Diff of /sml/trunk/HISTORY

Parent Directory | Revision Log | Patch

	revision 972, Sun Oct 28 00:37:02 2001 UTC	revision 1179, Tue Mar 26 03:04:46 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: 2002/03/25 22:06:00 EST
18		Tag: blume-20020325-cprotobug
19		Description:
20
21		Fixed a bug in cproto (c prototype decoder).
22
23		----------------------------------------------------------------------
24		Name: Matthias Blume
25		Date: 2002/03/25 16:00:00 EST
26		Tag: blume-20020325-raw-primops
27		Description:
28
29		I did some cleanup to Allen's new primop code and
30		replaced yesterday's bootfiles with new ones.
31		(But they are stored in the same place.)
32
33		----------------------------------------------------------------------
34		Name: Matthias Blume
35		Date: 2002/03/24 22:40:00 EST
36		Tag: blume-20020324-bootfiles
37		Description:
38
39		Made the bootfiles that Allen asked for.
40
41		----------------------------------------------------------------------
42		Name: Allen Leung
43		Date: 2002/03/23 15:50:00 EST
44		Tag: leunga-20020323-flint-cps-rcc-primops
45		Description:
46
47		1. Changes to FLINT primops:
48
49		(* make a call to a C-function;
50		* The primop carries C function prototype information and specifies
51		* which of its (ML-) arguments are floating point. C prototype
52		* information is for use by the backend, ML information is for
53		* use by the CPS converter. *)
54		| RAW_CCALL of { c_proto: CTypes.c_proto,
55		ml_args: ccall_type list,
56		ml_res_opt: ccall_type option,
57		reentrant : bool
58		} option
59		(* Allocate uninitialized storage on the heap.
60		* The record is meant to hold short-lived C objects, i.e., they
61		* are not ML pointers.  With the tag, the representation is
62		* the same as RECORD with tag tag_raw32 (sz=4), or tag_fblock (sz=8)
63		*)
64		| RAW_RECORD of {tag:bool,sz:int}
65		and ccall_type = CCALL_INT32 | CCALL_REAL64 | CCALL_ML_PTR
66
67		2.  These CPS primops are now overloaded:
68
69		rawload of {kind:numkind}
70		rawstore of {kind:numkind}
71
72		The one argument form is:
73
74		rawload {kind} address
75
76		The two argument form is:
77
78		rawload {kind} [ml object, byte-offset]
79
80		3. RAW_CCALL/RCC now takes two extra arguments:
81
82		a. The first is whether the C call is reentrant, i.e., whether
83		ML state should be saved and restored.
84		b. The second argument is a string argument specifying the name of
85		library and the C function.
86
87		These things are currently not handled in the code generator, yet.
88
89		4. In CProto,
90
91		An encoding type of "bool" means "ml object" and is mapped into
92		C prototype of PTR.  Note that "bool" is different than "string",
93		even though "string" is also mapped into PTR, because "bool"
94		is assigned an CPS type of BOGt, while "string" is assigned INT32t.
95
96		5. Pickler/unpicker
97
98		Changed to handle RAW_RECORD and newest RAW_CCALL
99
100		6. MLRiscGen,
101
102		1. Changed to handle the new rawload/rawstore/rawrecord operators.
103		2. Code for handling C Calls has been moved to a new module CPSCCalls,
104		in the file CodeGen/cpscompile/cps-c-calls.sml
105
106		7. Added the conditional move operator
107
108		condmove of branch
109
110		to cps.  Generation of this is still buggy so it is currently
111		disabled.
112
113		----------------------------------------------------------------------
114		Name: Lal George
115		Date: 2002/03/22 14:18:25 EST
116		Tag: george-20020322-cps-branch-prob
117		Description:
118
119		Implemented the Ball-Larus branch prediction-heuristics, and
120		incorporated graphical viewers for control flow graphs.
121
122		Ball-Larus Heuristics:
123		---------------------
124		See the file compiler/CodeGen/cpscompile/cpsBranchProb.sml.
125
126		By design it uses the Dempster-Shafer theory for combining
127		probabilities.  For example, in the function:
128
129		fun f(n,acc) = if n = 0 then acc else f(n-1, n*acc)
130
131		the ball-larus heuristics predicts that the n=0 is unlikely
132		(OH-heuristic), and the 'then' branch is unlikely because of the
133		RH-heuristic -- giving the 'then' branch an even lower combined
134		probability using the Dempster-Shafer theory.
135
136		Finally, John Reppy's loop analysis in MLRISC, further lowers the
137		probability of the 'then' branch because of the loop in the else
138		branch.
139
140
141		Graphical Viewing:
142		------------------
143		I merely plugged in Allen's graphical viewers into the compiler. The
144		additional code is not much. At the top level, saying:
145
146		Control.MLRISC.getFlag "cfg-graphical-view" := true;
147
148		will display the graphical view of the control flow graph just before
149		back-patching.  daVinci must be in your path for this to work. If
150		daVinci is not available, then the default viewer can be changed
151		using:
152
153		Control.MLRISC.getString "viewer"
154
155		which can be set to "dot" or "vcg" for the corresponding viewers. Of
156		course, these viewers must be in your path.
157
158		The above will display the compilation unit at the level of clusters,
159		many of which are small, boring, and un-interesting. Also setting:
160
161		Control.MLRISC.getInt "cfg-graphical-view_size"
162
163		will display clusters that are larger than the value set by the above.
164
165
166		----------------------------------------------------------------------
167		Name: Matthias Blume
168		Date: 2002/03/21 22:20:00 EST
169		Tag: blume-20020321-kmp-bugfix
170		Description:
171
172		Changed the interface to the KMP routine in PreString and fixed
173		a minor bug in one place where it was used.
174
175		----------------------------------------------------------------------
176		Name: Allen Leung
177		Date: 2002/03/21 20:30:00 EST
178		Tag: leunga-20020321-cfg
179		Description:
180
181		Fixed a potential problem in cfg edge splitting.
182
183		----------------------------------------------------------------------
184		Name: Allen Leung
185		Date: 2002/03/21 17:15:00 EST
186		Tag: leunga-20020321-x86-fp-cfg
187		Description:
188
189		1. Recoded the buggy parts of x86-fp.
190
191		a. All the block reordering code has been removed.
192		We now depend on the block placement phases to do this work.
193
194		b. Critical edge splitting code has been simplified and moved into the
195		CFG modules, as where they belong.
196
197		Both of these were quite buggy and complex.  The code is now much, much
198		simpler.
199
200		2. X86 backend.
201
202		a. Added instructions for 64-bit support.  Instruction selection for
203		64-bit has not been committed, however, since that
204		requires changes to MLTREE which haven't been approved by
205		Lal and John.
206
207		b. Added support for FUCOMI and FUCOMIP when generating code for
208		PentiumPro and above.  We only generate these instructions in
209		the fast-fp mode.
210
211		c. Added cases for JP and JNP in X86FreqProps.
212
213		3. CFG
214
215		CFG now has a bunch of methods for edge splitting and merging.
216
217		4. Machine description.
218
219		John's simplification of MLTREE_BASIS.fcond broke a few machine
220		description things:
221
222		rtl-build.{sig,sml} and hppa.mdl fixed.
223
224		NOTE: the machine description stuff in the repository is still broken.
225		Again, I can't put my fixes in because that involves
226		changes to MLTREE.
227
228		----------------------------------------------------------------------
229		Name: Matthias Blume
230		Date: 2002/03/20 15:55:00 EST
231		Tag: blume-20020320-kmp
232		Description:
233
234		Implemented Knuth-Morris-Pratt string matching in PreString and used
235		it for String.isSubstring, Substring.isSubstring, and
236		Substring.position.
237
238		(Might need some stress-testing.  Simple examples worked fine.)
