Diff of /sml/trunk/HISTORY

Parent Directory | Revision Log | Patch

	revision 824, Tue May 8 21:31:05 2001 UTC	revision 1184, Fri Mar 29 22:22:22 2002 UTC
#	Line 8	Line 8
8	The form of an entry should be:	The form of an entry should be:
9
10	Name:	Name:
11	Date:	Date: yyyy/mm/dd
12	Tag: <post-commit CVS tag>	Tag: <post-commit CVS tag>
13	Description:	Description:
14
15	----------------------------------------------------------------------	----------------------------------------------------------------------
16		Name: Matthias Blume
17		Date: 2002/03/29 17:22:00
18		Tag: blume-20020329-div
19		Description:
20
21		Fixed my broken div/mod logic.  Unfortunately, this means that the
22		inline code for div/mod now has one more comparison than before.
23		Fast paths (quotient > 0 or remainder = 0) are not affected, though.
24		The problem was with quotient = 0, because that alone does not tell
25		us which way the rounding went.  One then has to look at whether
26		remainder and divisor have the same sign...  :(
27
28		Anyway, I replaced the bootfiles with fresh ones...
29
30		----------------------------------------------------------------------
31		Name: Matthias Blume
32		Date: 2002/03/29 14:10:00 EST
33		Tag: blume-20020329-inlprims
34		Description:
35
36		NEW BOOTFILES!!!    Version number bumped to 110.39.3.
37
38		Primops have changed. This means that the bin/boot-file formats have
39		changed as well.
40
41		To make sure that there is no confusion, I made a new version.
42
43
44		CHANGES:
45
46		* removed REMT from mltree (remainder should never overflow).
47
48		* added primops to deal with divisions of all flavors to the frontend
49
50		* handled these primops all the way through so they map to their respective
51		MLRISC support
52
53		* used these primops in the implementation of Int, Int32, Word, Word32
54
55		* removed INLDIV, INLMOD, and INLREM as they are no longer necessary
56
57		* parameterized INLMIN, INLMAX, and INLABS by a numkind
58
59		* translate.sml now deals with all flavors of INL{MIN,MAX,ABS}, including
60		floating point
61
62		* used INL{MIN,MAX,ABS} in the implementation of Int, Int32, Word, Word32,
63		and Real (but Real.abs maps to a separate floating-point-only primop)
64
65
66		TODO items:
67
68		* Hacked Alpha32 instruction selection, disabling the selection of REMx
69		instructions because the machine instruction encoder cannot handle
70		them.  (Hppa, PPC, and Sparc instruction selection did not handle
71		REM in the first place, and REM is supported by the x86 machine coder.)
72
73		* Handle DIV and MOD with DIV_TO_NEGINF directly in the x86 instruction
74		selection phase.  (The two can be streamlined because the hardware
75		delivers both quotient and remainder at the same time anyway.)
76
77		* Think about what to do with "valOf(Int32.minInt) div ~1" and friends.
78		(Currently the behavior is inconsistent both across architectures and
79		wrt. the draft Basis spec.)
80
81		* Word8 should eventually be handled natively, too.
82
83		* There seems to be one serious bug in mltree-gen.sml.  It appears, though,
84		as if there currently is no execution path that could trigger it in
85		SML/NJ.  (The assumptions underlying functions arith and promotable do not
86		hold for things like multiplication and division.)
87
88		----------------------------------------------------------------------
89		Name: Matthias Blume
90		Date: 2002/03/27 16:27:00 EST
91		Tag: blume-20020327-mlrisc-divisions
92		Description:
93
94		Added support for all four division operations (ML's div, mod, quot,
95		and rem) to MLRISC.  In the course of doing so, I also rationalized
96		the naming (no more annoying switch-around of DIV and QUOT), by
97		parameterizing the operation by div_rounding_mode (which can be either
98		DIV_TO_ZERO or DIV_TO_NEGINF).
99
100		The generic MLTreeGen functor takes care of compiling all four
101		operations down to only round-to-zero div.
102
103		Missing pieces:
104
105		* Doing something smarter than relying on MLTreeGen on architectures
106		like, e.g., the x86 where hardware division delivers both quotient and
107		remainder at the same time.  With this, the implementation of the
108		round-to-neginf operations could be further streamlined.
109
110		* Remove inlining support for div/mod/rem from the frontend and replace it
111		with primops that get carried through to the backend.  Do this for all
112		int and word types.
113
114		----------------------------------------------------------------------
115		Name: Matthias Blume
116		Date: 2002/03/25 17:25:00 EST
117		Tag: blume-20020325-divmod
118		Description:
119
120		I improved (hopefully without breaking them) the implementation of Int.div,
121		Int.mod, and Int.rem.   For this, the code in translate.sml now takes
122		advantage of the following observations:
123
124		Let  q = x quot y      r = x rem y
125		d = x div  y      m = x mod y
126
127		where "quot" is the round-to-zero version of integer division that
128		hardware usually provides.  Then we have:
129
130		r = x - q * y        where neither the * nor the - will overflow
131		d = if q >= 0 orelse x = q * y then q else q - 1
132		where neither the * nor the - will overflow
133		m = if q >= 0 orelse r = 0 then r else r + y
134		where the + will not overflow
135
136		This results in substantial simplification of the generated code.
137		The following table shows the number of CFG nodes and edges generated
138		for
139		fun f (x, y) = x OPER y
140		(* with OPER \in div, mod, quot, rem *)
141
142
143		OPER | nodes(old) | edges(old) | nodes(new) | edges(new)
144		--------------------------------------------------------
145		div |         24 |         39 |         12 |         16
146		mod |         41 |         71 |         12 |         16
147		quot |          8 |         10 |          8 |         10
148		rem |         10 |         14 |          8 |         10
149
150
151		----------------------------------------------------------------------
152		Name: Matthias Blume
153		Date: 2002/03/25 22:06:00 EST
154		Tag: blume-20020325-cprotobug
155		Description:
156
157		Fixed a bug in cproto (c prototype decoder).
158
159		----------------------------------------------------------------------
160		Name: Matthias Blume
161		Date: 2002/03/25 16:00:00 EST
162		Tag: blume-20020325-raw-primops
163		Description:
164
165		I did some cleanup to Allen's new primop code and
166		replaced yesterday's bootfiles with new ones.
167		(But they are stored in the same place.)
168
169		----------------------------------------------------------------------
170		Name: Matthias Blume
171		Date: 2002/03/24 22:40:00 EST
172		Tag: blume-20020324-bootfiles
173		Description:
174
175		Made the bootfiles that Allen asked for.
176
177		----------------------------------------------------------------------
178		Name: Allen Leung
179		Date: 2002/03/23 15:50:00 EST
180		Tag: leunga-20020323-flint-cps-rcc-primops
181		Description:
182
183		1. Changes to FLINT primops:
184
185		(* make a call to a C-function;
186		* The primop carries C function prototype information and specifies
187		* which of its (ML-) arguments are floating point. C prototype
188		* information is for use by the backend, ML information is for
189		* use by the CPS converter. *)
190		| RAW_CCALL of { c_proto: CTypes.c_proto,
191		ml_args: ccall_type list,
192		ml_res_opt: ccall_type option,
193		reentrant : bool
194		} option
195		(* Allocate uninitialized storage on the heap.
196		* The record is meant to hold short-lived C objects, i.e., they
197		* are not ML pointers.  With the tag, the representation is
198		* the same as RECORD with tag tag_raw32 (sz=4), or tag_fblock (sz=8)
199		*)
200		| RAW_RECORD of {tag:bool,sz:int}
201		and ccall_type = CCALL_INT32 | CCALL_REAL64 | CCALL_ML_PTR
202
203		2.  These CPS primops are now overloaded:
204
205		rawload of {kind:numkind}
206		rawstore of {kind:numkind}
207
208		The one argument form is:
209
210		rawload {kind} address
211
212		The two argument form is:
213
214		rawload {kind} [ml object, byte-offset]
215
216		3. RAW_CCALL/RCC now takes two extra arguments:
217
218		a. The first is whether the C call is reentrant, i.e., whether
219		ML state should be saved and restored.
220		b. The second argument is a string argument specifying the name of
221		library and the C function.
222
223		These things are currently not handled in the code generator, yet.
224
225		4. In CProto,
226
227		An encoding type of "bool" means "ml object" and is mapped into
228		C prototype of PTR.  Note that "bool" is different than "string",
229		even though "string" is also mapped into PTR, because "bool"
230		is assigned an CPS type of BOGt, while "string" is assigned INT32t.
231
232		5. Pickler/unpicker
233
234		Changed to handle RAW_RECORD and newest RAW_CCALL
235
236		6. MLRiscGen,
237
238		1. Changed to handle the new rawload/rawstore/rawrecord operators.
239		2. Code for handling C Calls has been moved to a new module CPSCCalls,
240		in the file CodeGen/cpscompile/cps-c-calls.sml
241
242		7. Added the conditional move operator
243
244		condmove of branch
245
246		to cps.  Generation of this is still buggy so it is currently
247		disabled.
248
249		----------------------------------------------------------------------
250		Name: Lal George
251		Date: 2002/03/22 14:18:25 EST
252		Tag: george-20020322-cps-branch-prob
253		Description:
254
255		Implemented the Ball-Larus branch prediction-heuristics, and
256		incorporated graphical viewers for control flow graphs.
257
258		Ball-Larus Heuristics:
259		---------------------
260		See the file compiler/CodeGen/cpscompile/cpsBranchProb.sml.
261
262		By design it uses the Dempster-Shafer theory for combining
263		probabilities.  For example, in the function:
264
265		fun f(n,acc) = if n = 0 then acc else f(n-1, n*acc)
266
267		the ball-larus heuristics predicts that the n=0 is unlikely
268		(OH-heuristic), and the 'then' branch is unlikely because of the
269		RH-heuristic -- giving the 'then' branch an even lower combined
270		probability using the Dempster-Shafer theory.
271
272		Finally, John Reppy's loop analysis in MLRISC, further lowers the
273		probability of the 'then' branch because of the loop in the else
274		branch.
275
276
277		Graphical Viewing:
278		------------------
279		I merely plugged in Allen's graphical viewers into the compiler. The
280		additional code is not much. At the top level, saying:
281
282		Control.MLRISC.getFlag "cfg-graphical-view" := true;
283
284		will display the graphical view of the control flow graph just before
285		back-patching.  daVinci must be in your path for this to work. If
286		daVinci is not available, then the default viewer can be changed
287		using:
288
289		Control.MLRISC.getString "viewer"
290
291		which can be set to "dot" or "vcg" for the corresponding viewers. Of
292		course, these viewers must be in your path.
293
294		The above will display the compilation unit at the level of clusters,
295		many of which are small, boring, and un-interesting. Also setting:
296
297		Control.MLRISC.getInt "cfg-graphical-view_size"
298
299		will display clusters that are larger than the value set by the above.
300
301
302		----------------------------------------------------------------------
303		Name: Matthias Blume
304		Date: 2002/03/21 22:20:00 EST
305		Tag: blume-20020321-kmp-bugfix
306		Description:
307
308		Changed the interface to the KMP routine in PreString and fixed
309		a minor bug in one place where it was used.
310
311		----------------------------------------------------------------------
312		Name: Allen Leung
313		Date: 2002/03/21 20:30:00 EST
314		Tag: leunga-20020321-cfg
315		Description:
316
317		Fixed a potential problem in cfg edge splitting.
318
319		----------------------------------------------------------------------
320		Name: Allen Leung
321		Date: 2002/03/21 17:15:00 EST
322		Tag: leunga-20020321-x86-fp-cfg
323		Description:
324
325		1. Recoded the buggy parts of x86-fp.
326
327		a. All the block reordering code has been removed.
328		We now depend on the block placement phases to do this work.
329
330		b. Critical edge splitting code has been simplified and moved into the
331		CFG modules, as where they belong.
332
333		Both of these were quite buggy and complex.  The code is now much, much
334		simpler.
335
336		2. X86 backend.
337
338		a. Added instructions for 64-bit support.  Instruction selection for
339		64-bit has not been committed, however, since that
340		requires changes to MLTREE which haven't been approved by
341		Lal and John.
342
343		b. Added support for FUCOMI and FUCOMIP when generating code for
344		PentiumPro and above.  We only generate these instructions in
345		the fast-fp mode.
346
347		c. Added cases for JP and JNP in X86FreqProps.
348
349		3. CFG
350
351		CFG now has a bunch of methods for edge splitting and merging.
352
353		4. Machine description.
354
355		John's simplification of MLTREE_BASIS.fcond broke a few machine
356		description things:
357
358		rtl-build.{sig,sml} and hppa.mdl fixed.
