Diff of /sml/trunk/HISTORY

Parent Directory | Revision Log | Patch

	revision 815, Fri May 4 05:09:10 2001 UTC	revision 1249, Tue Jun 25 21:23:40 2002 UTC
#	Line 8	Line 8
8	The form of an entry should be:	The form of an entry should be:
9
10	Name:	Name:
11	Date:	Date: yyyy/mm/dd
12	Tag: <post-commit CVS tag>	Tag: <post-commit CVS tag>
13	Description:	Description:
14
15		----------------------------------------------------------------------
16		Name: Matthias Blume
17		Date: 2002/06/25 17:23:00 EDT
18		Tag: blume-20020625-fpmax
19		Description:
20
21		Fixed a bug in translation of INLMAX (and INLMIN) for the floating-point
22		case.  (The sense of the isNaN test was reversed -- which made min and
23		max always return their first argument.)
24
25		----------------------------------------------------------------------
26		Name: Matthias Blume
27		Date: 2002/06/11
28		Tag: blume-20020611-unixpath
29		Description:
30
31		Back-ported OS.Path.{from,to}UnixPath from idlbasis-devel branch.
32
33		----------------------------------------------------------------------
34		Name: Matthias Blume
35		Date: 2002/06/10 16:35:00 EDT
36		Tag: blume-20020610-ieeereal
37		Description:
38
39		I back-ported my implementation of IEEEReal.fromString from the
40		idlbasis-devel branch so that we can test it.
41
42		Another small change is that ppDec tries to give more information
43		than just "<sig>" in the case of functors.  However, this code is
44		broken in some mysterious way if the functor's body's signature
45		has not been declared by ascription but gets inferred from the
46		implementation.  This needs fixing...
47
48		----------------------------------------------------------------------
49		Name: Matthias Blume
50		Date: 2002/05/31
51		Tag: blume-20020531-btrace-mode
52		Description:
53
54		Resurrected SMLofNJ.Internals.BTrace.mode.  (It accidentally fell by
55		the wayside when I switched over to using Controls everywhere.)
56
57		----------------------------------------------------------------------
58		Name: Lal George
59		Date: 2002/05/23 12:21:40 EDT
60		Tag: george-20020523-visual-labels
61		Description:
62
63		Labels are now displayed in the graphical output to make
64		the fall-through and target blocks obvious.
65
66		----------------------------------------------------------------------
67		Name: Matthias Blume
68		Date: 2002/05/22 11:03:00 EDT
69		Tag: blume-20020522-shrink
70		Description:
71
72		John tweaked yesterday's fix for 1131 to handle an out-of-memory
73		situation that comes up when allocating huge arrays.
74
75		----------------------------------------------------------------------
76		Name: Matthias Blume
77		Date: 2002/05/21 16:00:00 EDT
78		Tag: Release_110_40
79		Description:
80
81		New working release (110.40).  New bootfiles.
82
83		[Also: John Reppy fixed GC bug 1131.]
84
85		----------------------------------------------------------------------
86		Name: Matthias Blume
87		Date: 2002/05/21 12:35:00 EDT
88		Tag: blume-20020521-cmdoc
89		Description:
90
91		CM documentation update.
92
93		----------------------------------------------------------------------
94		Name: Matthias Blume
95		Date: 2002/05/21 10:55:00 EDT
96		Tag: blume-20020521-misc
97		Description:
98
99		- John tweaked runtime to be silent on heap export (except when
100		GC messages are on).
101		- I added a few more things (cross-compiling versions of CMB) to
102		config/preloads (as suggestions).
103
104		----------------------------------------------------------------------
105		Name: Matthias Blume
106		Date: 2002/05/20 22:25:00 EDT
107		Tag: blume-20020520-controls
108		Description:
109
110		- Added ControlUtil structure to control-lib.cm.
111		- Use it throughout.
112		- Used Controls facility to define MLRISC controls (as opposed to
113		registering MLRISC control ref cells with Controls after the fact)
114		- Fixed messed-up controls priorities.
115
116		* Removed again all the stuff from config/preloads that one wouldn't
117		be able to preload at the time the initial heap image is built.
118		(Many libraries, e.g., CML, do not exist yet at this time.  The
119		only libraries that can be preloaded via config/preloads are those
120		that come bundled with the bootfiles.)
121
122		----------------------------------------------------------------------
123		Name: Matthias Blume
124		Date: 2002/05/20 10:59:00 EDT
125		Tag: blume-20020520-preloads
126		Description:
127
128		Added a lot of commented-out suggestions for things to be included
129		in config/preloads.
130
131		----------------------------------------------------------------------
132		Name: Allen Leung
133		Date: 2002/05/18 14:20:00 EDT
134		Tag: leunga-20020518-mdl
135		Description:
136
137		o Made the mdl tool stuff compile and run again.
138		o I've disabled all the stuff that depends on RTL specifications; they
139		are all badly broken anyway.
140
141		----------------------------------------------------------------------
142		Name: Matthias Blume
143		Date: 2002/05/17 16:49:00 EDT
144		Tag: blume-20020517-controls
145		Description:
146
147		0. John Reppy made several modifications to the SML/NJ library.
148		In particular, there is a shiny new controls-lib.cm.
149
150		1. Pushed new controls interface through compiler so that everything
151		compiles again.
152
153		2. Added FormatComb and FORMAT_COMB to the CML version of the
154		SML/NJ library (so that CML compiles again).
155
156		3. Modified init scripts because XXX_DEFAULT environment variables
157		are no longer with us.  (Boot-time initialization is now done
158		using the same environment variables that are also used for
159		startup-time initialization of controls.)
160
161		----------------------------------------------------------------------
162		Name: Lal George
163		Date: 2002/05/15 09:20:10 EDT
164		Tag: george-20020515-pseudo-op-decls
165		Description:
166
167		All pseudo-ops emitted before the first segment declaration
168		such as TEXT, DATA, and BSS directives are assumed to be global
169		declarations and are emitted first in the assembly file. This is
170		useful in a number of situations where one has pseudo-ops that are not
171		specific to any segment, and also works around the constraint that one
172		cannot have client pseudo-ops in the TEXT segment.
173
174		Because no segment is associated with these declarations it is
175		an error to allocate any space or objects before the first segment
176		directive and an exception will be raised. However, we cannot make
177		this check for client pseudo-ops.
178
179		These top level declarations are a field in the CFG graph_info.
180		In theory you can continue to add to this field after the CFG has been
181		built -- provided you know what you are doing;-)
182
183		----------------------------------------------------------------------
184		Name: Matthias Blume
185		Date: 2002/05/13 16:40:00 EDT
186		Tag: blume-20020513-pp-etc
187		Description:
188
189		A few minor bugfixes:
190
191		- Stopgap measure for bug recently reported by Elsa Gunter (ppDec).
192		(Bogus printouts for redefined bindings still occur.  Compiler
193		bug should no longer occur now.  We need to redo the prettyprinter
194		from scratch.)
195
196		- CM pathname printer now also adds escape sequences for ( and )
197
198		- commend and docu fixes for ml-nlffi
199
200		----------------------------------------------------------------------
201		Name: Matthias Blume
202		Date: 2002/05/10 16:40:00 EDT
203		Tag: blume-20020510-erg-textio
204		Description:
205
206		Applied the following bugfix provided by Emden Gansner:
207
208		Output is corrupted when outputSubstr is used rather than output.
209
210		The problem occurs when a substring
211
212		ss = (s, dataStart, dataLen)
213
214		where dataStart > 0, fills a stream buffer with avail bytes left.
215		avail bytes of s, starting at index dataStart, are copied into the
216		buffer, the buffer is flushed, and then the remaining dataLen-avail
217		bytes of ss are copied into the beginning of the buffer. Instead of
218		starting this copy at index dataStart+avail in s, the current code
219		starts the copy at index avail.
220
221		Fix:
222		In text-io-fn.sml, change line 695 from
223		val needsFlush = copyVec(v, avail, dataLen-avail, buf, 0)
224		to
225		val needsFlush = copyVec(v, dataStart+avail, dataLen-avail, buf, 0)
226
227		----------------------------------------------------------------------
228		Name: Matthias Blume
229		Date: 2002/04/12 13:55:00 EDT
230		Tag: blume-20020412-assyntax
231		Description:
232
233		1. Grabbed newer assyntax.h from the XFree86 project.
234		2. Fiddled with how to compile X86.prim.asm without warnings.
235		3. (Very) Minor cleanup in CM.
236
237		----------------------------------------------------------------------
238		Name: Matthias Blume
239		Date: 2002/04/01 (no joke!) 17:07:00 EST
240		Tag: blume-20020401-x86div
241		Description:
242
243		Added full support for div/mod/rem/quot on the x86, using the machine
244		instruction's two results (without clumsily recomputing the remainder)
245		directly where appropriate.
246
247		Some more extensive power-of-two support was added to the x86 instruction
248		selector (avoiding expensive divs, mods, and muls where they can be
249		replaced with cheaper shifts and masks).  However, this sort of thing
250		ought to be done earlier, e.g., within the CPS optimizer so that
251		all architectures benefit from it.
252
253		The compiler compiles to a fixed point, but changes might be somewhat
254		fragile nevertheless.  Please, report any strange things that you might
255		see wrt. div/mod/quot/rem...
256
257		----------------------------------------------------------------------
258		Name: Matthias Blume
259		Date: 2002/03/29 17:22:00
260		Tag: blume-20020329-div
261		Description:
262
263		Fixed my broken div/mod logic.  Unfortunately, this means that the
264		inline code for div/mod now has one more comparison than before.
265		Fast paths (quotient > 0 or remainder = 0) are not affected, though.
266		The problem was with quotient = 0, because that alone does not tell
267		us which way the rounding went.  One then has to look at whether
268		remainder and divisor have the same sign...  :(
269
270		Anyway, I replaced the bootfiles with fresh ones...
271
272		----------------------------------------------------------------------
273		Name: Matthias Blume
274		Date: 2002/03/29 14:10:00 EST
275		Tag: blume-20020329-inlprims
276		Description:
277
278		NEW BOOTFILES!!!    Version number bumped to 110.39.3.
279
280		Primops have changed. This means that the bin/boot-file formats have
281		changed as well.
282
283		To make sure that there is no confusion, I made a new version.
284
285
286		CHANGES:
287
288		* removed REMT from mltree (remainder should never overflow).
289
290		* added primops to deal with divisions of all flavors to the frontend
291
292		* handled these primops all the way through so they map to their respective
293		MLRISC support
294
295		* used these primops in the implementation of Int, Int32, Word, Word32
296
297		* removed INLDIV, INLMOD, and INLREM as they are no longer necessary
298
299		* parameterized INLMIN, INLMAX, and INLABS by a numkind
300
301		* translate.sml now deals with all flavors of INL{MIN,MAX,ABS}, including
302		floating point
303
304		* used INL{MIN,MAX,ABS} in the implementation of Int, Int32, Word, Word32,
305		and Real (but Real.abs maps to a separate floating-point-only primop)
306
307
308		TODO items:
309
310		* Hacked Alpha32 instruction selection, disabling the selection of REMx
311		instructions because the machine instruction encoder cannot handle
312		them.  (Hppa, PPC, and Sparc instruction selection did not handle
313		REM in the first place, and REM is supported by the x86 machine coder.)
314
315		* Handle DIV and MOD with DIV_TO_NEGINF directly in the x86 instruction
316		selection phase.  (The two can be streamlined because the hardware
317		delivers both quotient and remainder at the same time anyway.)
318
319		* Think about what to do with "valOf(Int32.minInt) div ~1" and friends.
320		(Currently the behavior is inconsistent both across architectures and
321		wrt. the draft Basis spec.)
322
323		* Word8 should eventually be handled natively, too.
324
325		* There seems to be one serious bug in mltree-gen.sml.  It appears, though,
326		as if there currently is no execution path that could trigger it in
327		SML/NJ.  (The assumptions underlying functions arith and promotable do not
328		hold for things like multiplication and division.)
329
330		----------------------------------------------------------------------
331		Name: Matthias Blume
332		Date: 2002/03/27 16:27:00 EST
333		Tag: blume-20020327-mlrisc-divisions
334		Description:
335
336		Added support for all four division operations (ML's div, mod, quot,
337		and rem) to MLRISC.  In the course of doing so, I also rationalized
338		the naming (no more annoying switch-around of DIV and QUOT), by
339		parameterizing the operation by div_rounding_mode (which can be either
340		DIV_TO_ZERO or DIV_TO_NEGINF).
341
342		The generic MLTreeGen functor takes care of compiling all four
343		operations down to only round-to-zero div.
344
345		Missing pieces:
346
347		* Doing something smarter than relying on MLTreeGen on architectures
348		like, e.g., the x86 where hardware division delivers both quotient and
349		remainder at the same time.  With this, the implementation of the
350		round-to-neginf operations could be further streamlined.
351
352		* Remove inlining support for div/mod/rem from the frontend and replace it
353		with primops that get carried through to the backend.  Do this for all
354		int and word types.
355
356		----------------------------------------------------------------------
357		Name: Matthias Blume
358		Date: 2002/03/25 17:25:00 EST
359		Tag: blume-20020325-divmod
360		Description:
361
362		I improved (hopefully without breaking them) the implementation of Int.div,
363		Int.mod, and Int.rem.   For this, the code in translate.sml now takes
364		advantage of the following observations:
365
366		Let  q = x quot y      r = x rem y
367		d = x div  y      m = x mod y
368
369		where "quot" is the round-to-zero version of integer division that
370		hardware usually provides.  Then we have:
371
372		r = x - q * y        where neither the * nor the - will overflow
373		d = if q >= 0 orelse x = q * y then q else q - 1
374		where neither the * nor the - will overflow
375		m = if q >= 0 orelse r = 0 then r else r + y
376		where the + will not overflow
377
378		This results in substantial simplification of the generated code.
379		The following table shows the number of CFG nodes and edges generated
380		for
381		fun f (x, y) = x OPER y
382		(* with OPER \in div, mod, quot, rem *)
383
384
385		OPER | nodes(old) | edges(old) | nodes(new) | edges(new)
386		--------------------------------------------------------
387		div |         24 |         39 |         12 |         16
388		mod |         41 |         71 |         12 |         16
389		quot |          8 |         10 |          8 |         10
390		rem |         10 |         14 |          8 |         10
391
392
393		----------------------------------------------------------------------
394		Name: Matthias Blume
395		Date: 2002/03/25 22:06:00 EST
396		Tag: blume-20020325-cprotobug
397		Description:
398
399		Fixed a bug in cproto (c prototype decoder).
400
401		----------------------------------------------------------------------
402		Name: Matthias Blume
403		Date: 2002/03/25 16:00:00 EST
404		Tag: blume-20020325-raw-primops
405		Description:
406
407		I did some cleanup to Allen's new primop code and
408		replaced yesterday's bootfiles with new ones.
