPro and above.  We only generate these instructions in
345            the fast-fp mode.
346    
347         c. Added cases for JP and JNP in X86FreqProps.
348    
349      3. CFG
350    
351         CFG now has a bunch of methods for edge splitting and merging.
352    
353      4. Machine description.
354    
355         John's simplification of MLTREE_BASIS.fcond broke a few machine
356         description things:
357    
358         rtl-build.{sig,sml} and hppa.mdl fixed.
359    
360         NOTE: the machine description stuff in the repository is still broken.
361               Again, I can't put my fixes in because that involves
362               changes to MLTREE.
363    
364    ----------------------------------------------------------------------
365    Name: Matthias Blume
366    Date: 2002/03/20 15:55:00 EST
367    Tag: blume-20020320-kmp
368    Description:
369    
370    Implemented Knuth-Morris-Pratt string matching in PreString and used
371    it for String.isSubstring, Substring.isSubstring, and
372    Substring.position.
373    
374    (Might need some stress-testing.  Simple examples worked fine.)
375    
376    ----------------------------------------------------------------------
377    Name: Matthias Blume
378    Date: 2002/03/19 16:37:00 EST
379    Tag: blume-20020319-witnesses
380    Description:
381    
382    Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
383    
384    This implements a generic mechanism for changing constness qualifiers
385    anywhere within big C types without resorting to outright "casts".
386    (So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
387    modify the constness at the outermost level.)
388    The implementation of "convert" is based on the idea of "witness"
389    values -- values that are not used by the operation but whose types
390    "testify" to their applicability.  On the implementation side, "convert"
391    is simply a projection (returning its second curried argument).  With
392    cross-module inlining, it should not result in any machine code being
393    generated.
394    
395    ----------------------------------------------------------------------
396    Name: Matthias Blume
397    Date: 2002/03/15 16:40:00 EST
398    Tag: blume-20020315-basis
399    Description:
400    
401    Provided (preliminary?) implementations for
402    
403      {String,Substring}.{concatWith,isSuffix,isSubstring}
404    
405    and
406    
407      Substring.full
408    
409    Those are in the Basis spec but they were missing in SML/NJ.
410    
411    ----------------------------------------------------------------------
412    Name: Matthias Blume
413    Date: 2002/03/14 21:30:00 EST
414    Tag: blume-20020314-controls
415    Description:
416    
417    Controls:
418    ---------
419    
420    1. Factored out the recently-added Controls : CONTROLS stuff and put
421       it into its own library $/controls-lib.cm.  The source tree for
422       this is under src/smlnj-lib/Controls.
423    
424    2. Changed the names of types and functions in this interface, so they
425       make a bit more "sense":
426    
427          module -> registry
428          'a registry -> 'a group
429    
430    3. The interface now deals in ref cells only.  The getter/setter interface
431       is (mostly) gone.
432    
433    4. Added a function that lets one register an already-existing ref cell.
434    
435    5. Made the corresponding modifications to the rest of the code so that
436       everything compiles again.
437    
438    6. Changed the implementation of Controls.MLRISC back to something closer
439       to the original.  In particular, this module (and therefore MLRISC)
440       does not depend on Controls.  There now is some link-time code in
441       int-sys.sml that registers the MLRISC controls with the Controls
442       module.
443    
444    CM:
445    ---
446    
447      * One can now specify the lambda-split aggressiveness in init.cmi.
448    
449    ----------------------------------------------------------------------
450    Name: Allen Leung
451    Date: 2002/03/13 17:30:00 EST
452    Tag: leunga-20020313-x86-fp-unary
453    Description:
454    
455    Bug fix for:
456    
457    > leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
458    > Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
459    > - fun f(x,(y,z)) = Real.~ y;
460    > [autoloading]
461    > [autoloading done]
462    >       fchsl   (%eax), 184(%esp)
463    > Error: MLRisc bug: X86MCEmitter.emitInstr
464    >
465    > uncaught exception Error
466    >   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
467    
468    The problem was that the code generator did not generate any fp registers
469    in this case, and the ra didn't know that it needed to run the X86FP phase to
470    translate the pseudo fp instruction.   This only happened with unary fp
471    operators in certain situations.
472    
473    ----------------------------------------------------------------------
474    Name: Matthias Blume
475    Date: 2002/03/13 14:00:00 EST
476    Tag: blume-20020313-overload-etc
477    Description:
478    
479    1. Added _overload as a synonym for overload for backward compatibility.
