Diff of /sml/trunk/HISTORY

Parent Directory | Revision Log | Patch

	revision 583, Thu Mar 23 21:52:30 2000 UTC	revision 648, Thu May 25 21:28:21 2000 UTC
#	Line 13	Line 13
13	Description:	Description:
14	----------------------------------------------------------------------	----------------------------------------------------------------------
15	Name: Allen Leung	Name: Allen Leung
16		Date: 2000/05/25 17:28 EDT
17		Tag: leunga-20000525-ra
18		Description:
19
20		Fixed a bug in freezing phase of the register allocator.
21
22		----------------------------------------------------------------------
23		Name: Allen Leung
24		Date: 2000/05/15 22:53 EDT
25		Tag: leunga-20000515-alpha-x86-ra
26		Description:
27
28		1. Alpha
29
30		Slight cleanup.  Removed the instruction SGNXL
31
32		2. X86
33
34		Added the following instructions to the instruction set:
35
36		ROLx, RORx,
37		BTx, BTSx, BTLx, BTRx,
38		XCHGx, and variants with the LOCK prefix
39
40		3. Register Allocation
41
42		The module ra-rewrite-with-renaming has been improved.
43
44		These have no effect on SML/NJ.
45
46		----------------------------------------------------------------------
47		Name: Matthias Blume
48		Date: 2000/05/15 16:20:00 JST
49		Tag: blume-20000515-lightrebuild
50		Description:
51
52		1. I added an alternative to "-rebuild" to "makeml".  The difference is
53		that prior to calling CMB.make' the CM-variable "LIGHT" will be
54		defined.  In effect, the command will not build any cross-compiler
55		backends and therefore finish more quickly.
56
57		The "fixpt" script also takes a "-light" switch to be able to use
58		this new facility while compiling for a fixpoint.
59
60		2. I replaced all mentions of anchored paths in group owner specifications
61		with simple relative paths (usually starting with "..").
62		The rationale is that a library's internal workings should not be
63		compromised by the lack of some anchor.  (An anchor is necessary
64		for someone who wants to refer to the library by an anchored path,
65		but it should not be necessary to build the same library in the first
66		place.)
67
68		3. I changed the way CM's tool mechanism determines the shell command
69		string used for things like ml-yacc etc. so that it does not break
70		when CM.Control.implicit_anchors is turned off.
71
72		----------------------------------------------------------------------
73		Name: Matthias Blume
74		Date: 2000/05/12 18:20:00 JST
75		Tag: blume-20000512-ml-build
76		Description:
77
78		Fixed a bug in config/_ml-build that prevented ml-yacc and ml-lex from
79		getting installed properly (by config/install.sh).
80
81		----------------------------------------------------------------------
82		Name: Matthias Blume
83		Date: 2000/05/12 17:30:00 JST
84		Tag: blume-20000512-anchors
85		Description:
86
87		!!! NEW BOOT FILES !!!
88
89		This change is in preparation of fading out support for "implicitly
90		anchored path names".  I went through all sources and used the
91		explicit (and relatively new) $-notation.  See system/README and the
92		CM manual for more info on this.
93
94		I also modified the anchoring scheme for some things such as "smlnj",
95		"MLRISC", "cm", etc. to take advantage of the fact that explicit
96		anchors are more expressive: anchor name and first arc do not have to
97		coincide.  This entails the following user-visible change:
98
99		You have to write $smlnj/foo/bar instead of smlnj/foo/bar.  In
100		particular, when you fire up sml with a command-line argument, say,
101		e.g.:
102
103		sml '$smlnj/cmb.cm'
104
105		At the ML toplevel prompt:
106
107		CM.autoload "$smlnj/cmb.cm";
108
109		There is also a new controller in CM.Control that can be used to turn
110		off all remaining support for implicit anchors by saying:
111
112		CM.autoload "$smlnj/
113		#set CM.Control.implicit_anchors false;
114
115		This causes CM to reject implicitly anchored paths.  This is (for the
116		time being) less permissive than the "final" version where there will
117		be no more such implicit anchors and relative paths will be just that:
118		relative.
119
120		The next step (version after next version?) will be to make the
121		default for CM.Control.implicit_anchors false.  After the dust has
122		settled, I can then produce the "final" version of this...
123
124		Note: Since bootstrapping is a bit tricky, I provided new boot files.
125
126		----------------------------------------------------------------------
127		Name: Matthias Blume
128		Date: 2000/05/11 16:30:00 JST
129		Tag: blume-20000511-sources
130		Description:
131
132		The main change is that I added function CM.sources as a generalized
133		version of the earlier CM.makedepend.  This entails the following
134		additional changes:
135
136		- CM.makedepend has been dropped.