409		(But they are stored in the same place.)
410
411		----------------------------------------------------------------------
412		Name: Matthias Blume
413		Date: 2002/03/24 22:40:00 EST
414		Tag: blume-20020324-bootfiles
415		Description:
416
417		Made the bootfiles that Allen asked for.
418
419		----------------------------------------------------------------------
420		Name: Allen Leung
421		Date: 2002/03/23 15:50:00 EST
422		Tag: leunga-20020323-flint-cps-rcc-primops
423		Description:
424
425		1. Changes to FLINT primops:
426
427		(* make a call to a C-function;
428		* The primop carries C function prototype information and specifies
429		* which of its (ML-) arguments are floating point. C prototype
430		* information is for use by the backend, ML information is for
431		* use by the CPS converter. *)
432		| RAW_CCALL of { c_proto: CTypes.c_proto,
433		ml_args: ccall_type list,
434		ml_res_opt: ccall_type option,
435		reentrant : bool
436		} option
437		(* Allocate uninitialized storage on the heap.
438		* The record is meant to hold short-lived C objects, i.e., they
439		* are not ML pointers.  With the tag, the representation is
440		* the same as RECORD with tag tag_raw32 (sz=4), or tag_fblock (sz=8)
441		*)
442		| RAW_RECORD of {tag:bool,sz:int}
443		and ccall_type = CCALL_INT32 | CCALL_REAL64 | CCALL_ML_PTR
444
445		2.  These CPS primops are now overloaded:
446
447		rawload of {kind:numkind}
448		rawstore of {kind:numkind}
449
450		The one argument form is:
451
452		rawload {kind} address
453
454		The two argument form is:
455
456		rawload {kind} [ml object, byte-offset]
457
458		3. RAW_CCALL/RCC now takes two extra arguments:
459
460		a. The first is whether the C call is reentrant, i.e., whether
461		ML state should be saved and restored.
462		b. The second argument is a string argument specifying the name of
463		library and the C function.
464
465		These things are currently not handled in the code generator, yet.
466
467		4. In CProto,
468
469		An encoding type of "bool" means "ml object" and is mapped into
470		C prototype of PTR.  Note that "bool" is different than "string",
471		even though "string" is also mapped into PTR, because "bool"
472		is assigned an CPS type of BOGt, while "string" is assigned INT32t.
473
474		5. Pickler/unpicker
475
476		Changed to handle RAW_RECORD and newest RAW_CCALL
477
478		6. MLRiscGen,
479
480		1. Changed to handle the new rawload/rawstore/rawrecord operators.
481		2. Code for handling C Calls has been moved to a new module CPSCCalls,
482		in the file CodeGen/cpscompile/cps-c-calls.sml
483
484		7. Added the conditional move operator
485
486		condmove of branch
487
488		to cps.  Generation of this is still buggy so it is currently
489		disabled.
490
491		----------------------------------------------------------------------
492		Name: Lal George
493		Date: 2002/03/22 14:18:25 EST
494		Tag: george-20020322-cps-branch-prob
495		Description:
496
497		Implemented the Ball-Larus branch prediction-heuristics, and
498		incorporated graphical viewers for control flow graphs.
499
500		Ball-Larus Heuristics:
501		---------------------
502		See the file compiler/CodeGen/cpscompile/cpsBranchProb.sml.
503
504		By design it uses the Dempster-Shafer theory for combining
505		probabilities.  For example, in the function:
506
507		fun f(n,acc) = if n = 0 then acc else f(n-1, n*acc)
508
509		the ball-larus heuristics predicts that the n=0 is unlikely
510		(OH-heuristic), and the 'then' branch is unlikely because of the
511		RH-heuristic -- giving the 'then' branch an even lower combined
512		probability using the Dempster-Shafer theory.
513
514		Finally, John Reppy's loop analysis in MLRISC, further lowers the
515		probability of the 'then' branch because of the loop in the else
516		branch.
517
518
519		Graphical Viewing:
520		------------------
521		I merely plugged in Allen's graphical viewers into the compiler. The
522		additional code is not much. At the top level, saying:
523
524		Control.MLRISC.getFlag "cfg-graphical-view" := true;
525
526		will display the graphical view of the control flow graph just before
527		back-patching.  daVinci must be in your path for this to work. If
528		daVinci is not available, then the default viewer can be changed
529		using:
530
531		Control.MLRISC.getString "viewer"
532
533		which can be set to "dot" or "vcg" for the corresponding viewers. Of
534		course, these viewers must be in your path.
535
536		The above will display the compilation unit at the level of clusters,
537		many of which are small, boring, and un-interesting. Also setting:
538
539		Control.MLRISC.getInt "cfg-graphical-view_size"
540
541		will display clusters that are larger than the value set by the above.
542
543
544		----------------------------------------------------------------------
545		Name: Matthias Blume
546		Date: 2002/03/21 22:20:00 EST
547		Tag: blume-20020321-kmp-bugfix
548		Description:
549
550		Changed the interface to the KMP routine in PreString and fixed
551		a minor bug in one place where it was used.
552
553		----------------------------------------------------------------------
554		Name: Allen Leung
555		Date: 2002/03/21 20:30:00 EST
556		Tag: leunga-20020321-cfg
557		Description:
558
559		Fixed a potential problem in cfg edge splitting.
560
561		----------------------------------------------------------------------
562		Name: Allen Leung
563		Date: 2002/03/21 17:15:00 EST
564		Tag: leunga-20020321-x86-fp-cfg
565		Description:
566
567		1. Recoded the buggy parts of x86-fp.
568
569		a. All the block reordering code has been removed.
570		We now depend on the block placement phases to do this work.
571
572		b. Critical edge splitting code has been simplified and moved into the
573		CFG modules, as where they belong.
574
575		Both of these were quite buggy and complex.  The code is now much, much
576		simpler.
577
578		2. X86 backend.
579
580		a. Added instructions for 64-bit support.  Instruction selection for
581		64-bit has not been committed, however, since that
582		requires changes to MLTREE which haven't been approved by
583		Lal and John.
584
585		b. Added support for FUCOMI and FUCOMIP when generating code for
586		PentiumPro and above.  We only generate these instructions in
587		the fast-fp mode.
588
589		c. Added cases for JP and JNP in X86FreqProps.
590
591		3. CFG
592
593		CFG now has a bunch of methods for edge splitting and merging.
594
595		4. Machine description.
596
597		John's simplification of MLTREE_BASIS.fcond broke a few machine
598		description things:
599
600		rtl-build.{sig,sml} and hppa.mdl fixed.
601
602		NOTE: the machine description stuff in the repository is still broken.
603		Again, I can't put my fixes in because that involves
604		changes to MLTREE.
605
606		----------------------------------------------------------------------
607		Name: Matthias Blume
608		Date: 2002/03/20 15:55:00 EST
609		Tag: blume-20020320-kmp
610		Description:
611
612		Implemented Knuth-Morris-Pratt string matching in PreString and used
613		it for String.isSubstring, Substring.isSubstring, and
614		Substring.position.
615
616		(Might need some stress-testing.  Simple examples worked fine.)
617
618		----------------------------------------------------------------------
619		Name: Matthias Blume
620		Date: 2002/03/19 16:37:00 EST
621		Tag: blume-20020319-witnesses
622		Description:
623
624		Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
625
626		This implements a generic mechanism for changing constness qualifiers
627		anywhere within big C types without resorting to outright "casts".
628		(So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
629		modify the constness at the outermost level.)
630		The implementation of "convert" is based on the idea of "witness"
631		values -- values that are not used by the operation but whose types
632		"testify" to their applicability.  On the implementation side, "convert"
633		is simply a projection (returning its second curried argument).  With
634		cross-module inlining, it should not result in any machine code being
635		generated.
636
637		----------------------------------------------------------------------
638		Name: Matthias Blume
639		Date: 2002/03/15 16:40:00 EST
640		Tag: blume-20020315-basis
641		Description:
642
643		Provided (preliminary?) implementations for
644
645		{String,Substring}.{concatWith,isSuffix,isSubstring}
646
647		and
648
649		Substring.full
650
651		Those are in the Basis spec but they were missing in SML/NJ.
652
653		----------------------------------------------------------------------
654		Name: Matthias Blume
655		Date: 2002/03/14 21:30:00 EST
656		Tag: blume-20020314-controls
657		Description:
658
659		Controls:
660		---------
661
662		1. Factored out the recently-added Controls : CONTROLS stuff and put
663		it into its own library $/controls-lib.cm.  The source tree for
664		this is under src/smlnj-lib/Controls.
665
666		2. Changed the names of types and functions in this interface, so they
667		make a bit more "sense":
668
669		module -> registry
670		'a registry -> 'a group
671
672		3. The interface now deals in ref cells only.  The getter/setter interface
673		is (mostly) gone.
674
675		4. Added a function that lets one register an already-existing ref cell.
676
677		5. Made the corresponding modifications to the rest of the code so that
678		everything compiles again.
679
680		6. Changed the implementation of Controls.MLRISC back to something closer
681		to the original.  In particular, this module (and therefore MLRISC)
682		does not depend on Controls.  There now is some link-time code in
683		int-sys.sml that registers the MLRISC controls with the Controls
684		module.
685
686		CM:
687		---
688
689		* One can now specify the lambda-split aggressiveness in init.cmi.
690
691		----------------------------------------------------------------------
692		Name: Allen Leung
693		Date: 2002/03/13 17:30:00 EST
694		Tag: leunga-20020313-x86-fp-unary
695		Description:
696
697		Bug fix for:
698
699		> leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
700		> Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
701		> - fun f(x,(y,z)) = Real.~ y;
702		> [autoloading]
703		> [autoloading done]
704		>       fchsl   (%eax), 184(%esp)
705		> Error: MLRisc bug: X86MCEmitter.emitInstr
706		>
707		> uncaught exception Error
708		>   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
709
710		The problem was that the code generator did not generate any fp registers
711		in this case, and the ra didn't know that it needed to run the X86FP phase to
712		translate the pseudo fp instruction.   This only happened with unary fp
713		operators in certain situations.
714
715		----------------------------------------------------------------------
716		Name: Matthias Blume
717		Date: 2002/03/13 14:00:00 EST
718		Tag: blume-20020313-overload-etc
719		Description:
720
721		1. Added _overload as a synonym for overload for backward compatibility.
722		(Control.overloadKW must be true for either version to be accepted.)
723
724		2. Fixed bug in install script that caused more things to be installed
725		than what was requested in config/targets.
726
727		3. Made CM aware of the (_)overload construct so that autoloading
728		works.
729
730		----------------------------------------------------------------------
731		Name: Matthias Blume
732		Date: 2002/03/12 22:03:00 EST
733		Tag: blume-20020312-url
734		Description:
735
736		Forgot to update BOOT and srcarchiveurl.
737
738		----------------------------------------------------------------------
739		Name: Matthias Blume
740		Date: 2002/03/12 17:30:00 EST
741		Tag: blume-20020312-version110392
742		Description:
743
744		Yet another version number bump (because of small changes to the
745		binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
746
747		Changes:
748
749		The new pid generation scheme described a few weeks ago was overly
750		complicated.  I implemented a new mechanism that is simpler and
751		provides a bit more "stability":  Once CM has seen a compilation
752		unit, it keeps its identity constant (as long as you do not delete
753		those crucial CM/GUID/* files).  This means that when you change
754		an interface, compile, then go back to the old interface, and
755		compile again, you arrive at the original pid.
756
757		There now also is a mechanism that instructs CM to use the plain
758		environment hash as a module's pid (effectively making its GUID
759		the empty string).  For this, "noguid" must be specified as an
760		option to the .sml file in question within its .cm file.
761		This is most useful for code that is being generated by tools such
762		as ml-nlffigen (because during development programmers tend to
763		erase the tool's entire output directory tree including CM's cached
764		GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
765		revert to the old, broken behavior of SML/NJ, but in specific cases
766		where there is no danger of interface confusion, its use is ok
767		(I think).
768
769		ml-nlffigen by default generates "noguid" annotations.  They can be
770		turned off by specifying -guid in its command line.
771
772		----------------------------------------------------------------------
773		Name: Lal George
774		Date: 2002/03/12 12 14:42:36 EST
775		Tag: george-20020312-frequency-computation
776		Description:
777
778		Integrated jump chaining and static block frequency into the
779		compiler. More details and numbers later.
780
781		----------------------------------------------------------------------
782		Name: Lal George
783		Date: 2002/03/11 11 22:38:53 EST
784		Tag: george-20020311-jump-chain-elim
785		Description:
786
787		Tested the jump chain elimination on all architectures (except the
788		hppa).  This is on by default right now and is profitable for the
789		alpha and x86, however, it may not be profitable for the sparc and ppc
790		when compiling the compiler.
791
792		The gc test will typically jump to a label at the end of the cluster,
793		where there is another jump to an external cluster containing the actual
794		code to invoke gc. This is to allow factoring of common gc invocation
795		sequences. That is to say, we generate:
796
797		f:
798		testgc
799		ja   L1      % jump if above to L1
800
801		L1:
802		jmp L2
803
804
805		After jump chain elimination the 'ja L1' instructions is converted to
806		'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
807		up being implemented in their long form (if L2 is far away) using:
808
809		jbe     L3      % jump if below or equal to L3
810		jmp     L2
811		L3:
812		...
813
814
815		For large compilation units L2  may be far away.
816
817
818		----------------------------------------------------------------------
819		Name: Matthias Blume
820		Date: 2002/03/11 13:30:00 EST
821		Tag: blume-20020311-mltreeeval
822		Description:
823
824		A functor parameter was missing.
825
826		----------------------------------------------------------------------
827		Name: Allen Leung
828		Date: 2002/03/11 10:30:00 EST
829		Tag: leunga-20020311-runtime-string0
830		Description:
831
832		The representation of the empty string now points to a
833		legal null terminated C string instead of unit.  It is now possible
834		to convert an ML string into C string with InlineT.CharVector.getData.