239
240		----------------------------------------------------------------------
241		Name: Matthias Blume
242		Date: 2002/03/19 16:37:00 EST
243		Tag: blume-20020319-witnesses
244		Description:
245
246		Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
247
248		This implements a generic mechanism for changing constness qualifiers
249		anywhere within big C types without resorting to outright "casts".
250		(So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
251		modify the constness at the outermost level.)
252		The implementation of "convert" is based on the idea of "witness"
253		values -- values that are not used by the operation but whose types
254		"testify" to their applicability.  On the implementation side, "convert"
255		is simply a projection (returning its second curried argument).  With
256		cross-module inlining, it should not result in any machine code being
257		generated.
258
259		----------------------------------------------------------------------
260		Name: Matthias Blume
261		Date: 2002/03/15 16:40:00 EST
262		Tag: blume-20020315-basis
263		Description:
264
265		Provided (preliminary?) implementations for
266
267		{String,Substring}.{concatWith,isSuffix,isSubstring}
268
269		and
270
271		Substring.full
272
273		Those are in the Basis spec but they were missing in SML/NJ.
274
275		----------------------------------------------------------------------
276		Name: Matthias Blume
277		Date: 2002/03/14 21:30:00 EST
278		Tag: blume-20020314-controls
279		Description:
280
281		Controls:
282		---------
283
284		1. Factored out the recently-added Controls : CONTROLS stuff and put
285		it into its own library $/controls-lib.cm.  The source tree for
286		this is under src/smlnj-lib/Controls.
287
288		2. Changed the names of types and functions in this interface, so they
289		make a bit more "sense":
290
291		module -> registry
292		'a registry -> 'a group
293
294		3. The interface now deals in ref cells only.  The getter/setter interface
295		is (mostly) gone.
296
297		4. Added a function that lets one register an already-existing ref cell.
298
299		5. Made the corresponding modifications to the rest of the code so that
300		everything compiles again.
301
302		6. Changed the implementation of Controls.MLRISC back to something closer
303		to the original.  In particular, this module (and therefore MLRISC)
304		does not depend on Controls.  There now is some link-time code in
305		int-sys.sml that registers the MLRISC controls with the Controls
306		module.
307
308		CM:
309		---
310
311		* One can now specify the lambda-split aggressiveness in init.cmi.
312
313		----------------------------------------------------------------------
314		Name: Allen Leung
315		Date: 2002/03/13 17:30:00 EST
316		Tag: leunga-20020313-x86-fp-unary
317		Description:
318
319		Bug fix for:
320
321		> leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
322		> Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
323		> - fun f(x,(y,z)) = Real.~ y;
324		> [autoloading]
325		> [autoloading done]
326		>       fchsl   (%eax), 184(%esp)
327		> Error: MLRisc bug: X86MCEmitter.emitInstr
328		>
329		> uncaught exception Error
330		>   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
331
332		The problem was that the code generator did not generate any fp registers
333		in this case, and the ra didn't know that it needed to run the X86FP phase to
334		translate the pseudo fp instruction.   This only happened with unary fp
335		operators in certain situations.
336
337		----------------------------------------------------------------------
338		Name: Matthias Blume
339		Date: 2002/03/13 14:00:00 EST
340		Tag: blume-20020313-overload-etc
341		Description:
342
343		1. Added _overload as a synonym for overload for backward compatibility.
344		(Control.overloadKW must be true for either version to be accepted.)
345
346		2. Fixed bug in install script that caused more things to be installed
347		than what was requested in config/targets.
348
349		3. Made CM aware of the (_)overload construct so that autoloading
350		works.
351
352		----------------------------------------------------------------------
353		Name: Matthias Blume
354		Date: 2002/03/12 22:03:00 EST
355		Tag: blume-20020312-url
356		Description:
357
358		Forgot to update BOOT and srcarchiveurl.
359
360		----------------------------------------------------------------------
361		Name: Matthias Blume
362		Date: 2002/03/12 17:30:00 EST
363		Tag: blume-20020312-version110392
364		Description:
365
366		Yet another version number bump (because of small changes to the
367		binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
368
369		Changes:
370
371		The new pid generation scheme described a few weeks ago was overly
372		complicated.  I implemented a new mechanism that is simpler and
373		provides a bit more "stability":  Once CM has seen a compilation
374		unit, it keeps its identity constant (as long as you do not delete
375		those crucial CM/GUID/* files).  This means that when you change
376		an interface, compile, then go back to the old interface, and
377		compile again, you arrive at the original pid.
378
379		There now also is a mechanism that instructs CM to use the plain
380		environment hash as a module's pid (effectively making its GUID
381		the empty string).  For this, "noguid" must be specified as an
382		option to the .sml file in question within its .cm file.
383		This is most useful for code that is being generated by tools such
384		as ml-nlffigen (because during development programmers tend to
385		erase the tool's entire output directory tree including CM's cached
386		GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
387		revert to the old, broken behavior of SML/NJ, but in specific cases
388		where there is no danger of interface confusion, its use is ok
389		(I think).
390
391		ml-nlffigen by default generates "noguid" annotations.  They can be
392		turned off by specifying -guid in its command line.
393
394		----------------------------------------------------------------------
395		Name: Lal George
396		Date: 2002/03/12 12 14:42:36 EST
397		Tag: george-20020312-frequency-computation
398		Description:
399
400		Integrated jump chaining and static block frequency into the
401		compiler. More details and numbers later.
402
403		----------------------------------------------------------------------
404		Name: Lal George
405		Date: 2002/03/11 11 22:38:53 EST
406		Tag: george-20020311-jump-chain-elim
407		Description:
408
409		Tested the jump chain elimination on all architectures (except the
410		hppa).  This is on by default right now and is profitable for the
411		alpha and x86, however, it may not be profitable for the sparc and ppc
412		when compiling the compiler.
413
414		The gc test will typically jump to a label at the end of the cluster,
415		where there is another jump to an external cluster containing the actual
416		code to invoke gc. This is to allow factoring of common gc invocation
417		sequences. That is to say, we generate:
418
419		f:
420		testgc
421		ja   L1      % jump if above to L1
422
423		L1:
424		jmp L2
425
426
427		After jump chain elimination the 'ja L1' instructions is converted to
428		'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
429		up being implemented in their long form (if L2 is far away) using:
430
431		jbe     L3      % jump if below or equal to L3
432		jmp     L2
433		L3:
434		...
435
436
437		For large compilation units L2  may be far away.
438
439
440		----------------------------------------------------------------------
441		Name: Matthias Blume
442		Date: 2002/03/11 13:30:00 EST
443		Tag: blume-20020311-mltreeeval
444		Description:
445
446		A functor parameter was missing.
447
448		----------------------------------------------------------------------
449		Name: Allen Leung
450		Date: 2002/03/11 10:30:00 EST
451		Tag: leunga-20020311-runtime-string0
452		Description:
453
454		The representation of the empty string now points to a
455		legal null terminated C string instead of unit.  It is now possible
456		to convert an ML string into C string with InlineT.CharVector.getData.
457		This compiles into one single machine instruction.
458
459		----------------------------------------------------------------------
460		Name: Allen Leung
461		Date: 2002/03/10 23:55:00 EST
462		Tag: leunga-20020310-x86-call
463		Description:
464
465		Added machine generation for CALL instruction (relative displacement mode)
466
467		----------------------------------------------------------------------
468		Name: Matthias Blume
469		Date: 2002/03/08 16:05:00
470		Tag: blume-20020308-entrypoints
471		Description:
472
473		Version number bumped to 110.39.1.  NEW BOOTFILES!
474
475		Entrypoints: non-zero offset into a code object where execution should begin.
476
477		- Added the notion of an entrypoint to CodeObj.