137
138		- CM manual has been updated.
139
140		- TOOLS signature and API have been changed.
141
142		----------------------------------------------------------------------
143		Name: Allen Leung
144		Date: 2000/05/10 21:17 EDT
145		Tag: leunga-20000510-moby-c--ssa
146		Description:
147
148		Various bug fixes and new features for C--, Moby and MLRISC optimizations.
149		None of these affect SML/NJ.
150
151		1. Register Allocation
152
153		a. A new ra spilling module (ra/ra-spill-with-renaming) is implemented.
154		This module tries to remove local (i.e. basic block level) redundancies
155		during spilling.
156
157		b. A new framework for performing region based register allocation.
158		Not yet entirely functional.
159
160		2. X86
161
162		a. DefUse for POP was missing the stack pointer [found by Lal]
163		b. Reload for CALL was incorrect in X86Spill [found by John]
164		c. Various fixes in X86Spill so that it can be used correctly for
165		the new spilling module.
166
167		3. SSA/IR
168
169		a. New module ir/dj-dataflow.sml implements elimination based
170		data flow analysis.
171
172		4. MLRiscGen
173
174		a. Fix for gc type annotation
175
176		5. MDGen
177
178		Various fixes for machine description -> ml code translation.  For ssa
179		only.
180
181		----------------------------------------------------------------------
182		Name: Allen Leung
183		Date: 2000/05/08 22:17 EDT
184		Tag: leunga-20000508-labexp
185		Description:
186
187		Fermin has found a few assembly problems with constant expressions
188		generated in LabelExp.  Mostly, the problems involve extra parentheses,
189		which choke on dumb assemblers.  This is his fix.
190
191		----------------------------------------------------------------------
192		Name: Dave MacQueen
193		Date: 2000/04/09 14:00 EDT
194		Tag: dbm-20000502-Version_110_28
195		Description:
196
197		1. Updated src/compiler/TopLevel/main/version.sml to version 110.28
198
199		2. Updated config/version to 110.28
200
201		3. Updated config/srcarchiveurl
202
203		3. New boot files!
204		ftp://ftp.research.bell-labs.com/dist/smlnj/working/110.28/
205
206		----------------------------------------------------------------------
207		Name: Matthias Blume
208		Date: 2000/05/01 19:05:00 JST
209		Tag: blume-20000501-noweb
210		Description:
211
212		A new noweb tool has been added.  The existing system is entirely
213		unaffected by this, but some CM users have asked for renewed noweb
214		support.  Everything is documented in the CM manual.
215
216		New (plugin) libraries:
217
218		noweb-tool.cm
219		nw-ext.cm
220
221		----------------------------------------------------------------------
222		Name: Dave MacQueen
223		Date: 2000/04/30 12:40PM EDT
224		Tag: dbm-20000430-bug_fixes
225		Description:
226
227		1. Fix for bug 1498
228		smlnj/src/system/Basis/Implementation/Unsafe/object.sig
229		smlnj/src/system/Basis/Implementation/Unsafe/object.sml
230		added toRealArray function
231		smlnj/src/compiler/MiscUtil/print/ppobj.sml
232		added check for tag Obj.RealArray to array printing case in ppObj
233
234		2. Fix for bug 1510
235		smlnj/src/compiler/Semant/types/typesutil.sml
236		fixed definition of dummyargs (used by equalTycon) so that
237		dummy args are distinct types
238
239		----------------------------------------------------------------------
240		Name: Matthias Blume
241		Date: 2000/04/30 01:00:00 JST
242		Tag: blume-20000430-versions
243		Description:
244
245		1. CM version numbering added.  This is an implementation of Lal's
246		proposal for adding version numbers and version checking to .cm
247		files.  Lal said that his proposal was just that -- a proposal.
248		For the time being I went ahead and implemented it so that people
249		can comment on it.  Everything is completely backward-compatible
250		(except for the stable library format, i.e., new bootfiles!).
251
252		As usual, see the CM manual for details.
253
254		2. An alternative syntax for anchored paths has been implemented.
255		Dave has recently voiced the same concerns that I had when I did
256		this, so there should be some support.  My take is that eventually
257		I will let support for the current syntax (where anchors are
258		"implicit") fade out in favor of the new, explicit syntax.
259		In order to be backward-compatible, both old and new syntax are
260		currently supported.