835		This compiles into one single machine instruction.
836
837		----------------------------------------------------------------------
838		Name: Allen Leung
839		Date: 2002/03/10 23:55:00 EST
840		Tag: leunga-20020310-x86-call
841		Description:
842
843		Added machine generation for CALL instruction (relative displacement mode)
844
845		----------------------------------------------------------------------
846		Name: Matthias Blume
847		Date: 2002/03/08 16:05:00
848		Tag: blume-20020308-entrypoints
849		Description:
850
851		Version number bumped to 110.39.1.  NEW BOOTFILES!
852
853		Entrypoints: non-zero offset into a code object where execution should begin.
854
855		- Added the notion of an entrypoint to CodeObj.
856		- Added reading/writing of entrypoint info to Binfile.
857		- Made runtime system bootloader aware of entrypoints.
858		- Use the address of the label of the first function given to mlriscGen
859		as the entrypoint.  This address is currently always 0, but it will
860		not be 0 once we turn on block placement.
861		- Removed the linkage cluster code (which was The Other Way(tm) of dealing
862		with entry points) from mlriscGen.
863
864		----------------------------------------------------------------------
865		Name: Allen Leung
866		Date: 2002/03/07 20:45:00 EST
867		Tag: leunga-20020307-x86-cmov
868		Description:
869
870		Bug fixes for CMOVcc on x86.
871
872		1. Added machine code generation for CMOVcc
873		2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
874		3. CMOVcc cannot have an immediate operand as argument.
875
876		----------------------------------------------------------------------
877		Name: Matthias Blume
878		Date: 2002/03/07 16:15:00 EST
879		Tag: blume-20020307-controls
880		Description:
881
882		This is a very large but mostly boring patch which makes (almost)
883		every tuneable compiler knob (i.e., pretty much everything under
884		Control.* plus a few other things) configurable via both the command
885		line and environment variables in the style CM did its configuration
886		until now.
887
888		Try starting sml with '-h' (or, if you are brave, '-H')
889
890		To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
891		implements the underlying generic mechanism.
892
893		The interface to some of the existing such facilities has changed somewhat.
894		For example, the MLRiscControl module now provides mkFoo instead of getFoo.
895		(The getFoo interface is still there for backward-compatibility, but its
896		use is deprecated.)
897
898		The ml-build script passes -Cxxx=yyy command-line arguments through so
899		that one can now twiddle the compiler settings when using this "batch"
900		compiler.
901
902		TODO items:
903
904		We should go through and throw out all controls that are no longer
905		connected to anything.  Moreover, we should go through and provide
906		meaningful (and correct!) documentation strings for those controls
907		that still are connected.
908
909		Currently, multiple calls to Controls.new are accepted (only the first
910		has any effect).  Eventually we should make sure that every control
911		is being made (via Controls.new) exactly once.  Future access can then
912		be done using Controls.acc.
913
914		Finally, it would probably be a good idea to use the getter-setter
915		interface to controls rather than ref cells.  For the time being, both
916		styles are provided by the Controls module, but getter-setter pairs are
917		better if thread-safety is of any concern because they can be wrapped.
918
919		*****************************************
920
921		One bug fix: The function blockPlacement in three of the MLRISC
922		backpatch files used to be hard-wired to one of two possibilities at
923		link time (according to the value of the placementFlag).  But (I
924		think) it should rather sense the flag every time.
925
926		*****************************************
927
928		Other assorted changes (by other people who did not supply a HISTORY entry):
929
930		1. the cross-module inliner now works much better (Monnier)
931		2. representation of weights, frequencies, and probabilities in MLRISC
932		changed in preparation of using those for weighted block placement
933		(Reppy, George)
934
935		----------------------------------------------------------------------
936		Name: Lal George
937		Date: 2002/03/07 14:44:24 EST 2002
938		Tag: george-20020307-weighted-block-placement
939
940		Tested the weighted block placement optimization on all architectures
941		(except the hppa) using AMPL to generate the block and edge frequencies.
942		Changes were required in the machine properties to correctly
943		categorize trap instructions. There is an MLRISC flag
944		"weighted-block-placement" that can be used to enable weighted block
945		placement, but this will be ineffective without block/edge
946		frequencies (coming soon).
947
948
949		----------------------------------------------------------------------
950		Name: Lal George
951		Date: 2002/03/05 17:24:48 EST
952		Tag: george-20020305-linkage-cluster
953
954		In order to support the block placement optimization, a new cluster
955		is generated as the very first cluster (called the linkage cluster).
956		It contains a single jump to the 'real' entry point for the compilation
957		unit. Block placement has no effect on the linkage cluster itself, but
958		all the other clusters  have full freedom in the manner in which they
959		reorder blocks or functions.
960
961		On the x86 the typical linkage code that is generated is:
962		----------------------
963		.align 2
964		L0:
965		addl    $L1-L0, 72(%esp)
966		jmp     L1
967
968
969		.align  2
970		L1:
971		----------------------
972
973		72(%esp) is the memory location for the stdlink register. This
974		must contain the address of the CPS function being called. In the
975		above example, it contains the address of  L0; before
976		calling L1 (the real entry point for the compilation unit), it
977		must contain the address for L1, and hence
978
979		addl $L1-L0, 72(%esp)
980
981		I have tested this on all architectures except the hppa.The increase
982		in code size is of course negligible
983
984		----------------------------------------------------------------------
985		Name: Allen Leung
986		Date: 2002/03/03 13:20:00 EST
987		Tag: leunga-20020303-mlrisc-tools
988
989		Added #[ ... ] expressions to mlrisc tools
990
991		----------------------------------------------------------------------
992		Name: Matthias Blume
993		Date: 2002/02/27 12:29:00 EST
994		Tag: blume-20020227-cdebug
995		Description:
996
997		- made types in structure C and C_Debug to be equal
998		- got rid of code duplication (c-int.sml vs. c-int-debug.sml)
999		- there no longer is a C_Int_Debug (C_Debug is directly derived from C)
1000
1001		----------------------------------------------------------------------
1002		Name: Matthias Blume
1003		Date: 2002/02/26 12:00:00 EST
1004		Tag: blume-20020226-ffi
1005		Description:
1006
1007		1. Fixed a minor bug in CM's "noweb" tool:
1008		If numbering is turned off, then truly don't number (i.e., do not
1009		supply the -L option to noweb).  The previous behavior was to supply
1010		-L'' -- which caused noweb to use the "default" line numbering scheme.
1011		Thanks to Chris Richards for pointing this out (and supplying the fix).
1012
1013		2. Once again, I reworked some aspects of the FFI:
1014
1015		A. The incomplete/complete type business:
1016
1017		- Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
1018		gone!
1019		- ML types representing an incomplete type are now *equal* to
1020		ML types representing their corresponding complete types (just like
1021		in C).  This is still safe because ml-nlffigen will not generate
1022		RTTI for incomplete types, nor will it generate functions that
1023		require access to such RTTI.   But when ML code generated from both
1024		incomplete and complete versions of the C type meet, the ML types
1025		are trivially interoperable.
1026
1027		NOTE:  These changes restore the full generality of the translation
1028		(which was previously lost when I eliminated functorization)!
1029
1030		B. Enum types:
1031
1032		- Structure C now has a type constructor "enum" that is similar to
1033		how the "su" constructor works.  However, "enum" is not a phantom
1034		type because each "T enum" has values (and is isomorphic to
1035		MLRep.Signed.int).
1036		- There are generic access operations for enum objects (using
1037		MLRep.Signed.int).
1038		- ml-nlffigen will generate a structure E_foo for each "enum foo".
1039		* The structure contains the definition of type "mlrep" (the ML-side
1040		representation type of the enum).  Normally, mlrep is the same
1041		as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
1042		then mlrep will be defined as a datatype -- thus facilitating
1043		pattern matching on mlrep values.
1044		("-ec" will be suppressed if there are duplicate values in an
1045		enumeration.)
1046		* Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
1047		will be generated for each C enum constant xxx.
1048		* Conversion functions m2i and i2m convert between mlrep and
1049		MLRep.Signed.int.  (Without "-ec", these functions are identities.)
1050		* Coversion functions c and ml convert between mlrep and "tag enum".
1051		* Access functions (get/set) fetch and store mlrep values.
1052		- By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
1053		enumerations are merged into one single enumeration represented by
1054		structure E_'.
1055
1056		----------------------------------------------------------------------
1057		Name: Allen Leung
1058		Date: 2002/02/25 04:45:00 EST
1059		Tag: leunga-20020225-cps-spill
1060
1061		This is a new implementation of the CPS spill phase.
1062		The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
1063		In case of problems, replace it with the old file spill.sml
1064
1065		The current compiler runs into some serious performance problems when
1066		constructing a large record.  This can happen when we try to compile a
1067		structure with many items.  Even a very simple structure like the following
1068		makes the compiler slow down.
1069
1070		structure Foo = struct
1071		val x_1 = 0w1 : Word32.int
1072		val x_2 = 0w2 : Word32.int
1073		val x_3 = 0w3 : Word32.int
1074		...
1075		val x_N = 0wN : Word32.int
1076		end
1077
1078		The following table shows the compile time, from N=1000 to N=4000,
1079		with the old compiler:
1080
1081		N
1082		1000   CPS 100 spill                           0.04u  0.00s  0.00g
1083		MLRISC ra                               0.06u  0.00s  0.05g
1084		(spills = 0 reloads = 0)
1085		TOTAL                                   0.63u  0.07s  0.21g
1086
1087		1100   CPS 100 spill                           8.25u  0.32s  0.64g
1088		MLRISC ra                               5.68u  0.59s  3.93g
1089		(spills = 0 reloads = 0)
1090		TOTAL                                   14.71u  0.99s  4.81g
1091
1092		1500   CPS 100 spill                           58.55u  2.34s  1.74g
1093		MLRISC ra                               5.54u  0.65s  3.91g
1094		(spills = 543 reloads = 1082)
1095		TOTAL                                   65.40u  3.13s  6.00g
1096
1097		2000   CPS 100 spill                           126.69u  4.84s  3.08g
1098		MLRISC ra                               0.80u  0.10s  0.55g
1099		(spills = 42 reloads = 84)
1100		TOTAL                                   129.42u  5.10s  4.13g
1101
1102		3000   CPS 100 spill                           675.59u  19.03s  11.64g
1103		MLRISC ra                               2.69u  0.27s  1.38g
1104		(spills = 62 reloads = 124)
1105		TOTAL                                   682.48u  19.61s  13.99g
1106
1107		4000   CPS 100 spill                           2362.82u  56.28s  43.60g
1108		MLRISC ra                               4.96u  0.27s  2.72g
1109		(spills = 85 reloads = 170)
1110		TOTAL                                   2375.26u  57.21s  48.00g
1111
1112		As you can see the old cps spill module suffers from some serious
1113		performance problem.  But since I cannot decipher the old code fully,
1114		instead of patching the problems up, I'm reimplementing it
1115		with a different algorithm.  The new code is more modular,
1116		smaller when compiled, and substantially faster
1117		(O(n log n) time and O(n) space).  Timing of the new spill module:
1118
1119		4000  CPS 100 spill                           0.02u  0.00s  0.00g
1120		MLRISC ra                               0.25u  0.02s  0.15g
1121		(spills=1 reloads=3)
1122		TOTAL                                   7.74u  0.34s  1.62g
1123
1124		Implementation details:
1125
1126		As far as I can tell, the purpose of the CPS spill module is to make sure the
1127		number of live variables at any program point (the bandwidth)
1128		does not exceed a certain limit, which is determined by the
1129		size of the spill area.
1130
1131		When the bandwidth is too large, we decrease the register pressure by
1132		packing live variables into spill records.  How we achieve this is
1133		completely different than what we did in the old code.
1134
1135		First, there is something about the MLRiscGen code generator
1136		that we should be aware of:
1137
1138		o MLRiscGen performs code motion!
1139
1140		In particular, it will move floating point computations and
1141		address computations involving only the heap pointer to
1142		their use sites (if there is only a single use).
1143		What this means is that if we have a CPS record construction
1144		statement
1145
1146		RECORD(k,vl,w,e)
1147
1148		we should never count the new record address w as live if w
1149		has only one use (which is often the case).
1150
1151		We should do something similar to floating point, but the transformation
1152		there is much more complex, so I won't deal with that.
1153
1154		Secondly, there are now two new cps primops at our disposal:
1155
1156		1. rawrecord of record_kind option
1157		This pure operator allocates some uninitialized storage from the heap.
1158		There are two forms:
1159
1160		rawrecord NONE [INT n]  allocates a tagless record of length n
1161		rawrecord (SOME rk) [INT n] allocates a tagged record of length n
1162		and initializes the tag.
1163
1164		2. rawupdate of cty
1165		rawupdate cty (v,i,x)
1166		Assigns to x to the ith component of record v.
1167		The storelist is not updated.
1168
1169		We use these new primops for both spilling and increment record construction.
1170
1171		1. Spilling.
1172
1173		This is implemented with a linear scan algorithm (but generalized
1174		to trees).  The algorithm will create a single spill record at the
1175		beginning of the cps function and use rawupdate to spill to it,
1176		and SELECT or SELp to reload from it.  So both spills and reloads
1177		are fine-grain operations.  In contrast, in the old algorithm
1178		"spills" have to be bundled together in records.
1179
1180		Ideally, we should sink the spill record construction to where
1181		it is needed.  We can even split the spill record into multiple ones
1182		at the places where they are needed.  But CPS is not a good
1183		representation for global code motion, so I'll keep it simple and
1184		am not attempting this.
1185
1186		2. Incremental record construction (aka record splitting).
1187
1188		Long records with many component values which are simulatenously live
1189		(recall that single use record addresses are not considered to
1190		be live) are constructed with rawrecord and rawupdate.
1191		We allocate space on the heap with rawrecord first, then gradually
1192		fill it in with rawupdate.  This is the technique suggested to me
1193		by Matthias.