478		- Added reading/writing of entrypoint info to Binfile.
479		- Made runtime system bootloader aware of entrypoints.
480		- Use the address of the label of the first function given to mlriscGen
481		as the entrypoint.  This address is currently always 0, but it will
482		not be 0 once we turn on block placement.
483		- Removed the linkage cluster code (which was The Other Way(tm) of dealing
484		with entry points) from mlriscGen.
485
486		----------------------------------------------------------------------
487		Name: Allen Leung
488		Date: 2002/03/07 20:45:00 EST
489		Tag: leunga-20020307-x86-cmov
490		Description:
491
492		Bug fixes for CMOVcc on x86.
493
494		1. Added machine code generation for CMOVcc
495		2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
496		3. CMOVcc cannot have an immediate operand as argument.
497
498		----------------------------------------------------------------------
499		Name: Matthias Blume
500		Date: 2002/03/07 16:15:00 EST
501		Tag: blume-20020307-controls
502		Description:
503
504		This is a very large but mostly boring patch which makes (almost)
505		every tuneable compiler knob (i.e., pretty much everything under
506		Control.* plus a few other things) configurable via both the command
507		line and environment variables in the style CM did its configuration
508		until now.
509
510		Try starting sml with '-h' (or, if you are brave, '-H')
511
512		To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
513		implements the underlying generic mechanism.
514
515		The interface to some of the existing such facilities has changed somewhat.
516		For example, the MLRiscControl module now provides mkFoo instead of getFoo.
517		(The getFoo interface is still there for backward-compatibility, but its
518		use is deprecated.)
519
520		The ml-build script passes -Cxxx=yyy command-line arguments through so
521		that one can now twiddle the compiler settings when using this "batch"
522		compiler.
523
524		TODO items:
525
526		We should go through and throw out all controls that are no longer
527		connected to anything.  Moreover, we should go through and provide
528		meaningful (and correct!) documentation strings for those controls
529		that still are connected.
530
531		Currently, multiple calls to Controls.new are accepted (only the first
532		has any effect).  Eventually we should make sure that every control
533		is being made (via Controls.new) exactly once.  Future access can then
534		be done using Controls.acc.
535
536		Finally, it would probably be a good idea to use the getter-setter
537		interface to controls rather than ref cells.  For the time being, both
538		styles are provided by the Controls module, but getter-setter pairs are
539		better if thread-safety is of any concern because they can be wrapped.
540
541		*****************************************
542
543		One bug fix: The function blockPlacement in three of the MLRISC
544		backpatch files used to be hard-wired to one of two possibilities at
545		link time (according to the value of the placementFlag).  But (I
546		think) it should rather sense the flag every time.
547
548		*****************************************
549
550		Other assorted changes (by other people who did not supply a HISTORY entry):
551
552		1. the cross-module inliner now works much better (Monnier)
553		2. representation of weights, frequencies, and probabilities in MLRISC
554		changed in preparation of using those for weighted block placement
555		(Reppy, George)
556
557		----------------------------------------------------------------------
558		Name: Lal George
559		Date: 2002/03/07 14:44:24 EST 2002
560		Tag: george-20020307-weighted-block-placement
561
562		Tested the weighted block placement optimization on all architectures
563		(except the hppa) using AMPL to generate the block and edge frequencies.
564		Changes were required in the machine properties to correctly
565		categorize trap instructions. There is an MLRISC flag
566		"weighted-block-placement" that can be used to enable weighted block
567		placement, but this will be ineffective without block/edge
568		frequencies (coming soon).
569
570
571		----------------------------------------------------------------------
572		Name: Lal George
573		Date: 2002/03/05 17:24:48 EST
574		Tag: george-20020305-linkage-cluster
575
576		In order to support the block placement optimization, a new cluster
577		is generated as the very first cluster (called the linkage cluster).
578		It contains a single jump to the 'real' entry point for the compilation
579		unit. Block placement has no effect on the linkage cluster itself, but
580		all the other clusters  have full freedom in the manner in which they
581		reorder blocks or functions.
582
583		On the x86 the typical linkage code that is generated is:
584		----------------------
585		.align 2
586		L0:
587		addl    $L1-L0, 72(%esp)
588		jmp     L1
589
590
591		.align  2
592		L1:
593		----------------------
594
595		72(%esp) is the memory location for the stdlink register. This
596		must contain the address of the CPS function being called. In the
597		above example, it contains the address of  L0; before
598		calling L1 (the real entry point for the compilation unit), it
599		must contain the address for L1, and hence
600
601		addl $L1-L0, 72(%esp)
602
603		I have tested this on all architectures except the hppa.The increase
604		in code size is of course negligible
605
606		----------------------------------------------------------------------
607		Name: Allen Leung
608		Date: 2002/03/03 13:20:00 EST
609		Tag: leunga-20020303-mlrisc-tools
610
611		Added #[ ... ] expressions to mlrisc tools
612
613		----------------------------------------------------------------------
614		Name: Matthias Blume
615		Date: 2002/02/27 12:29:00 EST
616		Tag: blume-20020227-cdebug
617		Description:
618
619		- made types in structure C and C_Debug to be equal
620		- got rid of code duplication (c-int.sml vs. c-int-debug.sml)
621		- there no longer is a C_Int_Debug (C_Debug is directly derived from C)
622
623		----------------------------------------------------------------------
624		Name: Matthias Blume
625		Date: 2002/02/26 12:00:00 EST
626		Tag: blume-20020226-ffi
627		Description:
628
629		1. Fixed a minor bug in CM's "noweb" tool:
630		If numbering is turned off, then truly don't number (i.e., do not
631		supply the -L option to noweb).  The previous behavior was to supply
632		-L'' -- which caused noweb to use the "default" line numbering scheme.
633		Thanks to Chris Richards for pointing this out (and supplying the fix).
634
635		2. Once again, I reworked some aspects of the FFI:
636
637		A. The incomplete/complete type business:
638
639		- Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
640		gone!
641		- ML types representing an incomplete type are now *equal* to
642		ML types representing their corresponding complete types (just like
643		in C).  This is still safe because ml-nlffigen will not generate
644		RTTI for incomplete types, nor will it generate functions that
645		require access to such RTTI.   But when ML code generated from both
646		incomplete and complete versions of the C type meet, the ML types
647		are trivially interoperable.
648
649		NOTE:  These changes restore the full generality of the translation
650		(which was previously lost when I eliminated functorization)!
651
652		B. Enum types:
653
654		- Structure C now has a type constructor "enum" that is similar to
655		how the "su" constructor works.  However, "enum" is not a phantom
656		type because each "T enum" has values (and is isomorphic to
657		MLRep.Signed.int).
658		- There are generic access operations for enum objects (using
659		MLRep.Signed.int).
660		- ml-nlffigen will generate a structure E_foo for each "enum foo".
661		* The structure contains the definition of type "mlrep" (the ML-side
662		representation type of the enum).  Normally, mlrep is the same
663		as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
664		then mlrep will be defined as a datatype -- thus facilitating
665		pattern matching on mlrep values.
666		("-ec" will be suppressed if there are duplicate values in an
667		enumeration.)
668		* Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
669		will be generated for each C enum constant xxx.
670		* Conversion functions m2i and i2m convert between mlrep and
671		MLRep.Signed.int.  (Without "-ec", these functions are identities.)
672		* Coversion functions c and ml convert between mlrep and "tag enum".
673		* Access functions (get/set) fetch and store mlrep values.
