Diff of /sml/trunk/HISTORY

Parent Directory | Revision Log | Patch

	revision 1094, Mon Feb 25 09:58:56 2002 UTC	revision 1211, Tue May 21 14:54:32 2002 UTC
#	Line 13	Line 13
13	Description:	Description:
14
15	----------------------------------------------------------------------	----------------------------------------------------------------------
16		Name: Matthias Blume
17		Date: 2002/05/21 10:55:00 EDT
18		Tag: blume-20020521-misc
19		Description:
20
21		- John tweaked runtime to be silent on heap export (except when
22		GC messages are on).
23		- I added a few more things (cross-compiling versions of CMB) to
24		config/preloads (as suggestions).
25
26		----------------------------------------------------------------------
27		Name: Matthias Blume
28		Date: 2002/05/20 22:25:00 EDT
29		Tag: blume-20020520-controls
30		Description:
31
32		- Added ControlUtil structure to control-lib.cm.
33		- Use it throughout.
34		- Used Controls facility to define MLRISC controls (as opposed to
35		registering MLRISC control ref cells with Controls after the fact)
36		- Fixed messed-up controls priorities.
37
38		* Removed again all the stuff from config/preloads that one wouldn't
39		be able to preload at the time the initial heap image is built.
40		(Many libraries, e.g., CML, do not exist yet at this time.  The
41		only libraries that can be preloaded via config/preloads are those
42		that come bundled with the bootfiles.)
43
44		----------------------------------------------------------------------
45		Name: Matthias Blume
46		Date: 2002/05/20 10:59:00 EDT
47		Tag: blume-20020520-preloads
48		Description:
49
50		Added a lot of commented-out suggestions for things to be included
51		in config/preloads.
52
53		----------------------------------------------------------------------
54		Name: Allen Leung
55		Date: 2002/05/18 14:20:00 EDT
56		Tag: leunga-20020518-mdl
57		Description:
58
59		o Made the mdl tool stuff compile and run again.
60		o I've disabled all the stuff that depends on RTL specifications; they
61		are all badly broken anyway.
62
63		----------------------------------------------------------------------
64		Name: Matthias Blume
65		Date: 2002/05/17 16:49:00 EDT
66		Tag: blume-20020517-controls
67		Description:
68
69		0. John Reppy made several modifications to the SML/NJ library.
70		In particular, there is a shiny new controls-lib.cm.
71
72		1. Pushed new controls interface through compiler so that everything
73		compiles again.
74
75		2. Added FormatComb and FORMAT_COMB to the CML version of the
76		SML/NJ library (so that CML compiles again).
77
78		3. Modified init scripts because XXX_DEFAULT environment variables
79		are no longer with us.  (Boot-time initialization is now done
80		using the same environment variables that are also used for
81		startup-time initialization of controls.)
82
83		----------------------------------------------------------------------
84		Name: Lal George
85		Date: 2002/05/15 09:20:10 EDT
86		Tag: george-20020515-pseudo-op-decls
87		Description:
88
89		All pseudo-ops emitted before the first segment declaration
90		such as TEXT, DATA, and BSS directives are assumed to be global
91		declarations and are emitted first in the assembly file. This is
92		useful in a number of situations where one has pseudo-ops that are not
93		specific to any segment, and also works around the constraint that one
94		cannot have client pseudo-ops in the TEXT segment.
95
96		Because no segment is associated with these declarations it is
97		an error to allocate any space or objects before the first segment
98		directive and an exception will be raised. However, we cannot make
99		this check for client pseudo-ops.
100
101		These top level declarations are a field in the CFG graph_info.
102		In theory you can continue to add to this field after the CFG has been
103		built -- provided you know what you are doing;-)
104
105		----------------------------------------------------------------------
106		Name: Matthias Blume
107		Date: 2002/05/13 16:40:00 EDT
108		Tag: blume-20020513-pp-etc
109		Description:
110
111		A few minor bugfixes:
112
113		- Stopgap measure for bug recently reported by Elsa Gunter (ppDec).
114		(Bogus printouts for redefined bindings still occur.  Compiler
115		bug should no longer occur now.  We need to redo the prettyprinter
116		from scratch.)
117
118		- CM pathname printer now also adds escape sequences for ( and )
119
120		- commend and docu fixes for ml-nlffi
121
122		----------------------------------------------------------------------
123		Name: Matthias Blume
124		Date: 2002/05/10 16:40:00 EDT
125		Tag: blume-20020510-erg-textio
126		Description:
127
128		Applied the following bugfix provided by Emden Gansner:
129
130		Output is corrupted when outputSubstr is used rather than output.
131
132		The problem occurs when a substring
133
134		ss = (s, dataStart, dataLen)
135
136		where dataStart > 0, fills a stream buffer with avail bytes left.