261
262		Again, see the CM manual for details.
263
264		3. Parallel make is trying to be slightly smarter:  When the master
265		process finds a "bottleneck", i.e., when there is only one
266		compilation unit that can be compiled and everybody else is
267		waiting on it, then it will simply compile it directly instead
268		of clumsily telling one of the slaves to do it.
269
270		4. Support for "unsharing" added.  This is necessary in order to be
271		able to have two different versions of the same library running
272		at the same time (e.g., for trying out a new MLRISC while still
273		having the old MLRISC linked into the current compiler, etc.)
274		See the CM manual.
275
276		5. Simple "makedepend" functionality added for generating Makefile
277		dependency information.  (This is rather crude at the moment.
278		Expect some changes here in the future.)
279
280		6. ".fun" added as a recognized suffix for ML files. Also documented
281		explicitly in the manual that the fallback behavior (unknown suffix
282		-> ML file) is not an official feature!
283
284		7. Small changes to the pickler for stable libraries.
285
286		8. Several internal changes to CM (for cleanup/improvement).
287
288
289		!!!! NEW BINFILES !!!!
290
291		----------------------------------------------------------------------
292		Name: Matthias Blume
293		Date: 2000/04/28 17:30:00 JST
294		Tag: blume-20000428-pathconfig
295		Description:
296
297		1. I changed config/install.sh to remove duplicate entries from the
298		lib/pathconfig file at the end.  Moreover, the final version of
299		lib/pathconfig is sorted alphabetically.  The same (sorting) is done
300		in src/system/installml.
301
302		2. The config/install.sh script now consistently uses relative
303		pathnames in lib/pathconfig whenever the anchor is in the lib
304		directory.  (So far this was true for the libraries that come
305		pre-compiled and bundled as part of the bootfiles but not for
306		libraries that are compiled by the script itself.)
307
308		----------------------------------------------------------------------
309		Name: Matthias Blume
310		Date: 2000/04/26 13:10:00 JST
311		Tag: blume-20000426-fun_suffix
312		Description:
313
314		Added ".fun" as a recognized file name suffix (for ML code).
315
316		----------------------------------------------------------------------
317		Name: Allen Leung
318		Date: 2000/04/25 17:00:00 EST
319		Tag: leunga-20000425-alpha-ra
320		Description:
321
322		1. Alpha
323
324		PSEUDOARITH was missing in AlphaRewrite.  This causes an endless loop
325		in C--.
326
327		2. RA
328
329		Added a flag "ra-dump-size" to print out the size of the flowgraph
330		and the interference graph.
331
332		----------------------------------------------------------------------
333		Name: Dave MacQueen
334		Date: 2000/04/25/
335		Tag: dbm-20000425-mlyacc_doc_examples
336		Description:
337		Updated mlyacc.tex sections 5 and 7 for SML '97 and CM.
338		Updated all three examples in src/ml-yacc/examples to run
339		under 110.* using CM.make.
340
341		----------------------------------------------------------------------
342		Name: Allen Leung
343		Date: 2000/04/20 23:04:00 EST
344		Tag: leunga-20000420-ssa-c---stuff
345		Description:
346
347		This update synchronizes my repository with Yale's.  Most of these
348		changes, however, do not affect SML/NJ at all (the RA is an exception).
349
350		1. Register Allocator
351
352		a. An improvement in the interference graph construction:
353		Given a copy
354
355		s <- t
356
357		no interference edge between s and t is added for this definition of s.
358
359		b. I've added two new spill heuristic modules that Fermin and I developed
360		(in the new library RA.cm). These are unused in SML/NJ but maybe
361		useful for others (Moby?)
362
363		2. X86
364
365		a. Various fixes in the backend provided by Fermin [C--] and Lal.
366
367		3. Alpha
368
369		a. Added the BSR instruction and code generation that goes with it [C--]
370		b. Other fixes too numerous to recount provided by Fermin [C--]
371
372		4. Regmaps
373
374		a. The regmaps are not initialized with the identity physical bindings
375		at creation time.  This is unneeded.
376
377		5. MLRISC Optimizations
378
379		a. The DJ-Graph module can now compute the iterated dominance frontiers
380		intersects with liveness incrementally in linear time! Woohoo!
381		This is now used in my new SSA construction algorithm.
382
383		b. THe branch reorganization module is now smarter about linear chains of
384		basic blocks.