359
360		NOTE: the machine description stuff in the repository is still broken.
361		Again, I can't put my fixes in because that involves
362		changes to MLTREE.
363
364		----------------------------------------------------------------------
365		Name: Matthias Blume
366		Date: 2002/03/20 15:55:00 EST
367		Tag: blume-20020320-kmp
368		Description:
369
370		Implemented Knuth-Morris-Pratt string matching in PreString and used
371		it for String.isSubstring, Substring.isSubstring, and
372		Substring.position.
373
374		(Might need some stress-testing.  Simple examples worked fine.)
375
376		----------------------------------------------------------------------
377		Name: Matthias Blume
378		Date: 2002/03/19 16:37:00 EST
379		Tag: blume-20020319-witnesses
380		Description:
381
382		Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
383
384		This implements a generic mechanism for changing constness qualifiers
385		anywhere within big C types without resorting to outright "casts".
386		(So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
387		modify the constness at the outermost level.)
388		The implementation of "convert" is based on the idea of "witness"
389		values -- values that are not used by the operation but whose types
390		"testify" to their applicability.  On the implementation side, "convert"
391		is simply a projection (returning its second curried argument).  With
392		cross-module inlining, it should not result in any machine code being
393		generated.
394
395		----------------------------------------------------------------------
396		Name: Matthias Blume
397		Date: 2002/03/15 16:40:00 EST
398		Tag: blume-20020315-basis
399		Description:
400
401		Provided (preliminary?) implementations for
402
403		{String,Substring}.{concatWith,isSuffix,isSubstring}
404
405		and
406
407		Substring.full
408
409		Those are in the Basis spec but they were missing in SML/NJ.
410
411		----------------------------------------------------------------------
412		Name: Matthias Blume
413		Date: 2002/03/14 21:30:00 EST
414		Tag: blume-20020314-controls
415		Description:
416
417		Controls:
418		---------
419
420		1. Factored out the recently-added Controls : CONTROLS stuff and put
421		it into its own library $/controls-lib.cm.  The source tree for
422		this is under src/smlnj-lib/Controls.
423
424		2. Changed the names of types and functions in this interface, so they
425		make a bit more "sense":
426
427		module -> registry
428		'a registry -> 'a group
429
430		3. The interface now deals in ref cells only.  The getter/setter interface
431		is (mostly) gone.
432
433		4. Added a function that lets one register an already-existing ref cell.
434
435		5. Made the corresponding modifications to the rest of the code so that
436		everything compiles again.
437
438		6. Changed the implementation of Controls.MLRISC back to something closer
439		to the original.  In particular, this module (and therefore MLRISC)
440		does not depend on Controls.  There now is some link-time code in
441		int-sys.sml that registers the MLRISC controls with the Controls
442		module.
443
444		CM:
445		---
446
447		* One can now specify the lambda-split aggressiveness in init.cmi.
448
449		----------------------------------------------------------------------
450		Name: Allen Leung
451		Date: 2002/03/13 17:30:00 EST
452		Tag: leunga-20020313-x86-fp-unary
453		Description:
454
455		Bug fix for:
456
457		> leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
458		> Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
459		> - fun f(x,(y,z)) = Real.~ y;
460		> [autoloading]
461		> [autoloading done]
462		>       fchsl   (%eax), 184(%esp)
463		> Error: MLRisc bug: X86MCEmitter.emitInstr
464		>
465		> uncaught exception Error
466		>   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
467
468		The problem was that the code generator did not generate any fp registers
469		in this case, and the ra didn't know that it needed to run the X86FP phase to
470		translate the pseudo fp instruction.   This only happened with unary fp
471		operators in certain situations.
472
473		----------------------------------------------------------------------
474		Name: Matthias Blume
475		Date: 2002/03/13 14:00:00 EST
476		Tag: blume-20020313-overload-etc
477		Description:
478
479		1. Added _overload as a synonym for overload for backward compatibility.
480		(Control.overloadKW must be true for either version to be accepted.)
481
482		2. Fixed bug in install script that caused more things to be installed
483		than what was requested in config/targets.
484
485		3. Made CM aware of the (_)overload construct so that autoloading
486		works.
487
488		----------------------------------------------------------------------
489		Name: Matthias Blume
490		Date: 2002/03/12 22:03:00 EST
491		Tag: blume-20020312-url
492		Description:
493
494		Forgot to update BOOT and srcarchiveurl.
495
496		----------------------------------------------------------------------
497		Name: Matthias Blume
498		Date: 2002/03/12 17:30:00 EST
499		Tag: blume-20020312-version110392
500		Description:
501
502		Yet another version number bump (because of small changes to the
503		binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
504
505		Changes:
506
507		The new pid generation scheme described a few weeks ago was overly
508		complicated.  I implemented a new mechanism that is simpler and
509		provides a bit more "stability":  Once CM has seen a compilation
510		unit, it keeps its identity constant (as long as you do not delete
511		those crucial CM/GUID/* files).  This means that when you change
512		an interface, compile, then go back to the old interface, and
513		compile again, you arrive at the original pid.
514
515		There now also is a mechanism that instructs CM to use the plain
516		environment hash as a module's pid (effectively making its GUID
517		the empty string).  For this, "noguid" must be specified as an
518		option to the .sml file in question within its .cm file.
519		This is most useful for code that is being generated by tools such
520		as ml-nlffigen (because during development programmers tend to
521		erase the tool's entire output directory tree including CM's cached
522		GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
523		revert to the old, broken behavior of SML/NJ, but in specific cases
524		where there is no danger of interface confusion, its use is ok
525		(I think).
526
527		ml-nlffigen by default generates "noguid" annotations.  They can be
528		turned off by specifying -guid in its command line.
529
530		----------------------------------------------------------------------
531		Name: Lal George
532		Date: 2002/03/12 12 14:42:36 EST
533		Tag: george-20020312-frequency-computation
534		Description:
535
536		Integrated jump chaining and static block frequency into the
537		compiler. More details and numbers later.
538
539		----------------------------------------------------------------------
540		Name: Lal George
541		Date: 2002/03/11 11 22:38:53 EST
542		Tag: george-20020311-jump-chain-elim
543		Description:
544
545		Tested the jump chain elimination on all architectures (except the
546		hppa).  This is on by default right now and is profitable for the
547		alpha and x86, however, it may not be profitable for the sparc and ppc
548		when compiling the compiler.
549
550		The gc test will typically jump to a label at the end of the cluster,
551		where there is another jump to an external cluster containing the actual
552		code to invoke gc. This is to allow factoring of common gc invocation
553		sequences. That is to say, we generate:
554
555		f:
556		testgc
557		ja   L1      % jump if above to L1
558
559		L1:
560		jmp L2
561
562
563		After jump chain elimination the 'ja L1' instructions is converted to
564		'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
565		up being implemented in their long form (if L2 is far away) using:
566
567		jbe     L3      % jump if below or equal to L3
568		jmp     L2
569		L3:
570		...
571
572
573		For large compilation units L2  may be far away.
574
575
576		----------------------------------------------------------------------
577		Name: Matthias Blume
578		Date: 2002/03/11 13:30:00 EST
579		Tag: blume-20020311-mltreeeval
580		Description:
581
582		A functor parameter was missing.
583
584		----------------------------------------------------------------------
585		Name: Allen Leung
586		Date: 2002/03/11 10:30:00 EST
587		Tag: leunga-20020311-runtime-string0
588		Description:
589
590		The representation of the empty string now points to a
591		legal null terminated C string instead of unit.  It is now possible
592		to convert an ML string into C string with InlineT.CharVector.getData.
593		This compiles into one single machine instruction.
594
595		----------------------------------------------------------------------
596		Name: Allen Leung
597		Date: 2002/03/10 23:55:00 EST
598		Tag: leunga-20020310-x86-call
599		Description:
600
601		Added machine generation for CALL instruction (relative displacement mode)
602
603		----------------------------------------------------------------------
604		Name: Matthias Blume
605		Date: 2002/03/08 16:05:00
606		Tag: blume-20020308-entrypoints
607		Description:
608
609		Version number bumped to 110.39.1.  NEW BOOTFILES!
610
611		Entrypoints: non-zero offset into a code object where execution should begin.
612
613		- Added the notion of an entrypoint to CodeObj.
614		- Added reading/writing of entrypoint info to Binfile.
615		- Made runtime system bootloader aware of entrypoints.
616		- Use the address of the label of the first function given to mlriscGen
617		as the entrypoint.  This address is currently always 0, but it will
618		not be 0 once we turn on block placement.
619		- Removed the linkage cluster code (which was The Other Way(tm) of dealing
620		with entry points) from mlriscGen.
621
622		----------------------------------------------------------------------
623		Name: Allen Leung
624		Date: 2002/03/07 20:45:00 EST
625		Tag: leunga-20020307-x86-cmov
626		Description:
627
628		Bug fixes for CMOVcc on x86.
629
630		1. Added machine code generation for CMOVcc
631		2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
632		3. CMOVcc cannot have an immediate operand as argument.
633
634		----------------------------------------------------------------------
635		Name: Matthias Blume
636		Date: 2002/03/07 16:15:00 EST
637		Tag: blume-20020307-controls
638		Description:
639
640		This is a very large but mostly boring patch which makes (almost)
641		every tuneable compiler knob (i.e., pretty much everything under
642		Control.* plus a few other things) configurable via both the command
643		line and environment variables in the style CM did its configuration
644		until now.
645
646		Try starting sml with '-h' (or, if you are brave, '-H')
647
648		To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
649		implements the underlying generic mechanism.
650
651		The interface to some of the existing such facilities has changed somewhat.
652		For example, the MLRiscControl module now provides mkFoo instead of getFoo.
653		(The getFoo interface is still there for backward-compatibility, but its
654		use is deprecated.)
655
656		The ml-build script passes -Cxxx=yyy command-line arguments through so
657		that one can now twiddle the compiler settings when using this "batch"
658		compiler.
659
660		TODO items:
661
662		We should go through and throw out all controls that are no longer
663		connected to anything.  Moreover, we should go through and provide
664		meaningful (and correct!) documentation strings for those controls
665		that still are connected.
666
667		Currently, multiple calls to Controls.new are accepted (only the first
668		has any effect).  Eventually we should make sure that every control
669		is being made (via Controls.new) exactly once.  Future access can then
670		be done using Controls.acc.
671
672		Finally, it would probably be a good idea to use the getter-setter
673		interface to controls rather than ref cells.  For the time being, both
674		styles are provided by the Controls module, but getter-setter pairs are
675		better if thread-safety is of any concern because they can be wrapped.
676
677		*****************************************
678
679		One bug fix: The function blockPlacement in three of the MLRISC
680		backpatch files used to be hard-wired to one of two possibilities at
681		link time (according to the value of the placementFlag).  But (I
682		think) it should rather sense the flag every time.
683
684		*****************************************
685
686		Other assorted changes (by other people who did not supply a HISTORY entry):
687
688		1. the cross-module inliner now works much better (Monnier)
689		2. representation of weights, frequencies, and probabilities in MLRISC
690		changed in preparation of using those for weighted block placement
691		(Reppy, George)
692
693		----------------------------------------------------------------------
694		Name: Lal George
695		Date: 2002/03/07 14:44:24 EST 2002
696		Tag: george-20020307-weighted-block-placement
697
698		Tested the weighted block placement optimization on all architectures
699		(except the hppa) using AMPL to generate the block and edge frequencies.
700		Changes were required in the machine properties to correctly
701		categorize trap instructions. There is an MLRISC flag
702		"weighted-block-placement" that can be used to enable weighted block
703		placement, but this will be ineffective without block/edge
704		frequencies (coming soon).
705
706
707		----------------------------------------------------------------------
708		Name: Lal George
709		Date: 2002/03/05 17:24:48 EST
710		Tag: george-20020305-linkage-cluster
711
712		In order to support the block placement optimization, a new cluster
713		is generated as the very first cluster (called the linkage cluster).
714		It contains a single jump to the 'real' entry point for the compilation
715		unit. Block placement has no effect on the linkage cluster itself, but
716		all the other clusters  have full freedom in the manner in which they
717		reorder blocks or functions.
