Diff of /sml/trunk/HISTORY

Parent Directory | Revision Log | Patch

	revision 1048, Thu Jan 31 06:36:17 2002 UTC	revision 1142, Wed Mar 13 22:25:37 2002 UTC
#	Line 14	Line 14
14
15	----------------------------------------------------------------------	----------------------------------------------------------------------
16	Name: Allen Leung	Name: Allen Leung
17		Date: 2002/03/13 17:30:00 EST
18		Tag: leunga-20020313-x86-fp-unary
19		Description:
20
21		Bug fix for:
22
23		> leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
24		> Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
25		> - fun f(x,(y,z)) = Real.~ y;
26		> [autoloading]
27		> [autoloading done]
28		>       fchsl   (%eax), 184(%esp)
29		> Error: MLRisc bug: X86MCEmitter.emitInstr
30		>
31		> uncaught exception Error
32		>   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
33
34		The problem was that the code generator did not generate any fp registers
35		in this case, and the ra didn't know that it needed to run the X86FP phase to
36		translate the pseudo fp instruction.   This only happened with unary fp
37		operators in certain situations.
38
39		----------------------------------------------------------------------
40		Name: Matthias Blume
41		Date: 2002/03/13 14:00:00 EST
42		Tag: blume-20020313-overload-etc
43		Description:
44
45		1. Added _overload as a synonym for overload for backward compatibility.
46		(Control.overloadKW must be true for either version to be accepted.)
47
48		2. Fixed bug in install script that caused more things to be installed
49		than what was requested in config/targets.
50
51		3. Made CM aware of the (_)overload construct so that autoloading
52		works.
53
54		----------------------------------------------------------------------
55		Name: Matthias Blume
56		Date: 2002/03/12 22:03:00 EST
57		Tag: blume-20020312-url
58		Description:
59
60		Forgot to update BOOT and srcarchiveurl.
61
62		----------------------------------------------------------------------
63		Name: Matthias Blume
64		Date: 2002/03/12 17:30:00 EST
65		Tag: blume-20020312-version110392
66		Description:
67
68		Yet another version number bump (because of small changes to the
69		binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
70
71		Changes:
72
73		The new pid generation scheme described a few weeks ago was overly
74		complicated.  I implemented a new mechanism that is simpler and
75		provides a bit more "stability":  Once CM has seen a compilation
76		unit, it keeps its identity constant (as long as you do not delete
77		those crucial CM/GUID/* files).  This means that when you change
78		and interface, compiler, then go back to the old interface, and
79		compile again, you arrive at the original pid.
80
81		There now also is a mechanism that instructs CM to use the plain
82		environment hash as a module's pid (effectively making its GUID
83		the empty string).  For this, "noguid" must be specified as an
84		option to the .sml file in question within its .cm file.
85		This is most useful for code that is being generated by tools such
86		as ml-nlffigen (because during development programmers tend to
87		erase the tool's entire output directory tree including CM's cached
88		GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
89		revert to the old, broken behavior of SML/NJ, but in specific cases
90		where there is no danger of interface confusion, its use is ok
91		(I think).
92
93		ml-nlffigen by default generates "noguid" annotations.  They can be
94		turned off by specifying -guid in its command line.
95
96		----------------------------------------------------------------------
97		Name: Lal George
98		Date: 2002/03/12 12 14:42:36 EST
99		Tag: george-20020312-frequency-computation
100		Description:
101
102		Integrated jump chaining and static block frequency into the
103		compiler. More details and numbers later.
104
105		----------------------------------------------------------------------
106		Name: Lal George
107		Date: 2002/03/11 11 22:38:53 EST
108		Tag: george-20020311-jump-chain-elim
109		Description:
110
111		Tested the jump chain elimination on all architectures (except the
112		hppa).  This is on by default right now and is profitable for the
113		alpha and x86, however, it may not be profitable for the sparc and ppc
114		when compiling the compiler.
115
116		The gc test will typically jump to a label at the end of the cluster,
117		where there is another jump to an external cluster containing the actual
118		code to invoke gc. This is to allow factoring of common gc invocation
119		sequences. That is to say, we generate:
120
121		f:
122		testgc
123		ja   L1      % jump if above to L1
124
125		L1:
126		jmp L2
127
128
129		After jump chain elimination the 'ja L1' instructions is converted to
130		'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
131		up being implemented in their long form (if L2 is far away) using:
132
133		jbe     L3      % jump if below or equal to L3
134		jmp     L2
135		L3:
136		...