385
386
387		----------------------------------------------------------------------
388		Name: Matthias Blume
389		Date: 2000/04/12 13:52:00 JST
390		Tag: blume_main_v110p27_1
391		Description:
392
393		Changed install.sh script to handle archive files without version number
394		and to use "boot.<arch>-<os>" instead of "sml.boot.<arch>-<os>" for the
395		name of the boot file archive.
396
397		----------------------------------------------------------------------
398		Name: Dave MacQueen
399		Date: 2000/04/09 14:00 EDT
400		Tag: dbm-20000410-Version_110_27
401		Description:
402
403		1. Updated src/compiler/TopLevel/main/version.sml to version 110.27
404
405		2. Updated src/config/version to 110.27
406
407		3. New boot files!
408
409		----------------------------------------------------------------------
410		Name: Allen Leung
411		Date: 2000/04/09 19:09:00 EST
412		Tag: leunga-20000409-misc
413		Description:
414
415		1.  Yet another fix for x86 assembly for idivl, imull, mull and friends.
416
417		2.  Miscellaneous improvements to MLRISC (unused in sml/nj)
418
419		----------------------------------------------------------------------
420		Name: Stefan
421		Date: 2000/04/07 10:00:00 EDT
422		Tag: monnier-20000406-branch-handling
423		Description:
424
425		Improved handling of branches (mostly those generated from
426		polymorphic equality), removed switchoff and changed the
427		default optimization settings (more cpsopt and less flintopt).
428
429		----------------------------------------------------------------------
430		Name: Allen Leung
431		Date: 2000/04/06 01:30:00 EST
432		Tag: leunga-20000406-peephole-x86-SSA-2
433		Description:
434
435		Forgot a few files.
436
437		----------------------------------------------------------------------
438		Name: Allen Leung
439		Date: 2000/04/06 00:36:00 EST
440		Tag: leunga-20000406-peephole-x86-SSA
441		Description:
442
443		1.  New Peephole code
444
445		2.  Minor improvement to X86 instruction selection
446
447		3.  Various fixes to SSA and machine description -> code translator
448
449		----------------------------------------------------------------------
450		Name: Matthias Blume
451		Date: 2000/04/05 12:30:00 JST
452		Tag: blume_main_v110p26p2_3
453		Description:
454
455		This update just merges three minor cosmetic updates to CM's sources
456		to get ready for the 110.27 code freeze on Friday.  No functionality
457		has changed.
458
459		----------------------------------------------------------------------
460		Name: Allen Leung
461		Date: 2000/04/04 19:39:00 EST
462		Tag: leunga-20000404-x86-asm
463		Description:
464
465		1.  Fixed a problem in X86 assembly.
466
467		Things like
468
469		jmp %eax
470		jmp (%eax)
471
472		should be output as
473
474		jmp *%eax
475		jmp *(%eax)
476
477		2.  Assembly output
478
479		Added a new flag
480
481		"asm-indent-copies" (default to false)
482
483		When this flag is on, parallel copies will be indented an extra level.
484
485		----------------------------------------------------------------------
486		Name: Allen Leung
487		Date: 2000/04/04 03:18:00 EST
488		Tag: leunga-20000404-C--Moby
489		Description:
490
491		All of these fixes are related to C--, Moby, and my own optimization
492		stuff; so they shouldn't affect SML/NJ.
493
494		1.  X86
495
496		Various fixes related floating point, and extensions.
497
498		2.  Alpha
499
500		Some extra patterns related to loads with signed/zero extension
501		provided by Fermin.
502
503		3.  Assembly
504
505		When generating assemby, resolve the value of client defined constants,
506		instead of generating symbolic values.  This is controlled by the
507		new flag "asm-resolve-constants", which is default to true.
508
509		4.  Machine Descriptions
510
511		a. The precedence parser was slightly broken when parsing infixr symbols.
512		b. The type generalizing code had the bound variables reversed, resulting
513		in a problem during arity raising.
514		c. Various fixes in machine descriptions.
515
516		----------------------------------------------------------------------
517		Name: Matthias Blume
518		Date: 2000/04/03 16:05:00 JST
519		Tag: blume_main_v110p26p2_2
520		Description:
521
522		I eliminated coreEnv from compInfo.  Access to the "Core" structure is
523		now done via the ordinary static environment that is context to each
524		compilation unit.
525
526		To this end, I arranged that instead of "structure Core" as "structure
527		_Core" is bound in the pervasive environment.  Core access is done via
528		_Core (which can never be accidentially rebound because _Core is not a
529		legal surface-syntax symbol).