137		avail bytes of s, starting at index dataStart, are copied into the
138		buffer, the buffer is flushed, and then the remaining dataLen-avail
139		bytes of ss are copied into the beginning of the buffer. Instead of
140		starting this copy at index dataStart+avail in s, the current code
141		starts the copy at index avail.
142
143		Fix:
144		In text-io-fn.sml, change line 695 from
145		val needsFlush = copyVec(v, avail, dataLen-avail, buf, 0)
146		to
147		val needsFlush = copyVec(v, dataStart+avail, dataLen-avail, buf, 0)
148
149		----------------------------------------------------------------------
150		Name: Matthias Blume
151		Date: 2002/04/12 13:55:00 EDT
152		Tag: blume-20020412-assyntax
153		Description:
154
155		1. Grabbed newer assyntax.h from the XFree86 project.
156		2. Fiddled with how to compile X86.prim.asm without warnings.
157		3. (Very) Minor cleanup in CM.
158
159		----------------------------------------------------------------------
160		Name: Matthias Blume
161		Date: 2002/04/01 (no joke!) 17:07:00 EST
162		Tag: blume-20020401-x86div
163		Description:
164
165		Added full support for div/mod/rem/quot on the x86, using the machine
166		instruction's two results (without clumsily recomputing the remainder)
167		directly where appropriate.
168
169		Some more extensive power-of-two support was added to the x86 instruction
170		selector (avoiding expensive divs, mods, and muls where they can be
171		replaced with cheaper shifts and masks).  However, this sort of thing
172		ought to be done earlier, e.g., within the CPS optimizer so that
173		all architectures benefit from it.
174
175		The compiler compiles to a fixed point, but changes might be somewhat
176		fragile nevertheless.  Please, report any strange things that you might
177		see wrt. div/mod/quot/rem...
178
179		----------------------------------------------------------------------
180		Name: Matthias Blume
181		Date: 2002/03/29 17:22:00
182		Tag: blume-20020329-div
183		Description:
184
185		Fixed my broken div/mod logic.  Unfortunately, this means that the
186		inline code for div/mod now has one more comparison than before.
187		Fast paths (quotient > 0 or remainder = 0) are not affected, though.
188		The problem was with quotient = 0, because that alone does not tell
189		us which way the rounding went.  One then has to look at whether
190		remainder and divisor have the same sign...  :(
191
192		Anyway, I replaced the bootfiles with fresh ones...
193
194		----------------------------------------------------------------------
195		Name: Matthias Blume
196		Date: 2002/03/29 14:10:00 EST
197		Tag: blume-20020329-inlprims
198		Description:
199
200		NEW BOOTFILES!!!    Version number bumped to 110.39.3.
201
202		Primops have changed. This means that the bin/boot-file formats have
203		changed as well.
204
205		To make sure that there is no confusion, I made a new version.
206
207
208		CHANGES:
209
210		* removed REMT from mltree (remainder should never overflow).
211
212		* added primops to deal with divisions of all flavors to the frontend
213
214		* handled these primops all the way through so they map to their respective
215		MLRISC support
216
217		* used these primops in the implementation of Int, Int32, Word, Word32
218
219		* removed INLDIV, INLMOD, and INLREM as they are no longer necessary
220
221		* parameterized INLMIN, INLMAX, and INLABS by a numkind
222
223		* translate.sml now deals with all flavors of INL{MIN,MAX,ABS}, including
224		floating point
225
226		* used INL{MIN,MAX,ABS} in the implementation of Int, Int32, Word, Word32,
227		and Real (but Real.abs maps to a separate floating-point-only primop)
228
229
230		TODO items:
231
232		* Hacked Alpha32 instruction selection, disabling the selection of REMx
233		instructions because the machine instruction encoder cannot handle
234		them.  (Hppa, PPC, and Sparc instruction selection did not handle
235		REM in the first place, and REM is supported by the x86 machine coder.)
236
237		* Handle DIV and MOD with DIV_TO_NEGINF directly in the x86 instruction
238		selection phase.  (The two can be streamlined because the hardware
239		delivers both quotient and remainder at the same time anyway.)
240
241		* Think about what to do with "valOf(Int32.minInt) div ~1" and friends.
242		(Currently the behavior is inconsistent both across architectures and
243		wrt. the draft Basis spec.)