480       (Control.overloadKW must be true for either version to be accepted.)
481    
482    2. Fixed bug in install script that caused more things to be installed
483       than what was requested in config/targets.
484    
485    3. Made CM aware of the (_)overload construct so that autoloading
486       works.
487    
488    ----------------------------------------------------------------------
489    Name: Matthias Blume
490    Date: 2002/03/12 22:03:00 EST
491    Tag: blume-20020312-url
492    Description:
493    
494    Forgot to update BOOT and srcarchiveurl.
495    
496    ----------------------------------------------------------------------
497    Name: Matthias Blume
498    Date: 2002/03/12 17:30:00 EST
499    Tag: blume-20020312-version110392
500    Description:
501    
502    Yet another version number bump (because of small changes to the
503    binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
504    
505    Changes:
506    
507      The new pid generation scheme described a few weeks ago was overly
508      complicated.  I implemented a new mechanism that is simpler and
509      provides a bit more "stability":  Once CM has seen a compilation
510      unit, it keeps its identity constant (as long as you do not delete
511      those crucial CM/GUID/* files).  This means that when you change
512      an interface, compile, then go back to the old interface, and
513      compile again, you arrive at the original pid.
514    
515      There now also is a mechanism that instructs CM to use the plain
516      environment hash as a module's pid (effectively making its GUID
517      the empty string).  For this, "noguid" must be specified as an
518      option to the .sml file in question within its .cm file.
519      This is most useful for code that is being generated by tools such
520      as ml-nlffigen (because during development programmers tend to
521      erase the tool's entire output directory tree including CM's cached
522      GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
523      revert to the old, broken behavior of SML/NJ, but in specific cases
524      where there is no danger of interface confusion, its use is ok
525      (I think).
526    
527      ml-nlffigen by default generates "noguid" annotations.  They can be
528      turned off by specifying -guid in its command line.
529    
530    ----------------------------------------------------------------------
531    Name: Lal George
532    Date: 2002/03/12 12 14:42:36 EST
533    Tag: george-20020312-frequency-computation
534    Description:
535    
536    Integrated jump chaining and static block frequency into the
537    compiler. More details and numbers later.
538    
539    ----------------------------------------------------------------------
540    Name: Lal George
541    Date: 2002/03/11 11 22:38:53 EST
542    Tag: george-20020311-jump-chain-elim
543    Description:
544    
545    Tested the jump chain elimination on all architectures (except the
546    hppa).  This is on by default right now and is profitable for the
547    alpha and x86, however, it may not be profitable for the sparc and ppc
548    when compiling the compiler.
549    
550    The gc test will typically jump to a label at the end of the cluster,
551    where there is another jump to an external cluster containing the actual
552    code to invoke gc. This is to allow factoring of common gc invocation
553    sequences. That is to say, we generate:
554    
555            f:
556               testgc
557               ja   L1      % jump if above to L1
558    
559            L1:
560               jmp L2
561    
562    
563    After jump chain elimination the 'ja L1' instructions is converted to
564    'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
565    up being implemented in their long form (if L2 is far away) using:
566    
567            jbe     L3      % jump if below or equal to L3
568            jmp     L2
569         L3:
570            ...
571    
572    
573    For large compilation units L2  may be far away.
574    
575    
576    ----------------------------------------------------------------------
577    Name: Matthias Blume
578    Date: 2002/03/11 13:30:00 EST
579    Tag: blume-20020311-mltreeeval
580    Description:
581    
582    A functor parameter was missing.
583    
584    ----------------------------------------------------------------------
585    Name: Allen Leung
586    Date: 2002/03/11 10:30:00 EST
587    Tag: leunga-20020311-runtime-string0
588    Description:
589    
590       The representation of the empty string now points to a
591    legal null terminated C string instead of unit.  It is now possible
592    to convert an ML string into C string with InlineT.CharVector.getData.
593    This compiles into one single machine instruction.
594    
595    ----------------------------------------------------------------------
596    Name: Allen Leung
597    Date: 2002/03/10 23:55:00 EST
598    Tag: leunga-20020310-x86-call
599    Description:
600    
601       Added machine generation for CALL instruction (relative displacement mode)
602    
603    ----------------------------------------------------------------------
604    Name: Matthias Blume
605    Date: 2002/03/08 16:05:00
606    Tag: blume-20020308-entrypoints
607    Description:
608    
609    Version number bumped to 110.39.1.  NEW BOOTFILES!
610    
611    Entrypoints: non-zero offset into a code object where execution should begin.
612    
613    - Added the notion of an entrypoint to CodeObj.