674		- By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
675		enumerations are merged into one single enumeration represented by
676		structure E_'.
677
678		----------------------------------------------------------------------
679		Name: Allen Leung
680		Date: 2002/02/25 04:45:00 EST
681		Tag: leunga-20020225-cps-spill
682
683		This is a new implementation of the CPS spill phase.
684		The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
685		In case of problems, replace it with the old file spill.sml
686
687		The current compiler runs into some serious performance problems when
688		constructing a large record.  This can happen when we try to compile a
689		structure with many items.  Even a very simple structure like the following
690		makes the compiler slow down.
691
692		structure Foo = struct
693		val x_1 = 0w1 : Word32.int
694		val x_2 = 0w2 : Word32.int
695		val x_3 = 0w3 : Word32.int
696		...
697		val x_N = 0wN : Word32.int
698		end
699
700		The following table shows the compile time, from N=1000 to N=4000,
701		with the old compiler:
702
703		N
704		1000   CPS 100 spill                           0.04u  0.00s  0.00g
705		MLRISC ra                               0.06u  0.00s  0.05g
706		(spills = 0 reloads = 0)
707		TOTAL                                   0.63u  0.07s  0.21g
708
709		1100   CPS 100 spill                           8.25u  0.32s  0.64g
710		MLRISC ra                               5.68u  0.59s  3.93g
711		(spills = 0 reloads = 0)
712		TOTAL                                   14.71u  0.99s  4.81g
713
714		1500   CPS 100 spill                           58.55u  2.34s  1.74g
715		MLRISC ra                               5.54u  0.65s  3.91g
716		(spills = 543 reloads = 1082)
717		TOTAL                                   65.40u  3.13s  6.00g
718
719		2000   CPS 100 spill                           126.69u  4.84s  3.08g
720		MLRISC ra                               0.80u  0.10s  0.55g
721		(spills = 42 reloads = 84)
722		TOTAL                                   129.42u  5.10s  4.13g
723
724		3000   CPS 100 spill                           675.59u  19.03s  11.64g
725		MLRISC ra                               2.69u  0.27s  1.38g
726		(spills = 62 reloads = 124)
727		TOTAL                                   682.48u  19.61s  13.99g
728
729		4000   CPS 100 spill                           2362.82u  56.28s  43.60g
730		MLRISC ra                               4.96u  0.27s  2.72g
731		(spills = 85 reloads = 170)
732		TOTAL                                   2375.26u  57.21s  48.00g
733
734		As you can see the old cps spill module suffers from some serious
735		performance problem.  But since I cannot decipher the old code fully,
736		instead of patching the problems up, I'm reimplementing it
737		with a different algorithm.  The new code is more modular,
738		smaller when compiled, and substantially faster
739		(O(n log n) time and O(n) space).  Timing of the new spill module:
740
741		4000  CPS 100 spill                           0.02u  0.00s  0.00g
742		MLRISC ra                               0.25u  0.02s  0.15g
743		(spills=1 reloads=3)
744		TOTAL                                   7.74u  0.34s  1.62g
745
746		Implementation details:
747
748		As far as I can tell, the purpose of the CPS spill module is to make sure the
749		number of live variables at any program point (the bandwidth)
750		does not exceed a certain limit, which is determined by the
751		size of the spill area.
752
753		When the bandwidth is too large, we decrease the register pressure by
754		packing live variables into spill records.  How we achieve this is
755		completely different than what we did in the old code.
756
757		First, there is something about the MLRiscGen code generator
758		that we should be aware of:
759
760		o MLRiscGen performs code motion!
761
762		In particular, it will move floating point computations and
763		address computations involving only the heap pointer to
764		their use sites (if there is only a single use).
765		What this means is that if we have a CPS record construction
766		statement
767
768		RECORD(k,vl,w,e)
769
770		we should never count the new record address w as live if w
771		has only one use (which is often the case).
772
773		We should do something similar to floating point, but the transformation
774		there is much more complex, so I won't deal with that.
775
776		Secondly, there are now two new cps primops at our disposal:
777
778		1. rawrecord of record_kind option
779		This pure operator allocates some uninitialized storage from the heap.
780		There are two forms:
781
782		rawrecord NONE [INT n]  allocates a tagless record of length n
783		rawrecord (SOME rk) [INT n] allocates a tagged record of length n
784		and initializes the tag.
785
786		2. rawupdate of cty
787		rawupdate cty (v,i,x)
788		Assigns to x to the ith component of record v.
789		The storelist is not updated.
790
791		We use these new primops for both spilling and increment record construction.
792
793		1. Spilling.
794
795		This is implemented with a linear scan algorithm (but generalized
796		to trees).  The algorithm will create a single spill record at the
797		beginning of the cps function and use rawupdate to spill to it,
798		and SELECT or SELp to reload from it.  So both spills and reloads
799		are fine-grain operations.  In contrast, in the old algorithm
800		"spills" have to be bundled together in records.
801
802		Ideally, we should sink the spill record construction to where
803		it is needed.  We can even split the spill record into multiple ones
804		at the places where they are needed.  But CPS is not a good
805		representation for global code motion, so I'll keep it simple and
806		am not attempting this.
807
808		2. Incremental record construction (aka record splitting).
809
810		Long records with many component values which are simulatenously live
811		(recall that single use record addresses are not considered to
812		be live) are constructed with rawrecord and rawupdate.
813		We allocate space on the heap with rawrecord first, then gradually
814		fill it in with rawupdate.  This is the technique suggested to me
815		by Matthias.
816
817		Some restrictions on when this is applicable:
818		1. It is not a VECTOR record.  The code generator currently does not handle
819		this case. VECTOR record uses double indirection like arrays.
820		2. All the record component values are defined in the same "basic block"
821		as the record constructor.  This is to prevent speculative
822		record construction.
823
824		----------------------------------------------------------------------
825		Name: Allen Leung
826		Date: 2002/02/22 01:02:00 EST
827		Tag: leunga-20020222-mlrisc-tools
828
829		Minor bug fixes in the parser and rewriter
830
831		----------------------------------------------------------------------
832		Name: Allen Leung
833		Date: 2002/02/21 20:20:00 EST
834		Tag: leunga-20020221-peephole
835
836		Regenerated the peephole files.  Some contained typos in the specification
837		and some didn't compile because of pretty printing bugs in the old version
838		of 'nowhere'.
839
840		----------------------------------------------------------------------
841		Name: Allen Leung
842		Date: 2002/02/19 20:20:00 EST
843		Tag: leunga-20020219-mlrisc-tools
844		Description:
845
846		Minor bug fixes to the mlrisc-tools library:
847
848		1.  Fixed up parsing colon suffixed keywords
849		2.  Added the ability to shut the error messages up
850		3.  Reimplemented the pretty printer and fixed up/improved
851		the pretty printing of handle and -> types.
852		4.  Fixed up generation of literal symbols in the nowhere tool.
853		5.  Added some SML keywords to to sml.sty
854
855		----------------------------------------------------------------------
856		Name: Matthias Blume
857		Date: 2002/02/19 16:20:00 EST
858		Tag: blume-20020219-cmffi
859		Description:
860
861		A wild mix of changes, some minor, some major:
862
863		* All C FFI-related libraries are now anchored under $c:
864		$/c.cm      --> $c/c.cm
865		$/c-int.cm  --> $c/internals/c-int.cm
866		$/memory.cm --> $c/memory/memory.cm
867
868		* "make" tool (in CM) now treats its argument pathname slightly
869		differently:
870		1. If the native expansion is an absolute name, then before invoking
871		the "make" command on it, CM will apply OS.Path.mkRelative
872		(with relativeTo = OS.FileSys.getDir()) to it.