244
245		* Word8 should eventually be handled natively, too.
246
247		* There seems to be one serious bug in mltree-gen.sml.  It appears, though,
248		as if there currently is no execution path that could trigger it in
249		SML/NJ.  (The assumptions underlying functions arith and promotable do not
250		hold for things like multiplication and division.)
251
252		----------------------------------------------------------------------
253		Name: Matthias Blume
254		Date: 2002/03/27 16:27:00 EST
255		Tag: blume-20020327-mlrisc-divisions
256		Description:
257
258		Added support for all four division operations (ML's div, mod, quot,
259		and rem) to MLRISC.  In the course of doing so, I also rationalized
260		the naming (no more annoying switch-around of DIV and QUOT), by
261		parameterizing the operation by div_rounding_mode (which can be either
262		DIV_TO_ZERO or DIV_TO_NEGINF).
263
264		The generic MLTreeGen functor takes care of compiling all four
265		operations down to only round-to-zero div.
266
267		Missing pieces:
268
269		* Doing something smarter than relying on MLTreeGen on architectures
270		like, e.g., the x86 where hardware division delivers both quotient and
271		remainder at the same time.  With this, the implementation of the
272		round-to-neginf operations could be further streamlined.
273
274		* Remove inlining support for div/mod/rem from the frontend and replace it
275		with primops that get carried through to the backend.  Do this for all
276		int and word types.
277
278		----------------------------------------------------------------------
279		Name: Matthias Blume
280		Date: 2002/03/25 17:25:00 EST
281		Tag: blume-20020325-divmod
282		Description:
283
284		I improved (hopefully without breaking them) the implementation of Int.div,
285		Int.mod, and Int.rem.   For this, the code in translate.sml now takes
286		advantage of the following observations:
287
288		Let  q = x quot y      r = x rem y
289		d = x div  y      m = x mod y
290
291		where "quot" is the round-to-zero version of integer division that
292		hardware usually provides.  Then we have:
293
294		r = x - q * y        where neither the * nor the - will overflow
295		d = if q >= 0 orelse x = q * y then q else q - 1
296		where neither the * nor the - will overflow
297		m = if q >= 0 orelse r = 0 then r else r + y
298		where the + will not overflow
299
300		This results in substantial simplification of the generated code.
301		The following table shows the number of CFG nodes and edges generated
302		for
303		fun f (x, y) = x OPER y
304		(* with OPER \in div, mod, quot, rem *)
305
306
307		OPER | nodes(old) | edges(old) | nodes(new) | edges(new)
308		--------------------------------------------------------
309		div |         24 |         39 |         12 |         16
310		mod |         41 |         71 |         12 |         16
311		quot |          8 |         10 |          8 |         10
312		rem |         10 |         14 |          8 |         10
313
314
315		----------------------------------------------------------------------
316		Name: Matthias Blume
317		Date: 2002/03/25 22:06:00 EST
318		Tag: blume-20020325-cprotobug
319		Description:
320
321		Fixed a bug in cproto (c prototype decoder).
322
323		----------------------------------------------------------------------
324		Name: Matthias Blume
325		Date: 2002/03/25 16:00:00 EST
326		Tag: blume-20020325-raw-primops
327		Description:
328
329		I did some cleanup to Allen's new primop code and
330		replaced yesterday's bootfiles with new ones.
331		(But they are stored in the same place.)
332
333		----------------------------------------------------------------------
334		Name: Matthias Blume
335		Date: 2002/03/24 22:40:00 EST
336		Tag: blume-20020324-bootfiles
337		Description:
338
339		Made the bootfiles that Allen asked for.
340
341		----------------------------------------------------------------------
342		Name: Allen Leung
343		Date: 2002/03/23 15:50:00 EST
344		Tag: leunga-20020323-flint-cps-rcc-primops
345		Description:
346
347		1. Changes to FLINT primops:
348
349		(* make a call to a C-function;
350		* The primop carries C function prototype information and specifies
351		* which of its (ML-) arguments are floating point. C prototype
352		* information is for use by the backend, ML information is for
353		* use by the CPS converter. *)
354		| RAW_CCALL of { c_proto: CTypes.c_proto,
355		ml_args: ccall_type list,
356		ml_res_opt: ccall_type option,
357		reentrant : bool
358		} option
359		(* Allocate uninitialized storage on the heap.