530
531		The current solution is much cleaner because the core environment is
532		now simply part of the pervasive environment which is part of every
533		compilation unit's context anyway.  In particular, this eliminates all
534		special-case handling that was necessary until now in order to deal
535		with dynamic and symbolic parts of the core environment.
536
537		Remaining hackery (to bind the "magic" symbol _Core) is localized in the
538		compilation mananger's bootstrap compiler (actually: in the "init group"
539		handling).  See the comments in src/system/smlnj/init/init.cmi for
540		more details.
541
542		I also tried to track down all mentions of "Core" (as string argument
543		to Symbol.strSymbol) in the compiler and replaced them with a
544		reference to the new CoreSym.coreSym.  Seems cleaner since the actual
545		name appears in one place only.
546
547		Binfile and bootfile format have not changed, but the switchover from
548		the old "init.cmi" to the new one is a bit tricky, so I supplied new
549		bootfiles anyway.
550
551		----------------------------------------------------------------------
552		Name: Allen Leung
553		Date: 2000/04/02 21:17:00 EST
554		Tag: leunga-20000402-mltree
555		Description:
556
557		1. Renamed the constructor CALL in MLTREE by popular demand.
558		2. Added a bunch of files from my repository.  These are currently
559		used by other non-SMLNJ backends.
560
561		----------------------------------------------------------------------
562		Name: Allen Leung
563		Date: 2000/03/31 21:15:00 EST
564		Tag: leunga-20000331-aliasing
565		Description:
566
567		This update contains a rewritten (and hopefully more correct) module
568		for extracting aliasing information from CPS.
569
570		To turn on this feature:
571
572		Compiler.Control.CG.memDisambiguate := true
573
574		To pretty print the region information with assembly
575
576		Compiler.Control.MLRISC.getFlag "asm-show-region" := true;
577
578		To control how many levels of aliasing information are printed, use:
579
580		Compiler.Control.MLRISC.getInt "points-to-show-level" := n
581
582		The default of n is 3.
583
584		----------------------------------------------------------------------
585		Name: David MacQueen
586		Date: 2000/03/31 11:15:00 EST
587		Tag: dbm-20000331-runtime_fix
588		Description:
589
590		This update contains:
591
592		1. runtime/c-lib/c-libraries.c
593		includes added in revision 1.2 caused compilation errors on hppa-hpux
594
595		2. fix for bug 1556
596		system/Basis/Implementation/NJ/internal-signals.sml
597
598		----------------------------------------------------------------------
599		Name: Matthias Blume
600		Date: 2000/03/31 18:00:00 JST
601		Tag: blume_main_v110p26p2_1
602		Description:
603
604		This update contains:
605
606		1. A small change to CM's handling of stable libraries:
607		CM now maintains one "global" modmap that is used for all stable
608		libraries.  The use of such a global modmap maximizes sharing and
609		minimizes the need for re-traversing parts of environments during
610		modmap construction.  (However, this has minor impact since modmap
611		construction seems to account for just one percent or less of total
612		compile time.)
613
614		2. I added a "genmap" phase to the statistics.  This is where I got the
615		"one percent" number (see above).
616
617		3. CM's new tool parameter mechanism just became _even_ better. :)
618		- The parser understands named parameters and recursive options.
619		- The "make" and "shell" tools use these new features.
620		(This makes it a lot easier to cascade these tools.)
621		- There is a small syntax change: named parameters use a
622
623		<name> : ( <option> ... )            or
624		<name> : <string>
625
626		syntax.  Previously, named parameters were implemented in an
627		ad-hoc fashion by each tool individually (by parsing strings)
628		and had the form
629
630		<name>=<string>
631
632		See the CM manual for a full description of these issues.
633
634		----------------------------------------------------------------------
635		Name: Matthias Blume
636		Date: 2000/03/30 18:00:00 JST
637		Tag: blume_main_v110p26p2_0
638		Description:
639
640		!!!!! WARNING !!!!!!
641		!!  New binfiles  !!
642		!!!!!!!!!!!!!!!!!!!!
643
644		This update contains:
645
646		1. Moderate changes to CM:
647
648		- Changes to CM's tools mechanism.  In particular, it is now possible
649		to have tools that accept additional "command line" parameters
650		(specified in the .cm file at each instance where the tool's class is
651		used).
652
653		This was done to accomodate the new "make" and "shell" tools which
654		facilitate fairly seemless hookup to portions of code managed using
655		Makefiles or Shell scripts.
656
657		There are no classes "shared" or "private" anymore.  Instead, the
658		sharing annotation is now a parameter to the "sml" class.