137
138
139		For large compilation units L2  may be far away.
140
141
142		----------------------------------------------------------------------
143		Name: Matthias Blume
144		Date: 2002/03/11 13:30:00 EST
145		Tag: blume-20020311-mltreeeval
146		Description:
147
148		A functor parameter was missing.
149
150		----------------------------------------------------------------------
151		Name: Allen Leung
152		Date: 2002/03/11 10:30:00 EST
153		Tag: leunga-20020311-runtime-string0
154		Description:
155
156		The representation of the empty string now points to a
157		legal null terminated C string instead of unit.  It is now possible
158		to convert an ML string into C string with InlineT.CharVector.getData.
159		This compiles into one single machine instruction.
160
161		----------------------------------------------------------------------
162		Name: Allen Leung
163		Date: 2002/03/10 23:55:00 EST
164		Tag: leunga-20020310-x86-call
165		Description:
166
167		Added machine generation for CALL instruction (relative displacement mode)
168
169		----------------------------------------------------------------------
170		Name: Matthias Blume
171		Date: 2002/03/08 16:05:00
172		Tag: blume-20020308-entrypoints
173		Description:
174
175		Version number bumped to 110.39.1.  NEW BOOTFILES!
176
177		Entrypoints: non-zero offset into a code object where execution should begin.
178
179		- Added the notion of an entrypoint to CodeObj.
180		- Added reading/writing of entrypoint info to Binfile.
181		- Made runtime system bootloader aware of entrypoints.
182		- Use the address of the label of the first function given to mlriscGen
183		as the entrypoint.  This address is currently always 0, but it will
184		not be 0 once we turn on block placement.
185		- Removed the linkage cluster code (which was The Other Way(tm) of dealing
186		with entry points) from mlriscGen.
187
188		----------------------------------------------------------------------
189		Name: Allen Leung
190		Date: 2002/03/07 20:45:00 EST
191		Tag: leunga-20020307-x86-cmov
192		Description:
193
194		Bug fixes for CMOVcc on x86.
195
196		1. Added machine code generation for CMOVcc
197		2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
198		3. CMOVcc cannot have an immediate operand as argument.
199
200		----------------------------------------------------------------------
201		Name: Matthias Blume
202		Date: 2002/03/07 16:15:00 EST
203		Tag: blume-20020307-controls
204		Description:
205
206		This is a very large but mostly boring patch which makes (almost)
207		every tuneable compiler knob (i.e., pretty much everything under
208		Control.* plus a few other things) configurable via both the command
209		line and environment variables in the style CM did its configuration
210		until now.
211
212		Try starting sml with '-h' (or, if you are brave, '-H')
213
214		To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
215		implements the underlying generic mechanism.
216
217		The interface to some of the existing such facilities has changed somewhat.
218		For example, the MLRiscControl module now provides mkFoo instead of getFoo.
219		(The getFoo interface is still there for backward-compatibility, but its
220		use is deprecated.)
221
222		The ml-build script passes -Cxxx=yyy command-line arguments through so
223		that one can now twiddle the compiler settings when using this "batch"
224		compiler.
225
226		TODO items:
227
228		We should go through and throw out all controls that are no longer
229		connected to anything.  Moreover, we should go through and provide
230		meaningful (and correct!) documentation strings for those controls
231		that still are connected.
232
233		Currently, multiple calls to Controls.new are accepted (only the first
234		has any effect).  Eventually we should make sure that every control
235		is being made (via Controls.new) exactly once.  Future access can then
236		be done using Controls.acc.
237
238		Finally, it would probably be a good idea to use the getter-setter
239		interface to controls rather than ref cells.  For the time being, both
240		styles are provided by the Controls module, but getter-setter pairs are
241		better if thread-safety is of any concern because they can be wrapped.
242
243		*****************************************
244
245		One bug fix: The function blockPlacement in three of the MLRISC
246		backpatch files used to be hard-wired to one of two possibilities at
247		link time (according to the value of the placementFlag).  But (I
248		think) it should rather sense the flag every time.
249
250		*****************************************
251
252		Other assorted changes (by other people who did not supply a HISTORY entry):
253
254		1. the cross-module inliner now works much better (Monnier)
255		2. representation of weights, frequencies, and probabilities in MLRISC
256		changed in preparation of using those for weighted block placement
257		(Reppy, George)
258
259		----------------------------------------------------------------------
260		Name: Lal George
261		Date: 2002/03/07 14:44:24 EST 2002
262		Tag: george-20020307-weighted-block-placement
263
264		Tested the weighted block placement optimization on all architectures
265		(except the hppa) using AMPL to generate the block and edge frequencies.