360		* The record is meant to hold short-lived C objects, i.e., they
361		* are not ML pointers.  With the tag, the representation is
362		* the same as RECORD with tag tag_raw32 (sz=4), or tag_fblock (sz=8)
363		*)
364		| RAW_RECORD of {tag:bool,sz:int}
365		and ccall_type = CCALL_INT32 | CCALL_REAL64 | CCALL_ML_PTR
366
367		2.  These CPS primops are now overloaded:
368
369		rawload of {kind:numkind}
370		rawstore of {kind:numkind}
371
372		The one argument form is:
373
374		rawload {kind} address
375
376		The two argument form is:
377
378		rawload {kind} [ml object, byte-offset]
379
380		3. RAW_CCALL/RCC now takes two extra arguments:
381
382		a. The first is whether the C call is reentrant, i.e., whether
383		ML state should be saved and restored.
384		b. The second argument is a string argument specifying the name of
385		library and the C function.
386
387		These things are currently not handled in the code generator, yet.
388
389		4. In CProto,
390
391		An encoding type of "bool" means "ml object" and is mapped into
392		C prototype of PTR.  Note that "bool" is different than "string",
393		even though "string" is also mapped into PTR, because "bool"
394		is assigned an CPS type of BOGt, while "string" is assigned INT32t.
395
396		5. Pickler/unpicker
397
398		Changed to handle RAW_RECORD and newest RAW_CCALL
399
400		6. MLRiscGen,
401
402		1. Changed to handle the new rawload/rawstore/rawrecord operators.
403		2. Code for handling C Calls has been moved to a new module CPSCCalls,
404		in the file CodeGen/cpscompile/cps-c-calls.sml
405
406		7. Added the conditional move operator
407
408		condmove of branch
409
410		to cps.  Generation of this is still buggy so it is currently
411		disabled.
412
413		----------------------------------------------------------------------
414		Name: Lal George
415		Date: 2002/03/22 14:18:25 EST
416		Tag: george-20020322-cps-branch-prob
417		Description:
418
419		Implemented the Ball-Larus branch prediction-heuristics, and
420		incorporated graphical viewers for control flow graphs.
421
422		Ball-Larus Heuristics:
423		---------------------
424		See the file compiler/CodeGen/cpscompile/cpsBranchProb.sml.
425
426		By design it uses the Dempster-Shafer theory for combining
427		probabilities.  For example, in the function:
428
429		fun f(n,acc) = if n = 0 then acc else f(n-1, n*acc)
430
431		the ball-larus heuristics predicts that the n=0 is unlikely
432		(OH-heuristic), and the 'then' branch is unlikely because of the
433		RH-heuristic -- giving the 'then' branch an even lower combined
434		probability using the Dempster-Shafer theory.
435
436		Finally, John Reppy's loop analysis in MLRISC, further lowers the
437		probability of the 'then' branch because of the loop in the else
438		branch.
439
440
441		Graphical Viewing:
442		------------------
443		I merely plugged in Allen's graphical viewers into the compiler. The
444		additional code is not much. At the top level, saying:
445
446		Control.MLRISC.getFlag "cfg-graphical-view" := true;
447
448		will display the graphical view of the control flow graph just before
449		back-patching.  daVinci must be in your path for this to work. If
450		daVinci is not available, then the default viewer can be changed
451		using:
452
453		Control.MLRISC.getString "viewer"
454
455		which can be set to "dot" or "vcg" for the corresponding viewers. Of
456		course, these viewers must be in your path.
457
458		The above will display the compilation unit at the level of clusters,
459		many of which are small, boring, and un-interesting. Also setting:
460
461		Control.MLRISC.getInt "cfg-graphical-view_size"
462
463		will display clusters that are larger than the value set by the above.
464
465
466		----------------------------------------------------------------------
467		Name: Matthias Blume
468		Date: 2002/03/21 22:20:00 EST
469		Tag: blume-20020321-kmp-bugfix
470		Description:
471
472		Changed the interface to the KMP routine in PreString and fixed
473		a minor bug in one place where it was used.
474
475		----------------------------------------------------------------------
476		Name: Allen Leung
477		Date: 2002/03/21 20:30:00 EST
478		Tag: leunga-20020321-cfg
479		Description:
480
481		Fixed a potential problem in cfg edge splitting.
482
483		----------------------------------------------------------------------
484		Name: Allen Leung
485		Date: 2002/03/21 17:15:00 EST
486		Tag: leunga-20020321-x86-fp-cfg
487		Description:
488
489		1. Recoded the buggy parts of x86-fp.
490
491		a. All the block reordering code has been removed.