1194
1195		Some restrictions on when this is applicable:
1196		1. It is not a VECTOR record.  The code generator currently does not handle
1197		this case. VECTOR record uses double indirection like arrays.
1198		2. All the record component values are defined in the same "basic block"
1199		as the record constructor.  This is to prevent speculative
1200		record construction.
1201
1202		----------------------------------------------------------------------
1203		Name: Allen Leung
1204		Date: 2002/02/22 01:02:00 EST
1205		Tag: leunga-20020222-mlrisc-tools
1206
1207		Minor bug fixes in the parser and rewriter
1208
1209		----------------------------------------------------------------------
1210		Name: Allen Leung
1211		Date: 2002/02/21 20:20:00 EST
1212		Tag: leunga-20020221-peephole
1213
1214		Regenerated the peephole files.  Some contained typos in the specification
1215		and some didn't compile because of pretty printing bugs in the old version
1216		of 'nowhere'.
1217
1218		----------------------------------------------------------------------
1219		Name: Allen Leung
1220		Date: 2002/02/19 20:20:00 EST
1221		Tag: leunga-20020219-mlrisc-tools
1222		Description:
1223
1224		Minor bug fixes to the mlrisc-tools library:
1225
1226		1.  Fixed up parsing colon suffixed keywords
1227		2.  Added the ability to shut the error messages up
1228		3.  Reimplemented the pretty printer and fixed up/improved
1229		the pretty printing of handle and -> types.
1230		4.  Fixed up generation of literal symbols in the nowhere tool.
1231		5.  Added some SML keywords to to sml.sty
1232
1233		----------------------------------------------------------------------
1234		Name: Matthias Blume
1235		Date: 2002/02/19 16:20:00 EST
1236		Tag: blume-20020219-cmffi
1237		Description:
1238
1239		A wild mix of changes, some minor, some major:
1240
1241		* All C FFI-related libraries are now anchored under $c:
1242		$/c.cm      --> $c/c.cm
1243		$/c-int.cm  --> $c/internals/c-int.cm
1244		$/memory.cm --> $c/memory/memory.cm
1245
1246		* "make" tool (in CM) now treats its argument pathname slightly
1247		differently:
1248		1. If the native expansion is an absolute name, then before invoking
1249		the "make" command on it, CM will apply OS.Path.mkRelative
1250		(with relativeTo = OS.FileSys.getDir()) to it.
1251		2. The argument will be passed through to subsequent phases of CM
1252		processing without "going native".  In particular, if the argument
1253		was an anchored path, then "make" will not lose track of that anchor.
1254
1255		* Compiler backends now "know" their respective C calling conventions
1256		instead of having to be told about it by ml-nlffigen.  This relieves
1257		ml-nlffigen from one of its burdens.
1258
1259		* The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
1260
1261		* Export C_DEBUG and C_Debug from $c/c.cm.
1262
1263		* C type encoding in ml-nlffi-lib has been improved to model the conceptual
1264		subtyping relationship between incomplete pointers and their complete
1265		counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
1266		with the convention of instantiating 'o with ('t, 'c) obj whenever
1267		the pointer target type is complete.  In the incomplete case, 'o
1268		will be instantiated with some "'c iobj" -- a type obtained by
1269		using one of the functors PointerToIncompleteType or PointerToCompleteType.
1270
1271		Operations that work on both incomplete and complete pointer types are
1272		typed as taking an 'o ptr while operations that require the target to
1273		be known are typed as taking some ('t, 'c) obj ptr.
1274
1275		voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
1276		where void is an eqtype without any values.  This makes it possible
1277		to work on voidptr values using functions meant to operate on light
1278		incomplete pointers.
1279
1280		* As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
1281		been vastly simplified.
1282
1283		----------------------------------------------------------------------
1284		Name: Matthias Blume
1285		Date: 2002/02/19 10:48:00 EST
1286		Tag: blume-20020219-pqfix
1287		Description:
1288
1289		Applied Chris Okasaki's bug fix for priority queues.
1290
1291		----------------------------------------------------------------------
1292		Name: Matthias Blume
1293		Date: 2002/02/15 17:05:00
1294		Tag: Release_110_39
1295		Description:
1296
1297		Last-minute retagging is becoming a tradition... :-(
1298
1299		This is the working release 110.39.
1300
1301		----------------------------------------------------------------------
1302		Name: Matthias Blume
1303		Date: 2002/02/15 16:00:00 EST
1304		Tag: Release_110_39-orig
1305		Description:
1306
1307		Working release 110.39.  New bootfiles.
1308
1309		(Update: There was a small bug in the installer so it wouldn't work
1310		with all shells.  So I retagged. -Matthias)
1311
1312		----------------------------------------------------------------------
1313		Name: Matthias Blume
1314		Date: 2002/02/15 14:17:00 EST
1315		Tag: blume-20020215-showbindings
1316		Description:
1317
1318		Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
1319		the latter can be useful for exploring what bindings are available at
1320		the interactive prompt.  (The first function returns only the list
1321		of symbols that are really bound, the second prints those but also the
1322		ones that CM's autoloading mechanism knows about.)
1323
1324		----------------------------------------------------------------------
1325		Name: Matthias Blume
1326		Date: 2002/02/15 12:08:00 EST
1327		Tag: blume-20020215-iptrs
1328		Description:
1329
1330		Two improvements to ml-nlffigen:
1331
1332		1. Write files only if they do not exist or if their current contents
1333		do not coincide with what's being written.  (That is, avoid messing
1334		with the time stamps unless absolutely necessary.)
1335
1336		2. Implement a "repository" mechanism for generated files related
1337		to "incomplete pointer types".   See the README file for details.
1338
1339		----------------------------------------------------------------------
1340		Name: Matthias Blume
1341		Date: 2002/02/14 11:50:00 EST
1342		Tag: blume-20020214-quote
1343		Description:
1344
1345		Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
1346		because of the new and improved tag generation scheme.  (Thanks to Allen
1347		Leung for pointing it out.)
1348
1349		----------------------------------------------------------------------
1350		Name: Lal George
1351		Date: 2002/02/14 09:55:27 EST 2002
1352		Tag: george-20020214-isabelle-bug
1353		Description:
1354
1355		Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
1356		of Isabelle on the x86.
1357
1358		From Allen:
1359		-----------
1360		I've found the problem:
1361
1362		in ra-core.sml, I use the counter "blocked" to keep track of the
1363		true number of elements in the freeze queue.  When the counter goes
1364		to zero, I skip examining the queue.  But I've messed up the
1365		bookkeeping in combine():
1366
1367		else ();
1368		case !ucol of
1369		PSEUDO => (if !cntv > 0 then
1370		(if !cntu > 0 then blocked := !blocked - 1 else ();
1371		^^^^^^^^^^^^^^^^^^^^^^^
1372		moveu := mergeMoveList(!movev, !moveu)
1373		)
1374		else ();
1375
1376		combine() is called to coalesce two nodes u and v.
1377		I think I was thinking that if the move counts of u and v are both
1378		greater than zero then after they are coalesced then one node is
1379		removed from the freeze queue.  Apparently I was thinking that
1380		both u and v are of low degree, but that's clearly not necessarily true.
1381
1382
1383		02/12/2002:
1384		Here's the patch.  HOL now compiles.
1385
1386		I don't know how this impact on performance (compile
1387		time or runtime).  This bug caused the RA (especially on the x86)
1388		to go thru the potential spill phase when there are still nodes on the
1389		freeze queue.
1390
1391
1392
1393
1394		----------------------------------------------------------------------
1395		Name: Matthias Blume
1396		Date: 2002/02/13 22:40:00 EST
1397		Tag: blume-20020213-fptr-rtti
1398		Description:
1399
1400		Fixed a bug in ml-nlffigen that was introduced with one of the previous
1401		updates.
1402
1403		----------------------------------------------------------------------
1404		Name: Matthias Blume
1405		Date: 2002/02/13 16:41:00 EST
1406		Tag: blume-20020213-cmlpq
1407		Description:
1408
1409		Added new priority queue export symbols (which have just been added to
1410		smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
1411		not compile and the installer would choke.)
1412
1413		----------------------------------------------------------------------
1414		Name: Matthias Blume
1415		Date: 2002/02/13 16:15:00 EST
1416		Tag: blume-20020213-various
1417		Description:
1418
1419		1. More tweaks to ml-nlffigen:
1420
1421		- better internal datastructures (resulting in slight speedup)
1422		- "-match" option requires exact match
1423		- "localized" gensym counters (untagged structs/unions nested within
1424		other structs/unions or within typedefs get a fresh counter; their
1425		tag will be prefixed by a concatenation of their parents' tags)
1426		- bug fixes (related to calculation of transitive closure of types
1427		to be included in the output)
1428
1429		2. Minor Basis updates:
1430
1431		- added implementations for List.collate and Option.app
1432
1433		----------------------------------------------------------------------
1434		Name: Matthias Blume
1435		Date: 2002/02/11 15:55:00 EST
1436		Tag: blume-20020211-gensym
1437		Description:
1438
1439		Added a "-gensym" option to command line of ml-nlffigen.  This can be
1440		used to specify a "stem" -- a string that is inserted in all "gensym'd"
1441		names (ML structure names that correspond to unnamed C structs, unions,
1442		and enums), so that separate runs of ml-nlffigen do not clash.
1443
1444		----------------------------------------------------------------------
1445		Name: Matthias Blume
1446		Date: 2002/02/11 12:05:00 EST
1447		Tag: blume-20020211-gensml
1448		Description:
1449
1450		A quick fix for a problem with GenSML (in the pgraph-util library):
1451		Make generation of toplevel "local" optional.  (Strictly speaking,
1452		signature definitions within "local" are not legal SML.)
1453
1454		Other than that: updates to INSTALL and cm/TODO.
1455
1456		----------------------------------------------------------------------
1457		Name: Matthias Blume
1458		Date: 2002/02/08 15:00:00 EST
1459		Tag: blume-20020208-uniquepid
1460		Description:
1461
1462		0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
1463
1464		1. The installer (config/install.sh) has gotten smarter:
1465
1466		- Configuration options are a bit easier to specify now
1467		(in config/targets).
1468		- Bug in recognizing .tar.bz2 files fixed.
1469		- Installer automatically resolves dependencies between
1470		configuration options (e.g., if you ask for eXene, you will
1471		also get cml -- regardless whether you asked for it or not).
1472		- Installer can run in "quieter mode" by setting the environment
1473		variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
1474		"completely silent", though.
1475		- Build HashCons library as part of smlnj-lib.
1476
1477		2. A new scheme for assigning persistent identifiers to compilation
1478		units (and, by extension, to types etc.) has been put into place.
1479		This fixes a long-standing bug where types and even dynamic values
1480		can get internally confused, thereby compromising type safety
1481		(abstraction) and dynamic correctness.  See
1482
1483		http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
1484
1485		for an example of how things could go wrong until now.
1486
1487		The downside of the new scheme is that pids are not quite as
1488		persistent as they used to be: CM will generate a fresh pid
1489		for every compilation unit that it thinks it sees for the first
1490		time.  That means that if you compile starting from a clean, fresh
1491		source tree at two different times, you end up with different
1492		binaries.
1493
1494		Cutoff recompilation, however, has not been compromised because
1495		CM keeps pid information in special caches between runs.
1496
1497		----------------------------------------------------------------------
1498		Name: Lal George
1499		Date: 2002/02/07 15:34:13 EST 2002
1500		Tag: <none>
1501		Description:
1502
1503		Compilers that generate assembly code may produce  global labels
1504		whose value is resolved at link time. The various peephole optimization
1505		modules did not take this in account.
1506
1507		TODO. The Labels.addrOf function should really return an option
1508		type so that clients are forced to deal with this issue, rather
1509		than an exception being raised.
1510
1511		----------------------------------------------------------------------
1512		Name: Lal George
1513		Date: 2002/02/06 13:55:02 EST
1514		Tag: george-20020206-ra-breakup
1515		Description:
1516
1517		1. A bug fix from Allen.
1518
1519		A typo causes extra fstp %st(0)'s to be generated at compensation
1520		edges, which might cause stack underflow traps at runtime.  This
1521		occurs in fft where there are extraneous fstps right before the 'into'
1522		trap instruction (in this case they are harmless since none of the
1523		integers overflow.)
1524
1525		2. Pulled out various utility modules that were embedded in the modules
1526		of the register allocator. I need these modules for other purposes, but
1527		they are not complete enough to put into a library (just yet).
1528		----------------------------------------------------------------------
1529		Name: Matthias Blume
1530		Date: 2002/01/31 16:05:00 EST
1531		Tag: blume-20020131-sparc-ccalls
1532		Description:
1533
1534		1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
1535		of extra stack space by mistake.  Fixed.
1536
1537		2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
1538
1539		----------------------------------------------------------------------
1540		Name: Allen Leung
1541		Date: 2002/01/30
1542		Tag: leunga-20020130-nowhere-bug-fix
1543		Description:
1544
1545		MLRISC bug fixes:
1546		1. Fixed a bindings computation bug in the 'nowhere' program generator tool.
1547		2. MachineInt.fromString was negating its value.
1548
1549		----------------------------------------------------------------------
1550		Name: Matthias Blume
1551		Date: 2002/01/29
1552		Tag: blume-20020129-INSTALL
1553		Description:
1554
1555		- Added somewhat detailed installation instructions (file INSTALL).
1556		- Fixed curl-detection bug in config/install.sh.
1557		- It is now possible to select the URL getter using the URLGETTER
1558		environment variable:
1559
1560		not set / "unknown"      --> automatic detection (script tries wget,
1561		curl, and lynx)
1562		"wget" / "curl" / "lynx" --> use the specified program (script "knows"
1563		how to properly invoke them)
1564		other                    --> use $URLGETTER directly, it must take
1565		precisely two command-line arguments
1566		(source URL and destination file name)
1567
1568		----------------------------------------------------------------------
1569		Name: Matthias Blume
1570		Date: 2002/01/28
1571		Tag: blume-20020128-sparc-ccalls
1572		Description:
1573
1574		- Fixed problem with calculation of "used" registers in sparc-c-calls.
1575		- Make use of the allocParam argument in sparc-c-calls.