266		Changes were required in the machine properties to correctly
267		categorize trap instructions. There is an MLRISC flag
268		"weighted-block-placement" that can be used to enable weighted block
269		placement, but this will be ineffective without block/edge
270		frequencies (coming soon).
271
272
273		----------------------------------------------------------------------
274		Name: Lal George
275		Date: 2002/03/05 17:24:48 EST
276		Tag: george-20020305-linkage-cluster
277
278		In order to support the block placement optimization, a new cluster
279		is generated as the very first cluster (called the linkage cluster).
280		It contains a single jump to the 'real' entry point for the compilation
281		unit. Block placement has no effect on the linkage cluster itself, but
282		all the other clusters  have full freedom in the manner in which they
283		reorder blocks or functions.
284
285		On the x86 the typical linkage code that is generated is:
286		----------------------
287		.align 2
288		L0:
289		addl    $L1-L0, 72(%esp)
290		jmp     L1
291
292
293		.align  2
294		L1:
295		----------------------
296
297		72(%esp) is the memory location for the stdlink register. This
298		must contain the address of the CPS function being called. In the
299		above example, it contains the address of  L0; before
300		calling L1 (the real entry point for the compilation unit), it
301		must contain the address for L1, and hence
302
303		addl $L1-L0, 72(%esp)
304
305		I have tested this on all architectures except the hppa.The increase
306		in code size is of course negligible
307
308		----------------------------------------------------------------------
309		Name: Allen Leung
310		Date: 2002/03/03 13:20:00 EST
311		Tag: leunga-20020303-mlrisc-tools
312
313		Added #[ ... ] expressions to mlrisc tools
314
315		----------------------------------------------------------------------
316		Name: Matthias Blume
317		Date: 2002/02/27 12:29:00 EST
318		Tag: blume-20020227-cdebug
319		Description:
320
321		- made types in structure C and C_Debug to be equal
322		- got rid of code duplication (c-int.sml vs. c-int-debug.sml)
323		- there no longer is a C_Int_Debug (C_Debug is directly derived from C)
324
325		----------------------------------------------------------------------
326		Name: Matthias Blume
327		Date: 2002/02/26 12:00:00 EST
328		Tag: blume-20020226-ffi
329		Description:
330
331		1. Fixed a minor bug in CM's "noweb" tool:
332		If numbering is turned off, then truly don't number (i.e., do not
333		supply the -L option to noweb).  The previous behavior was to supply
334		-L'' -- which caused noweb to use the "default" line numbering scheme.
335		Thanks to Chris Richards for pointing this out (and supplying the fix).
336
337		2. Once again, I reworked some aspects of the FFI:
338
339		A. The incomplete/complete type business:
340
341		- Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
342		gone!
343		- ML types representing an incomplete type are now *equal* to
344		ML types representing their corresponding complete types (just like
345		in C).  This is still safe because ml-nlffigen will not generate
346		RTTI for incomplete types, nor will it generate functions that
347		require access to such RTTI.   But when ML code generated from both
348		incomplete and complete versions of the C type meet, the ML types
349		are trivially interoperable.
350
351		NOTE:  These changes restore the full generality of the translation
352		(which was previously lost when I eliminated functorization)!
353
354		B. Enum types:
355
356		- Structure C now has a type constructor "enum" that is similar to
357		how the "su" constructor works.  However, "enum" is not a phantom
358		type because each "T enum" has values (and is isomorphic to
359		MLRep.Signed.int).
360		- There are generic access operations for enum objects (using
361		MLRep.Signed.int).
362		- ml-nlffigen will generate a structure E_foo for each "enum foo".
363		* The structure contains the definition of type "mlrep" (the ML-side
364		representation type of the enum).  Normally, mlrep is the same
365		as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
366		then mlrep will be defined as a datatype -- thus facilitating
367		pattern matching on mlrep values.
368		("-ec" will be suppressed if there are duplicate values in an
369		enumeration.)
370		* Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
371		will be generated for each C enum constant xxx.
372		* Conversion functions m2i and i2m convert between mlrep and
373		MLRep.Signed.int.  (Without "-ec", these functions are identities.)
374		* Coversion functions c and ml convert between mlrep and "tag enum".
375		* Access functions (get/set) fetch and store mlrep values.
376		- By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
377		enumerations are merged into one single enumeration represented by
378		structure E_'.
379
380		----------------------------------------------------------------------
381		Name: Allen Leung
382		Date: 2002/02/25 04:45:00 EST
383		Tag: leunga-20020225-cps-spill
384
385		This is a new implementation of the CPS spill phase.
386		The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
387		In case of problems, replace it with the old file spill.sml
388
389		The current compiler runs into some serious performance problems when
390		constructing a large record.  This can happen when we try to compile a
391		structure with many items.  Even a very simple structure like the following
392		makes the compiler slow down.