492		We now depend on the block placement phases to do this work.
493
494		b. Critical edge splitting code has been simplified and moved into the
495		CFG modules, as where they belong.
496
497		Both of these were quite buggy and complex.  The code is now much, much
498		simpler.
499
500		2. X86 backend.
501
502		a. Added instructions for 64-bit support.  Instruction selection for
503		64-bit has not been committed, however, since that
504		requires changes to MLTREE which haven't been approved by
505		Lal and John.
506
507		b. Added support for FUCOMI and FUCOMIP when generating code for
508		PentiumPro and above.  We only generate these instructions in
509		the fast-fp mode.
510
511		c. Added cases for JP and JNP in X86FreqProps.
512
513		3. CFG
514
515		CFG now has a bunch of methods for edge splitting and merging.
516
517		4. Machine description.
518
519		John's simplification of MLTREE_BASIS.fcond broke a few machine
520		description things:
521
522		rtl-build.{sig,sml} and hppa.mdl fixed.
523
524		NOTE: the machine description stuff in the repository is still broken.
525		Again, I can't put my fixes in because that involves
526		changes to MLTREE.
527
528		----------------------------------------------------------------------
529		Name: Matthias Blume
530		Date: 2002/03/20 15:55:00 EST
531		Tag: blume-20020320-kmp
532		Description:
533
534		Implemented Knuth-Morris-Pratt string matching in PreString and used
535		it for String.isSubstring, Substring.isSubstring, and
536		Substring.position.
537
538		(Might need some stress-testing.  Simple examples worked fine.)
539
540		----------------------------------------------------------------------
541		Name: Matthias Blume
542		Date: 2002/03/19 16:37:00 EST
543		Tag: blume-20020319-witnesses
544		Description:
545
546		Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
547
548		This implements a generic mechanism for changing constness qualifiers
549		anywhere within big C types without resorting to outright "casts".
550		(So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
551		modify the constness at the outermost level.)
552		The implementation of "convert" is based on the idea of "witness"
553		values -- values that are not used by the operation but whose types
554		"testify" to their applicability.  On the implementation side, "convert"
555		is simply a projection (returning its second curried argument).  With
556		cross-module inlining, it should not result in any machine code being
557		generated.
558
559		----------------------------------------------------------------------
560		Name: Matthias Blume
561		Date: 2002/03/15 16:40:00 EST
562		Tag: blume-20020315-basis
563		Description:
564
565		Provided (preliminary?) implementations for
566
567		{String,Substring}.{concatWith,isSuffix,isSubstring}
568
569		and
570
571		Substring.full
572
573		Those are in the Basis spec but they were missing in SML/NJ.
574
575		----------------------------------------------------------------------
576		Name: Matthias Blume
577		Date: 2002/03/14 21:30:00 EST
578		Tag: blume-20020314-controls
579		Description:
580
581		Controls:
582		---------
583
584		1. Factored out the recently-added Controls : CONTROLS stuff and put
585		it into its own library $/controls-lib.cm.  The source tree for
586		this is under src/smlnj-lib/Controls.
587
588		2. Changed the names of types and functions in this interface, so they
589		make a bit more "sense":
590
591		module -> registry
592		'a registry -> 'a group
593
594		3. The interface now deals in ref cells only.  The getter/setter interface
595		is (mostly) gone.
596
597		4. Added a function that lets one register an already-existing ref cell.
598
599		5. Made the corresponding modifications to the rest of the code so that
600		everything compiles again.
601
602		6. Changed the implementation of Controls.MLRISC back to something closer
603		to the original.  In particular, this module (and therefore MLRISC)
604		does not depend on Controls.  There now is some link-time code in
605		int-sys.sml that registers the MLRISC controls with the Controls
606		module.
607
608		CM:
609		---
610
611		* One can now specify the lambda-split aggressiveness in init.cmi.
612
613		----------------------------------------------------------------------
614		Name: Allen Leung
615		Date: 2002/03/13 17:30:00 EST
616		Tag: leunga-20020313-x86-fp-unary
617		Description:
618
619		Bug fix for:
620
621		> leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
622		> Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
623		> - fun f(x,(y,z)) = Real.~ y;
624		> [autoloading]
625		> [autoloading done]
626		>       fchsl   (%eax), 184(%esp)
627		> Error: MLRisc bug: X86MCEmitter.emitInstr
628		>
629		> uncaught exception Error
630		>   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
631
632		The problem was that the code generator did not generate any fp registers
633		in this case, and the ra didn't know that it needed to run the X86FP phase to
634		translate the pseudo fp instruction.   This only happened with unary fp
635		operators in certain situations.