718
719		On the x86 the typical linkage code that is generated is:
720		----------------------
721		.align 2
722		L0:
723		addl    $L1-L0, 72(%esp)
724		jmp     L1
725
726
727		.align  2
728		L1:
729		----------------------
730
731		72(%esp) is the memory location for the stdlink register. This
732		must contain the address of the CPS function being called. In the
733		above example, it contains the address of  L0; before
734		calling L1 (the real entry point for the compilation unit), it
735		must contain the address for L1, and hence
736
737		addl $L1-L0, 72(%esp)
738
739		I have tested this on all architectures except the hppa.The increase
740		in code size is of course negligible
741
742		----------------------------------------------------------------------
743		Name: Allen Leung
744		Date: 2002/03/03 13:20:00 EST
745		Tag: leunga-20020303-mlrisc-tools
746
747		Added #[ ... ] expressions to mlrisc tools
748
749		----------------------------------------------------------------------
750		Name: Matthias Blume
751		Date: 2002/02/27 12:29:00 EST
752		Tag: blume-20020227-cdebug
753		Description:
754
755		- made types in structure C and C_Debug to be equal
756		- got rid of code duplication (c-int.sml vs. c-int-debug.sml)
757		- there no longer is a C_Int_Debug (C_Debug is directly derived from C)
758
759		----------------------------------------------------------------------
760		Name: Matthias Blume
761		Date: 2002/02/26 12:00:00 EST
762		Tag: blume-20020226-ffi
763		Description:
764
765		1. Fixed a minor bug in CM's "noweb" tool:
766		If numbering is turned off, then truly don't number (i.e., do not
767		supply the -L option to noweb).  The previous behavior was to supply
768		-L'' -- which caused noweb to use the "default" line numbering scheme.
769		Thanks to Chris Richards for pointing this out (and supplying the fix).
770
771		2. Once again, I reworked some aspects of the FFI:
772
773		A. The incomplete/complete type business:
774
775		- Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
776		gone!
777		- ML types representing an incomplete type are now *equal* to
778		ML types representing their corresponding complete types (just like
779		in C).  This is still safe because ml-nlffigen will not generate
780		RTTI for incomplete types, nor will it generate functions that
781		require access to such RTTI.   But when ML code generated from both
782		incomplete and complete versions of the C type meet, the ML types
783		are trivially interoperable.
784
785		NOTE:  These changes restore the full generality of the translation
786		(which was previously lost when I eliminated functorization)!
787
788		B. Enum types:
789
790		- Structure C now has a type constructor "enum" that is similar to
791		how the "su" constructor works.  However, "enum" is not a phantom
792		type because each "T enum" has values (and is isomorphic to
793		MLRep.Signed.int).
794		- There are generic access operations for enum objects (using
795		MLRep.Signed.int).
796		- ml-nlffigen will generate a structure E_foo for each "enum foo".
797		* The structure contains the definition of type "mlrep" (the ML-side
798		representation type of the enum).  Normally, mlrep is the same
799		as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
800		then mlrep will be defined as a datatype -- thus facilitating
801		pattern matching on mlrep values.
802		("-ec" will be suppressed if there are duplicate values in an
803		enumeration.)
804		* Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
805		will be generated for each C enum constant xxx.
806		* Conversion functions m2i and i2m convert between mlrep and
807		MLRep.Signed.int.  (Without "-ec", these functions are identities.)
808		* Coversion functions c and ml convert between mlrep and "tag enum".
809		* Access functions (get/set) fetch and store mlrep values.
810		- By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
811		enumerations are merged into one single enumeration represented by
812		structure E_'.
813
814		----------------------------------------------------------------------
815		Name: Allen Leung
816		Date: 2002/02/25 04:45:00 EST
817		Tag: leunga-20020225-cps-spill
818
819		This is a new implementation of the CPS spill phase.
820		The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
821		In case of problems, replace it with the old file spill.sml
822
823		The current compiler runs into some serious performance problems when
824		constructing a large record.  This can happen when we try to compile a
825		structure with many items.  Even a very simple structure like the following
826		makes the compiler slow down.
827
828		structure Foo = struct
829		val x_1 = 0w1 : Word32.int
830		val x_2 = 0w2 : Word32.int
831		val x_3 = 0w3 : Word32.int
832		...
833		val x_N = 0wN : Word32.int
834		end
835
836		The following table shows the compile time, from N=1000 to N=4000,
837		with the old compiler:
838
839		N
840		1000   CPS 100 spill                           0.04u  0.00s  0.00g
841		MLRISC ra                               0.06u  0.00s  0.05g
842		(spills = 0 reloads = 0)
843		TOTAL                                   0.63u  0.07s  0.21g
844
845		1100   CPS 100 spill                           8.25u  0.32s  0.64g
846		MLRISC ra                               5.68u  0.59s  3.93g
847		(spills = 0 reloads = 0)
848		TOTAL                                   14.71u  0.99s  4.81g
849
850		1500   CPS 100 spill                           58.55u  2.34s  1.74g
851		MLRISC ra                               5.54u  0.65s  3.91g
852		(spills = 543 reloads = 1082)
853		TOTAL                                   65.40u  3.13s  6.00g
854
855		2000   CPS 100 spill                           126.69u  4.84s  3.08g
856		MLRISC ra                               0.80u  0.10s  0.55g
857		(spills = 42 reloads = 84)
858		TOTAL                                   129.42u  5.10s  4.13g
859
860		3000   CPS 100 spill                           675.59u  19.03s  11.64g
861		MLRISC ra                               2.69u  0.27s  1.38g
862		(spills = 62 reloads = 124)
863		TOTAL                                   682.48u  19.61s  13.99g
864
865		4000   CPS 100 spill                           2362.82u  56.28s  43.60g
866		MLRISC ra                               4.96u  0.27s  2.72g
867		(spills = 85 reloads = 170)
868		TOTAL                                   2375.26u  57.21s  48.00g
869
870		As you can see the old cps spill module suffers from some serious
871		performance problem.  But since I cannot decipher the old code fully,
872		instead of patching the problems up, I'm reimplementing it
873		with a different algorithm.  The new code is more modular,
874		smaller when compiled, and substantially faster
875		(O(n log n) time and O(n) space).  Timing of the new spill module:
876
877		4000  CPS 100 spill                           0.02u  0.00s  0.00g
878		MLRISC ra                               0.25u  0.02s  0.15g
879		(spills=1 reloads=3)
880		TOTAL                                   7.74u  0.34s  1.62g
881
882		Implementation details:
883
884		As far as I can tell, the purpose of the CPS spill module is to make sure the
885		number of live variables at any program point (the bandwidth)
886		does not exceed a certain limit, which is determined by the
887		size of the spill area.
888
889		When the bandwidth is too large, we decrease the register pressure by
890		packing live variables into spill records.  How we achieve this is
891		completely different than what we did in the old code.
892
893		First, there is something about the MLRiscGen code generator
894		that we should be aware of:
895
896		o MLRiscGen performs code motion!
897
898		In particular, it will move floating point computations and
899		address computations involving only the heap pointer to
900		their use sites (if there is only a single use).
901		What this means is that if we have a CPS record construction
902		statement
903
904		RECORD(k,vl,w,e)
905
906		we should never count the new record address w as live if w
907		has only one use (which is often the case).
908
909		We should do something similar to floating point, but the transformation
910		there is much more complex, so I won't deal with that.
911
912		Secondly, there are now two new cps primops at our disposal:
913
914		1. rawrecord of record_kind option
915		This pure operator allocates some uninitialized storage from the heap.
916		There are two forms:
917
918		rawrecord NONE [INT n]  allocates a tagless record of length n
919		rawrecord (SOME rk) [INT n] allocates a tagged record of length n
920		and initializes the tag.
921
922		2. rawupdate of cty
923		rawupdate cty (v,i,x)
924		Assigns to x to the ith component of record v.
925		The storelist is not updated.
926
927		We use these new primops for both spilling and increment record construction.
928
929		1. Spilling.
930
931		This is implemented with a linear scan algorithm (but generalized
932		to trees).  The algorithm will create a single spill record at the
933		beginning of the cps function and use rawupdate to spill to it,
934		and SELECT or SELp to reload from it.  So both spills and reloads
935		are fine-grain operations.  In contrast, in the old algorithm
936		"spills" have to be bundled together in records.
937
938		Ideally, we should sink the spill record construction to where
939		it is needed.  We can even split the spill record into multiple ones
940		at the places where they are needed.  But CPS is not a good
941		representation for global code motion, so I'll keep it simple and
942		am not attempting this.
943
944		2. Incremental record construction (aka record splitting).
945
946		Long records with many component values which are simulatenously live
947		(recall that single use record addresses are not considered to
948		be live) are constructed with rawrecord and rawupdate.
949		We allocate space on the heap with rawrecord first, then gradually
950		fill it in with rawupdate.  This is the technique suggested to me
951		by Matthias.
952
953		Some restrictions on when this is applicable:
954		1. It is not a VECTOR record.  The code generator currently does not handle
955		this case. VECTOR record uses double indirection like arrays.
956		2. All the record component values are defined in the same "basic block"
957		as the record constructor.  This is to prevent speculative
958		record construction.
959
960		----------------------------------------------------------------------
961		Name: Allen Leung
962		Date: 2002/02/22 01:02:00 EST
963		Tag: leunga-20020222-mlrisc-tools
964
965		Minor bug fixes in the parser and rewriter
966
967		----------------------------------------------------------------------
968		Name: Allen Leung
969		Date: 2002/02/21 20:20:00 EST
970		Tag: leunga-20020221-peephole
971
972		Regenerated the peephole files.  Some contained typos in the specification
973		and some didn't compile because of pretty printing bugs in the old version
974		of 'nowhere'.
975
976		----------------------------------------------------------------------
977		Name: Allen Leung
978		Date: 2002/02/19 20:20:00 EST
979		Tag: leunga-20020219-mlrisc-tools
980		Description:
981
982		Minor bug fixes to the mlrisc-tools library:
983
984		1.  Fixed up parsing colon suffixed keywords
985		2.  Added the ability to shut the error messages up
986		3.  Reimplemented the pretty printer and fixed up/improved
987		the pretty printing of handle and -> types.
988		4.  Fixed up generation of literal symbols in the nowhere tool.
989		5.  Added some SML keywords to to sml.sty
990
991		----------------------------------------------------------------------
992		Name: Matthias Blume
993		Date: 2002/02/19 16:20:00 EST
994		Tag: blume-20020219-cmffi
995		Description:
996
997		A wild mix of changes, some minor, some major:
998
999		* All C FFI-related libraries are now anchored under $c:
1000		$/c.cm      --> $c/c.cm
1001		$/c-int.cm  --> $c/internals/c-int.cm
1002		$/memory.cm --> $c/memory/memory.cm
1003
1004		* "make" tool (in CM) now treats its argument pathname slightly
1005		differently:
1006		1. If the native expansion is an absolute name, then before invoking
1007		the "make" command on it, CM will apply OS.Path.mkRelative
1008		(with relativeTo = OS.FileSys.getDir()) to it.
1009		2. The argument will be passed through to subsequent phases of CM
1010		processing without "going native".  In particular, if the argument
1011		was an anchored path, then "make" will not lose track of that anchor.
1012
1013		* Compiler backends now "know" their respective C calling conventions
1014		instead of having to be told about it by ml-nlffigen.  This relieves
1015		ml-nlffigen from one of its burdens.
1016
1017		* The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
1018
1019		* Export C_DEBUG and C_Debug from $c/c.cm.
1020
1021		* C type encoding in ml-nlffi-lib has been improved to model the conceptual
1022		subtyping relationship between incomplete pointers and their complete
1023		counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
1024		with the convention of instantiating 'o with ('t, 'c) obj whenever
1025		the pointer target type is complete.  In the incomplete case, 'o
1026		will be instantiated with some "'c iobj" -- a type obtained by
1027		using one of the functors PointerToIncompleteType or PointerToCompleteType.
1028
1029		Operations that work on both incomplete and complete pointer types are
1030		typed as taking an 'o ptr while operations that require the target to
1031		be known are typed as taking some ('t, 'c) obj ptr.
1032
1033		voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
1034		where void is an eqtype without any values.  This makes it possible
1035		to work on voidptr values using functions meant to operate on light
1036		incomplete pointers.
1037
1038		* As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
1039		been vastly simplified.
1040
1041		----------------------------------------------------------------------
1042		Name: Matthias Blume
1043		Date: 2002/02/19 10:48:00 EST
1044		Tag: blume-20020219-pqfix
1045		Description:
1046
1047		Applied Chris Okasaki's bug fix for priority queues.
1048
1049		----------------------------------------------------------------------
1050		Name: Matthias Blume
1051		Date: 2002/02/15 17:05:00
1052		Tag: Release_110_39
1053		Description:
1054
1055		Last-minute retagging is becoming a tradition... :-(
1056
1057		This is the working release 110.39.
1058
1059		----------------------------------------------------------------------
1060		Name: Matthias Blume
1061		Date: 2002/02/15 16:00:00 EST
1062		Tag: Release_110_39-orig
1063		Description:
1064
1065		Working release 110.39.  New bootfiles.
1066
1067		(Update: There was a small bug in the installer so it wouldn't work
1068		with all shells.  So I retagged. -Matthias)
1069
1070		----------------------------------------------------------------------
1071		Name: Matthias Blume
1072		Date: 2002/02/15 14:17:00 EST
1073		Tag: blume-20020215-showbindings
1074		Description:
1075
1076		Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
1077		the latter can be useful for exploring what bindings are available at
1078		the interactive prompt.  (The first function returns only the list
1079		of symbols that are really bound, the second prints those but also the
1080		ones that CM's autoloading mechanism knows about.)