614    - Added reading/writing of entrypoint info to Binfile.
615    - Made runtime system bootloader aware of entrypoints.
616    - Use the address of the label of the first function given to mlriscGen
617      as the entrypoint.  This address is currently always 0, but it will
618      not be 0 once we turn on block placement.
619    - Removed the linkage cluster code (which was The Other Way(tm) of dealing
620      with entry points) from mlriscGen.
621    
622    ----------------------------------------------------------------------
623    Name: Allen Leung
624    Date: 2002/03/07 20:45:00 EST
625    Tag: leunga-20020307-x86-cmov
626    Description:
627    
628       Bug fixes for CMOVcc on x86.
629    
630       1. Added machine code generation for CMOVcc
631       2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
632       3. CMOVcc cannot have an immediate operand as argument.
633    
634    ----------------------------------------------------------------------
635    Name: Matthias Blume
636    Date: 2002/03/07 16:15:00 EST
637    Tag: blume-20020307-controls
638    Description:
639    
640    This is a very large but mostly boring patch which makes (almost)
641    every tuneable compiler knob (i.e., pretty much everything under
642    Control.* plus a few other things) configurable via both the command
643    line and environment variables in the style CM did its configuration
644    until now.
645    
646    Try starting sml with '-h' (or, if you are brave, '-H')
647    
648    To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
649    implements the underlying generic mechanism.
650    
651    The interface to some of the existing such facilities has changed somewhat.
652    For example, the MLRiscControl module now provides mkFoo instead of getFoo.
653    (The getFoo interface is still there for backward-compatibility, but its
654    use is deprecated.)
655    
656    The ml-build script passes -Cxxx=yyy command-line arguments through so
657    that one can now twiddle the compiler settings when using this "batch"
658    compiler.
659    
660    TODO items:
661    
662    We should go through and throw out all controls that are no longer
663    connected to anything.  Moreover, we should go through and provide
664    meaningful (and correct!) documentation strings for those controls
665    that still are connected.
666    
667    Currently, multiple calls to Controls.new are accepted (only the first
668    has any effect).  Eventually we should make sure that every control
669    is being made (via Controls.new) exactly once.  Future access can then
670    be done using Controls.acc.
671    
672    Finally, it would probably be a good idea to use the getter-setter
673    interface to controls rather than ref cells.  For the time being, both
674    styles are provided by the Controls module, but getter-setter pairs are
675    better if thread-safety is of any concern because they can be wrapped.
676    
677    *****************************************
678    
679    One bug fix: The function blockPlacement in three of the MLRISC
680    backpatch files used to be hard-wired to one of two possibilities at
681    link time (according to the value of the placementFlag).  But (I
682    think) it should rather sense the flag every time.
683    
684    *****************************************
685    
686    Other assorted changes (by other people who did not supply a HISTORY entry):
687    
688    1. the cross-module inliner now works much better (Monnier)
689    2. representation of weights, frequencies, and probabilities in MLRISC
690       changed in preparation of using those for weighted block placement
691       (Reppy, George)
692    
693    ----------------------------------------------------------------------
694    Name: Lal George
695    Date: 2002/03/07 14:44:24 EST 2002
696    Tag: george-20020307-weighted-block-placement
697    
698    Tested the weighted block placement optimization on all architectures
699    (except the hppa) using AMPL to generate the block and edge frequencies.
700    Changes were required in the machine properties to correctly
701    categorize trap instructions. There is an MLRISC flag
702    "weighted-block-placement" that can be used to enable weighted block
703    placement, but this will be ineffective without block/edge
704    frequencies (coming soon).
705    
706    
707    ----------------------------------------------------------------------
708    Name: Lal George
709    Date: 2002/03/05 17:24:48 EST
710    Tag: george-20020305-linkage-cluster
711    
712    In order to support the block placement optimization, a new cluster
713    is generated as the very first cluster (called the linkage cluster).
714    It contains a single jump to the 'real' entry point for the compilation
715    unit. Block placement has no effect on the linkage cluster itself, but
716    all the other clusters  have full freedom in the manner in which they
717    reorder blocks or functions.