636
637		----------------------------------------------------------------------
638		Name: Matthias Blume
639		Date: 2002/03/13 14:00:00 EST
640		Tag: blume-20020313-overload-etc
641		Description:
642
643		1. Added _overload as a synonym for overload for backward compatibility.
644		(Control.overloadKW must be true for either version to be accepted.)
645
646		2. Fixed bug in install script that caused more things to be installed
647		than what was requested in config/targets.
648
649		3. Made CM aware of the (_)overload construct so that autoloading
650		works.
651
652		----------------------------------------------------------------------
653		Name: Matthias Blume
654		Date: 2002/03/12 22:03:00 EST
655		Tag: blume-20020312-url
656		Description:
657
658		Forgot to update BOOT and srcarchiveurl.
659
660		----------------------------------------------------------------------
661		Name: Matthias Blume
662		Date: 2002/03/12 17:30:00 EST
663		Tag: blume-20020312-version110392
664		Description:
665
666		Yet another version number bump (because of small changes to the
667		binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
668
669		Changes:
670
671		The new pid generation scheme described a few weeks ago was overly
672		complicated.  I implemented a new mechanism that is simpler and
673		provides a bit more "stability":  Once CM has seen a compilation
674		unit, it keeps its identity constant (as long as you do not delete
675		those crucial CM/GUID/* files).  This means that when you change
676		an interface, compile, then go back to the old interface, and
677		compile again, you arrive at the original pid.
678
679		There now also is a mechanism that instructs CM to use the plain
680		environment hash as a module's pid (effectively making its GUID
681		the empty string).  For this, "noguid" must be specified as an
682		option to the .sml file in question within its .cm file.
683		This is most useful for code that is being generated by tools such
684		as ml-nlffigen (because during development programmers tend to
685		erase the tool's entire output directory tree including CM's cached
686		GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
687		revert to the old, broken behavior of SML/NJ, but in specific cases
688		where there is no danger of interface confusion, its use is ok
689		(I think).
690
691		ml-nlffigen by default generates "noguid" annotations.  They can be
692		turned off by specifying -guid in its command line.
693
694		----------------------------------------------------------------------
695		Name: Lal George
696		Date: 2002/03/12 12 14:42:36 EST
697		Tag: george-20020312-frequency-computation
698		Description:
699
700		Integrated jump chaining and static block frequency into the
701		compiler. More details and numbers later.
702
703		----------------------------------------------------------------------
704		Name: Lal George
705		Date: 2002/03/11 11 22:38:53 EST
706		Tag: george-20020311-jump-chain-elim
707		Description:
708
709		Tested the jump chain elimination on all architectures (except the
710		hppa).  This is on by default right now and is profitable for the
711		alpha and x86, however, it may not be profitable for the sparc and ppc
712		when compiling the compiler.
713
714		The gc test will typically jump to a label at the end of the cluster,
715		where there is another jump to an external cluster containing the actual
716		code to invoke gc. This is to allow factoring of common gc invocation
717		sequences. That is to say, we generate:
718
719		f:
720		testgc
721		ja   L1      % jump if above to L1
722
723		L1:
724		jmp L2
725
726
727		After jump chain elimination the 'ja L1' instructions is converted to
728		'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
729		up being implemented in their long form (if L2 is far away) using:
730
731		jbe     L3      % jump if below or equal to L3
732		jmp     L2
733		L3:
734		...
735
736
737		For large compilation units L2  may be far away.
738
739
740		----------------------------------------------------------------------
741		Name: Matthias Blume
742		Date: 2002/03/11 13:30:00 EST
743		Tag: blume-20020311-mltreeeval
744		Description:
745
746		A functor parameter was missing.
747
748		----------------------------------------------------------------------
749		Name: Allen Leung
750		Date: 2002/03/11 10:30:00 EST
751		Tag: leunga-20020311-runtime-string0
752		Description:
753
754		The representation of the empty string now points to a
755		legal null terminated C string instead of unit.  It is now possible
756		to convert an ML string into C string with InlineT.CharVector.getData.
757		This compiles into one single machine instruction.
758
759		----------------------------------------------------------------------
760		Name: Allen Leung
761		Date: 2002/03/10 23:55:00 EST
762		Tag: leunga-20020310-x86-call
763		Description:
764
765		Added machine generation for CALL instruction (relative displacement mode)
766
767		----------------------------------------------------------------------
768		Name: Matthias Blume
769		Date: 2002/03/08 16:05:00
770		Tag: blume-20020308-entrypoints
771		Description:
772
773		Version number bumped to 110.39.1.  NEW BOOTFILES!
774
775		Entrypoints: non-zero offset into a code object where execution should begin.
776
777		- Added the notion of an entrypoint to CodeObj.