1081
1082		----------------------------------------------------------------------
1083		Name: Matthias Blume
1084		Date: 2002/02/15 12:08:00 EST
1085		Tag: blume-20020215-iptrs
1086		Description:
1087
1088		Two improvements to ml-nlffigen:
1089
1090		1. Write files only if they do not exist or if their current contents
1091		do not coincide with what's being written.  (That is, avoid messing
1092		with the time stamps unless absolutely necessary.)
1093
1094		2. Implement a "repository" mechanism for generated files related
1095		to "incomplete pointer types".   See the README file for details.
1096
1097		----------------------------------------------------------------------
1098		Name: Matthias Blume
1099		Date: 2002/02/14 11:50:00 EST
1100		Tag: blume-20020214-quote
1101		Description:
1102
1103		Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
1104		because of the new and improved tag generation scheme.  (Thanks to Allen
1105		Leung for pointing it out.)
1106
1107		----------------------------------------------------------------------
1108		Name: Lal George
1109		Date: 2002/02/14 09:55:27 EST 2002
1110		Tag: george-20020214-isabelle-bug
1111		Description:
1112
1113		Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
1114		of Isabelle on the x86.
1115
1116		From Allen:
1117		-----------
1118		I've found the problem:
1119
1120		in ra-core.sml, I use the counter "blocked" to keep track of the
1121		true number of elements in the freeze queue.  When the counter goes
1122		to zero, I skip examining the queue.  But I've messed up the
1123		bookkeeping in combine():
1124
1125		else ();
1126		case !ucol of
1127		PSEUDO => (if !cntv > 0 then
1128		(if !cntu > 0 then blocked := !blocked - 1 else ();
1129		^^^^^^^^^^^^^^^^^^^^^^^
1130		moveu := mergeMoveList(!movev, !moveu)
1131		)
1132		else ();
1133
1134		combine() is called to coalesce two nodes u and v.
1135		I think I was thinking that if the move counts of u and v are both
1136		greater than zero then after they are coalesced then one node is
1137		removed from the freeze queue.  Apparently I was thinking that
1138		both u and v are of low degree, but that's clearly not necessarily true.
1139
1140
1141		02/12/2002:
1142		Here's the patch.  HOL now compiles.
1143
1144		I don't know how this impact on performance (compile
1145		time or runtime).  This bug caused the RA (especially on the x86)
1146		to go thru the potential spill phase when there are still nodes on the
1147		freeze queue.
1148
1149
1150
1151
1152		----------------------------------------------------------------------
1153		Name: Matthias Blume
1154		Date: 2002/02/13 22:40:00 EST
1155		Tag: blume-20020213-fptr-rtti
1156		Description:
1157
1158		Fixed a bug in ml-nlffigen that was introduced with one of the previous
1159		updates.
1160
1161		----------------------------------------------------------------------
1162		Name: Matthias Blume
1163		Date: 2002/02/13 16:41:00 EST
1164		Tag: blume-20020213-cmlpq
1165		Description:
1166
1167		Added new priority queue export symbols (which have just been added to
1168		smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
1169		not compile and the installer would choke.)
1170
1171		----------------------------------------------------------------------
1172		Name: Matthias Blume
1173		Date: 2002/02/13 16:15:00 EST
1174		Tag: blume-20020213-various
1175		Description:
1176
1177		1. More tweaks to ml-nlffigen:
1178
1179		- better internal datastructures (resulting in slight speedup)
1180		- "-match" option requires exact match
1181		- "localized" gensym counters (untagged structs/unions nested within
1182		other structs/unions or within typedefs get a fresh counter; their
1183		tag will be prefixed by a concatenation of their parents' tags)
1184		- bug fixes (related to calculation of transitive closure of types
1185		to be included in the output)
1186
1187		2. Minor Basis updates:
1188
1189		- added implementations for List.collate and Option.app
1190
1191		----------------------------------------------------------------------
1192		Name: Matthias Blume
1193		Date: 2002/02/11 15:55:00 EST
1194		Tag: blume-20020211-gensym
1195		Description:
1196
1197		Added a "-gensym" option to command line of ml-nlffigen.  This can be
1198		used to specify a "stem" -- a string that is inserted in all "gensym'd"
1199		names (ML structure names that correspond to unnamed C structs, unions,
1200		and enums), so that separate runs of ml-nlffigen do not clash.
1201
1202		----------------------------------------------------------------------
1203		Name: Matthias Blume
1204		Date: 2002/02/11 12:05:00 EST
1205		Tag: blume-20020211-gensml
1206		Description:
1207
1208		A quick fix for a problem with GenSML (in the pgraph-util library):
1209		Make generation of toplevel "local" optional.  (Strictly speaking,
1210		signature definitions within "local" are not legal SML.)
1211
1212		Other than that: updates to INSTALL and cm/TODO.
1213
1214		----------------------------------------------------------------------
1215		Name: Matthias Blume
1216		Date: 2002/02/08 15:00:00 EST
1217		Tag: blume-20020208-uniquepid
1218		Description:
1219
1220		0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
1221
1222		1. The installer (config/install.sh) has gotten smarter:
1223
1224		- Configuration options are a bit easier to specify now
1225		(in config/targets).
1226		- Bug in recognizing .tar.bz2 files fixed.
1227		- Installer automatically resolves dependencies between
1228		configuration options (e.g., if you ask for eXene, you will
1229		also get cml -- regardless whether you asked for it or not).
1230		- Installer can run in "quieter mode" by setting the environment
1231		variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
1232		"completely silent", though.
1233		- Build HashCons library as part of smlnj-lib.
1234
1235		2. A new scheme for assigning persistent identifiers to compilation
1236		units (and, by extension, to types etc.) has been put into place.
1237		This fixes a long-standing bug where types and even dynamic values
1238		can get internally confused, thereby compromising type safety
1239		(abstraction) and dynamic correctness.  See
1240
1241		http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
1242
1243		for an example of how things could go wrong until now.
1244
1245		The downside of the new scheme is that pids are not quite as
1246		persistent as they used to be: CM will generate a fresh pid
1247		for every compilation unit that it thinks it sees for the first
1248		time.  That means that if you compile starting from a clean, fresh
1249		source tree at two different times, you end up with different
1250		binaries.
1251
1252		Cutoff recompilation, however, has not been compromised because
1253		CM keeps pid information in special caches between runs.
1254
1255		----------------------------------------------------------------------
1256		Name: Lal George
1257		Date: 2002/02/07 15:34:13 EST 2002
1258		Tag: <none>
1259		Description:
1260
1261		Compilers that generate assembly code may produce  global labels
1262		whose value is resolved at link time. The various peephole optimization
1263		modules did not take this in account.
1264
1265		TODO. The Labels.addrOf function should really return an option
1266		type so that clients are forced to deal with this issue, rather
1267		than an exception being raised.
1268
1269		----------------------------------------------------------------------
1270		Name: Lal George
1271		Date: 2002/02/06 13:55:02 EST
1272		Tag: george-20020206-ra-breakup
1273		Description:
1274
1275		1. A bug fix from Allen.
1276
1277		A typo causes extra fstp %st(0)'s to be generated at compensation
1278		edges, which might cause stack underflow traps at runtime.  This
1279		occurs in fft where there are extraneous fstps right before the 'into'
1280		trap instruction (in this case they are harmless since none of the
1281		integers overflow.)
1282
1283		2. Pulled out various utility modules that were embedded in the modules
1284		of the register allocator. I need these modules for other purposes, but
1285		they are not complete enough to put into a library (just yet).
1286		----------------------------------------------------------------------
1287		Name: Matthias Blume
1288		Date: 2002/01/31 16:05:00 EST
1289		Tag: blume-20020131-sparc-ccalls
1290		Description:
1291
1292		1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
1293		of extra stack space by mistake.  Fixed.
1294
1295		2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
1296
1297		----------------------------------------------------------------------
1298		Name: Allen Leung
1299		Date: 2002/01/30
1300		Tag: leunga-20020130-nowhere-bug-fix
1301		Description:
1302
1303		MLRISC bug fixes:
1304		1. Fixed a bindings computation bug in the 'nowhere' program generator tool.
1305		2. MachineInt.fromString was negating its value.
1306
1307		----------------------------------------------------------------------
1308		Name: Matthias Blume
1309		Date: 2002/01/29
1310		Tag: blume-20020129-INSTALL
1311		Description:
1312
1313		- Added somewhat detailed installation instructions (file INSTALL).
1314		- Fixed curl-detection bug in config/install.sh.
1315		- It is now possible to select the URL getter using the URLGETTER
1316		environment variable:
1317
1318		not set / "unknown"      --> automatic detection (script tries wget,
1319		curl, and lynx)
1320		"wget" / "curl" / "lynx" --> use the specified program (script "knows"
1321		how to properly invoke them)
1322		other                    --> use $URLGETTER directly, it must take
1323		precisely two command-line arguments
1324		(source URL and destination file name)
1325
1326		----------------------------------------------------------------------
1327		Name: Matthias Blume
1328		Date: 2002/01/28
1329		Tag: blume-20020128-sparc-ccalls
1330		Description:
1331
1332		- Fixed problem with calculation of "used" registers in sparc-c-calls.
1333		- Make use of the allocParam argument in sparc-c-calls.
1334
1335		----------------------------------------------------------------------
1336		Name: Matthias Blume
1337		Date: 2002/01/28
1338		Tag: blume-20020128-allocParam
1339		Description:
1340
1341		John Reppy:  Changes c-calls API to accept client-callback for
1342		allocating extra stack space.
1343		me: Corresponding changes to mlriscGen (using a dummy argument that
1344		does not change the current behavior).
1345
1346		----------------------------------------------------------------------
1347		Name: Matthias Blume
1348		Date: 2002/01/28 12:00:00
1349		Tag: Release_110_38
1350		Description:
1351
1352		This time for real!!!
1353
1354		----------------------------------------------------------------------
1355		Name: Matthias Blume
1356		Date: 2002/01/28 10:56:00 EST
1357		Tag: blume-20020128-retraction
1358		Description:
1359
1360		0. Retracted earlier 110.38.  (The Release_110_38 tag has been replaced
1361		with blume-Release_110_38-retracted.)
1362
1363		1. Fixed a problem with incorrect rounding modes in real64.sml.
1364		(Thanks to Andrew Mccreight <andrew.mccreight@yale.edu>.)
1365
1366		2. A bug in ml-nlffigen related to the handling of unnamed structs, unions,
1367		and enums fixed.  The naming of corresponding ML identifiers should
1368		now be consistent again.
1369
1370		----------------------------------------------------------------------
1371		Name: Allen Leung
1372		Date: 2002/01/27
1373		Tag: leunga-20020127-nowhere
1374		Description:
1375
1376		Added a target called nowhere in the configuration scripts.
1377		Enabling this will build the MLRISC 'nowhere' tool (for translating
1378		programs with where-clauses into legal SML code) during installation.
1379
1380		----------------------------------------------------------------------
1381		Name: Matthias Blume
1382		Date: 2002/01/25 21:27:00 EST
1383		Tag: blume-Release_110_38-retracted
1384		Description:
1385
1386		Call it a (working) release!  Version is 110.38. Bootfiles are ready.
1387
1388		README will be added later.
1389
1390		!!! NOTE:  Re-tagged as blume-Release_110_38-retracted. Original tag
1391		(Release_110_38) removed.  Reason: Last-minute bug fixes.
1392
1393		----------------------------------------------------------------------
1394		Name: Matthias Blume
1395		Date: 2002/01/25
1396		Tag: blume-20020125-ffi
1397		Description:
1398
1399		A large number of tweaks and improvements to ml-nlffi-lib and
1400		ml-nlffigen:
1401
1402		- ML represenation types have been streamlined
1403		- getter and setter functions work with concrete values, not abstract
1404		ones where possible
1405		- ml-nlffigen command line more flexible (see README file there)
1406		- some bugs have been fixed (hopefully)
1407
1408		----------------------------------------------------------------------
1409		Name: Lal George
1410		Date: 2002/01/24
1411		Tag: george-20020124-risc-ra-interface
1412		Description:
1413
1414		There is a dramatic simplification in the interface to the
1415		register allocator for RISC architectures as a result of making
1416		parallel copy instructions explicit.