659
660		There is a bit of generic machinery for implementing one's own
661		tools that accept command-line parameters.  However, I am not yet fully
662		satisfied with that part, so expect changes here in the future.
663
664		All existing tools are described in the CM manual.
665
666		- Slightly better error handling.  (CM now surpresses many followup
667		error messages that tended to be more annoying than helpful.)
668
669		2. Major changes to the compiler's static environment data structures.
670
671		- no CMStaticEnv anymore.
672		- no CMEnv, no "BareEnvironment" (actually, _only_ BareEnvironment,
673		but it is called Environment), no conversions between different
674		kinds of static environments
675
676		- There is still a notion of a "modmap", but such modmaps are generated
677		on demand at the time when they are needed.  This sounds slow, but I
678		sped up the code that generates modmaps enough for this not to lead to
679		a slowdown of the compiler (at least I didn't detect any).
680
681		- To facilitate rapid modmap generation, static environments now
682		contain an (optional) "modtree" structure.  Modtree annotations are
683		constructed by the unpickler during unpickling.  (This means that
684		the elaborator does not have to worry about modtrees at all.)
685		Modtrees have the advantage that they are compositional in the same
686		way as the environment data structure itself is compositional.
687		As a result, modtrees never hang on to parts of an environment that
688		has already been rendered "stale" by filtering or rebinding.
689
690		- I went through many, many trials and errors before arriving at the
691		current solution.  (The initial idea of "linkpaths" did not work.)
692		But the result of all this is that I have touched a lot of files that
693		depend on the "modules" and "types" data structures (most of the
694		elaborator). There were a lot of changes during my "linkpath" trials
695		that could have been reverted to their original state but weren't.
696		Please, don't be too harsh on me for messing with this code a bit more
697		than what was strictly necessary...  (I _did_ resist the tempation
698		of doing any "global reformatting" to avoid an untimely death at
699		Dave's hands. :)
700
701		- One positive aspect of the previous point:  At least I made sure that
702		all files that I touched now compile without warnings (other than
703		"polyEqual").
704
705		- compiler now tends to run "leaner" (i.e., ties up less memory in
706		redundant modmaps)
707
708		----------------------------------------------------------------------
709		Name: Allen Leung
710		Date: 2000/03/29 18:00:00
711		Tag: leunga-20000327-mlriscGen_hppa_alpha_x86
712		Boot files (optional): ftp://react-ilp.cs.nyu.edu/leunga/110.26.1-sml.boot.x86-unix-20000330.tar.gz
713		Description:
714
715		This update contains *MAJOR* changes to the way code is generated from CPS
716		in the module mlriscGen, and in various backend modules.
717
718		CHANGES
719		=======
720
721		1. MLRiscGen: forward propagation fix.
722
723		There was a bug in forward propagation introduced at about the same time
724		as the MLRISC x86 backend, which prohibits coalescing to be
725		performed effectively in loops.
726
727		Effect: speed up of loops in RISC architectures.
728		By itself, this actually slowed down certain benchmarks on the x86.
729
730		2. MLRiscGen:  forward propagating addresses from consing.
731
732		I've changed the way consing code is generated.  Basically I separated
733		out the initialization part:
734
735		store tag,   offset(allocptr)
736		store elem1, offset+4(allocptr)
737		store elem2, offset+8(allocptr)
738		...
739		store elemn, offset+4n(allocptr)
740
741		and the address computation part:
742
743		celladdr <- offset+4+alloctpr
744
745		and move the address computation part
746
747		Effect:  register pressure is generally lower as a result.  This
748		makes compilation of certain expressions much faster, such as
749		long lists with non-trivial elements.
750
751		[(0,0), (0,0), .... (0,0)]
752
753		3. MLRiscGen: base pointer elimination.
754
755		As part of the linkage mechanism, we generate the sequence:
756
757		L:  ...  <- start of the code fragment
758
759		L1:
760		base pointer <- linkreg - L1 + L
761
762		The base pointer was then used for computing relocatable addresses
763		in the code fragment.  Frequently (such as in lots of continuations)
764		this is not needed.  We now eliminate this sequence whenever possible.
765
766		For compile time efficiency, I'm using a very stupid local heuristic.
767		But in general, this should be done as a control flow analysis.
768
769		Effect:  Smaller code size.  Speed up of most programs.
770
771		4. Hppa back end
772
773		Long jumps in span dependence resolution used to depend on the existence
774		of the base pointer.