718    
719    On the x86 the typical linkage code that is generated is:
720       ----------------------
721            .align 2
722       L0:
723            addl    $L1-L0, 72(%esp)
724            jmp     L1
725    
726    
727            .align  2
728       L1:
729       ----------------------
730    
731    72(%esp) is the memory location for the stdlink register. This
732    must contain the address of the CPS function being called. In the
733    above example, it contains the address of  L0; before
734    calling L1 (the real entry point for the compilation unit), it
735    must contain the address for L1, and hence
736    
737            addl $L1-L0, 72(%esp)
738    
739    I have tested this on all architectures except the hppa.The increase
740    in code size is of course negligible
741    
742    ----------------------------------------------------------------------
743    Name: Allen Leung
744    Date: 2002/03/03 13:20:00 EST
745    Tag: leunga-20020303-mlrisc-tools
746    
747      Added #[ ... ] expressions to mlrisc tools
748    
749    ----------------------------------------------------------------------
750    Name: Matthias Blume
751    Date: 2002/02/27 12:29:00 EST
752    Tag: blume-20020227-cdebug
753    Description:
754    
755    - made types in structure C and C_Debug to be equal
756    - got rid of code duplication (c-int.sml vs. c-int-debug.sml)
757    - there no longer is a C_Int_Debug (C_Debug is directly derived from C)
758    
759    ----------------------------------------------------------------------
760    Name: Matthias Blume
761    Date: 2002/02/26 12:00:00 EST
762    Tag: blume-20020226-ffi
763    Description:
764    
765    1. Fixed a minor bug in CM's "noweb" tool:
766       If numbering is turned off, then truly don't number (i.e., do not
767       supply the -L option to noweb).  The previous behavior was to supply
768       -L'' -- which caused noweb to use the "default" line numbering scheme.
769       Thanks to Chris Richards for pointing this out (and supplying the fix).
770    
771    2. Once again, I reworked some aspects of the FFI:
772    
773       A. The incomplete/complete type business:
774    
775       - Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
776         gone!
777       - ML types representing an incomplete type are now *equal* to
778         ML types representing their corresponding complete types (just like
779         in C).  This is still safe because ml-nlffigen will not generate
780         RTTI for incomplete types, nor will it generate functions that
781         require access to such RTTI.   But when ML code generated from both
782         incomplete and complete versions of the C type meet, the ML types
783         are trivially interoperable.
784    
785         NOTE:  These changes restore the full generality of the translation
786         (which was previously lost when I eliminated functorization)!
787    
788       B. Enum types:
789    
790       - Structure C now has a type constructor "enum" that is similar to
791         how the "su" constructor works.  However, "enum" is not a phantom
792         type because each "T enum" has values (and is isomorphic to
793         MLRep.Signed.int).
794       - There are generic access operations for enum objects (using
795         MLRep.Signed.int).
796       - ml-nlffigen will generate a structure E_foo for each "enum foo".
797         * The structure contains the definition of type "mlrep" (the ML-side
798         representation type of the enum).  Normally, mlrep is the same
799         as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
800         then mlrep will be defined as a datatype -- thus facilitating
801         pattern matching on mlrep values.
802         ("-ec" will be suppressed if there are duplicate values in an
803          enumeration.)
804         * Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
805         will be generated for each C enum constant xxx.
806         * Conversion functions m2i and i2m convert between mlrep and
807         MLRep.Signed.int.  (Without "-ec", these functions are identities.)
808         * Coversion functions c and ml convert between mlrep and "tag enum".
809         * Access functions (get/set) fetch and store mlrep values.
810       - By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
811         enumerations are merged into one single enumeration represented by
812         structure E_'.
813    
814    ----------------------------------------------------------------------
815    Name: Allen Leung
816    Date: 2002/02/25 04:45:00 EST
817    Tag: leunga-20020225-cps-spill
818    
819    This is a new implementation of the CPS spill phase.
820    The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
821    In case of problems, replace it with the old file spill.sml
822    
823    The current compiler runs into some serious performance problems when
824    constructing a large record.  This can happen when we try to compile a
825    structure with many items.  Even a very simple structure like the following
826    makes the compiler slow down.
827    
828        structure Foo = struct
829           val x_1 = 0w1 : Word32.int
830           val x_2 = 0w2 : Word32.int
831           val x_3 = 0w3 : Word32.int
832           ...