1417
1418		----------------------------------------------------------------------
1419		Name: Matthias Blume
1420		Date: 2002/01/22
1421		Tag: blume-20020122-x86-ccalls
1422		Description:
1423
1424		Bug fix for c-calls on x86 (having to do with how char- and
1425		short-arguments are being handled).
1426
1427		----------------------------------------------------------------------
1428		Name: Matthias Blume
1429		Date: 2002/01/21
1430		Tag: blume-20020121-ff
1431		Description:
1432
1433		Another day of fiddling with the FFI...
1434
1435		1. Bug fix/workaround:  CKIT does not complain about negative array
1436		dimensions, so ml-nlffigen has to guard itself against this possibility.
1437		(Otherwise a negative dimension would send it into an infinite loop.)
1438
1439		2. Some of the abstract types (light objects, light pointers, most "base"
1440		types) in structure C are now eqtypes.
1441
1442		3. Added constructors and test functions for NULL function pointers.
1443
1444		----------------------------------------------------------------------
1445		Name: Matthias Blume
1446		Date: 2002/01/18
1447		Tag: blume-20020118-ready-for-new-release
1448		Description:
1449
1450		Made config/srcarchiveurl point to a new place.  (Will provide boot
1451		files shortly.)
1452
1453		Maybe we christen this to be 110.38?
1454
1455		----------------------------------------------------------------------
1456		Name: Matthias Blume
1457		Date: 2002/01/18
1458		Tag: blume-20020118-more-ffifiddle
1459		Description:
1460
1461		Today's FFI fiddling:
1462
1463		- Provided a structure CGetSet with "convenient" versions of C.Get.* and
1464		C.Set.* that use concrete (MLRep.*) arguments and results instead
1465		of abstract ones.
1466
1467		- Provided word-style bit operations etc. for "int" representation
1468		types in MLRep.S<Foo>Bitops where <Foo> ranges over Char, Int, Short,
1469		and Long.
1470
1471		----------------------------------------------------------------------
1472		Name: Matthias Blume
1473		Date: 2002/01/18
1474		Tag: blume-20020118-use-x86-fp
1475		Description:
1476
1477		Now that x86-fast-fp seems to be working, I turned it back on again
1478		by default.  (Seems to work fine now, even with the FFI.)
1479
1480		Other than that, I added some documentation about the FFI to
1481		src/ml-nlffigen/README and updated the FFI test examples in
1482		src/ml-nlffi-lib/Tests/*.
1483
1484		----------------------------------------------------------------------
1485		Name: Allen Leung
1486		Date: 2002/01/17
1487		Tag: leunga-20020117-x86-fast-fp-call
1488		Description:
1489
1490		1. Fixed a problem with handling return fp values when x86's fast fp
1491		mode is turned on.
1492
1493		2. Minor pretty printing fix for cellset.  Print %st(0) as %st(0) instead
1494		of %f32.
1495
1496		3. Added a constructor INT32lit to the ast of MLRISC tools.
1497
1498		----------------------------------------------------------------------
1499		Name: Matthias Blume
1500		Date: 2002/01/16
1501		Tag: blume-20020116-ffifiddle
1502		Description:
1503
1504		More fiddling with the FFI interface:
1505
1506		- Make constness 'c instead of rw wherever possible.  This eliminates
1507		the need for certain explicit coercions.  (However, due to ML's
1508		value polymorphism, there will still be many cases where explicit
1509		coercions are necessary.  Phantom types are not the whole answer
1510		to modeling a subtyping relationship in ML.)
1511
1512		- ro/rw coersions for pointers added.  (Avoids the detour through */&.)
1513
1514		- "printf" test example added to src/ml-nlffi-lib/Tests.  (Demonstrates
1515		clumsy workaround for varargs problem.)
1516
1517		----------------------------------------------------------------------
1518		Name: Lal George
1519		Date: 2002/01/15
1520		Tag: <none>
1521		Description:
1522
1523		1. Since COPY instructions are no longer native to the architecture,
1524		a generic functor can be used to implement the expandCopies function.
1525
1526		2. Allowed EXPORT and IMPORT pseudo-op declarations to appear inside a
1527		TEXT segment.
1528
1529		----------------------------------------------------------------------
1530		Name: Matthias Blume
1531		Date: 2002/01/15
1532		Tag: blume-20020115-ffiupdates
1533		Description:
1534
1535		1. Fix for bug resulting in single-precision float values being returned
1536		incorrectly from FFI calls.
1537
1538		2. Small modifications to C FFI API:
1539
1540		- memory-allocation routines return straight objects (no options)
1541		and raise an exception in out-of-memory situations
1542		- unsafe extensions to cast between function pointers and pointers
1543		from/to ints
1544		- added structure C_Debug as an alternative to structure C where
1545		pointer-dereferencing (|*| and |*!) always check for null-pointers
1546		- added open_lib' to DynLinkage;  open_lib' works like open_lib
1547		but also takes a (possibly empty) list of existing library handles
1548		that the current library depends on
1549
1550		----------------------------------------------------------------------
1551		Name: Matthias Blume
1552		Date: 2002/01/10
1553		Tag: blume-20020110-newffigen
1554		Description:
1555
1556		1. Updates to portable graph code.
1557
1558		2. Major update to ml-nlffigen and ml-nlffi-lib.  Things are much
1559		more scalable now so that even huge interfaces such as the one
1560		for GTK compile in finite time and space. :-)
1561		See src/ml-nlffigen/README for details on what's new.
1562
1563		----------------------------------------------------------------------
1564		Name: Lal George
1565		Date: 2001/01/09 14:31:35 EST 2002
1566		Tag: george-20011206-rm-native-copy
1567		Description:
1568
1569		Removed the native COPY and FCOPY instructions
1570		from all the architectures and replaced it with the
1571		explicit COPY instruction from the previous commit.
1572
1573		It is now possible to simplify many of the optimizations
1574		modules that manipulate copies. This has not been
1575		done in this change.
1576
1577		----------------------------------------------------------------------
1578		Name: Lal George
1579		Date: 2001/12/06 16:50:13 EST 2001
1580		Tag: george-20011206-mlrisc-instruction
1581		Description:
1582
1583		Changed the representation of instructions from being fully abstract
1584		to being partially concrete. That is to say:
1585
1586		from
1587		type instruction
1588
1589		to
1590		type instr                              (* machine instruction *)
1591
1592		datatype instruction =
1593		LIVE of {regs: C.cellset, spilled: C.cellset}
1594		| KILL of {regs: C.cellset, spilled: C.cellset}
1595		| COPYXXX of {k: CB.cellkind, dst: CB.cell list, src: CB.cell list}
1596		| ANNOTATION of {i: instruction, a: Annotations.annotation}
1597		| INSTR of instr
1598
1599		This makes the handling of certain special instructions that appear on
1600		all architectures easier and uniform.
1601
1602		LIVE and KILL say that a list of registers are live or killed at the
1603		program point where they appear. No spill code is generated when an
1604		element of the 'regs' field is spilled, but the register is moved to
1605		the 'spilled' (which is present, more for debugging than anything else).
1606
1607		LIVE replaces the (now deprecated) DEFFREG instruction on the alpha.
1608		We used to generate:
1609
1610		DEFFREG f1
1611		f1 := f2 + f3
1612		trapb
1613
1614		but now generate:
1615
1616		f1 := f2 + f3
1617		trapb
1618		LIVE {regs=[f1,f2,f3], spilled=[]}
1619
1620		Furthermore, the DEFFREG (hack) required that all floating point instruction
1621		use all registers mentioned in the instruction. Therefore f1 := f2 + f3,
1622		defines f1 and uses [f1,f2,f3]! This hack is no longer required resulting
1623		in a cleaner alpha implementation. (Hopefully, intel will not get rid of
1624		this architecture).
1625
1626		COPYXXX is intended to replace the parallel COPY and FCOPY  available on
1627		all the architectures. This will result in further simplification of the
1628		register allocator that must be aware of them for coalescing purposes, and
1629		will also simplify certain aspects of the machine description that provides
1630		callbacks related to parallel copies.
1631
1632		ANNOTATION should be obvious, and now INSTR represents the honest to God
1633		machine instruction set!
1634
1635		The <arch>/instructions/<arch>Instr.sml files define certain utility
1636		functions for making porting easier -- essentially converting upper case
1637		to lower case. All machine instructions (of type instr) are in upper case,
1638		and the lower case form generates an MLRISC instruction. For example on
1639		the alpha we have:
1640
1641		datatype instr =
1642		LDA of {r:cell, b:cell, d:operand}
1643		| ...
1644
1645		val lda : {r:cell, b:cell, d:operand} -> instruction
1646		...
1647
1648		where lda is just (INSTR o LDA), etc.
1649
1650		----------------------------------------------------------------------
1651		Name: Matthias Blume
1652		Date: 2001/11/22 21:40:00 EST
1653		Tag: Release_110_37
1654		Description:
1655
1656		Release 110.37.  This time for real.
1657
1658		----------------------------------------------------------------------
1659		Name: Matthias Blume
1660		Date: 2001/11/21 16:35:00 EST
1661		Tag: blume-20011121-foot-in-mouth
1662		Description:
1663
1664		Removed the "Release_110_37" tag because of a serious bug.
1665		This will be re-tagged once the bug is fixed.
1666
1667		----------------------------------------------------------------------
1668		Name: Matthias Blume
1669		Date: 2001/11/21 16:14:00 EST
1670		Tag: blume-20011121-forgottenfile
1671		Description:
1672
1673		Forgot to add a file.  (Just a .tex-file -- part of
1674		the CM manual source.)
1675
1676		----------------------------------------------------------------------
1677		Name: Matthias Blume
1678		Date: 2001/11/21 16:10:00 EST
1679		Tag: blume-20011121-invalid_110_37
1680		Description:
1681
1682		Note: I removed the original tag "Release_110_37" from this commit
1683		because we found a serious bug in all non-x86 backends.
1684		- Matthias
1685
1686		1. Modifications to the SML/NJ code generator and to the runtime system
1687		so that code object name strings are directly inserted into code
1688		objects at code generation time.  The only business the runtime system
1689		has with this is now to read the name strings on occasions.
1690		(The encoding of the name string has also changed somewhat.)
1691
1692		2. CM now implements a simple "set calculus" for specifying export lists.
1693		In particular, it is now possible to refer to the export lists of
1694		other libraries/groups/sources and form unions as well as differences.
1695		See the latest CM manual for details.
1696
1697		3. An separate notion of "proxy" libraries has again be eliminated from
1698		CM's model.  (Proxy libraries are now simply a special case of using
1699		the export list calculus.)
1700
1701		4. Some of the existing libraries now take advantage of the new set
1702		calculus.
1703		(Notice that not all libraries have been converted because some
1704		of the existing .cm-files are supposed to be backward compatible
1705		with 110.0.x.)
1706
1707		5. Some cleanup in stand-alone programs.  (Don't use "exnMessage" -- use
1708		"General.exnMessage"!  The former relies on a certain hook to be
1709		initialized, and that often does not happen in the stand-alone case.)
1710
1711		----------------------------------------------------------------------
1712		Name: Lal George
1713		Date: 2001/11/21  13:56:18 EST
1714		Tag: george-2001121-pseudo-ops
1715		Description:
1716
1717		Implemented a complete redesign of MLRISC pseudo-ops. Now there
1718		ought to never be any question of incompatabilities with
1719		pseudo-op syntax expected by host assemblers.
1720
1721		For now, only modules supporting GAS syntax are implemented
1722		but more should follow, such as MASM, and vendor assembler
1723		syntax, e.g. IBM as, Sun as, etc.
1724
1725		----------------------------------------------------------------------
1726		Name: Matthias Blume
1727		Date: 2001/11/14 11:52:00 EST
1728		Tag: blume-20011114-srcname
1729		Description:
1730
1731		1. Routed the name of the current source file to mlriscgen where it
1732		should be directly emitted into the code object.  (This last part
1733		is yet to be done.)
1734
1735		2. Some cleanup of the pgraph code to make it match the proposal that
1736		I put out the other day.  (The proposal notwithstanding, things are
1737		still in flux here.)
1738
1739		----------------------------------------------------------------------
1740		Name: Lal George
1741		Date: 2001/11/14 09:44:04 EST
1742		Tag:
1743		Description:
1744
1745		Fix for a backpatching bug reported by Allen.
1746
1747		Because the boundary between short and long span-dependent
1748		instructions is +/- 128, there are an astounding number of
1749		span-dependent instructions whose size is over estimated.
1750
1751		Allen came up with the idea of letting the size of span
1752		dependent instructions be non-monotonic, for a maxIter
1753		number of times, after which the size must be monotonically
1754		increasing.