775
776		A jump to a long label L was expanded into the following sequence:
777
778		LDIL %hi(L-8192), %r29
779		LDO  %lo(L-8192)(%r29), %r29
780		ADD  %r29, baseptr, %r29
781		BV,n %r0(%r29)
782
783		In the presence of change (3) above, this will not work.  I've changed
784		it so that the following sequence of instructions are generated, which
785		doesn't mention the base pointer at all:
786
787		BL,n  L', %r29           /* branch and link, L' + 4 -> %r29 */
788		L':  ADDIL L-(L'+4), %r29     /* Compute address of L */
789		BV,n  %r0(%r29)          /* Jump */
790
791		5. Alpha back end
792
793		New alpha instructions LDB/LDW have been added, as per Fermin's
794		suggestions.   This is unrelated to all other changes.
795
796		6. X86 back end
797
798		I've changed andl to testl in the floating point test sequence
799		whenever appropriate.  The Intel optimization guide states that
800		testl is perferable to andl.
801
802		7. RA (x86 only)
803
804		I've improved the spill propagation algorithm, using an approximation
805		of maximal weighted independent sets.   This seems to be necessary to
806		alleviate the negative effect in light of the slow down in (1).
807
808		I'll write down the algorithm one of these days.
809
810		8. MLRiscGen: frequencies
811
812		I've added an annotation that states that all call gc blocks have zero
813		execution frequencies.  This improves register allocation on the x86.
814
815		BENCHMARKS
816		==========
817
818		I've only perform the comparison on 110.25.
819
820		The platforms are:
821
822		HPPA  A four processor HP machine (E9000) with 5G of memory.
823		X86   A 300Hhz Pentium II with 128M of memory, and
824		SPARC An Ultra sparc 2 with 512M of memory.
825
826		I used the following parameters for the SML benchmarks:
827
828		@SMLalloc
829		HPPA    256k
830		SPARC   512k
831		X86     256k
832
833		COMPILATION TIME
834		----------------
835		Here are the numbers comparing the compilation times of the compilers.
836		I've only compared 110.25 compiling the new sources versus
837		a fixpoint version of the new compiler compiling the same.
838
839		110.25                                  New
840		Total  Time in RA  Spill+Reload   Total  Time In RA Spill+Reload
841		HPPA   627s    116s        2684+3584     599s    95s       1003+1879
842		SPARC  892s    173s        2891+3870     708s    116s      1004+1880
843		X86    999s    315s       94006+130691   987s    296s    108877+141957
844
845		110.25         New
846		Code Size      Code Size
847		HPPA   8596736         8561421
848		SPARC  8974299         8785143
849		X86    9029180         8716783
850
851		So in summary, things are at least as good as before.   Dramatic
852		reduction in compilation is obtained on the Sparc; I can't explain it,
853		but it is reproducible.  Perhaps someone should try to reproduce this
854		on their own machines.
855
856		SML BENCHMARKS
857		--------------
858
859		On the average, all benchmarks perform at least as well as before.
860
861		HPPA         Compilation Time     Spill+Reload      Run Time
862		110.25  New            110.25    New   110.25  New
863
864		barnesHut  3.158  3.015  4.75%    1+1       0+0   2.980  2.922   2.00%
865		boyer  6.152  5.708  7.77%    0+0       0+0   0.218  0.213   2.34%
866		count-graphs  1.168  1.120  4.32%    0+0       0+0  22.705 23.073  -1.60%
867		fft  0.877  0.792 10.74%    1+3       1+3   0.602  0.587   2.56%
868		knuthBendix  3.180  2.857 11.32%    0+0       0+0   0.675  0.662   2.02%
869		lexgen  6.190  5.290 17.01%    0+0       0+0   0.913  0.788  15.86%
870		life  0.803  0.703 14.22%   25+25      0+0   0.153  0.140   9.52%
871		logic  2.048  2.007  2.08%    6+6       1+1   4.133  4.008   3.12%
872		mandelbrot  0.077  0.080 -4.17%    0+0       0+0   0.765  0.712   7.49%
873		mlyacc 22.932 20.937  9.53%  154+181    32+57  0.468  0.430   8.91%
874		nucleic  5.183  5.060  2.44%    2+2       0+0   0.125  0.120   4.17%