1576
1577		----------------------------------------------------------------------
1578		Name: Matthias Blume
1579		Date: 2002/01/28
1580		Tag: blume-20020128-allocParam
1581		Description:
1582
1583		John Reppy:  Changes c-calls API to accept client-callback for
1584		allocating extra stack space.
1585		me: Corresponding changes to mlriscGen (using a dummy argument that
1586		does not change the current behavior).
1587
1588		----------------------------------------------------------------------
1589		Name: Matthias Blume
1590		Date: 2002/01/28 12:00:00
1591		Tag: Release_110_38
1592		Description:
1593
1594		This time for real!!!
1595
1596		----------------------------------------------------------------------
1597		Name: Matthias Blume
1598		Date: 2002/01/28 10:56:00 EST
1599		Tag: blume-20020128-retraction
1600		Description:
1601
1602		0. Retracted earlier 110.38.  (The Release_110_38 tag has been replaced
1603		with blume-Release_110_38-retracted.)
1604
1605		1. Fixed a problem with incorrect rounding modes in real64.sml.
1606		(Thanks to Andrew Mccreight <andrew.mccreight@yale.edu>.)
1607
1608		2. A bug in ml-nlffigen related to the handling of unnamed structs, unions,
1609		and enums fixed.  The naming of corresponding ML identifiers should
1610		now be consistent again.
1611
1612		----------------------------------------------------------------------
1613		Name: Allen Leung
1614		Date: 2002/01/27
1615		Tag: leunga-20020127-nowhere
1616		Description:
1617
1618		Added a target called nowhere in the configuration scripts.
1619		Enabling this will build the MLRISC 'nowhere' tool (for translating
1620		programs with where-clauses into legal SML code) during installation.
1621
1622		----------------------------------------------------------------------
1623		Name: Matthias Blume
1624		Date: 2002/01/25 21:27:00 EST
1625		Tag: blume-Release_110_38-retracted
1626		Description:
1627
1628		Call it a (working) release!  Version is 110.38. Bootfiles are ready.
1629
1630		README will be added later.
1631
1632		!!! NOTE:  Re-tagged as blume-Release_110_38-retracted. Original tag
1633		(Release_110_38) removed.  Reason: Last-minute bug fixes.
1634
1635		----------------------------------------------------------------------
1636		Name: Matthias Blume
1637		Date: 2002/01/25
1638		Tag: blume-20020125-ffi
1639		Description:
1640
1641		A large number of tweaks and improvements to ml-nlffi-lib and
1642		ml-nlffigen:
1643
1644		- ML represenation types have been streamlined
1645		- getter and setter functions work with concrete values, not abstract
1646		ones where possible
1647		- ml-nlffigen command line more flexible (see README file there)
1648		- some bugs have been fixed (hopefully)
1649
1650		----------------------------------------------------------------------
1651		Name: Lal George
1652		Date: 2002/01/24
1653		Tag: george-20020124-risc-ra-interface
1654		Description:
1655
1656		There is a dramatic simplification in the interface to the
1657		register allocator for RISC architectures as a result of making
1658		parallel copy instructions explicit.
1659
1660		----------------------------------------------------------------------
1661		Name: Matthias Blume
1662		Date: 2002/01/22
1663		Tag: blume-20020122-x86-ccalls
1664		Description:
1665
1666		Bug fix for c-calls on x86 (having to do with how char- and
1667		short-arguments are being handled).
1668
1669		----------------------------------------------------------------------
1670		Name: Matthias Blume
1671		Date: 2002/01/21
1672		Tag: blume-20020121-ff
1673		Description:
1674
1675		Another day of fiddling with the FFI...
1676
1677		1. Bug fix/workaround:  CKIT does not complain about negative array
1678		dimensions, so ml-nlffigen has to guard itself against this possibility.
1679		(Otherwise a negative dimension would send it into an infinite loop.)
1680
1681		2. Some of the abstract types (light objects, light pointers, most "base"
1682		types) in structure C are now eqtypes.
1683
1684		3. Added constructors and test functions for NULL function pointers.
1685
1686		----------------------------------------------------------------------
1687		Name: Matthias Blume
1688		Date: 2002/01/18
1689		Tag: blume-20020118-ready-for-new-release
1690		Description:
1691
1692		Made config/srcarchiveurl point to a new place.  (Will provide boot
1693		files shortly.)
1694
1695		Maybe we christen this to be 110.38?
1696
1697		----------------------------------------------------------------------
1698		Name: Matthias Blume
1699		Date: 2002/01/18
1700		Tag: blume-20020118-more-ffifiddle
1701		Description:
1702
1703		Today's FFI fiddling:
1704
1705		- Provided a structure CGetSet with "convenient" versions of C.Get.* and
1706		C.Set.* that use concrete (MLRep.*) arguments and results instead
1707		of abstract ones.
1708
1709		- Provided word-style bit operations etc. for "int" representation
1710		types in MLRep.S<Foo>Bitops where <Foo> ranges over Char, Int, Short,
1711		and Long.
1712
1713		----------------------------------------------------------------------
1714		Name: Matthias Blume
1715		Date: 2002/01/18
1716		Tag: blume-20020118-use-x86-fp
1717		Description:
1718
1719		Now that x86-fast-fp seems to be working, I turned it back on again
1720		by default.  (Seems to work fine now, even with the FFI.)
1721
1722		Other than that, I added some documentation about the FFI to
1723		src/ml-nlffigen/README and updated the FFI test examples in
1724		src/ml-nlffi-lib/Tests/*.
1725
1726		----------------------------------------------------------------------
1727		Name: Allen Leung
1728		Date: 2002/01/17
1729		Tag: leunga-20020117-x86-fast-fp-call
1730		Description:
1731
1732		1. Fixed a problem with handling return fp values when x86's fast fp
1733		mode is turned on.
1734
1735		2. Minor pretty printing fix for cellset.  Print %st(0) as %st(0) instead
1736		of %f32.
1737
1738		3. Added a constructor INT32lit to the ast of MLRISC tools.
1739
1740		----------------------------------------------------------------------
1741		Name: Matthias Blume
1742		Date: 2002/01/16
1743		Tag: blume-20020116-ffifiddle
1744		Description:
1745
1746		More fiddling with the FFI interface:
1747
1748		- Make constness 'c instead of rw wherever possible.  This eliminates
1749		the need for certain explicit coercions.  (However, due to ML's
1750		value polymorphism, there will still be many cases where explicit
1751		coercions are necessary.  Phantom types are not the whole answer
1752		to modeling a subtyping relationship in ML.)
1753
1754		- ro/rw coersions for pointers added.  (Avoids the detour through */&.)
1755
1756		- "printf" test example added to src/ml-nlffi-lib/Tests.  (Demonstrates
1757		clumsy workaround for varargs problem.)
1758
1759		----------------------------------------------------------------------
1760		Name: Lal George
1761		Date: 2002/01/15
1762		Tag: <none>
1763		Description:
1764
1765		1. Since COPY instructions are no longer native to the architecture,
1766		a generic functor can be used to implement the expandCopies function.
1767
1768		2. Allowed EXPORT and IMPORT pseudo-op declarations to appear inside a
1769		TEXT segment.
1770
1771		----------------------------------------------------------------------
1772		Name: Matthias Blume
1773		Date: 2002/01/15
1774		Tag: blume-20020115-ffiupdates
1775		Description:
1776
1777		1. Fix for bug resulting in single-precision float values being returned
1778		incorrectly from FFI calls.
1779
1780		2. Small modifications to C FFI API:
1781
1782		- memory-allocation routines return straight objects (no options)
1783		and raise an exception in out-of-memory situations
1784		- unsafe extensions to cast between function pointers and pointers
1785		from/to ints
1786		- added structure C_Debug as an alternative to structure C where
1787		pointer-dereferencing (|*| and |*!) always check for null-pointers
1788		- added open_lib' to DynLinkage;  open_lib' works like open_lib
1789		but also takes a (possibly empty) list of existing library handles
1790		that the current library depends on
1791
1792		----------------------------------------------------------------------
1793		Name: Matthias Blume
1794		Date: 2002/01/10
1795		Tag: blume-20020110-newffigen
1796		Description:
1797
1798		1. Updates to portable graph code.
1799
1800		2. Major update to ml-nlffigen and ml-nlffi-lib.  Things are much
1801		more scalable now so that even huge interfaces such as the one
1802		for GTK compile in finite time and space. :-)
1803		See src/ml-nlffigen/README for details on what's new.
1804
1805		----------------------------------------------------------------------
1806		Name: Lal George
1807		Date: 2001/01/09 14:31:35 EST 2002
1808		Tag: george-20011206-rm-native-copy
1809		Description:
1810
1811		Removed the native COPY and FCOPY instructions
1812		from all the architectures and replaced it with the
1813		explicit COPY instruction from the previous commit.
1814
1815		It is now possible to simplify many of the optimizations
1816		modules that manipulate copies. This has not been
1817		done in this change.
1818
1819		----------------------------------------------------------------------
1820		Name: Lal George
1821		Date: 2001/12/06 16:50:13 EST 2001
1822		Tag: george-20011206-mlrisc-instruction
1823		Description:
1824
1825		Changed the representation of instructions from being fully abstract
1826		to being partially concrete. That is to say:
1827
1828		from
1829		type instruction
1830
1831		to
1832		type instr                              (* machine instruction *)
1833
1834		datatype instruction =
1835		LIVE of {regs: C.cellset, spilled: C.cellset}
1836		| KILL of {regs: C.cellset, spilled: C.cellset}
1837		| COPYXXX of {k: CB.cellkind, dst: CB.cell list, src: CB.cell list}
1838		| ANNOTATION of {i: instruction, a: Annotations.annotation}
1839		| INSTR of instr
1840
1841		This makes the handling of certain special instructions that appear on
1842		all architectures easier and uniform.
1843
1844		LIVE and KILL say that a list of registers are live or killed at the
1845		program point where they appear. No spill code is generated when an
1846		element of the 'regs' field is spilled, but the register is moved to
1847		the 'spilled' (which is present, more for debugging than anything else).
1848
1849		LIVE replaces the (now deprecated) DEFFREG instruction on the alpha.
1850		We used to generate:
1851
1852		DEFFREG f1
1853		f1 := f2 + f3
1854		trapb
1855
1856		but now generate:
1857
1858		f1 := f2 + f3
1859		trapb
1860		LIVE {regs=[f1,f2,f3], spilled=[]}
1861
1862		Furthermore, the DEFFREG (hack) required that all floating point instruction
1863		use all registers mentioned in the instruction. Therefore f1 := f2 + f3,
1864		defines f1 and uses [f1,f2,f3]! This hack is no longer required resulting
1865		in a cleaner alpha implementation. (Hopefully, intel will not get rid of
1866		this architecture).
1867
1868		COPYXXX is intended to replace the parallel COPY and FCOPY  available on
1869		all the architectures. This will result in further simplification of the
1870		register allocator that must be aware of them for coalescing purposes, and
1871		will also simplify certain aspects of the machine description that provides
1872		callbacks related to parallel copies.
1873
1874		ANNOTATION should be obvious, and now INSTR represents the honest to God
1875		machine instruction set!
1876
1877		The <arch>/instructions/<arch>Instr.sml files define certain utility
1878		functions for making porting easier -- essentially converting upper case
1879		to lower case. All machine instructions (of type instr) are in upper case,
1880		and the lower case form generates an MLRISC instruction. For example on
1881		the alpha we have:
1882
1883		datatype instr =
1884		LDA of {r:cell, b:cell, d:operand}
1885		| ...
1886
1887		val lda : {r:cell, b:cell, d:operand} -> instruction
1888		...
1889
1890		where lda is just (INSTR o LDA), etc.
1891
1892		----------------------------------------------------------------------
1893		Name: Matthias Blume
1894		Date: 2001/11/22 21:40:00 EST
1895		Tag: Release_110_37
1896		Description:
1897
1898		Release 110.37.  This time for real.
1899
1900		----------------------------------------------------------------------
1901		Name: Matthias Blume
1902		Date: 2001/11/21 16:35:00 EST
1903		Tag: blume-20011121-foot-in-mouth
1904		Description:
1905
1906		Removed the "Release_110_37" tag because of a serious bug.
1907		This will be re-tagged once the bug is fixed.
1908
1909		----------------------------------------------------------------------
1910		Name: Matthias Blume
1911		Date: 2001/11/21 16:14:00 EST
1912		Tag: blume-20011121-forgottenfile
1913		Description:
1914
1915		Forgot to add a file.  (Just a .tex-file -- part of
1916		the CM manual source.)
1917
1918		----------------------------------------------------------------------
1919		Name: Matthias Blume
1920		Date: 2001/11/21 16:10:00 EST
1921		Tag: blume-20011121-invalid_110_37
1922		Description:
1923
1924		Note: I removed the original tag "Release_110_37" from this commit
1925		because we found a serious bug in all non-x86 backends.
1926		- Matthias
1927
1928		1. Modifications to the SML/NJ code generator and to the runtime system
1929		so that code object name strings are directly inserted into code
1930		objects at code generation time.  The only business the runtime system
1931		has with this is now to read the name strings on occasions.
1932		(The encoding of the name string has also changed somewhat.)
1933
1934		2. CM now implements a simple "set calculus" for specifying export lists.
1935		In particular, it is now possible to refer to the export lists of
1936		other libraries/groups/sources and form unions as well as differences.
1937		See the latest CM manual for details.
1938
1939		3. An separate notion of "proxy" libraries has again be eliminated from
1940		CM's model.  (Proxy libraries are now simply a special case of using
1941		the export list calculus.)
1942
1943		4. Some of the existing libraries now take advantage of the new set
1944		calculus.
1945		(Notice that not all libraries have been converted because some
1946		of the existing .cm-files are supposed to be backward compatible
1947		with 110.0.x.)
1948
1949		5. Some cleanup in stand-alone programs.  (Don't use "exnMessage" -- use
1950		"General.exnMessage"!  The former relies on a certain hook to be
1951		initialized, and that often does not happen in the stand-alone case.)
1952
1953		----------------------------------------------------------------------
1954		Name: Lal George
1955		Date: 2001/11/21  13:56:18 EST
1956		Tag: george-2001121-pseudo-ops
1957		Description:
1958
1959		Implemented a complete redesign of MLRISC pseudo-ops. Now there
1960		ought to never be any question of incompatabilities with
1961		pseudo-op syntax expected by host assemblers.