1755
1756		This table shows the number of span-dependent instructions
1757		whose size was over-estimated as a function of maxIter, for the
1758		file Parse/parse/ml.grm.sml:
1759
1760		maxIter            # of instructions:
1761		10                      687
1762		20                      438
1763		30                      198
1764		40                        0
1765
1766		In compiling the compiler, there is no significant difference in
1767		compilation speed between maxIter=10 and maxIter=40. Actually,
1768		my measurements showed that maxIter=40 was a tad faster than
1769		maxIter=10! Also 96% of the  files in the compiler reach a fix
1770		point within 13 iterations, so fixing maxIter at 40, while high,
1771		is okay.
1772
1773		----------------------------------------------------------------------
1774		Name: Matthias Blume
1775		Date: 2001/10/31 15:25:00 EST
1776		Tag: blume-20011031-pgraph
1777		Description:
1778
1779		CKIT:
1780		* Changed the "Function" constructor of type Ast.ctype to carry optional
1781		argument identifiers.
1782		* Changed the return type of TypeUtil.getFunction accordingly.
1783		* Type equality ignores the argument names.
1784		* TypeUtil.composite tries to preserve argument names but gives up quickly
1785		if there is a mismatch.
1786
1787		installation script:
1788		* attempts to use "curl" if available (unless "wget" is available as well)
1789
1790		CM:
1791		* has an experimental implementation of "portable graphs" which I will
1792		soon propose as an implementation-independent library format
1793		* there are also new libraries $/pgraph.cm and $/pgraph-util.cm
1794
1795		NLFFI-LIB:
1796		* some cleanup (all cosmetic)
1797
1798		NLFFIGEN:
1799		* temporarily disabled the mechanism that suppresses ML output for
1800		C definitions whose identifiers start with an underscore character
1801		* generate val bindings for enum constants
1802		* user can request that only one style (light or heavy) is being used;
1803		default is to use both (command-line arguments: -heavy and -light)
1804		* fixed bug in handling of function types involving incomplete pointers
1805		* generate ML entry points that take record arguments (i.e., using
1806		named arguments) for C functions that have a prototype with named
1807		arguments
1808		(see changes to CKIT)
1809
1810		----------------------------------------------------------------------
1811		Name: Allen Leung
1812		Date: 2001/10/27 20:34:00 EDT
1813		Tag: leunga-20011027-x86-fast-fp-call
1814		Description:
1815
1816		Fixed the bug described in blume-20010920-slowfp.
1817
1818		The fix involves
1819		1. generating FCOPYs in FSTP in ia32-svid
1820		2. marking a CALL with the appropriate annotation
1821
1822		----------------------------------------------------------------------
1823		Name: Matthias Blume
1824		Date: 2001/10/16 11:32:00 EDT
1825		Tag: blume-20011016-netbsd
1826		Description:
1827
1828		Underscore patch from Chris Richards (fixing problem with compiling
1829		runtime system under recent NetBSD).
1830
1831		----------------------------------------------------------------------
1832		Name: Allen Leung
1833		Date: 2001/10/12 17:18:32 EDT 2001
1834		Tag: leung-20011012-x86-printflowgraph
1835		Description:
1836
1837		X86RA now uses a valid (instead of dummy) PrintFlowgraph module.
1838
1839		----------------------------------------------------------------------
1840		Name: Lal George
1841		Date: 2001/10/11 23:51:34 EDT
1842		Tag: george-20011011-too-many-instrs
1843		Description:
1844
1845		The representation of a program point never expected to see more
1846		than 65536 instructions in a basic block!
1847
1848		----------------------------------------------------------------------
1849		Name: Lal George
1850		Date: 2001/10/09 09:41:37 EDT
1851		Tag: george-20011008-mlrisc-labels
1852		Description:
1853
1854		Changed the machine description files to support printing of
1855		local and global labels in assembly code, based on host assembler
1856		conventions.
1857
1858		----------------------------------------------------------------------
1859		Name: Matthias Blume
1860		Date: 2001/09/25 15:25:00 EDT
1861		Tag: blume-20010925-exninfo
1862		Description:
1863
1864		I provided a non-hook implementation of exnName (at the toplevel) and
1865		made the "dummy" implementation of exnMessage (at the toplevel) more
1866		useful: if nothing gets "hooked in", then at least you are going to
1867		see the exception name and a message indicating why you don't see more.
1868
1869		[For the time being, programs that need exnMessage and want to use
1870		ml-build should either use General.exnMessage (strongly recommended) or
1871		refer to structure General at some other point so that CM sees a
1872		static dependency.]
1873
1874		[Similar remarks go for "print" and "use":  If you want to use their
1875		functionality in stand-alone programs generated by ml-build, then use
1876		TextIO.output and Backend.Interact.useFile (from $smlnj/compiler.cm).]
1877
1878		----------------------------------------------------------------------
1879		Name: Matthias Blume
1880		Date: 2001/09/20 17:28:00 EDT
1881		Tag: blume-20010920-slowfp
1882		Description:
1883
1884		Allen says that x86-fast-fp is not safe yet, so I turned it off again...
1885
1886		----------------------------------------------------------------------
1887		Name: Matthias Blume
1888		Date: 2001/09/20 17:20:00 EDT
1889		Tag: blume-20010920-canonicalpaths
1890		Description:
1891
1892		0. Updated the BOOT file (something that I forgot to do earlier).
1893
1894		1. Small internal change to CM so that it avoids "/../" in filenames
1895		as much as possible (but only where it is safe).
1896
1897		2. Changed config/_run-sml (resulting in a changed bin/.run-sml) so
1898		that arguments that contain delimiters are passed through correctly.
1899		This change also means that all "special" arguments of the form
1900		@SMLxxx... must come first.
1901
1902		3. Changed install script to put relative anchor names for tool commands
1903		into pathconfig.
1904
1905		----------------------------------------------------------------------
1906		Name: Matthias Blume
1907		Date: 2001/09/18 15:35:00 EDT
1908		Tag: blume-20010918-readme11036
1909		Description:
1910
1911		Added README files.
1912
1913		----------------------------------------------------------------------
1914		Name: Matthias Blume
1915		Date: 2001/09/18 11:45:00 EDT
1916		Tag: Release_110_36 (retag)
1917		Description:
1918
1919		Fixed mistake in config/preloads. Retagged as 110.36.
1920
1921		----------------------------------------------------------------------
1922		Name: Matthias Blume
1923		Date: 2001/09/18 09:40:00 EDT
1924		Tag: Release_110_36_orig (tag changed)
1925		Description:
1926
1927		New version (110.36).  New bootfiles.
1928
1929		----------------------------------------------------------------------
1930		Name: Matthias Blume
1931		Date: 2001/09/14 16:15:00 EDT
1932		Tag: blume-20010914-x86fastfp
1933		Description:
1934
1935		John committed some changes that Allen made, in particular a (hopefully)
1936		correctly working version of the x86-fp module.
1937
1938		I changed the default setting of the Control.MLRISC.getFlag "x86-fast-fp"
1939		flag to "true".  Everything seems to compile to a fixpoint ok, and
1940		"mandelbrot" speeds up by about 15%.
1941
1942		----------------------------------------------------------------------
1943		Name: Matthias Blume
1944		Date: 2001/09/13 11:20:00 EDT
1945		Tag: blume-20010913-minimal
1946		Description:
1947
1948		1. Stefan Monnier's patch to fix a miscompilation problem that
1949		was brought to light by John Reppy's work on Moby.
1950
1951		2. Implemented a minimal "structure Compiler" that contains just
1952		"version" and "architecture".  The minimal version will be
1953		available when the full version is not.  This is for backward-
1954		compatibility with code that wants to test Compiler.version.
1955
1956		----------------------------------------------------------------------
1957		Name: Matthias Blume
1958		Date: 2001/08/28 14:03:00 EDT
1959		Tag: blume-20010828-ml-lex
1960		Description:
1961
1962		Fix for bug 1581, received from Neophytos Michael.
1963
1964		----------------------------------------------------------------------
1965		Name: Matthias Blume
1966		Date: 2001/08/27 11:20:00 EDT
1967		Tag: blume-20010827-readme11035
1968		Description:
1969
1970		Fleshed out the README file for 110.35.
1971
1972		----------------------------------------------------------------------
1973		Name: Matthias Blume
1974		Date: 2001/08/24 17:10:00 EDT
1975		Tag: Release_110_35
1976		Description:
1977
1978		New version number (110.35).  New bootfiles.
1979
1980		----------------------------------------------------------------------
1981		Name: Lal George
1982		Date: 2001/08/24 13:47:18 EDT 2001
1983		Tag: george-20010824-MLRISC-graphs
1984		Description:
1985
1986		removed clusters from MLRISC completely and replaced with graphs.
1987
1988		----------------------------------------------------------------------
1989		Name: Matthias Blume
1990		Date: 2001/08/23 17:50:00 EDT
1991		Tag: blume-20010823-toplevel
1992		Description:
1993
1994		- some reorganization of the code that implements various kinds of
1995		environments in the compiler (static, dynamic, symbolic, combined)
1996		- re-implemented the EnvRef module so that evalStream works properly
1997		(if the stream contains references to "use", "CM.make", etc.)
1998		- cleaned up evalloop.sml and interact.sml (but they need more cleaning)
1999
2000		----------------------------------------------------------------------
2001		Name: Matthias Blume
2002		Date: 2001/08/20 15:50 EDT
2003		Tag: blume20010820-slipup
2004		Description:
2005
2006		I forgot to commit a few files.  Here they are...
2007
2008		----------------------------------------------------------------------
2009		Name: Matthias Blume
2010		Date: 2001/08/20 15:35:00 EDT
2011		Tag: blume-20010820-debugprof
2012		Description:
2013
2014		!!!! NEW BOOTFILES !!!!
2015
2016		This is another round of reorganizing the compiler sources.  This
2017		time the main goal was to factor out all the "instrumentation"
2018		passes (for profiling and backtracing) into their own library.
2019		The difficulty was to do it in such a way that it does not depend
2020		on elaborate.cm but only on elabdata.cm.
2021
2022		Therefore there have been further changes to both elaborate.cm and
2023		elabdata.cm -- more "generic" things have been moved from the former
2024		to the latter.  As a result, I was forced to split the assignment
2025		of numbers indicating "primtyc"s into two portions: SML-generic and
2026		SML/NJ-specific.  Since it would have been awkward to maintain,
2027		I bit the bullet and actually _changed_ the mapping between these
2028		numbers and primtycs.  The bottom line of this is that you need
2029		a new set of bin- and bootfiles.
2030
2031		I have built new bootfiles for all architectures, so doing a fresh
2032		checkout and config/install.sh should be all you need.
2033
2034		The newly created library's name is
2035
2036		$smlnj/viscomp/debugprof.cm
2037
2038		and its sources live under
2039
2040		src/compiler/DebugProf
2041
2042		----------------------------------------------------------------------
2043		Name: Matthias Blume
2044		Date: 2001/08/15 17:15:00 EDT
2045		Tag: blume-20010815-compreorg
2046		Description:
2047
2048		This is a first cut at reorganizing the CM libraries that make up the
2049		core of the compiler.  The idea is to separate out pieces that could
2050		be used independently by tools, e.g., the parser, the typechecker, etc.
2051
2052		The current status is a step in this direction, but it is not quite
2053		satisfactory yet.  Expect more changes in the future.
2054
2055		Here is the current (new) organization...
2056
2057		What used to be $smlnj/viscomp/core.cm is now divided into
2058		six CM libraries:
2059
2060		$smlnj/viscomp/basics.cm
2061		/parser.cm
2062		/elabdata.cm
2063		/elaborate.cm
2064		/execute.cm
2065		/core.cm
2066
2067		The CM files for these libraries live under src/system/smlnj/viscomp.
2068		All these libraries are proxy libraries that contain precisely
2069		one CM library component.  Here are the locations of the components
2070		(all within the src/compiler tree):
2071
2072		Basics/basics.cm
2073		Parse/parser.cm
2074		ElabData/elabdata.cm
2075		Elaborator/elaborate.cm
2076		Execution/execute.cm
2077		core.cm
2078
2079		[This organization is the same that has been used already
2080		for a while for the architecture-specific parts of the visible
2081		compiler and for the old version of core.cm.]
2082
2083		As you will notice, many source files have been moved from their
2084		respective original locations to a new home in one of the above
2085		subtrees.
2086
2087		The division of labor between the new libraries is the following:
2088
2089		basics.cm:
2090		- Simple, basic definitions that pertain to many (or all) of
2091		the other libraries.
2092		parser.cm:
2093		- The SML parser, producing output of type Ast.dec.
2094		- The type family for Ast is also defined and exported here.