873		2. The argument will be passed through to subsequent phases of CM
874		processing without "going native".  In particular, if the argument
875		was an anchored path, then "make" will not lose track of that anchor.
876
877		* Compiler backends now "know" their respective C calling conventions
878		instead of having to be told about it by ml-nlffigen.  This relieves
879		ml-nlffigen from one of its burdens.
880
881		* The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
882
883		* Export C_DEBUG and C_Debug from $c/c.cm.
884
885		* C type encoding in ml-nlffi-lib has been improved to model the conceptual
886		subtyping relationship between incomplete pointers and their complete
887		counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
888		with the convention of instantiating 'o with ('t, 'c) obj whenever
889		the pointer target type is complete.  In the incomplete case, 'o
890		will be instantiated with some "'c iobj" -- a type obtained by
891		using one of the functors PointerToIncompleteType or PointerToCompleteType.
892
893		Operations that work on both incomplete and complete pointer types are
894		typed as taking an 'o ptr while operations that require the target to
895		be known are typed as taking some ('t, 'c) obj ptr.
896
897		voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
898		where void is an eqtype without any values.  This makes it possible
899		to work on voidptr values using functions meant to operate on light
900		incomplete pointers.
901
902		* As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
903		been vastly simplified.
904
905		----------------------------------------------------------------------
906		Name: Matthias Blume
907		Date: 2002/02/19 10:48:00 EST
908		Tag: blume-20020219-pqfix
909		Description:
910
911		Applied Chris Okasaki's bug fix for priority queues.
912
913		----------------------------------------------------------------------
914		Name: Matthias Blume
915		Date: 2002/02/15 17:05:00
916		Tag: Release_110_39
917		Description:
918
919		Last-minute retagging is becoming a tradition... :-(
920
921		This is the working release 110.39.
922
923		----------------------------------------------------------------------
924		Name: Matthias Blume
925		Date: 2002/02/15 16:00:00 EST
926		Tag: Release_110_39-orig
927		Description:
928
929		Working release 110.39.  New bootfiles.
930
931		(Update: There was a small bug in the installer so it wouldn't work
932		with all shells.  So I retagged. -Matthias)
933
934		----------------------------------------------------------------------
935		Name: Matthias Blume
936		Date: 2002/02/15 14:17:00 EST
937		Tag: blume-20020215-showbindings
938		Description:
939
940		Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
941		the latter can be useful for exploring what bindings are available at
942		the interactive prompt.  (The first function returns only the list
943		of symbols that are really bound, the second prints those but also the
944		ones that CM's autoloading mechanism knows about.)
945
946		----------------------------------------------------------------------
947		Name: Matthias Blume
948		Date: 2002/02/15 12:08:00 EST
949		Tag: blume-20020215-iptrs
950		Description:
951
952		Two improvements to ml-nlffigen:
953
954		1. Write files only if they do not exist or if their current contents
955		do not coincide with what's being written.  (That is, avoid messing
956		with the time stamps unless absolutely necessary.)
957
958		2. Implement a "repository" mechanism for generated files related
959		to "incomplete pointer types".   See the README file for details.
960
961		----------------------------------------------------------------------
962		Name: Matthias Blume
963		Date: 2002/02/14 11:50:00 EST
964		Tag: blume-20020214-quote
965		Description:
966
967		Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
968		because of the new and improved tag generation scheme.  (Thanks to Allen
969		Leung for pointing it out.)
970
971		----------------------------------------------------------------------
972		Name: Lal George
973		Date: 2002/02/14 09:55:27 EST 2002
974		Tag: george-20020214-isabelle-bug
975		Description:
976
977		Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
978		of Isabelle on the x86.
979
980		From Allen:
981		-----------
982		I've found the problem:
983
984		in ra-core.sml, I use the counter "blocked" to keep track of the
985		true number of elements in the freeze queue.  When the counter goes
986		to zero, I skip examining the queue.  But I've messed up the
987		bookkeeping in combine():
988
989		else ();
990		case !ucol of
991		PSEUDO => (if !cntv > 0 then
992		(if !cntu > 0 then blocked := !blocked - 1 else ();
993		^^^^^^^^^^^^^^^^^^^^^^^
994		moveu := mergeMoveList(!movev, !moveu)
995		)
996		else ();
997
998		combine() is called to coalesce two nodes u and v.
999		I think I was thinking that if the move counts of u and v are both
1000		greater than zero then after they are coalesced then one node is
1001		removed from the freeze queue.  Apparently I was thinking that
1002		both u and v are of low degree, but that's clearly not necessarily true.
1003
1004
1005		02/12/2002:
1006		Here's the patch.  HOL now compiles.
1007
1008		I don't know how this impact on performance (compile
1009		time or runtime).  This bug caused the RA (especially on the x86)
1010		to go thru the potential spill phase when there are still nodes on the
1011		freeze queue.
1012
1013
1014
1015
1016		----------------------------------------------------------------------
1017		Name: Matthias Blume
1018		Date: 2002/02/13 22:40:00 EST
1019		Tag: blume-20020213-fptr-rtti
1020		Description:
1021
1022		Fixed a bug in ml-nlffigen that was introduced with one of the previous
1023		updates.
1024
1025		----------------------------------------------------------------------
1026		Name: Matthias Blume
1027		Date: 2002/02/13 16:41:00 EST
1028		Tag: blume-20020213-cmlpq
1029		Description:
1030
1031		Added new priority queue export symbols (which have just been added to
1032		smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
1033		not compile and the installer would choke.)
1034
1035		----------------------------------------------------------------------
1036		Name: Matthias Blume
1037		Date: 2002/02/13 16:15:00 EST
1038		Tag: blume-20020213-various
1039		Description:
1040
1041		1. More tweaks to ml-nlffigen:
1042
1043		- better internal datastructures (resulting in slight speedup)
1044		- "-match" option requires exact match
1045		- "localized" gensym counters (untagged structs/unions nested within
1046		other structs/unions or within typedefs get a fresh counter; their
1047		tag will be prefixed by a concatenation of their parents' tags)
1048		- bug fixes (related to calculation of transitive closure of types
1049		to be included in the output)
1050
1051		2. Minor Basis updates:
1052
1053		- added implementations for List.collate and Option.app
1054
1055		----------------------------------------------------------------------
1056		Name: Matthias Blume
1057		Date: 2002/02/11 15:55:00 EST
1058		Tag: blume-20020211-gensym
1059		Description:
1060
1061		Added a "-gensym" option to command line of ml-nlffigen.  This can be
1062		used to specify a "stem" -- a string that is inserted in all "gensym'd"
1063		names (ML structure names that correspond to unnamed C structs, unions,
1064		and enums), so that separate runs of ml-nlffigen do not clash.
1065
1066		----------------------------------------------------------------------
1067		Name: Matthias Blume
1068		Date: 2002/02/11 12:05:00 EST
1069		Tag: blume-20020211-gensml
1070		Description:
1071
1072		A quick fix for a problem with GenSML (in the pgraph-util library):
1073		Make generation of toplevel "local" optional.  (Strictly speaking,
1074		signature definitions within "local" are not legal SML.)
1075
1076		Other than that: updates to INSTALL and cm/TODO.
1077
1078		----------------------------------------------------------------------
1079		Name: Matthias Blume
1080		Date: 2002/02/08 15:00:00 EST
1081		Tag: blume-20020208-uniquepid
1082		Description:
1083
1084		0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
1085
1086		1. The installer (config/install.sh) has gotten smarter:
1087
1088		- Configuration options are a bit easier to specify now
1089		(in config/targets).
1090		- Bug in recognizing .tar.bz2 files fixed.