1962
1963		For now, only modules supporting GAS syntax are implemented
1964		but more should follow, such as MASM, and vendor assembler
1965		syntax, e.g. IBM as, Sun as, etc.
1966
1967		----------------------------------------------------------------------
1968		Name: Matthias Blume
1969		Date: 2001/11/14 11:52:00 EST
1970		Tag: blume-20011114-srcname
1971		Description:
1972
1973		1. Routed the name of the current source file to mlriscgen where it
1974		should be directly emitted into the code object.  (This last part
1975		is yet to be done.)
1976
1977		2. Some cleanup of the pgraph code to make it match the proposal that
1978		I put out the other day.  (The proposal notwithstanding, things are
1979		still in flux here.)
1980
1981		----------------------------------------------------------------------
1982		Name: Lal George
1983		Date: 2001/11/14 09:44:04 EST
1984		Tag:
1985		Description:
1986
1987		Fix for a backpatching bug reported by Allen.
1988
1989		Because the boundary between short and long span-dependent
1990		instructions is +/- 128, there are an astounding number of
1991		span-dependent instructions whose size is over estimated.
1992
1993		Allen came up with the idea of letting the size of span
1994		dependent instructions be non-monotonic, for a maxIter
1995		number of times, after which the size must be monotonically
1996		increasing.
1997
1998		This table shows the number of span-dependent instructions
1999		whose size was over-estimated as a function of maxIter, for the
2000		file Parse/parse/ml.grm.sml:
2001
2002		maxIter            # of instructions:
2003		10                      687
2004		20                      438
2005		30                      198
2006		40                        0
2007
2008		In compiling the compiler, there is no significant difference in
2009		compilation speed between maxIter=10 and maxIter=40. Actually,
2010		my measurements showed that maxIter=40 was a tad faster than
2011		maxIter=10! Also 96% of the  files in the compiler reach a fix
2012		point within 13 iterations, so fixing maxIter at 40, while high,
2013		is okay.
2014
2015		----------------------------------------------------------------------
2016		Name: Matthias Blume
2017		Date: 2001/10/31 15:25:00 EST
2018		Tag: blume-20011031-pgraph
2019		Description:
2020
2021		CKIT:
2022		* Changed the "Function" constructor of type Ast.ctype to carry optional
2023		argument identifiers.
2024		* Changed the return type of TypeUtil.getFunction accordingly.
2025		* Type equality ignores the argument names.
2026		* TypeUtil.composite tries to preserve argument names but gives up quickly
2027		if there is a mismatch.
2028
2029		installation script:
2030		* attempts to use "curl" if available (unless "wget" is available as well)
2031
2032		CM:
2033		* has an experimental implementation of "portable graphs" which I will
2034		soon propose as an implementation-independent library format
2035		* there are also new libraries $/pgraph.cm and $/pgraph-util.cm
2036
2037		NLFFI-LIB:
2038		* some cleanup (all cosmetic)
2039
2040		NLFFIGEN:
2041		* temporarily disabled the mechanism that suppresses ML output for
2042		C definitions whose identifiers start with an underscore character
2043		* generate val bindings for enum constants
2044		* user can request that only one style (light or heavy) is being used;
2045		default is to use both (command-line arguments: -heavy and -light)
2046		* fixed bug in handling of function types involving incomplete pointers
2047		* generate ML entry points that take record arguments (i.e., using
2048		named arguments) for C functions that have a prototype with named
2049		arguments
2050		(see changes to CKIT)
2051
2052		----------------------------------------------------------------------
2053		Name: Allen Leung
2054		Date: 2001/10/27 20:34:00 EDT
2055		Tag: leunga-20011027-x86-fast-fp-call
2056		Description:
2057
2058		Fixed the bug described in blume-20010920-slowfp.
2059
2060		The fix involves
2061		1. generating FCOPYs in FSTP in ia32-svid
2062		2. marking a CALL with the appropriate annotation
2063
2064		----------------------------------------------------------------------
2065		Name: Matthias Blume
2066		Date: 2001/10/16 11:32:00 EDT
2067		Tag: blume-20011016-netbsd
2068		Description:
2069
2070		Underscore patch from Chris Richards (fixing problem with compiling
2071		runtime system under recent NetBSD).
2072
2073		----------------------------------------------------------------------
2074		Name: Allen Leung
2075		Date: 2001/10/12 17:18:32 EDT 2001
2076		Tag: leung-20011012-x86-printflowgraph
2077		Description:
2078
2079		X86RA now uses a valid (instead of dummy) PrintFlowgraph module.
2080
2081		----------------------------------------------------------------------
2082		Name: Lal George
2083		Date: 2001/10/11 23:51:34 EDT
2084		Tag: george-20011011-too-many-instrs
2085		Description:
2086
2087		The representation of a program point never expected to see more
2088		than 65536 instructions in a basic block!
2089
2090		----------------------------------------------------------------------
2091		Name: Lal George
2092		Date: 2001/10/09 09:41:37 EDT
2093		Tag: george-20011008-mlrisc-labels
2094		Description:
2095
2096		Changed the machine description files to support printing of
2097		local and global labels in assembly code, based on host assembler
2098		conventions.
2099
2100		----------------------------------------------------------------------
2101		Name: Matthias Blume
2102		Date: 2001/09/25 15:25:00 EDT
2103		Tag: blume-20010925-exninfo
2104		Description:
2105
2106		I provided a non-hook implementation of exnName (at the toplevel) and
2107		made the "dummy" implementation of exnMessage (at the toplevel) more
2108		useful: if nothing gets "hooked in", then at least you are going to
2109		see the exception name and a message indicating why you don't see more.
2110
2111		[For the time being, programs that need exnMessage and want to use
2112		ml-build should either use General.exnMessage (strongly recommended) or
2113		refer to structure General at some other point so that CM sees a
2114		static dependency.]
2115
2116		[Similar remarks go for "print" and "use":  If you want to use their
2117		functionality in stand-alone programs generated by ml-build, then use
2118		TextIO.output and Backend.Interact.useFile (from $smlnj/compiler.cm).]
2119
2120		----------------------------------------------------------------------
2121		Name: Matthias Blume
2122		Date: 2001/09/20 17:28:00 EDT
2123		Tag: blume-20010920-slowfp
2124		Description:
2125
2126		Allen says that x86-fast-fp is not safe yet, so I turned it off again...
2127
2128		----------------------------------------------------------------------
2129		Name: Matthias Blume
2130		Date: 2001/09/20 17:20:00 EDT
2131		Tag: blume-20010920-canonicalpaths
2132		Description:
2133
2134		0. Updated the BOOT file (something that I forgot to do earlier).
2135
2136		1. Small internal change to CM so that it avoids "/../" in filenames
2137		as much as possible (but only where it is safe).
2138
2139		2. Changed config/_run-sml (resulting in a changed bin/.run-sml) so
2140		that arguments that contain delimiters are passed through correctly.
2141		This change also means that all "special" arguments of the form
2142		@SMLxxx... must come first.
2143
2144		3. Changed install script to put relative anchor names for tool commands
2145		into pathconfig.
2146
2147		----------------------------------------------------------------------
2148		Name: Matthias Blume
2149		Date: 2001/09/18 15:35:00 EDT
2150		Tag: blume-20010918-readme11036
2151		Description:
2152
2153		Added README files.
2154
2155		----------------------------------------------------------------------
2156		Name: Matthias Blume
2157		Date: 2001/09/18 11:45:00 EDT
2158		Tag: Release_110_36 (retag)
2159		Description:
2160
2161		Fixed mistake in config/preloads. Retagged as 110.36.
2162
2163		----------------------------------------------------------------------
2164		Name: Matthias Blume
2165		Date: 2001/09/18 09:40:00 EDT
2166		Tag: Release_110_36_orig (tag changed)
2167		Description:
2168
2169		New version (110.36).  New bootfiles.
2170
2171		----------------------------------------------------------------------
2172		Name: Matthias Blume
2173		Date: 2001/09/14 16:15:00 EDT
2174		Tag: blume-20010914-x86fastfp
2175		Description:
2176
2177		John committed some changes that Allen made, in particular a (hopefully)
2178		correctly working version of the x86-fp module.
2179
2180		I changed the default setting of the Control.MLRISC.getFlag "x86-fast-fp"
2181		flag to "true".  Everything seems to compile to a fixpoint ok, and
2182		"mandelbrot" speeds up by about 15%.
2183
2184		----------------------------------------------------------------------
2185		Name: Matthias Blume
2186		Date: 2001/09/13 11:20:00 EDT
2187		Tag: blume-20010913-minimal
2188		Description:
2189
2190		1. Stefan Monnier's patch to fix a miscompilation problem that
2191		was brought to light by John Reppy's work on Moby.
2192
2193		2. Implemented a minimal "structure Compiler" that contains just
2194		"version" and "architecture".  The minimal version will be
2195		available when the full version is not.  This is for backward-
2196		compatibility with code that wants to test Compiler.version.
2197
2198		----------------------------------------------------------------------
2199		Name: Matthias Blume
2200		Date: 2001/08/28 14:03:00 EDT
2201		Tag: blume-20010828-ml-lex
2202		Description:
2203
2204		Fix for bug 1581, received from Neophytos Michael.
2205
2206		----------------------------------------------------------------------
2207		Name: Matthias Blume
2208		Date: 2001/08/27 11:20:00 EDT
2209		Tag: blume-20010827-readme11035
2210		Description:
2211
2212		Fleshed out the README file for 110.35.
2213
2214		----------------------------------------------------------------------
2215		Name: Matthias Blume
2216		Date: 2001/08/24 17:10:00 EDT
2217		Tag: Release_110_35
2218		Description:
2219
2220		New version number (110.35).  New bootfiles.
2221
2222		----------------------------------------------------------------------
2223		Name: Lal George
2224		Date: 2001/08/24 13:47:18 EDT 2001
2225		Tag: george-20010824-MLRISC-graphs
2226		Description:
2227
2228		removed clusters from MLRISC completely and replaced with graphs.
2229
2230		----------------------------------------------------------------------
2231		Name: Matthias Blume
2232		Date: 2001/08/23 17:50:00 EDT
2233		Tag: blume-20010823-toplevel
2234		Description:
2235
2236		- some reorganization of the code that implements various kinds of
2237		environments in the compiler (static, dynamic, symbolic, combined)
2238		- re-implemented the EnvRef module so that evalStream works properly
2239		(if the stream contains references to "use", "CM.make", etc.)
2240		- cleaned up evalloop.sml and interact.sml (but they need more cleaning)
2241
2242		----------------------------------------------------------------------
2243		Name: Matthias Blume
2244		Date: 2001/08/20 15:50 EDT
2245		Tag: blume20010820-slipup
2246		Description:
2247
2248		I forgot to commit a few files.  Here they are...
2249
2250		----------------------------------------------------------------------
2251		Name: Matthias Blume
2252		Date: 2001/08/20 15:35:00 EDT
2253		Tag: blume-20010820-debugprof
2254		Description:
2255
2256		!!!! NEW BOOTFILES !!!!
2257
2258		This is another round of reorganizing the compiler sources.  This
2259		time the main goal was to factor out all the "instrumentation"
2260		passes (for profiling and backtracing) into their own library.
2261		The difficulty was to do it in such a way that it does not depend
2262		on elaborate.cm but only on elabdata.cm.
2263
2264		Therefore there have been further changes to both elaborate.cm and
2265		elabdata.cm -- more "generic" things have been moved from the former
2266		to the latter.  As a result, I was forced to split the assignment
2267		of numbers indicating "primtyc"s into two portions: SML-generic and
2268		SML/NJ-specific.  Since it would have been awkward to maintain,
2269		I bit the bullet and actually _changed_ the mapping between these
2270		numbers and primtycs.  The bottom line of this is that you need
2271		a new set of bin- and bootfiles.
2272
2273		I have built new bootfiles for all architectures, so doing a fresh
2274		checkout and config/install.sh should be all you need.
2275
2276		The newly created library's name is
2277
2278		$smlnj/viscomp/debugprof.cm
2279
2280		and its sources live under
2281
2282		src/compiler/DebugProf
2283
2284		----------------------------------------------------------------------
2285		Name: Matthias Blume
2286		Date: 2001/08/15 17:15:00 EDT
2287		Tag: blume-20010815-compreorg
2288		Description:
2289
2290		This is a first cut at reorganizing the CM libraries that make up the
2291		core of the compiler.  The idea is to separate out pieces that could
2292		be used independently by tools, e.g., the parser, the typechecker, etc.
2293
2294		The current status is a step in this direction, but it is not quite
2295		satisfactory yet.  Expect more changes in the future.
2296
2297		Here is the current (new) organization...
2298
2299		What used to be $smlnj/viscomp/core.cm is now divided into
2300		six CM libraries:
2301
2302		$smlnj/viscomp/basics.cm
2303		/parser.cm
2304		/elabdata.cm
2305		/elaborate.cm
2306		/execute.cm
2307		/core.cm
2308
2309		The CM files for these libraries live under src/system/smlnj/viscomp.
2310		All these libraries are proxy libraries that contain precisely
2311		one CM library component.  Here are the locations of the components
2312		(all within the src/compiler tree):
2313
2314		Basics/basics.cm
2315		Parse/parser.cm
2316		ElabData/elabdata.cm
2317		Elaborator/elaborate.cm
2318		Execution/execute.cm
2319		core.cm
2320
2321		[This organization is the same that has been used already
2322		for a while for the architecture-specific parts of the visible
2323		compiler and for the old version of core.cm.]
2324
2325		As you will notice, many source files have been moved from their
2326		respective original locations to a new home in one of the above
2327		subtrees.
2328
2329		The division of labor between the new libraries is the following:
2330
2331		basics.cm:
2332		- Simple, basic definitions that pertain to many (or all) of
2333		the other libraries.
2334		parser.cm:
2335		- The SML parser, producing output of type Ast.dec.
2336		- The type family for Ast is also defined and exported here.
2337		elabdata.cm:
2338		- The datatypes that describe input and output of the elaborator.
2339		This includes types, absyn, and static environments.
2340		elaborator.cm:
2341		- The SML/NJ type checker and elaborator.
2342		This maps an Ast.dec (with a given static environment) to
2343		an Absyn.dec (with a new static environment).