833           val x_N = 0wN : Word32.int
834        end
835    
836    The following table shows the compile time, from N=1000 to N=4000,
837    with the old compiler:
838    
839    N
840    1000   CPS 100 spill                           0.04u  0.00s  0.00g
841           MLRISC ra                               0.06u  0.00s  0.05g
842              (spills = 0 reloads = 0)
843           TOTAL                                   0.63u  0.07s  0.21g
844    
845    1100   CPS 100 spill                           8.25u  0.32s  0.64g
846           MLRISC ra                               5.68u  0.59s  3.93g
847              (spills = 0 reloads = 0)
848           TOTAL                                   14.71u  0.99s  4.81g
849    
850    1500   CPS 100 spill                           58.55u  2.34s  1.74g
851           MLRISC ra                               5.54u  0.65s  3.91g
852              (spills = 543 reloads = 1082)
853           TOTAL                                   65.40u  3.13s  6.00g
854    
855    2000   CPS 100 spill                           126.69u  4.84s  3.08g
856           MLRISC ra                               0.80u  0.10s  0.55g
857              (spills = 42 reloads = 84)
858           TOTAL                                   129.42u  5.10s  4.13g
859    
860    3000   CPS 100 spill                           675.59u  19.03s  11.64g
861           MLRISC ra                               2.69u  0.27s  1.38g
862              (spills = 62 reloads = 124)
863           TOTAL                                   682.48u  19.61s  13.99g
864    
865    4000   CPS 100 spill                           2362.82u  56.28s  43.60g
866           MLRISC ra                               4.96u  0.27s  2.72g
867              (spills = 85 reloads = 170)
868           TOTAL                                   2375.26u  57.21s  48.00g
869    
870    As you can see the old cps spill module suffers from some serious
871    performance problem.  But since I cannot decipher the old code fully,
872    instead of patching the problems up, I'm reimplementing it
873    with a different algorithm.  The new code is more modular,
874    smaller when compiled, and substantially faster
875    (O(n log n) time and O(n) space).  Timing of the new spill module:
876    
877    4000  CPS 100 spill                           0.02u  0.00s  0.00g
878          MLRISC ra                               0.25u  0.02s  0.15g
879             (spills=1 reloads=3)
880          TOTAL                                   7.74u  0.34s  1.62g
881    
882    Implementation details:
883    
884    As far as I can tell, the purpose of the CPS spill module is to make sure the
885    number of live variables at any program point (the bandwidth)
886    does not exceed a certain limit, which is determined by the
887    size of the spill area.
888    
889    When the bandwidth is too large, we decrease the register pressure by
890    packing live variables into spill records.  How we achieve this is
891    completely different than what we did in the old code.
892    
893    First, there is something about the MLRiscGen code generator
894    that we should be aware of:
895    
896    o MLRiscGen performs code motion!
897    
898       In particular, it will move floating point computations and
899       address computations involving only the heap pointer to
900       their use sites (if there is only a single use).
901       What this means is that if we have a CPS record construction
902       statement
903    
904           RECORD(k,vl,w,e)
905    
906       we should never count the new record address w as live if w
907       has only one use (which is often the case).
908    
909       We should do something similar to floating point, but the transformation
910       there is much more complex, so I won't deal with that.
911    
912    Secondly, there are now two new cps primops at our disposal:
913    
914     1. rawrecord of record_kind option
915        This pure operator allocates some uninitialized storage from the heap.
916        There are two forms:
917    
918         rawrecord NONE [INT n]  allocates a tagless record of length n
919         rawrecord (SOME rk) [INT n] allocates a tagged record of length n
920                                     and initializes the tag.
921    
922     2. rawupdate of cty
923          rawupdate cty (v,i,x)
924          Assigns to x to the ith component of record v.
925          The storelist is not updated.
926    
927    We use these new primops for both spilling and increment record construction.
928    
929     1. Spilling.
930    
931        This is implemented with a linear scan algorithm (but generalized
932        to trees).  The algorithm will create a single spill record at the
933        beginning of the cps function and use rawupdate to spill to it,
934        and SELECT or SELp to reload from it.  So both spills and reloads
935        are fine-grain operations.  In contrast, in the old algorithm
936        "spills" have to be bundled together in records.
937    
938        Ideally, we should sink the spill record construction to where
939        it is needed.  We can even split the spill record into multiple ones
940        at the places where they are needed.  But CPS is not a good
941        representation for global code motion, so I'll keep it simple and
942        am not attempting this.