778		- Added reading/writing of entrypoint info to Binfile.
779		- Made runtime system bootloader aware of entrypoints.
780		- Use the address of the label of the first function given to mlriscGen
781		as the entrypoint.  This address is currently always 0, but it will
782		not be 0 once we turn on block placement.
783		- Removed the linkage cluster code (which was The Other Way(tm) of dealing
784		with entry points) from mlriscGen.
785
786		----------------------------------------------------------------------
787		Name: Allen Leung
788		Date: 2002/03/07 20:45:00 EST
789		Tag: leunga-20020307-x86-cmov
790		Description:
791
792		Bug fixes for CMOVcc on x86.
793
794		1. Added machine code generation for CMOVcc
795		2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
796		3. CMOVcc cannot have an immediate operand as argument.
797
798		----------------------------------------------------------------------
799		Name: Matthias Blume
800		Date: 2002/03/07 16:15:00 EST
801		Tag: blume-20020307-controls
802		Description:
803
804		This is a very large but mostly boring patch which makes (almost)
805		every tuneable compiler knob (i.e., pretty much everything under
806		Control.* plus a few other things) configurable via both the command
807		line and environment variables in the style CM did its configuration
808		until now.
809
810		Try starting sml with '-h' (or, if you are brave, '-H')
811
812		To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
813		implements the underlying generic mechanism.
814
815		The interface to some of the existing such facilities has changed somewhat.
816		For example, the MLRiscControl module now provides mkFoo instead of getFoo.
817		(The getFoo interface is still there for backward-compatibility, but its
818		use is deprecated.)
819
820		The ml-build script passes -Cxxx=yyy command-line arguments through so
821		that one can now twiddle the compiler settings when using this "batch"
822		compiler.
823
824		TODO items:
825
826		We should go through and throw out all controls that are no longer
827		connected to anything.  Moreover, we should go through and provide
828		meaningful (and correct!) documentation strings for those controls
829		that still are connected.
830
831		Currently, multiple calls to Controls.new are accepted (only the first
832		has any effect).  Eventually we should make sure that every control
833		is being made (via Controls.new) exactly once.  Future access can then
834		be done using Controls.acc.
835
836		Finally, it would probably be a good idea to use the getter-setter
837		interface to controls rather than ref cells.  For the time being, both
838		styles are provided by the Controls module, but getter-setter pairs are
839		better if thread-safety is of any concern because they can be wrapped.
840
841		*****************************************
842
843		One bug fix: The function blockPlacement in three of the MLRISC
844		backpatch files used to be hard-wired to one of two possibilities at
845		link time (according to the value of the placementFlag).  But (I
846		think) it should rather sense the flag every time.
847
848		*****************************************
849
850		Other assorted changes (by other people who did not supply a HISTORY entry):
851
852		1. the cross-module inliner now works much better (Monnier)
853		2. representation of weights, frequencies, and probabilities in MLRISC
854		changed in preparation of using those for weighted block placement
855		(Reppy, George)
856
857		----------------------------------------------------------------------
858		Name: Lal George
859		Date: 2002/03/07 14:44:24 EST 2002
860		Tag: george-20020307-weighted-block-placement
861
862		Tested the weighted block placement optimization on all architectures
863		(except the hppa) using AMPL to generate the block and edge frequencies.
864		Changes were required in the machine properties to correctly
865		categorize trap instructions. There is an MLRISC flag
866		"weighted-block-placement" that can be used to enable weighted block
867		placement, but this will be ineffective without block/edge
868		frequencies (coming soon).
869
870
871		----------------------------------------------------------------------
872		Name: Lal George
873		Date: 2002/03/05 17:24:48 EST
874		Tag: george-20020305-linkage-cluster
875
876		In order to support the block placement optimization, a new cluster
877		is generated as the very first cluster (called the linkage cluster).
878		It contains a single jump to the 'real' entry point for the compilation
879		unit. Block placement has no effect on the linkage cluster itself, but
880		all the other clusters  have full freedom in the manner in which they
881		reorder blocks or functions.