2344		- This libraries implements certain modules that used to be
2345		structures as functors (to remove dependencies on FLINT).
2346		execute.cm:
2347		- Everything having to do with executing binary code objects.
2348		- Dynamic environments.
2349		core.cm:
2350		- SML/NJ-specific instantiations of the elaborator and MLRISC.
2351		- Top-level modules.
2352		- FLINT (this should eventually become its own library)
2353
2354		Notes:
2355
2356		I am not 100% happy with the way I separated the elaborator (and its
2357		data structures) from FLINT.  Two instances of the same problem:
2358
2359		1. Data structures contain certain fields that carry FLINT-specific
2360		information.  I hacked around this using exn and the property list
2361		module from smlnj-lib.  But the fact that there are middle-end
2362		specific fields around at all is a bit annoying.
2363
2364		2. The elaborator calculates certain FLINT-related information.  I tried
2365		to make this as abstract as I could using functorization, but, again,
2366		the fact that the elaborator has to perform calculations on behalf
2367		of the middle-end at all is not nice.
2368
2369		3. Having to used exn and property lists is unfortunate because it
2370		weakens type checking.  The other alternative (parameterizing
2371		nearly *everything*) is not appealing, though.
2372
2373		I removed the "rebinding =" warning hack because due to the new organization
2374		it was awkward to maintain it.  As a result, the compiler now issues some of
2375		these warnings when compiling init.cmi during bootstrap compilation. On
2376		the plus side, you also get a warning when you do, for example:
2377		val op = = Int32.+
2378		which was not the case up to now.
2379
2380		I placed "assign" and "deref" into the _Core structure so that the
2381		code that deals with the "lazy" keyword can find them there.  This
2382		removes the need for having access to the primitive environment
2383		during elaboration.
2384
2385		----------------------------------------------------------------------
2386		Name: Matthias Blume
2387		Date: 2001/08/13
2388		Tag: blume-20010813-closures
2389		Description:
2390
2391		This fix was sent to us by Zhong Shao.  It is supposed to improve the
2392		performance of certain loops by avoiding needless closure allocation.
2393
2394		----------------------------------------------------------------------
2395		Name: Lal George
2396		Date: 2001/07/31 10:03:23 EDT 2001
2397		Tag: george-20010731-x86-fmalloc
2398		Description: Fixed bug in x86 calls
2399
2400		There was a bug where call instructions would mysteriously
2401		vanish. The call instruction had to be one that returned
2402		a floating point value.
2403
2404		----------------------------------------------------------------------
2405		Name: Lal George
2406		Date: 2001/07/19 16:36:29 EDT 2001
2407		Tag: george-20010719-simple-cells
2408		Description:
2409
2410		I have dramatically simplified the interface for CELLS in MLRISC.
2411
2412		In summary, the cells interface is broken up into three parts:
2413
2414		1. CellsBasis : CELLS_BASIS
2415
2416		CellsBasis is a top level structure and common for all
2417		architectures.  it contains the definitions of basic datatypes
2418		and utility  functions over these types.
2419
2420		2. functor Cells() : CELLS
2421
2422		Cells generates an interface for CELLS that incorporates the
2423		specific resources on the target architecture, such as the
2424		presence of special register classes, their number and size,
2425		and various useful substructures.
2426
2427		3. <ARCH>CELLS
2428
2429		e.g. SparcCells: SPARCCELLS
2430
2431		<ARCH>CELLS usually contains additional bindings for special
2432		registers  on the architecture, such as:
2433
2434		val r0 : cell           (* register zero *)
2435		val y : cell            (* Y register *)
2436		val psr : cell          (* processor status register *)
2437		...
2438
2439		The structure returned by applying the Cells functor is opened
2440		in this interface.
2441
2442		The main implication of all this is that the datatypes for cells is
2443		split between CellsBasis and CELLS -- a fairly simple change for user
2444		code.
2445
2446		In the old scheme the CELLS interface had a definitional binding of
2447		the form:
2448
2449		signature CELLS = sig
2450
2451		structure CellsBasis = CellsBasis
2452
2453		...
2454
2455		end
2456
2457		With all the sharing constraints that goes on in MLRISC, this old
2458		design  quickly leads to errors such as:
2459
2460		"structure definition spec inside of sharing ... "
2461
2462
2463		and appears to require an unacceptable amount of sharing and where
2464		constraint hackery.
2465
2466		I think this error message (the interaction of definitional specs and
2467		sharing) requires more explanation on our web page.
2468
2469		----------------------------------------------------------------------
2470		Name: Matthias Blume
2471		Date: 2001/07/19 15:00:00 EDT
2472		Tag: blume-20010719-libreorg
2473		Description:
2474
2475		This update puts together a fairly extensive but straightforward change
2476		to the way the libraries that implement the interactive system are
2477		organized:
2478
2479		The biggest change is the elimination of structure Compiler.  As a
2480		replacement for this structure, there is now a CM library
2481		(known as $smlnj/compiler.cm or $smlnj/compiler/current.cm)
2482		that exports all the substructures of the original structure Compiler
2483		directly.  So instead of saying Compiler.Foo.bar one now simply
2484		says Foo.bar.  (The CM libraries actually export a collection of
2485		structures that is richer than the collection of substructures of
2486		structure Compiler.)
2487
2488		To make the transition smooth, there is a separate library called
2489		$smlnj/compiler/compiler.cm which puts together and exports the
2490		original structure Compiler (or at least something very close to it).
2491
2492		There are five members of the original structure Compiler
2493		that are not exported directly but which instead became members
2494		of a new structure Backend (described by signature BACKEND).  These are:
2495		structure Profile (: PROFILE), structure Compile (: COMPILE), structure
2496		Interact (: INTERACT), structure Machine (: MACHINE), and val
2497		architecture (: string).
2498
2499		Structure Compiler.Version has become structure CompilerVersion.
2500
2501		Cross-compilers for alpha32, hppa, ppc, sparc, and x86 are provided
2502		by $smlnj/compiler/<arch>.cm where <arch> is alpha32, hppa, ppc, sparc,
2503		or x86, respectively.
2504		Each of these exports the same frontend structures that
2505		$smlnj/compiler.cm exports.  But they do not have a structure Backend
2506		and instead export some structure <Arch>Backend where <Arch> is Alpha32,
2507		Hppa, PPC, Sparc, or X86, respectively.
2508
2509		Library $smlnj/compiler/all.cm exports the union of the exports of
2510		$smlnj/compiler/<arch>.cm
2511
2512		There are no structures <Arch>Compiler anymore, use
2513		$smlnj/compiler/<arch>.cm instead.
2514
2515		Library host-compiler-0.cm is gone.  Instead, the internal library
2516		that instantiates CM is now called cm0.cm.  Selection of the host
2517		compiler (backend) is no longer done here but. (Responsibility for it
2518		now lies with $smlnj/compiler/current.cm.  This seems to be more
2519		logical.)
2520
2521		Many individual files have been moved or renamed.  Some files have
2522		been split into multiple files, and some "dead" files have been deleted.
2523
2524		Aside from these changes to library organization, there are also changes
2525		to the way the code itself is organized:
2526
2527		Structure Binfile has been re-implemented in such a way that it no
2528		longer needs any knowledge of the compiler.  It exclusively deals
2529		with the details of binfile layout.  It no longer invokes the
2530		compiler (for the purpose of creating new prospective binfile
2531		content), and it no longer has any knowledge of how to interpret
2532		pickles.
2533
2534		Structure Compile (: COMPILE) has been stripped down to the bare
2535		essentials of compilation.  It no longer deals with linking/execution.
2536		The interface has been cleaned up considerably.
2537
2538		Utility routines for dealing with linking and execution have been
2539		moved into their own substructures.
2540
2541		(The ultimate goal of these changes is to provide a light-weight
2542		binfile loader/linker (at least for, e.g., stable libraries) that
2543		does not require CM or the compiler to be present.)
2544
2545		CM documentation has been updated to reflect the changes to library
2546		organization.
2547
2548		----------------------------------------------------------------------
2549		Name: Matthias Blume
2550		Date: 2001/07/10 17:30:00 EDT
2551		Tag: Release_110_34
2552		Description:
2553
2554		Minor tweak to 110.34 (re-tagged):
2555
2556		- README.html file added to CVS repository
2557		- runtime compiles properly under FreeBSD 3.X and 4.X
2558
2559		----------------------------------------------------------------------
2560		Name: Matthias Blume
2561		Date: 2001/07/10 17:30:00 EDT
2562		Tag: Release_110_34
2563		Description:
2564
2565		New version number (110.34). New bootfiles.
2566
2567		----------------------------------------------------------------------
2568		Name: Matthias Blume
2569		Date: 2001/07/09 16:00:00 EDT
2570		Tag: blume-20010709-more-varargs
2571		Description:
2572
2573		I changed the handling of varargs in ml-nlffigen again:
2574		The ellipsis ... will now simply be ignored (with an accompanying warning).
2575
2576		The immediate effect is that you can actually call a varargs function
2577		from ML -- but you can't actually supply any arguments beyond the ones
2578		specified explicitly.  (For example, you can call printf with its format
2579		string, but you cannot pass additional arguments.)
2580
2581		This behavior is only marginally more useful than the one before, but
2582		it has the advantage that a function or, more importantly, a function
2583		type never gets dropped on the floor, thus avoiding follow-up problems with
2584		other types that refer to the offending one.
2585
2586		----------------------------------------------------------------------
2587		Name: Matthias Blume
2588		Date: 2001/07/09 11:25:00 EDT
2589		Tag: blume-20010709-varargs
2590		Description:
2591
2592		1. ckit-lib.cm now exports structure Error
2593		2. ml-nlffigen reports occurences of "..." (i.e., varargs function types)
2594		with a warning accompanied by a source location.  Moreover, it
2595		merely skips the offending function or type and proceeds with the
2596		rest of its work.u  As a result, one can safely feed C code containing
2597		"..." to ml-nlffigen.
2598		3. There are some internal improvements to CM, providing slightly
2599		more general string substitutions in the tools subsystem.
2600
2601		----------------------------------------------------------------------
2602		Name: Matthias Blume
2603		Date: 2001/06/27 15:10:00 EDT
2604		Tag: blume-20010627-concur
2605		Description:
2606
2607		Fixed a small bug in CM's handling of parallel compilation.
2608		(You could observe the bug by Control-C-interrupting an ordinary
2609		CMB.make or CM.stabilize and then attaching some compile servers.
2610		The result was that all of a sudden the previously interrupted
2611		compilation would continue on its own.  This was because of
2612		an over-optimization: CM did not bother to clean out certain queues
2613		when no servers were attached "anyway", resulting in the contents
2614		of these queues to grab control when new servers did get attached.)
2615
2616		There is also another minor update to the CM manual.
2617
2618		----------------------------------------------------------------------
2619		Name: Matthias Blume
2620		Date: 2001/06/26 16:15:00 EDT
2621		Tag: blume-20010626-cmdoc
2622		Description:
2623
2624		Minor typo fixed in CM manual (syntax diagram for libraries).
2625
2626		----------------------------------------------------------------------
2627		Name: Matthias Blume
2628		Date: 2001/06/25 22:55:00 EDT
2629		Tag: blume-20010625-x86pc
2630		Description:
2631
2632		Fixed a nasty bug in the X86 assembly code that caused signal
2633		handlers to fail (crash) randomly.
2634
2635		----------------------------------------------------------------------
2636		Name: Matthias Blume
2637		Date: 2001/06/25 12:05:00 EDT
2638		Tag: blume-20010625-nlffigen
2639		Description:
2640
2641		This update fixes a number of minor bugs in ml-nlffigen as reported by
2642		Nick Carter <nbc@andrew.cmu.edu>.
2643
2644		1. Silly but ok typedefs of the form "typedef void myvoid;" are now accepted.
2645		2. Default names for generated files are now derived from the name of
2646		the C file *without its directory*.  In particular, this causes generated
2647		files to be placed locally even if the C file is in some system directory.
2648		3. Default names for generated signatures and structures are also derived
2649		from the C file name without its directory.  This avoids silly things
2650		like "structure GL/GL".
2651		(Other silly names are still possible because ml-nlffigen does not do
2652		a thorough check of whether generated names are legal ML identifiers.
2653		When in doubt, use command line arguments to force particular names.)
2654
2655		----------------------------------------------------------------------
2656		Name: Matthias Blume
2657		Date: 2001/06/21 12:25:00 EDT
2658		Tag: blume-20010621-eXene
2659		Description:
2660
2661		eXene now compiles and (sort of) works again.
2662
2663		The library name (for version > 110.33) is $/eXene.cm.
2664
2665		I also added an new example in src/eXene/examples/nbody.  See the
2666		README file there for details.
2667
2668		----------------------------------------------------------------------
2669		Name: Matthias Blume
2670		Date: 2001/06/20 16:40:00 EDT
2671		Tag: blume-20010620-cml
2672		Description:
2673
2674		CML now compiles and works again.
2675
2676		Libraries (for version > 110.33):
2677
2678		$cml/cml.cm            Main CML library.
2679		$cml/basis.cm          CML's version of $/basis.cm.
2680		$cml/cml-internal.cm   Internal helper library.
2681		$cml/core-cml.cm       Internal helper library.
2682		$cml-lib/trace-cml.cm  Tracing facility.
2683		$cml-lib/smlnj-lib.cm  CML's version of $/smlnj-lib.cm
2684
2685		The installer (config/install.sh) has been taught how to properly
2686		install this stuff.
2687
2688		----------------------------------------------------------------------
2689		Name: Matthias Blume
2690		Date: 2001/06/19 17:55:00 EDT
2691		Tag: blume-20010619-instantiate
2692		Description:
2693
2694		This un-breaks the fix for bug 1432.
2695		(The bug was originally fixed in 110.9 but I broke it again some
2696		time after that.)
2697
2698		----------------------------------------------------------------------
2699		Name: Matthias Blume
2700		Date: 2001/06/19 17:25:00 EDT
2701		Tag: blume-20010619-signals
2702		Description:
2703
2704		This should (hopefully) fix the long-standing signal handling bug.
2705		(The runtime system was constructing a continuation record with an
2706		incorrect descriptor which would cause the GC to drop data on the floor...)