1091		- Installer automatically resolves dependencies between
1092		configuration options (e.g., if you ask for eXene, you will
1093		also get cml -- regardless whether you asked for it or not).
1094		- Installer can run in "quieter mode" by setting the environment
1095		variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
1096		"completely silent", though.
1097		- Build HashCons library as part of smlnj-lib.
1098
1099		2. A new scheme for assigning persistent identifiers to compilation
1100		units (and, by extension, to types etc.) has been put into place.
1101		This fixes a long-standing bug where types and even dynamic values
1102		can get internally confused, thereby compromising type safety
1103		(abstraction) and dynamic correctness.  See
1104
1105		http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
1106
1107		for an example of how things could go wrong until now.
1108
1109		The downside of the new scheme is that pids are not quite as
1110		persistent as they used to be: CM will generate a fresh pid
1111		for every compilation unit that it thinks it sees for the first
1112		time.  That means that if you compile starting from a clean, fresh
1113		source tree at two different times, you end up with different
1114		binaries.
1115
1116		Cutoff recompilation, however, has not been compromised because
1117		CM keeps pid information in special caches between runs.
1118
1119		----------------------------------------------------------------------
1120		Name: Lal George
1121		Date: 2002/02/07 15:34:13 EST 2002
1122		Tag: <none>
1123		Description:
1124
1125		Compilers that generate assembly code may produce  global labels
1126		whose value is resolved at link time. The various peephole optimization
1127		modules did not take this in account.
1128
1129		TODO. The Labels.addrOf function should really return an option
1130		type so that clients are forced to deal with this issue, rather
1131		than an exception being raised.
1132
1133		----------------------------------------------------------------------
1134		Name: Lal George
1135		Date: 2002/02/06 13:55:02 EST
1136		Tag: george-20020206-ra-breakup
1137		Description:
1138
1139		1. A bug fix from Allen.
1140
1141		A typo causes extra fstp %st(0)'s to be generated at compensation
1142		edges, which might cause stack underflow traps at runtime.  This
1143		occurs in fft where there are extraneous fstps right before the 'into'
1144		trap instruction (in this case they are harmless since none of the
1145		integers overflow.)
1146
1147		2. Pulled out various utility modules that were embedded in the modules
1148		of the register allocator. I need these modules for other purposes, but
1149		they are not complete enough to put into a library (just yet).
1150		----------------------------------------------------------------------
1151		Name: Matthias Blume
1152		Date: 2002/01/31 16:05:00 EST
1153		Tag: blume-20020131-sparc-ccalls
1154		Description:
1155
1156		1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
1157		of extra stack space by mistake.  Fixed.
1158
1159		2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
1160
1161		----------------------------------------------------------------------
1162		Name: Allen Leung
1163		Date: 2002/01/30
1164		Tag: leunga-20020130-nowhere-bug-fix
1165		Description:
1166
1167		MLRISC bug fixes:
1168		1. Fixed a bindings computation bug in the 'nowhere' program generator tool.
1169		2. MachineInt.fromString was negating its value.
1170
1171		----------------------------------------------------------------------
1172		Name: Matthias Blume
1173		Date: 2002/01/29
1174		Tag: blume-20020129-INSTALL
1175		Description:
1176
1177		- Added somewhat detailed installation instructions (file INSTALL).
1178		- Fixed curl-detection bug in config/install.sh.
1179		- It is now possible to select the URL getter using the URLGETTER
1180		environment variable:
1181
1182		not set / "unknown"      --> automatic detection (script tries wget,
1183		curl, and lynx)
1184		"wget" / "curl" / "lynx" --> use the specified program (script "knows"
1185		how to properly invoke them)
1186		other                    --> use $URLGETTER directly, it must take
1187		precisely two command-line arguments
1188		(source URL and destination file name)
1189
1190		----------------------------------------------------------------------
1191		Name: Matthias Blume
1192		Date: 2002/01/28
1193		Tag: blume-20020128-sparc-ccalls
1194		Description:
1195
1196		- Fixed problem with calculation of "used" registers in sparc-c-calls.
1197		- Make use of the allocParam argument in sparc-c-calls.
1198
1199		----------------------------------------------------------------------
1200		Name: Matthias Blume
1201		Date: 2002/01/28
1202		Tag: blume-20020128-allocParam
1203		Description:
1204
1205		John Reppy:  Changes c-calls API to accept client-callback for
1206		allocating extra stack space.
1207		me: Corresponding changes to mlriscGen (using a dummy argument that
1208		does not change the current behavior).
1209
1210		----------------------------------------------------------------------
1211		Name: Matthias Blume
1212		Date: 2002/01/28 12:00:00
1213		Tag: Release_110_38
1214		Description:
1215
1216		This time for real!!!
1217
1218		----------------------------------------------------------------------
1219		Name: Matthias Blume
1220		Date: 2002/01/28 10:56:00 EST
1221		Tag: blume-20020128-retraction
1222		Description:
1223
1224		0. Retracted earlier 110.38.  (The Release_110_38 tag has been replaced
1225		with blume-Release_110_38-retracted.)
1226
1227		1. Fixed a problem with incorrect rounding modes in real64.sml.
1228		(Thanks to Andrew Mccreight <andrew.mccreight@yale.edu>.)
1229
1230		2. A bug in ml-nlffigen related to the handling of unnamed structs, unions,
1231		and enums fixed.  The naming of corresponding ML identifiers should
1232		now be consistent again.
1233
1234		----------------------------------------------------------------------
1235		Name: Allen Leung
1236		Date: 2002/01/27
1237		Tag: leunga-20020127-nowhere
1238		Description:
1239
1240		Added a target called nowhere in the configuration scripts.
1241		Enabling this will build the MLRISC 'nowhere' tool (for translating
1242		programs with where-clauses into legal SML code) during installation.
1243
1244		----------------------------------------------------------------------
1245		Name: Matthias Blume
1246		Date: 2002/01/25 21:27:00 EST
1247		Tag: blume-Release_110_38-retracted
1248		Description:
1249
1250		Call it a (working) release!  Version is 110.38. Bootfiles are ready.
1251
1252		README will be added later.
1253
1254		!!! NOTE:  Re-tagged as blume-Release_110_38-retracted. Original tag
1255		(Release_110_38) removed.  Reason: Last-minute bug fixes.
1256
1257		----------------------------------------------------------------------
1258		Name: Matthias Blume
1259		Date: 2002/01/25
1260		Tag: blume-20020125-ffi
1261		Description:
1262
1263		A large number of tweaks and improvements to ml-nlffi-lib and
1264		ml-nlffigen:
1265
1266		- ML represenation types have been streamlined
1267		- getter and setter functions work with concrete values, not abstract
1268		ones where possible
1269		- ml-nlffigen command line more flexible (see README file there)
1270		- some bugs have been fixed (hopefully)
1271
1272		----------------------------------------------------------------------
1273		Name: Lal George
1274		Date: 2002/01/24
1275		Tag: george-20020124-risc-ra-interface
1276		Description:
1277
1278		There is a dramatic simplification in the interface to the
1279		register allocator for RISC architectures as a result of making
1280		parallel copy instructions explicit.
1281
1282		----------------------------------------------------------------------
1283		Name: Matthias Blume
1284		Date: 2002/01/22
1285		Tag: blume-20020122-x86-ccalls
1286		Description:
1287
1288		Bug fix for c-calls on x86 (having to do with how char- and
1289		short-arguments are being handled).
1290
1291		----------------------------------------------------------------------
1292		Name: Matthias Blume
1293		Date: 2002/01/21
1294		Tag: blume-20020121-ff
1295		Description:
1296
1297		Another day of fiddling with the FFI...
1298
1299		1. Bug fix/workaround:  CKIT does not complain about negative array
1300		dimensions, so ml-nlffigen has to guard itself against this possibility.