882
883		On the x86 the typical linkage code that is generated is:
884		----------------------
885		.align 2
886		L0:
887		addl    $L1-L0, 72(%esp)
888		jmp     L1
889
890
891		.align  2
892		L1:
893		----------------------
894
895		72(%esp) is the memory location for the stdlink register. This
896		must contain the address of the CPS function being called. In the
897		above example, it contains the address of  L0; before
898		calling L1 (the real entry point for the compilation unit), it
899		must contain the address for L1, and hence
900
901		addl $L1-L0, 72(%esp)
902
903		I have tested this on all architectures except the hppa.The increase
904		in code size is of course negligible
905
906		----------------------------------------------------------------------
907		Name: Allen Leung
908		Date: 2002/03/03 13:20:00 EST
909		Tag: leunga-20020303-mlrisc-tools
910
911		Added #[ ... ] expressions to mlrisc tools
912
913		----------------------------------------------------------------------
914		Name: Matthias Blume
915		Date: 2002/02/27 12:29:00 EST
916		Tag: blume-20020227-cdebug
917		Description:
918
919		- made types in structure C and C_Debug to be equal
920		- got rid of code duplication (c-int.sml vs. c-int-debug.sml)
921		- there no longer is a C_Int_Debug (C_Debug is directly derived from C)
922
923		----------------------------------------------------------------------
924		Name: Matthias Blume
925		Date: 2002/02/26 12:00:00 EST
926		Tag: blume-20020226-ffi
927		Description:
928
929		1. Fixed a minor bug in CM's "noweb" tool:
930		If numbering is turned off, then truly don't number (i.e., do not
931		supply the -L option to noweb).  The previous behavior was to supply
932		-L'' -- which caused noweb to use the "default" line numbering scheme.
933		Thanks to Chris Richards for pointing this out (and supplying the fix).
934
935		2. Once again, I reworked some aspects of the FFI:
936
937		A. The incomplete/complete type business:
938
939		- Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
940		gone!
941		- ML types representing an incomplete type are now *equal* to
942		ML types representing their corresponding complete types (just like
943		in C).  This is still safe because ml-nlffigen will not generate
944		RTTI for incomplete types, nor will it generate functions that
945		require access to such RTTI.   But when ML code generated from both
946		incomplete and complete versions of the C type meet, the ML types
947		are trivially interoperable.
948
949		NOTE:  These changes restore the full generality of the translation
950		(which was previously lost when I eliminated functorization)!
951
952		B. Enum types:
953
954		- Structure C now has a type constructor "enum" that is similar to
955		how the "su" constructor works.  However, "enum" is not a phantom
956		type because each "T enum" has values (and is isomorphic to
957		MLRep.Signed.int).
958		- There are generic access operations for enum objects (using
959		MLRep.Signed.int).
960		- ml-nlffigen will generate a structure E_foo for each "enum foo".
961		* The structure contains the definition of type "mlrep" (the ML-side
962		representation type of the enum).  Normally, mlrep is the same
963		as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
964		then mlrep will be defined as a datatype -- thus facilitating
965		pattern matching on mlrep values.
966		("-ec" will be suppressed if there are duplicate values in an
967		enumeration.)
968		* Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
969		will be generated for each C enum constant xxx.
970		* Conversion functions m2i and i2m convert between mlrep and
971		MLRep.Signed.int.  (Without "-ec", these functions are identities.)
972		* Coversion functions c and ml convert between mlrep and "tag enum".
973		* Access functions (get/set) fetch and store mlrep values.
974		- By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
975		enumerations are merged into one single enumeration represented by
976		structure E_'.
977
978		----------------------------------------------------------------------
979	Name: Allen Leung	Name: Allen Leung
980	Date: 2002/02/25 04:45:00 EST	Date: 2002/02/25 04:45:00 EST
981	Tag: leunga-20020225-cps-spill	Tag: leunga-20020225-cps-spill
#	Line 69	Line 1032
1032	TOTAL                                   2375.26u  57.21s  48.00g	TOTAL                                   2375.26u  57.21s  48.00g
1033
1034	As you can see the old cps spill module suffers from some serious	As you can see the old cps spill module suffers from some serious
1035	performance problem but since I cannot decipher the old code fully,	performance problem.  But since I cannot decipher the old code fully,
1036	innstead of patching the problems up, I'm reimplementing it	instead of patching the problems up, I'm reimplementing it
1037	with a different algorithm.  The new code is more modular,	with a different algorithm.  The new code is more modular,
1038	smaller when compiled, and substantially faster	smaller when compiled, and substantially faster
1039	(O(n log n) time and O(n) space).  Timing of the new spill module:	(O(n log n) time and O(n) space).  Timing of the new spill module:

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.1094   changed lines   Added in v.1211

---

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