2095		elabdata.cm:
2096		- The datatypes that describe input and output of the elaborator.
2097		This includes types, absyn, and static environments.
2098		elaborator.cm:
2099		- The SML/NJ type checker and elaborator.
2100		This maps an Ast.dec (with a given static environment) to
2101		an Absyn.dec (with a new static environment).
2102		- This libraries implements certain modules that used to be
2103		structures as functors (to remove dependencies on FLINT).
2104		execute.cm:
2105		- Everything having to do with executing binary code objects.
2106		- Dynamic environments.
2107		core.cm:
2108		- SML/NJ-specific instantiations of the elaborator and MLRISC.
2109		- Top-level modules.
2110		- FLINT (this should eventually become its own library)
2111
2112		Notes:
2113
2114		I am not 100% happy with the way I separated the elaborator (and its
2115		data structures) from FLINT.  Two instances of the same problem:
2116
2117		1. Data structures contain certain fields that carry FLINT-specific
2118		information.  I hacked around this using exn and the property list
2119		module from smlnj-lib.  But the fact that there are middle-end
2120		specific fields around at all is a bit annoying.
2121
2122		2. The elaborator calculates certain FLINT-related information.  I tried
2123		to make this as abstract as I could using functorization, but, again,
2124		the fact that the elaborator has to perform calculations on behalf
2125		of the middle-end at all is not nice.
2126
2127		3. Having to used exn and property lists is unfortunate because it
2128		weakens type checking.  The other alternative (parameterizing
2129		nearly *everything*) is not appealing, though.
2130
2131		I removed the "rebinding =" warning hack because due to the new organization
2132		it was awkward to maintain it.  As a result, the compiler now issues some of
2133		these warnings when compiling init.cmi during bootstrap compilation. On
2134		the plus side, you also get a warning when you do, for example:
2135		val op = = Int32.+
2136		which was not the case up to now.
2137
2138		I placed "assign" and "deref" into the _Core structure so that the
2139		code that deals with the "lazy" keyword can find them there.  This
2140		removes the need for having access to the primitive environment
2141		during elaboration.
2142
2143		----------------------------------------------------------------------
2144		Name: Matthias Blume
2145		Date: 2001/08/13
2146		Tag: blume-20010813-closures
2147		Description:
2148
2149		This fix was sent to us by Zhong Shao.  It is supposed to improve the
2150		performance of certain loops by avoiding needless closure allocation.
2151
2152		----------------------------------------------------------------------
2153		Name: Lal George
2154		Date: 2001/07/31 10:03:23 EDT 2001
2155		Tag: george-20010731-x86-fmalloc
2156		Description: Fixed bug in x86 calls
2157
2158		There was a bug where call instructions would mysteriously
2159		vanish. The call instruction had to be one that returned
2160		a floating point value.
2161
2162		----------------------------------------------------------------------
2163		Name: Lal George
2164		Date: 2001/07/19 16:36:29 EDT 2001
2165		Tag: george-20010719-simple-cells
2166		Description:
2167
2168		I have dramatically simplified the interface for CELLS in MLRISC.
2169
2170		In summary, the cells interface is broken up into three parts:
2171
2172		1. CellsBasis : CELLS_BASIS
2173
2174		CellsBasis is a top level structure and common for all
2175		architectures.  it contains the definitions of basic datatypes
2176		and utility  functions over these types.
2177
2178		2. functor Cells() : CELLS
2179
2180		Cells generates an interface for CELLS that incorporates the
2181		specific resources on the target architecture, such as the
2182		presence of special register classes, their number and size,
2183		and various useful substructures.
2184
2185		3. <ARCH>CELLS
2186
2187		e.g. SparcCells: SPARCCELLS
2188
2189		<ARCH>CELLS usually contains additional bindings for special
2190		registers  on the architecture, such as:
2191
2192		val r0 : cell           (* register zero *)
2193		val y : cell            (* Y register *)
2194		val psr : cell          (* processor status register *)
2195		...
2196
2197		The structure returned by applying the Cells functor is opened
2198		in this interface.
2199
2200		The main implication of all this is that the datatypes for cells is
2201		split between CellsBasis and CELLS -- a fairly simple change for user
2202		code.
2203
2204		In the old scheme the CELLS interface had a definitional binding of
2205		the form:
2206
2207		signature CELLS = sig
2208
2209		structure CellsBasis = CellsBasis
2210
2211		...
2212
2213		end
2214
2215		With all the sharing constraints that goes on in MLRISC, this old
2216		design  quickly leads to errors such as:
2217
2218		"structure definition spec inside of sharing ... "
2219
2220
2221		and appears to require an unacceptable amount of sharing and where
2222		constraint hackery.
2223
2224		I think this error message (the interaction of definitional specs and
2225		sharing) requires more explanation on our web page.
2226
2227		----------------------------------------------------------------------
2228		Name: Matthias Blume
2229		Date: 2001/07/19 15:00:00 EDT
2230		Tag: blume-20010719-libreorg
2231		Description:
2232
2233		This update puts together a fairly extensive but straightforward change
2234		to the way the libraries that implement the interactive system are
2235		organized:
2236
2237		The biggest change is the elimination of structure Compiler.  As a
2238		replacement for this structure, there is now a CM library
2239		(known as $smlnj/compiler.cm or $smlnj/compiler/current.cm)
2240		that exports all the substructures of the original structure Compiler
2241		directly.  So instead of saying Compiler.Foo.bar one now simply
2242		says Foo.bar.  (The CM libraries actually export a collection of
2243		structures that is richer than the collection of substructures of
2244		structure Compiler.)
2245
2246		To make the transition smooth, there is a separate library called
2247		$smlnj/compiler/compiler.cm which puts together and exports the
2248		original structure Compiler (or at least something very close to it).
2249
2250		There are five members of the original structure Compiler
2251		that are not exported directly but which instead became members
2252		of a new structure Backend (described by signature BACKEND).  These are:
2253		structure Profile (: PROFILE), structure Compile (: COMPILE), structure
2254		Interact (: INTERACT), structure Machine (: MACHINE), and val
2255		architecture (: string).
2256
2257		Structure Compiler.Version has become structure CompilerVersion.
2258
2259		Cross-compilers for alpha32, hppa, ppc, sparc, and x86 are provided
2260		by $smlnj/compiler/<arch>.cm where <arch> is alpha32, hppa, ppc, sparc,
2261		or x86, respectively.
2262		Each of these exports the same frontend structures that
2263		$smlnj/compiler.cm exports.  But they do not have a structure Backend
2264		and instead export some structure <Arch>Backend where <Arch> is Alpha32,
2265		Hppa, PPC, Sparc, or X86, respectively.
2266
2267		Library $smlnj/compiler/all.cm exports the union of the exports of
2268		$smlnj/compiler/<arch>.cm
2269
2270		There are no structures <Arch>Compiler anymore, use
2271		$smlnj/compiler/<arch>.cm instead.
2272
2273		Library host-compiler-0.cm is gone.  Instead, the internal library
2274		that instantiates CM is now called cm0.cm.  Selection of the host
2275		compiler (backend) is no longer done here but. (Responsibility for it
2276		now lies with $smlnj/compiler/current.cm.  This seems to be more
2277		logical.)
2278
2279		Many individual files have been moved or renamed.  Some files have
2280		been split into multiple files, and some "dead" files have been deleted.
2281
2282		Aside from these changes to library organization, there are also changes
2283		to the way the code itself is organized:
2284
2285		Structure Binfile has been re-implemented in such a way that it no
2286		longer needs any knowledge of the compiler.  It exclusively deals
2287		with the details of binfile layout.  It no longer invokes the
2288		compiler (for the purpose of creating new prospective binfile
2289		content), and it no longer has any knowledge of how to interpret
2290		pickles.
2291
2292		Structure Compile (: COMPILE) has been stripped down to the bare
2293		essentials of compilation.  It no longer deals with linking/execution.
2294		The interface has been cleaned up considerably.
2295
2296		Utility routines for dealing with linking and execution have been
2297		moved into their own substructures.
2298
2299		(The ultimate goal of these changes is to provide a light-weight
2300		binfile loader/linker (at least for, e.g., stable libraries) that
2301		does not require CM or the compiler to be present.)
2302
2303		CM documentation has been updated to reflect the changes to library
2304		organization.
2305
2306		----------------------------------------------------------------------
2307		Name: Matthias Blume
2308		Date: 2001/07/10 17:30:00 EDT
2309		Tag: Release_110_34
2310		Description:
2311
2312		Minor tweak to 110.34 (re-tagged):
2313
2314		- README.html file added to CVS repository
2315		- runtime compiles properly under FreeBSD 3.X and 4.X
2316
2317		----------------------------------------------------------------------
2318		Name: Matthias Blume
2319		Date: 2001/07/10 17:30:00 EDT
2320		Tag: Release_110_34
2321		Description:
2322
2323		New version number (110.34). New bootfiles.
2324
2325		----------------------------------------------------------------------
2326		Name: Matthias Blume
2327		Date: 2001/07/09 16:00:00 EDT
2328		Tag: blume-20010709-more-varargs
2329		Description:
2330
2331		I changed the handling of varargs in ml-nlffigen again:
2332		The ellipsis ... will now simply be ignored (with an accompanying warning).
2333
2334		The immediate effect is that you can actually call a varargs function
2335		from ML -- but you can't actually supply any arguments beyond the ones
2336		specified explicitly.  (For example, you can call printf with its format
2337		string, but you cannot pass additional arguments.)
2338
2339		This behavior is only marginally more useful than the one before, but
2340		it has the advantage that a function or, more importantly, a function
2341		type never gets dropped on the floor, thus avoiding follow-up problems with
2342		other types that refer to the offending one.
2343
2344		----------------------------------------------------------------------
2345		Name: Matthias Blume
2346		Date: 2001/07/09 11:25:00 EDT
2347		Tag: blume-20010709-varargs
2348		Description:
2349
2350		1. ckit-lib.cm now exports structure Error
2351		2. ml-nlffigen reports occurences of "..." (i.e., varargs function types)
2352		with a warning accompanied by a source location.  Moreover, it
2353		merely skips the offending function or type and proceeds with the
2354		rest of its work.u  As a result, one can safely feed C code containing
2355		"..." to ml-nlffigen.
2356		3. There are some internal improvements to CM, providing slightly
2357		more general string substitutions in the tools subsystem.
2358
2359		----------------------------------------------------------------------
2360		Name: Matthias Blume
2361		Date: 2001/06/27 15:10:00 EDT
2362		Tag: blume-20010627-concur
2363		Description:
2364
2365		Fixed a small bug in CM's handling of parallel compilation.
2366		(You could observe the bug by Control-C-interrupting an ordinary
2367		CMB.make or CM.stabilize and then attaching some compile servers.
2368		The result was that all of a sudden the previously interrupted
2369		compilation would continue on its own.  This was because of
2370		an over-optimization: CM did not bother to clean out certain queues
2371		when no servers were attached "anyway", resulting in the contents
2372		of these queues to grab control when new servers did get attached.)
2373
2374		There is also another minor update to the CM manual.
2375
2376		----------------------------------------------------------------------
2377		Name: Matthias Blume
2378		Date: 2001/06/26 16:15:00 EDT
2379		Tag: blume-20010626-cmdoc
2380		Description:
2381
2382		Minor typo fixed in CM manual (syntax diagram for libraries).
2383
2384		----------------------------------------------------------------------
2385		Name: Matthias Blume
2386		Date: 2001/06/25 22:55:00 EDT
2387		Tag: blume-20010625-x86pc
2388		Description:
2389
2390		Fixed a nasty bug in the X86 assembly code that caused signal
2391		handlers to fail (crash) randomly.
2392
2393		----------------------------------------------------------------------
2394		Name: Matthias Blume
2395		Date: 2001/06/25 12:05:00 EDT
2396		Tag: blume-20010625-nlffigen
2397		Description:
2398
2399		This update fixes a number of minor bugs in ml-nlffigen as reported by
2400		Nick Carter <nbc@andrew.cmu.edu>.
2401
2402		1. Silly but ok typedefs of the form "typedef void myvoid;" are now accepted.
2403		2. Default names for generated files are now derived from the name of
2404		the C file *without its directory*.  In particular, this causes generated
2405		files to be placed locally even if the C file is in some system directory.
2406		3. Default names for generated signatures and structures are also derived
2407		from the C file name without its directory.  This avoids silly things
2408		like "structure GL/GL".
2409		(Other silly names are still possible because ml-nlffigen does not do
2410		a thorough check of whether generated names are legal ML identifiers.