1301		(Otherwise a negative dimension would send it into an infinite loop.)
1302
1303		2. Some of the abstract types (light objects, light pointers, most "base"
1304		types) in structure C are now eqtypes.
1305
1306		3. Added constructors and test functions for NULL function pointers.
1307
1308		----------------------------------------------------------------------
1309		Name: Matthias Blume
1310		Date: 2002/01/18
1311		Tag: blume-20020118-ready-for-new-release
1312		Description:
1313
1314		Made config/srcarchiveurl point to a new place.  (Will provide boot
1315		files shortly.)
1316
1317		Maybe we christen this to be 110.38?
1318
1319		----------------------------------------------------------------------
1320		Name: Matthias Blume
1321		Date: 2002/01/18
1322		Tag: blume-20020118-more-ffifiddle
1323		Description:
1324
1325		Today's FFI fiddling:
1326
1327		- Provided a structure CGetSet with "convenient" versions of C.Get.* and
1328		C.Set.* that use concrete (MLRep.*) arguments and results instead
1329		of abstract ones.
1330
1331		- Provided word-style bit operations etc. for "int" representation
1332		types in MLRep.S<Foo>Bitops where <Foo> ranges over Char, Int, Short,
1333		and Long.
1334
1335		----------------------------------------------------------------------
1336		Name: Matthias Blume
1337		Date: 2002/01/18
1338		Tag: blume-20020118-use-x86-fp
1339		Description:
1340
1341		Now that x86-fast-fp seems to be working, I turned it back on again
1342		by default.  (Seems to work fine now, even with the FFI.)
1343
1344		Other than that, I added some documentation about the FFI to
1345		src/ml-nlffigen/README and updated the FFI test examples in
1346		src/ml-nlffi-lib/Tests/*.
1347
1348		----------------------------------------------------------------------
1349		Name: Allen Leung
1350		Date: 2002/01/17
1351		Tag: leunga-20020117-x86-fast-fp-call
1352		Description:
1353
1354		1. Fixed a problem with handling return fp values when x86's fast fp
1355		mode is turned on.
1356
1357		2. Minor pretty printing fix for cellset.  Print %st(0) as %st(0) instead
1358		of %f32.
1359
1360		3. Added a constructor INT32lit to the ast of MLRISC tools.
1361
1362		----------------------------------------------------------------------
1363		Name: Matthias Blume
1364		Date: 2002/01/16
1365		Tag: blume-20020116-ffifiddle
1366		Description:
1367
1368		More fiddling with the FFI interface:
1369
1370		- Make constness 'c instead of rw wherever possible.  This eliminates
1371		the need for certain explicit coercions.  (However, due to ML's
1372		value polymorphism, there will still be many cases where explicit
1373		coercions are necessary.  Phantom types are not the whole answer
1374		to modeling a subtyping relationship in ML.)
1375
1376		- ro/rw coersions for pointers added.  (Avoids the detour through */&.)
1377
1378		- "printf" test example added to src/ml-nlffi-lib/Tests.  (Demonstrates
1379		clumsy workaround for varargs problem.)
1380
1381		----------------------------------------------------------------------
1382		Name: Lal George
1383		Date: 2002/01/15
1384		Tag: <none>
1385		Description:
1386
1387		1. Since COPY instructions are no longer native to the architecture,
1388		a generic functor can be used to implement the expandCopies function.
1389
1390		2. Allowed EXPORT and IMPORT pseudo-op declarations to appear inside a
1391		TEXT segment.
1392
1393		----------------------------------------------------------------------
1394		Name: Matthias Blume
1395		Date: 2002/01/15
1396		Tag: blume-20020115-ffiupdates
1397		Description:
1398
1399		1. Fix for bug resulting in single-precision float values being returned
1400		incorrectly from FFI calls.
1401
1402		2. Small modifications to C FFI API:
1403
1404		- memory-allocation routines return straight objects (no options)
1405		and raise an exception in out-of-memory situations
1406		- unsafe extensions to cast between function pointers and pointers
1407		from/to ints
1408		- added structure C_Debug as an alternative to structure C where
1409		pointer-dereferencing (|*| and |*!) always check for null-pointers
1410		- added open_lib' to DynLinkage;  open_lib' works like open_lib
1411		but also takes a (possibly empty) list of existing library handles
1412		that the current library depends on
1413
1414		----------------------------------------------------------------------
1415		Name: Matthias Blume
1416		Date: 2002/01/10
1417		Tag: blume-20020110-newffigen
1418		Description:
1419
1420		1. Updates to portable graph code.
1421
1422		2. Major update to ml-nlffigen and ml-nlffi-lib.  Things are much
1423		more scalable now so that even huge interfaces such as the one
1424		for GTK compile in finite time and space. :-)
1425		See src/ml-nlffigen/README for details on what's new.
1426
1427		----------------------------------------------------------------------
1428		Name: Lal George
1429		Date: 2001/01/09 14:31:35 EST 2002
1430		Tag: george-20011206-rm-native-copy
1431		Description:
1432
1433		Removed the native COPY and FCOPY instructions
1434		from all the architectures and replaced it with the
1435		explicit COPY instruction from the previous commit.
1436
1437		It is now possible to simplify many of the optimizations
1438		modules that manipulate copies. This has not been
1439		done in this change.
1440
1441		----------------------------------------------------------------------
1442		Name: Lal George
1443		Date: 2001/12/06 16:50:13 EST 2001
1444		Tag: george-20011206-mlrisc-instruction
1445		Description:
1446
1447		Changed the representation of instructions from being fully abstract
1448		to being partially concrete. That is to say:
1449
1450		from
1451		type instruction
1452
1453		to
1454		type instr                              (* machine instruction *)
1455
1456		datatype instruction =
1457		LIVE of {regs: C.cellset, spilled: C.cellset}
1458		| KILL of {regs: C.cellset, spilled: C.cellset}
1459		| COPYXXX of {k: CB.cellkind, dst: CB.cell list, src: CB.cell list}
1460		| ANNOTATION of {i: instruction, a: Annotations.annotation}
1461		| INSTR of instr
1462
1463		This makes the handling of certain special instructions that appear on
1464		all architectures easier and uniform.
1465
1466		LIVE and KILL say that a list of registers are live or killed at the
1467		program point where they appear. No spill code is generated when an
1468		element of the 'regs' field is spilled, but the register is moved to
1469		the 'spilled' (which is present, more for debugging than anything else).
1470
1471		LIVE replaces the (now deprecated) DEFFREG instruction on the alpha.
1472		We used to generate:
1473
1474		DEFFREG f1
1475		f1 := f2 + f3
1476		trapb
1477
1478		but now generate:
1479
1480		f1 := f2 + f3
1481		trapb
1482		LIVE {regs=[f1,f2,f3], spilled=[]}
1483
1484		Furthermore, the DEFFREG (hack) required that all floating point instruction
1485		use all registers mentioned in the instruction. Therefore f1 := f2 + f3,
1486		defines f1 and uses [f1,f2,f3]! This hack is no longer required resulting
1487		in a cleaner alpha implementation. (Hopefully, intel will not get rid of
1488		this architecture).
1489
1490		COPYXXX is intended to replace the parallel COPY and FCOPY  available on
1491		all the architectures. This will result in further simplification of the
1492		register allocator that must be aware of them for coalescing purposes, and
1493		will also simplify certain aspects of the machine description that provides
1494		callbacks related to parallel copies.
1495
1496		ANNOTATION should be obvious, and now INSTR represents the honest to God
1497		machine instruction set!