943    
944     2. Incremental record construction (aka record splitting).
945    
946        Long records with many component values which are simulatenously live
947        (recall that single use record addresses are not considered to
948         be live) are constructed with rawrecord and rawupdate.
949        We allocate space on the heap with rawrecord first, then gradually
950        fill it in with rawupdate.  This is the technique suggested to me
951        by Matthias.
952    
953        Some restrictions on when this is applicable:
954        1. It is not a VECTOR record.  The code generator currently does not handle
955           this case. VECTOR record uses double indirection like arrays.
956        2. All the record component values are defined in the same "basic block"
957           as the record constructor.  This is to prevent speculative
958           record construction.
959    
960    ----------------------------------------------------------------------
961    Name: Allen Leung
962    Date: 2002/02/22 01:02:00 EST
963    Tag: leunga-20020222-mlrisc-tools
964    
965    Minor bug fixes in the parser and rewriter
966    
967    ----------------------------------------------------------------------
968    Name: Allen Leung
969    Date: 2002/02/21 20:20:00 EST
970    Tag: leunga-20020221-peephole
971    
972    Regenerated the peephole files.  Some contained typos in the specification
973    and some didn't compile because of pretty printing bugs in the old version
974    of 'nowhere'.
975    
976    ----------------------------------------------------------------------
977    Name: Allen Leung
978    Date: 2002/02/19 20:20:00 EST
979    Tag: leunga-20020219-mlrisc-tools
980    Description:
981    
982       Minor bug fixes to the mlrisc-tools library:
983    
984       1.  Fixed up parsing colon suffixed keywords
985       2.  Added the ability to shut the error messages up
986       3.  Reimplemented the pretty printer and fixed up/improved
987           the pretty printing of handle and -> types.
988       4.  Fixed up generation of literal symbols in the nowhere tool.
989       5.  Added some SML keywords to to sml.sty
990    
991    ----------------------------------------------------------------------
992    Name: Matthias Blume
993    Date: 2002/02/19 16:20:00 EST
994    Tag: blume-20020219-cmffi
995    Description:
996    
997    A wild mix of changes, some minor, some major:
998    
999    * All C FFI-related libraries are now anchored under $c:
1000        $/c.cm      --> $c/c.cm
1001        $/c-int.cm  --> $c/internals/c-int.cm
1002        $/memory.cm --> $c/memory/memory.cm
1003    
1004    * "make" tool (in CM) now treats its argument pathname slightly
1005      differently:
1006        1. If the native expansion is an absolute name, then before invoking
1007           the "make" command on it, CM will apply OS.Path.mkRelative
1008           (with relativeTo = OS.FileSys.getDir()) to it.
1009        2. The argument will be passed through to subsequent phases of CM
1010           processing without "going native".  In particular, if the argument
1011           was an anchored path, then "make" will not lose track of that anchor.
1012    
1013    * Compiler backends now "know" their respective C calling conventions
1014      instead of having to be told about it by ml-nlffigen.  This relieves
1015      ml-nlffigen from one of its burdens.
1016    
1017    * The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
1018    
1019    * Export C_DEBUG and C_Debug from $c/c.cm.
1020    
1021    * C type encoding in ml-nlffi-lib has been improved to model the conceptual
1022      subtyping relationship between incomplete pointers and their complete
1023      counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
1024      with the convention of instantiating 'o with ('t, 'c) obj whenever
1025      the pointer target type is complete.  In the incomplete case, 'o
1026      will be instantiated with some "'c iobj" -- a type obtained by
1027      using one of the functors PointerToIncompleteType or PointerToCompleteType.
1028    
1029      Operations that work on both incomplete and complete pointer types are
1030      typed as taking an 'o ptr while operations that require the target to
1031      be known are typed as taking some ('t, 'c) obj ptr.
1032    
1033      voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
1034      where void is an eqtype without any values.  This makes it possible
1035      to work on voidptr values using functions meant to operate on light
1036      incomplete pointers.
1037    
1038    * As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
1039      been vastly simplified.