2707
2708		----------------------------------------------------------------------
2709		Name: Matthias Blume
2710		Date: 2001/06/15 15:05:00 EDT
2711		Tag: blume-20010615-moresparc
2712		Description:
2713
2714		Here is a short late-hour update related to Sparc c-calls:
2715
2716		-- made handling of double-word arguments a bit smarter
2717
2718		-- instruction selection phase tries to collapse certain clumsily
2719		constructed ML-Trees; typical example:
2720
2721		ADD(ty,ADD(_,e,LI d1),LI d2)  ->  ADD(ty,e,LI(d1+d2))
2722
2723		This currently has no further impact on SML/NJ since mlriscGen does
2724		not seem to generate such patterns in the first place, and c-calls
2725		(which did generate them in the beginning) has meanwhile been fixed
2726		so as to avoid them as well.
2727
2728		----------------------------------------------------------------------
2729		Name: Matthias Blume
2730		Date: 2001/06/15 15:05:00 EDT
2731		Tag: blume-20010615-sparc
2732		Description:
2733
2734		The purpose of this update is to provide an implementation of NLFFI
2735		on Sparc machines.
2736
2737		Here are the changes in detail:
2738
2739		* src/MLRISC/sparc/c-calls/sparc-c-calls.sml is a new file containing
2740		the Sparc implementation of the c-calls API.
2741		* The Sparc backend of SML/NJ has been modified to uniformely use %fp
2742		for accessing the ML frame.  Thus, we have a real frame pointer and
2743		can freely modify %sp without need for an omit-frame-ptr phase.
2744		The vfp logic in src/compiler/CodeGen/* has been changed to accomodate
2745		this case.
2746		* ml-nlffigen has been taught to produce code for different architectures
2747		and calling conventions.
2748		* In a way similar to what was done in the x86 case, the Sparc
2749		backend uses its own specific extension to mltree.  (For example,
2750		it needs to be able to generate UNIMP instructions which are part
2751		of the calling convention.)
2752		* ml-nlffi-lib was reorganized to make it more modular (in particular,
2753		to make it easier to plug in new machine- and os-dependent parts).
2754
2755		There are some other fairly unrelated bug fixes and cleanups as well:
2756
2757		* I further hacked the .cm files for MLRISC tools (like MDLGen) so
2758		that they properly share their libraries with existing SML/NJ libraries.
2759		* I fixed a minor cosmetic bug in CM, supressing certain spurious
2760		follow-up error messages.
2761		* Updates to CM/CMB documentation.
2762
2763		TODO items:
2764
2765		* MLRISC should use a different register as its asmTemp on the Sparc.
2766		(The current %o2 is a really bad choice because it is part of the
2767		calling conventions, so things might interfere in unexpected ways.)
2768
2769		----------------------------------------------------------------------
2770		Name: Matthias Blume
2771		Date: 2001/06/07
2772		Tag: blume-20010607-calls
2773		Description:
2774
2775		A number of internal changes related to C calls and calling conventions:
2776
2777		1. ML-Tree CALL statements now carry a "pops" field.  It indicates the
2778		number of bytes popped implicitly (by the callee).  In most cases
2779		this field is 0 but on x86/win32 it is some non-zero value.  This
2780		is information provided for the benefit of the "omit-frameptr" pass.
2781		2. The CALL instruction on the x86 carries a similar "pops" field.
2782		The instruction selection phase copies its value from the ML-Tree
2783		CALL statement.
2784		3. On all other architectures, the instruction selection phase checks
2785		whether "pops=0" and complains if not.
2786		4. The c-calls implementation for x86 now accepts two calling conventions:
2787		"ccall" and "stdcall".  When "ccall" is selected, the caller cleans
2788		up after the call and pops is set to 0.  For "stdcall", the caller
2789		does nothing, leaving the cleanup to the callee; pops is set to
2790		the number of bytes that were pushed onto the stack.
2791		5. The cproto decoder (compiler/Semant/types/cproto.sml) now can
2792		distinguish between "ccall" and "stdcall".
2793		6. The UNIMP instruction has been added to the supported Sparc instruction
2794		set. (This is needed for implementing the official C calling convention
2795		on this architecture.)
2796		7. I fixed some of the .cm files under src/MLRISC/Tools to make them
2797		work with the latest CM.
2798
2799		----------------------------------------------------------------------
2800		Name: Matthias Blume
2801		Date: 2001/06/05 15:10:00 EDT
2802		Tag: blume-20010605-cm-index
2803		Description:
2804
2805		0. The "lambdasplit" parameter for class "sml" in CM has been documented.
2806
2807		1. CM can now generate "index files".  These are human-readable files
2808		that list on a per-.cm-file basis each toplevel symbol defined or
2809		imported.  The location of the index file for
2810		<p>/<d>.cm is <p>/CM/INDEX/<d>.cm.
2811		To enable index-file generation, set CM.Control.generate_index to true
2812		or export an environment-symbol: export CM_GENERATE_INDEX=true.
2813
2814		The CM manual has been updated accordingly.
2815
2816		2. I made some slight modifications to the c-calls API in MLRISC.
2817
2818		a) There is now a callback to support saving/restoring of
2819		dedicated but caller-save registers around the actual call
2820		instruction.
2821		b) One can optionally specify a comment-annotation for the
2822		call instruction.
2823
2824		3. SML/NJ (mlriscGen.sml) uses this new API for the rawccall primop.
2825		(For example, the comment annotation shows the C prototype of
2826		the function being called.)
2827
2828		----------------------------------------------------------------------
2829		Name: Matthias Blume
2830		Date: 2001/06/01 13:30:00 EDT
2831		Tag: blume-20010601-nlffi-cleanup
2832		Description:
2833
2834		This is mostly a cleanup of MLFFI stuff:
2835
2836		- some signature files have been put into a more exposed place
2837		- the ugly 'f type parameter is gone (simplifies types tremendously!)
2838		- ml-nlffigen changed accordingly
2839		- tutorial updated
2840
2841		Other changes:
2842
2843		- author's affiliation in CM manual(s) updated
2844		- some more recognized keywords added to Allen's sml.sty
2845
2846		----------------------------------------------------------------------
2847		Name: Matthias Blume
2848		Date: 2001/05/25 15:30:00 EDT
2849		Tag: blume-20010525-iptr
2850		Description:
2851
2852		- put the official 110.33-README (as it appears on the ftp server) under
2853		CVS
2854		- fixed a small bug related to incomplete pointer types in
2855		ml-nlffigen
2856		- small cosmetic change to the ml-nlffi-lib's "arr" type constructor
2857		(it does not need the 'f type parameter)
2858
2859		----------------------------------------------------------------------
2860		Name: Matthias Blume
2861		Date: 2001/05/23 14:30:00 EDT
2862		Tag: Release_110_33
2863		Description:
2864
2865		New version number (110.33).  New bootfiles.
2866
2867		----------------------------------------------------------------------
2868		Name: Matthias Blume
2869		Date: 2001/05/22 18:06:00 EDT
2870		Tag: blume-20010522-targets
2871		Description:
2872
2873		Made install.sh use file config/targets.customized if it exists, falling
2874		back to config/targets if it doesn't.  This way one can have a customized
2875		version of the targets file without touching the "real thing", thus
2876		eliminating the constant fear of accidentally checking something bogus
2877		back into the CVS repository...  (File config/targets.customized must
2878		not be added to the repository!)
2879
2880		----------------------------------------------------------------------
2881		Name: Matthias Blume
2882		Date: 2001/05/22 16:30:00 EDT
2883		Tag: blume-20010522-minitut
2884		Description:
2885
2886		1. Bug fix in ml-nlffigen; now (hopefully) correctly handling
2887		struct returns.
2888		2. Added src/ml-nlffi-lib/Doc/mini-tutorial.txt.  This is some very
2889		incomplete, preliminary documentation for NLFFI.
2890
2891		----------------------------------------------------------------------
2892		Name: Matthias Blume
2893		Date: 2001/05/14 11:30:00 EDT
2894		Tag: blume-20010514-script
2895		Description:
2896
2897		Some bugs in install script fixed.
2898
2899		In addition to that I also made a slight change to the NLFFI API:
2900		Functors generated by ml-nlffigen now take the dynamic library as a
2901		straight functor argument, not as a suspended one.  (The original
2902		functor code used to force the suspension right away anyway, so there
2903		was nothing gained by this complication of the interface.)
2904
2905		----------------------------------------------------------------------
2906		Name: Matthias Blume
2907		Date: 2001/05/11 14:35:00 EDT
2908		Tag: blume-20010511-ml-nlffi
2909		Description:
2910
2911		I finally took the plunge and added my new FFI code to the main
2912		repository.  For x86-linux it is now ready for prime-time.
2913
2914		There are two new subdirectories of "src":
2915
2916		- ml-nlffi-lib:
2917		The utility library for programs using the FFI interface.
2918		Here is the implementation of $/c.cm and its associated low-level
2919		partners $/c-int.cm and $/memory.cm.
2920		- ml-nlffigen:
2921		A stand-alone program for generating ML glue code from C source
2922		code.
2923
2924		Building ml-nlffigen requires $/ckit-lib.cm.
2925
2926		The config/install.sh script has been updates to do the Right Thing
2927		(hopefully).
2928
2929		Notice that the source tree for the C-Kit will not be put under "src"
2930		but directly under the installation root directory.  (This is the
2931		structure that currently exists on the CVS server when you check out
2932		module "sml".)  Fortunately, config/install.sh knows about this oddity.
2933
2934		Bugs: No documentation yet.
2935
2936		----------------------------------------------------------------------
2937		Name: Matthias Blume
2938		Date: 2001/05/09 16:35:00 EDT
2939		Tag: blume-20010509-cpscontract
2940		Description:
2941
2942		Fixed a bug in the accounting code in cpsopt/contract.sml.  (The
2943		wrapper/unwrapper elimination did not decrement usage counts and some
2944		dead variables got overlooked by the dead-up logic.)
2945
2946		----------------------------------------------------------------------
2947		Name: Lal George
2948		Date: 2001/05/08  17:26:09 EDT
2949		Tag: george-20010508-omit-frameptr
2950		Description:
2951
2952		Changes to implement the omit-frame-pointer optimization to support
2953		raw C calls. For now, there is only support on the Intel x86, but
2954		other architectures will follow as more experience is gained with this.
2955
2956
2957		----------------------------------------------------------------------
2958		Name: Matthias Blume
2959		Date: 2001/05/07 14:40:00 EDT
2960		Tag: blume-20010507-proxies
2961		Description:
2962
2963		I made into "proxy libraries" all libraries that qualify for such a
2964		change.  (A qualifying library is a library that has another library or
2965		groups as its sole member and repeats that member's export list
2966		verbatim.  A proxy library avoids this repetition by omitting its export
2967		list, effectively inheriting the list that its (only) member exports.
2968		See the CM manual for more explanation.)
2969		The main effect is that explicit export lists for these libraries
2970		do not have to be kepts in sync, making maintenance a bit easier.
2971
2972		I also added copyright notices to many .cm-files.
2973
2974		Last but not least, I made a new set of bootfiles.
2975
2976		----------------------------------------------------------------------
2977		Name: Matthias Blume
2978		Date: 2001/05/04 17:00:00 EDT
2979		Tag: blume-20010504-cm-lsplit
2980		Description:
2981
2982		0. John merged pending changes to $/smlnj-lib.cm
2983
2984		1. Allen's previous change accidentally backed out of one of Lal's
2985		earlier changes.  I undid this mistake (re-introducing Lal's change).
2986
2987		2. I used the new topOrder' function from graph-scc.sml (from $/smlnj-lib.cm)
2988		within the compiler where applicable.  There is some code simplification
2989		because of that.
2990
2991		3. The "split" phase (in FLINT) is now part of the default list of phases.
2992		Compiler.Control.LambdaSplitting.* can be used to globally control the
2993		lambda-splitting (cross-module-inlining) engine.  In addition to that,
2994		it can now also be controlled on a per-source basis: CM has been taught
2995		a new tool parameter applicable to ML source files.
2996
2997		- To turn lambda-splitting off completely:
2998		local open Compiler.Control.LambdaSplitting in
2999		val _ = set Off
3000		end
3001		- To make "no lambda-splitting" the global default (but allow per-source
3002		overriding); this is the initial setting:
3003		local open Compiler.Control.LambdaSplitting in
3004		val _ = set (Default NONE)
3005		end
3006		- To make "lambda-splitting with aggressiveness a" the global default
3007		(and allow per-source overriding):
3008		local open Compiler.Control.LambdaSplitting in
3009		val _ = set (Default (SOME a))
3010		end
3011
3012		- To turn lambda-splitting off for a given ML souce file (say: a.sml)
3013		write (in the respective .cm-file):
3014		a.sml (lambdasplitting:off)
3015		- To turn lambda-splitting for a.sml on with minimal aggressiveness:
3016		a.sml (lambdasplitting:on)
3017		- To turn lambda-splitting for a.sml on with aggressiveness <a> (where
3018		<a> is a decimal non-negative integer):
3019		a.sml (lambdasplitting:<a>)
3020		- To turn lambda-splitting for a.sml on with maximal aggressiveness:
3021		a.sml (lambdasplitting:infinity)
3022		- To use the global default for a.sml:
3023		a.sml (lambdasplitting:default)
3024		or simply
3025		a.sml
3026
3027	----------------------------------------------------------------------	----------------------------------------------------------------------
3028	Name: Allen Leung	Name: Allen Leung
3029	Date: 2001/05/04 01:57:00 EDT	Date: 2001/05/04 01:57:00 EDT
#	Line 2109	Line 5122
5122	elaborator). There were a lot of changes during my "linkpath" trials	elaborator). There were a lot of changes during my "linkpath" trials
5123	that could have been reverted to their original state but weren't.	that could have been reverted to their original state but weren't.
5124	Please, don't be too harsh on me for messing with this code a bit more	Please, don't be too harsh on me for messing with this code a bit more
5125	than what was strictly necessary...  (I _did_ resist the tempation	than what was strictly necessary...  (I _did_ resist the temptation
5126	of doing any "global reformatting" to avoid an untimely death at	of doing any "global reformatting" to avoid an untimely death at
5127	Dave's hands. :)	Dave's hands. :)
5128

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.815   changed lines   Added in v.1249

---

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