1498
1499		The <arch>/instructions/<arch>Instr.sml files define certain utility
1500		functions for making porting easier -- essentially converting upper case
1501		to lower case. All machine instructions (of type instr) are in upper case,
1502		and the lower case form generates an MLRISC instruction. For example on
1503		the alpha we have:
1504
1505		datatype instr =
1506		LDA of {r:cell, b:cell, d:operand}
1507		| ...
1508
1509		val lda : {r:cell, b:cell, d:operand} -> instruction
1510		...
1511
1512		where lda is just (INSTR o LDA), etc.
1513
1514		----------------------------------------------------------------------
1515		Name: Matthias Blume
1516		Date: 2001/11/22 21:40:00 EST
1517		Tag: Release_110_37
1518		Description:
1519
1520		Release 110.37.  This time for real.
1521
1522		----------------------------------------------------------------------
1523		Name: Matthias Blume
1524		Date: 2001/11/21 16:35:00 EST
1525		Tag: blume-20011121-foot-in-mouth
1526		Description:
1527
1528		Removed the "Release_110_37" tag because of a serious bug.
1529		This will be re-tagged once the bug is fixed.
1530
1531		----------------------------------------------------------------------
1532		Name: Matthias Blume
1533		Date: 2001/11/21 16:14:00 EST
1534		Tag: blume-20011121-forgottenfile
1535		Description:
1536
1537		Forgot to add a file.  (Just a .tex-file -- part of
1538		the CM manual source.)
1539
1540		----------------------------------------------------------------------
1541		Name: Matthias Blume
1542		Date: 2001/11/21 16:10:00 EST
1543		Tag: blume-20011121-invalid_110_37
1544		Description:
1545
1546		Note: I removed the original tag "Release_110_37" from this commit
1547		because we found a serious bug in all non-x86 backends.
1548		- Matthias
1549
1550		1. Modifications to the SML/NJ code generator and to the runtime system
1551		so that code object name strings are directly inserted into code
1552		objects at code generation time.  The only business the runtime system
1553		has with this is now to read the name strings on occasions.
1554		(The encoding of the name string has also changed somewhat.)
1555
1556		2. CM now implements a simple "set calculus" for specifying export lists.
1557		In particular, it is now possible to refer to the export lists of
1558		other libraries/groups/sources and form unions as well as differences.
1559		See the latest CM manual for details.
1560
1561		3. An separate notion of "proxy" libraries has again be eliminated from
1562		CM's model.  (Proxy libraries are now simply a special case of using
1563		the export list calculus.)
1564
1565		4. Some of the existing libraries now take advantage of the new set
1566		calculus.
1567		(Notice that not all libraries have been converted because some
1568		of the existing .cm-files are supposed to be backward compatible
1569		with 110.0.x.)
1570
1571		5. Some cleanup in stand-alone programs.  (Don't use "exnMessage" -- use
1572		"General.exnMessage"!  The former relies on a certain hook to be
1573		initialized, and that often does not happen in the stand-alone case.)
1574
1575		----------------------------------------------------------------------
1576		Name: Lal George
1577		Date: 2001/11/21  13:56:18 EST
1578		Tag: george-2001121-pseudo-ops
1579		Description:
1580
1581		Implemented a complete redesign of MLRISC pseudo-ops. Now there
1582		ought to never be any question of incompatabilities with
1583		pseudo-op syntax expected by host assemblers.
1584
1585		For now, only modules supporting GAS syntax are implemented
1586		but more should follow, such as MASM, and vendor assembler
1587		syntax, e.g. IBM as, Sun as, etc.
1588
1589		----------------------------------------------------------------------
1590		Name: Matthias Blume
1591		Date: 2001/11/14 11:52:00 EST
1592		Tag: blume-20011114-srcname
1593		Description:
1594
1595		1. Routed the name of the current source file to mlriscgen where it
1596		should be directly emitted into the code object.  (This last part
1597		is yet to be done.)
1598
1599		2. Some cleanup of the pgraph code to make it match the proposal that
1600		I put out the other day.  (The proposal notwithstanding, things are
1601		still in flux here.)
1602
1603		----------------------------------------------------------------------
1604		Name: Lal George
1605		Date: 2001/11/14 09:44:04 EST
1606		Tag:
1607		Description:
1608
1609		Fix for a backpatching bug reported by Allen.
1610
1611		Because the boundary between short and long span-dependent
1612		instructions is +/- 128, there are an astounding number of
1613		span-dependent instructions whose size is over estimated.
1614
1615		Allen came up with the idea of letting the size of span
1616		dependent instructions be non-monotonic, for a maxIter
1617		number of times, after which the size must be monotonically
1618		increasing.
1619
1620		This table shows the number of span-dependent instructions
1621		whose size was over-estimated as a function of maxIter, for the
1622		file Parse/parse/ml.grm.sml:
1623
1624		maxIter            # of instructions:
1625		10                      687
1626		20                      438
1627		30                      198
1628		40                        0
1629
1630		In compiling the compiler, there is no significant difference in
1631		compilation speed between maxIter=10 and maxIter=40. Actually,
1632		my measurements showed that maxIter=40 was a tad faster than
1633		maxIter=10! Also 96% of the  files in the compiler reach a fix
1634		point within 13 iterations, so fixing maxIter at 40, while high,
1635		is okay.
1636
1637		----------------------------------------------------------------------
1638		Name: Matthias Blume
1639		Date: 2001/10/31 15:25:00 EST
1640		Tag: blume-20011031-pgraph
1641		Description:
1642
1643		CKIT:
1644		* Changed the "Function" constructor of type Ast.ctype to carry optional
1645		argument identifiers.
1646		* Changed the return type of TypeUtil.getFunction accordingly.
1647		* Type equality ignores the argument names.
1648		* TypeUtil.composite tries to preserve argument names but gives up quickly
1649		if there is a mismatch.
1650
1651		installation script:
1652		* attempts to use "curl" if available (unless "wget" is available as well)
1653
1654		CM:
1655		* has an experimental implementation of "portable graphs" which I will
1656		soon propose as an implementation-independent library format
1657		* there are also new libraries $/pgraph.cm and $/pgraph-util.cm
1658
1659		NLFFI-LIB:
1660		* some cleanup (all cosmetic)
1661
1662		NLFFIGEN:
1663		* temporarily disabled the mechanism that suppresses ML output for
1664		C definitions whose identifiers start with an underscore character
1665		* generate val bindings for enum constants
1666		* user can request that only one style (light or heavy) is being used;
1667		default is to use both (command-line arguments: -heavy and -light)
1668		* fixed bug in handling of function types involving incomplete pointers
1669		* generate ML entry points that take record arguments (i.e., using
1670		named arguments) for C functions that have a prototype with named
1671		arguments
1672		(see changes to CKIT)
1673
1674		----------------------------------------------------------------------
1675	Name: Allen Leung	Name: Allen Leung
1676	Date: 2001/10/27 20:34:00 EDT	Date: 2001/10/27 20:34:00 EDT
1677	Tag: leunga-20011027-x86-fast-fp-call	Tag: leunga-20011027-x86-fast-fp-call
#	Line 109	Line 1768
1768
1769	----------------------------------------------------------------------	----------------------------------------------------------------------
1770	Name: Matthias Blume	Name: Matthias Blume
	>>>>>>> 1.169
1771	Date: 2001/09/18 15:35:00 EDT	Date: 2001/09/18 15:35:00 EDT
1772	Tag: blume-20010918-readme11036	Tag: blume-20010918-readme11036
1773	Description:	Description:

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.972   changed lines   Added in v.1179

---

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