2411		When in doubt, use command line arguments to force particular names.)
2412
2413		----------------------------------------------------------------------
2414		Name: Matthias Blume
2415		Date: 2001/06/21 12:25:00 EDT
2416		Tag: blume-20010621-eXene
2417		Description:
2418
2419		eXene now compiles and (sort of) works again.
2420
2421		The library name (for version > 110.33) is $/eXene.cm.
2422
2423		I also added an new example in src/eXene/examples/nbody.  See the
2424		README file there for details.
2425
2426		----------------------------------------------------------------------
2427		Name: Matthias Blume
2428		Date: 2001/06/20 16:40:00 EDT
2429		Tag: blume-20010620-cml
2430		Description:
2431
2432		CML now compiles and works again.
2433
2434		Libraries (for version > 110.33):
2435
2436		$cml/cml.cm            Main CML library.
2437		$cml/basis.cm          CML's version of $/basis.cm.
2438		$cml/cml-internal.cm   Internal helper library.
2439		$cml/core-cml.cm       Internal helper library.
2440		$cml-lib/trace-cml.cm  Tracing facility.
2441		$cml-lib/smlnj-lib.cm  CML's version of $/smlnj-lib.cm
2442
2443		The installer (config/install.sh) has been taught how to properly
2444		install this stuff.
2445
2446		----------------------------------------------------------------------
2447		Name: Matthias Blume
2448		Date: 2001/06/19 17:55:00 EDT
2449		Tag: blume-20010619-instantiate
2450		Description:
2451
2452		This un-breaks the fix for bug 1432.
2453		(The bug was originally fixed in 110.9 but I broke it again some
2454		time after that.)
2455
2456		----------------------------------------------------------------------
2457		Name: Matthias Blume
2458		Date: 2001/06/19 17:25:00 EDT
2459		Tag: blume-20010619-signals
2460		Description:
2461
2462		This should (hopefully) fix the long-standing signal handling bug.
2463		(The runtime system was constructing a continuation record with an
2464		incorrect descriptor which would cause the GC to drop data on the floor...)
2465
2466		----------------------------------------------------------------------
2467		Name: Matthias Blume
2468		Date: 2001/06/15 15:05:00 EDT
2469		Tag: blume-20010615-moresparc
2470		Description:
2471
2472		Here is a short late-hour update related to Sparc c-calls:
2473
2474		-- made handling of double-word arguments a bit smarter
2475
2476		-- instruction selection phase tries to collapse certain clumsily
2477		constructed ML-Trees; typical example:
2478
2479		ADD(ty,ADD(_,e,LI d1),LI d2)  ->  ADD(ty,e,LI(d1+d2))
2480
2481		This currently has no further impact on SML/NJ since mlriscGen does
2482		not seem to generate such patterns in the first place, and c-calls
2483		(which did generate them in the beginning) has meanwhile been fixed
2484		so as to avoid them as well.
2485
2486		----------------------------------------------------------------------
2487		Name: Matthias Blume
2488		Date: 2001/06/15 15:05:00 EDT
2489		Tag: blume-20010615-sparc
2490		Description:
2491
2492		The purpose of this update is to provide an implementation of NLFFI
2493		on Sparc machines.
2494
2495		Here are the changes in detail:
2496
2497		* src/MLRISC/sparc/c-calls/sparc-c-calls.sml is a new file containing
2498		the Sparc implementation of the c-calls API.
2499		* The Sparc backend of SML/NJ has been modified to uniformely use %fp
2500		for accessing the ML frame.  Thus, we have a real frame pointer and
2501		can freely modify %sp without need for an omit-frame-ptr phase.
2502		The vfp logic in src/compiler/CodeGen/* has been changed to accomodate
2503		this case.
2504		* ml-nlffigen has been taught to produce code for different architectures
2505		and calling conventions.
2506		* In a way similar to what was done in the x86 case, the Sparc
2507		backend uses its own specific extension to mltree.  (For example,
2508		it needs to be able to generate UNIMP instructions which are part
2509		of the calling convention.)
2510		* ml-nlffi-lib was reorganized to make it more modular (in particular,
2511		to make it easier to plug in new machine- and os-dependent parts).
2512
2513		There are some other fairly unrelated bug fixes and cleanups as well:
2514
2515		* I further hacked the .cm files for MLRISC tools (like MDLGen) so
2516		that they properly share their libraries with existing SML/NJ libraries.
2517		* I fixed a minor cosmetic bug in CM, supressing certain spurious
2518		follow-up error messages.
2519		* Updates to CM/CMB documentation.
2520
2521		TODO items:
2522
2523		* MLRISC should use a different register as its asmTemp on the Sparc.
2524		(The current %o2 is a really bad choice because it is part of the
2525		calling conventions, so things might interfere in unexpected ways.)
2526
2527		----------------------------------------------------------------------
2528		Name: Matthias Blume
2529		Date: 2001/06/07
2530		Tag: blume-20010607-calls
2531		Description:
2532
2533		A number of internal changes related to C calls and calling conventions:
2534
2535		1. ML-Tree CALL statements now carry a "pops" field.  It indicates the
2536		number of bytes popped implicitly (by the callee).  In most cases
2537		this field is 0 but on x86/win32 it is some non-zero value.  This
2538		is information provided for the benefit of the "omit-frameptr" pass.
2539		2. The CALL instruction on the x86 carries a similar "pops" field.
2540		The instruction selection phase copies its value from the ML-Tree
2541		CALL statement.
2542		3. On all other architectures, the instruction selection phase checks
2543		whether "pops=0" and complains if not.
2544		4. The c-calls implementation for x86 now accepts two calling conventions:
2545		"ccall" and "stdcall".  When "ccall" is selected, the caller cleans
2546		up after the call and pops is set to 0.  For "stdcall", the caller
2547		does nothing, leaving the cleanup to the callee; pops is set to
2548		the number of bytes that were pushed onto the stack.
2549		5. The cproto decoder (compiler/Semant/types/cproto.sml) now can
2550		distinguish between "ccall" and "stdcall".
2551		6. The UNIMP instruction has been added to the supported Sparc instruction
2552		set. (This is needed for implementing the official C calling convention
2553		on this architecture.)
2554		7. I fixed some of the .cm files under src/MLRISC/Tools to make them
2555		work with the latest CM.
2556
2557		----------------------------------------------------------------------
2558		Name: Matthias Blume
2559		Date: 2001/06/05 15:10:00 EDT
2560		Tag: blume-20010605-cm-index
2561		Description:
2562
2563		0. The "lambdasplit" parameter for class "sml" in CM has been documented.
2564
2565		1. CM can now generate "index files".  These are human-readable files
2566		that list on a per-.cm-file basis each toplevel symbol defined or
2567		imported.  The location of the index file for
2568		<p>/<d>.cm is <p>/CM/INDEX/<d>.cm.
2569		To enable index-file generation, set CM.Control.generate_index to true
2570		or export an environment-symbol: export CM_GENERATE_INDEX=true.
2571
2572		The CM manual has been updated accordingly.
2573
2574		2. I made some slight modifications to the c-calls API in MLRISC.
2575
2576		a) There is now a callback to support saving/restoring of
2577		dedicated but caller-save registers around the actual call
2578		instruction.
2579		b) One can optionally specify a comment-annotation for the
2580		call instruction.
2581
2582		3. SML/NJ (mlriscGen.sml) uses this new API for the rawccall primop.
2583		(For example, the comment annotation shows the C prototype of
2584		the function being called.)
2585
2586		----------------------------------------------------------------------
2587		Name: Matthias Blume
2588		Date: 2001/06/01 13:30:00 EDT
2589		Tag: blume-20010601-nlffi-cleanup
2590		Description:
2591
2592		This is mostly a cleanup of MLFFI stuff:
2593
2594		- some signature files have been put into a more exposed place
2595		- the ugly 'f type parameter is gone (simplifies types tremendously!)
2596		- ml-nlffigen changed accordingly
2597		- tutorial updated
2598
2599		Other changes:
2600
2601		- author's affiliation in CM manual(s) updated
2602		- some more recognized keywords added to Allen's sml.sty
2603
2604		----------------------------------------------------------------------
2605		Name: Matthias Blume
2606		Date: 2001/05/25 15:30:00 EDT
2607		Tag: blume-20010525-iptr
2608		Description:
2609
2610		- put the official 110.33-README (as it appears on the ftp server) under
2611		CVS
2612		- fixed a small bug related to incomplete pointer types in
2613		ml-nlffigen
2614		- small cosmetic change to the ml-nlffi-lib's "arr" type constructor
2615		(it does not need the 'f type parameter)
2616
2617		----------------------------------------------------------------------
2618		Name: Matthias Blume
2619		Date: 2001/05/23 14:30:00 EDT
2620		Tag: Release_110_33
2621		Description:
2622
2623		New version number (110.33).  New bootfiles.
2624
2625		----------------------------------------------------------------------
2626		Name: Matthias Blume
2627		Date: 2001/05/22 18:06:00 EDT
2628		Tag: blume-20010522-targets
2629		Description:
2630
2631		Made install.sh use file config/targets.customized if it exists, falling
2632		back to config/targets if it doesn't.  This way one can have a customized
2633		version of the targets file without touching the "real thing", thus
2634		eliminating the constant fear of accidentally checking something bogus
2635		back into the CVS repository...  (File config/targets.customized must
2636		not be added to the repository!)
2637
2638		----------------------------------------------------------------------
2639		Name: Matthias Blume
2640		Date: 2001/05/22 16:30:00 EDT
2641		Tag: blume-20010522-minitut
2642		Description:
2643
2644		1. Bug fix in ml-nlffigen; now (hopefully) correctly handling
2645		struct returns.
2646		2. Added src/ml-nlffi-lib/Doc/mini-tutorial.txt.  This is some very
2647		incomplete, preliminary documentation for NLFFI.
2648
2649		----------------------------------------------------------------------
2650		Name: Matthias Blume
2651		Date: 2001/05/14 11:30:00 EDT
2652		Tag: blume-20010514-script
2653		Description:
2654
2655		Some bugs in install script fixed.
2656
2657		In addition to that I also made a slight change to the NLFFI API:
2658		Functors generated by ml-nlffigen now take the dynamic library as a
2659		straight functor argument, not as a suspended one.  (The original
2660		functor code used to force the suspension right away anyway, so there
2661		was nothing gained by this complication of the interface.)
2662
2663		----------------------------------------------------------------------
2664		Name: Matthias Blume
2665		Date: 2001/05/11 14:35:00 EDT
2666		Tag: blume-20010511-ml-nlffi
2667		Description:
2668
2669		I finally took the plunge and added my new FFI code to the main
2670		repository.  For x86-linux it is now ready for prime-time.
2671
2672		There are two new subdirectories of "src":
2673
2674		- ml-nlffi-lib:
2675		The utility library for programs using the FFI interface.
2676		Here is the implementation of $/c.cm and its associated low-level
2677		partners $/c-int.cm and $/memory.cm.
2678		- ml-nlffigen:
2679		A stand-alone program for generating ML glue code from C source
2680		code.
2681
2682		Building ml-nlffigen requires $/ckit-lib.cm.
2683
2684		The config/install.sh script has been updates to do the Right Thing
2685		(hopefully).
2686
2687		Notice that the source tree for the C-Kit will not be put under "src"
2688		but directly under the installation root directory.  (This is the
2689		structure that currently exists on the CVS server when you check out
2690		module "sml".)  Fortunately, config/install.sh knows about this oddity.
2691
2692		Bugs: No documentation yet.
2693
2694		----------------------------------------------------------------------
2695		Name: Matthias Blume
2696		Date: 2001/05/09 16:35:00 EDT
2697		Tag: blume-20010509-cpscontract
2698		Description:
2699
2700		Fixed a bug in the accounting code in cpsopt/contract.sml.  (The
2701		wrapper/unwrapper elimination did not decrement usage counts and some
2702		dead variables got overlooked by the dead-up logic.)
2703
2704		----------------------------------------------------------------------
2705	Name: Lal George	Name: Lal George
2706	Date: 2001/05/08  17:26:09 EDT	Date: 2001/05/08  17:26:09 EDT
2707	Tag: george-20010508-omit-frameptr	Tag: george-20010508-omit-frameptr
#	Line 2191	Line 4880
4880	elaborator). There were a lot of changes during my "linkpath" trials	elaborator). There were a lot of changes during my "linkpath" trials
4881	that could have been reverted to their original state but weren't.	that could have been reverted to their original state but weren't.
4882	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
4883	than what was strictly necessary...  (I _did_ resist the tempation	than what was strictly necessary...  (I _did_ resist the temptation
4884	of doing any "global reformatting" to avoid an untimely death at	of doing any "global reformatting" to avoid an untimely death at
4885	Dave's hands. :)	Dave's hands. :)
4886

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.824   changed lines   Added in v.1184

---

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