1040    
1041    ----------------------------------------------------------------------
1042    Name: Matthias Blume
1043    Date: 2002/02/19 10:48:00 EST
1044    Tag: blume-20020219-pqfix
1045    Description:
1046    
1047    Applied Chris Okasaki's bug fix for priority queues.
1048    
1049    ----------------------------------------------------------------------
1050    Name: Matthias Blume
1051    Date: 2002/02/15 17:05:00
1052    Tag: Release_110_39
1053    Description:
1054    
1055    Last-minute retagging is becoming a tradition... :-(
1056    
1057    This is the working release 110.39.
1058    
1059    ----------------------------------------------------------------------
1060    Name: Matthias Blume
1061    Date: 2002/02/15 16:00:00 EST
1062    Tag: Release_110_39-orig
1063    Description:
1064    
1065    Working release 110.39.  New bootfiles.
1066    
1067    (Update: There was a small bug in the installer so it wouldn't work
1068    with all shells.  So I retagged. -Matthias)
1069    
1070    ----------------------------------------------------------------------
1071    Name: Matthias Blume
1072    Date: 2002/02/15 14:17:00 EST
1073    Tag: blume-20020215-showbindings
1074    Description:
1075    
1076    Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
1077    the latter can be useful for exploring what bindings are available at
1078    the interactive prompt.  (The first function returns only the list
1079    of symbols that are really bound, the second prints those but also the
1080    ones that CM's autoloading mechanism knows about.)
1081    
1082    ----------------------------------------------------------------------
1083    Name: Matthias Blume
1084    Date: 2002/02/15 12:08:00 EST
1085    Tag: blume-20020215-iptrs
1086    Description:
1087    
1088    Two improvements to ml-nlffigen:
1089    
1090      1. Write files only if they do not exist or if their current contents
1091         do not coincide with what's being written.  (That is, avoid messing
1092         with the time stamps unless absolutely necessary.)
1093    
1094      2. Implement a "repository" mechanism for generated files related
1095         to "incomplete pointer types".   See the README file for details.
1096    
1097    ----------------------------------------------------------------------
1098    Name: Matthias Blume
1099    Date: 2002/02/14 11:50:00 EST
1100    Tag: blume-20020214-quote
1101    Description:
1102    
1103    Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
1104    because of the new and improved tag generation scheme.  (Thanks to Allen
1105    Leung for pointing it out.)
1106    
1107    ----------------------------------------------------------------------
1108    Name: Lal George
1109    Date: 2002/02/14 09:55:27 EST 2002
1110    Tag: george-20020214-isabelle-bug
1111    Description:
1112    
1113    Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
1114    of Isabelle on the x86.
1115    
1116    From Allen:
1117    -----------
1118     I've found the problem:
1119    
1120         in ra-core.sml, I use the counter "blocked" to keep track of the
1121         true number of elements in the freeze queue.  When the counter goes
1122         to zero, I skip examining the queue.  But I've messed up the
1123         bookkeeping in combine():
1124    
1125             else ();
1126             case !ucol of
1127               PSEUDO => (if !cntv > 0 then
1128                     (if !cntu > 0 then blocked := !blocked - 1 else ();
1129                                        ^^^^^^^^^^^^^^^^^^^^^^^
1130                      moveu := mergeMoveList(!movev, !moveu)
1131                     )
1132                  else ();
1133    
1134         combine() is called to coalesce two nodes u and v.
1135         I think I was thinking that if the move counts of u and v are both
1136         greater than zero then after they are coalesced then one node is
1137         removed from the freeze queue.  Apparently I was thinking that
1138         both u and v are of low degree, but that's clearly not necessarily true.
1139    
1140    
1141    02/12/2002:
1142        Here's the patch.  HOL now compiles.
1143    
1144        I don't know how this impact on performance (compile
1145        time or runtime).  This bug caused the RA (especially on the x86)
1146        to go thru the potential spill phase when there are still nodes on the
1147        freeze queue.
1148    
1149    
1150    
1151    
1152    ----------------------------------------------------------------------
1153    Name: Matthias Blume
1154    Date: 2002/02/13 22:40:00 EST
1155    Tag: blume-20020213-fptr-rtti
1156    Description:
1157    
1158    Fixed a bug in ml-nlffigen that was introduced with one of the previous
1159    updates.
1160    
1161    ----------------------------------------------------------------------
1162    Name: Matthias Blume
1163  Date: 2002/02/13 16:41:00 EST  Date: 2002/02/13 16:41:00 EST
1164  Tag: blume-20020213-cmlpq  Tag: blume-20020213-cmlpq
1165  Description:  Description:

Legend:
Removed from v.1063  
changed lines
  Added in v.1184

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