875		ratio-regions  3.357  3.142  6.84%    0+0       0+0  116.225 113.173 2.70%
876		ray  1.283  1.290 -0.52%    0+0       0+0   2.887  2.855   1.11%
877		simple  6.307  6.032  4.56%   28+30      5+7   3.705  3.658   1.28%
878		tsp  0.888  0.862  3.09%    0+0       0+0   7.040  6.893   2.13%
879		vliw 24.378 23.455  3.94%  106+127    25+45  2.758  2.707   1.91%
880		--------------------------------------------------------------------------
881		Average                     6.12%                                   4.09%
882
883		SPARC        Compilation Time     Spill+Reload      Run Time
884		110.25  New            110.25    New   110.25  New
885
886		barnesHut  3.778  3.592  5.20%    2+2       0+0   3.648  3.453    5.65%
887		boyer  6.632  6.110  8.54%    0+0       0+0   0.258  0.242    6.90%
888		count-graphs  1.435  1.325  8.30%    0+0       0+0  33.672 34.737   -3.07%
889		fft  0.980  0.940  4.26%    3+9       2+6   0.838  0.827    1.41%
890		knuthBendix  3.590  3.138 14.39%    0+0       0+0   0.962  0.967   -0.52%
891		lexgen  6.593  6.072  8.59%    1+1       0+0   1.077  1.078   -0.15%
892		life  0.972  0.868 11.90%   26+26      0+0   0.143  0.140    2.38%
893		logic  2.525  2.387  5.80%    7+7       1+1   5.625  5.158    9.05%
894		mandelbrot  0.090  0.093 -3.57%    0+0       0+0   0.855  0.728   17.39%
895		mlyacc 26.732 23.827 12.19%  162+189    32+57  0.550  0.560   -1.79%
896		nucleic  6.233  6.197  0.59%    3+3       0+0   0.163  0.173   -5.77%
897		ratio-regions  3.780  3.507  7.79%    0+0       0+0 133.993 131.035   2.26%
898		ray  1.595  1.550  2.90%    1+1       0+0   3.440  3.418    0.63%
899		simple  6.972  6.487  7.48%   29+32      5+7   3.523  3.525   -0.05%
900		tsp  1.115  1.063  4.86%    0+0       0+0   7.393  7.265    1.77%
901		vliw 27.765 24.818 11.87%  110+135    25+45  2.265  2.135    6.09%
902		----------------------------------------------------------------------------
903		Average                     6.94%                                    2.64%
904
905		X86          Compilation Time     Spill+Reload      Run Time
906		110.25  New            110.25    New   110.25  New
907
908		barnesHut  5.530  5.420  2.03%  593+893   597+915   3.532  3.440   2.66%
909		boyer  8.768  7.747 13.19%  493+199   301+289   0.327  0.297  10.11%
910		count-graphs  2.040  2.010  1.49%  298+394   315+457  26.578 28.660  -7.26%
911		fft  1.327  1.302  1.92%  112+209   115+210   1.055  0.962   9.71%
912		knuthBendix  5.218  5.475 -4.69%  451+598   510+650   0.928  0.932  -0.36%
913		lexgen  9.970  9.623  3.60% 1014+841  1157+885   0.947  0.928   1.97%
914		life  1.183  1.183  0.00%  162+182   145+148   0.127  0.103  22.58%
915		logic  3.285  3.512 -6.45%  514+684   591+836   5.682  5.577   1.88%
916		mandelbrot  0.147  0.143  2.33%   38+41     33+54    0.703  0.690   1.93%
917		mlyacc 35.457 32.763  8.22% 3496+4564 3611+4860  0.552  0.550   0.30%
918		nucleic  7.100  6.888  3.07%  239+168   201+158   0.175  0.173   0.96%
919		ratio-regions  6.388  6.843 -6.65% 1182+257   981+300  120.142 120.345 -0.17%
920		ray  2.332  2.338 -0.29%  346+398   402+494   3.593  3.540   1.51%
921		simple  9.912  9.903  0.08% 1475+941  1579+1168  3.057  3.178  -3.83%
922		tsp  1.623  1.532  5.98%  266+200   250+211   8.045  7.878   2.12%
923		vliw 33.947 35.470 -4.29% 2629+2774 2877+3171  2.072  1.890   9.61%
924		----------------------------------------------------------------------------
925		Average                     1.22%                                     3.36%
926
927		----------------------------------------------------------------------
928		Name: Allen Leung
929	Date: 2000/03/23 16:25:00	Date: 2000/03/23 16:25:00
930	Tag: leunga-20000323-fix_x86_alpha	Tag: leunga-20000323-fix_x86_alpha
931	Description:	Description:

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.583   changed lines   Added in v.648

---

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