393
394		structure Foo = struct
395		val x_1 = 0w1 : Word32.int
396		val x_2 = 0w2 : Word32.int
397		val x_3 = 0w3 : Word32.int
398		...
399		val x_N = 0wN : Word32.int
400		end
401
402		The following table shows the compile time, from N=1000 to N=4000,
403		with the old compiler:
404
405		N
406		1000   CPS 100 spill                           0.04u  0.00s  0.00g
407		MLRISC ra                               0.06u  0.00s  0.05g
408		(spills = 0 reloads = 0)
409		TOTAL                                   0.63u  0.07s  0.21g
410
411		1100   CPS 100 spill                           8.25u  0.32s  0.64g
412		MLRISC ra                               5.68u  0.59s  3.93g
413		(spills = 0 reloads = 0)
414		TOTAL                                   14.71u  0.99s  4.81g
415
416		1500   CPS 100 spill                           58.55u  2.34s  1.74g
417		MLRISC ra                               5.54u  0.65s  3.91g
418		(spills = 543 reloads = 1082)
419		TOTAL                                   65.40u  3.13s  6.00g
420
421		2000   CPS 100 spill                           126.69u  4.84s  3.08g
422		MLRISC ra                               0.80u  0.10s  0.55g
423		(spills = 42 reloads = 84)
424		TOTAL                                   129.42u  5.10s  4.13g
425
426		3000   CPS 100 spill                           675.59u  19.03s  11.64g
427		MLRISC ra                               2.69u  0.27s  1.38g
428		(spills = 62 reloads = 124)
429		TOTAL                                   682.48u  19.61s  13.99g
430
431		4000   CPS 100 spill                           2362.82u  56.28s  43.60g
432		MLRISC ra                               4.96u  0.27s  2.72g
433		(spills = 85 reloads = 170)
434		TOTAL                                   2375.26u  57.21s  48.00g
435
436		As you can see the old cps spill module suffers from some serious
437		performance problem.  But since I cannot decipher the old code fully,
438		instead of patching the problems up, I'm reimplementing it
439		with a different algorithm.  The new code is more modular,
440		smaller when compiled, and substantially faster
441		(O(n log n) time and O(n) space).  Timing of the new spill module:
442
443		4000  CPS 100 spill                           0.02u  0.00s  0.00g
444		MLRISC ra                               0.25u  0.02s  0.15g
445		(spills=1 reloads=3)
446		TOTAL                                   7.74u  0.34s  1.62g
447
448		Implementation details:
449
450		As far as I can tell, the purpose of the CPS spill module is to make sure the
451		number of live variables at any program point (the bandwidth)
452		does not exceed a certain limit, which is determined by the
453		size of the spill area.
454
455		When the bandwidth is too large, we decrease the register pressure by
456		packing live variables into spill records.  How we achieve this is
457		completely different than what we did in the old code.
458
459		First, there is something about the MLRiscGen code generator
460		that we should be aware of:
461
462		o MLRiscGen performs code motion!
463
464		In particular, it will move floating point computations and
465		address computations involving only the heap pointer to
466		their use sites (if there is only a single use).
467		What this means is that if we have a CPS record construction
468		statement
469
470		RECORD(k,vl,w,e)
471
472		we should never count the new record address w as live if w
473		has only one use (which is often the case).
474
475		We should do something similar to floating point, but the transformation
476		there is much more complex, so I won't deal with that.
477
478		Secondly, there are now two new cps primops at our disposal:
479
480		1. rawrecord of record_kind option
481		This pure operator allocates some uninitialized storage from the heap.
482		There are two forms:
483
484		rawrecord NONE [INT n]  allocates a tagless record of length n
485		rawrecord (SOME rk) [INT n] allocates a tagged record of length n
486		and initializes the tag.
487
488		2. rawupdate of cty
489		rawupdate cty (v,i,x)
490		Assigns to x to the ith component of record v.
491		The storelist is not updated.
492
493		We use these new primops for both spilling and increment record construction.
494
495		1. Spilling.
496
497		This is implemented with a linear scan algorithm (but generalized
498		to trees).  The algorithm will create a single spill record at the
499		beginning of the cps function and use rawupdate to spill to it,
500		and SELECT or SELp to reload from it.  So both spills and reloads
501		are fine-grain operations.  In contrast, in the old algorithm
502		"spills" have to be bundled together in records.
503
504		Ideally, we should sink the spill record construction to where
505		it is needed.  We can even split the spill record into multiple ones
506		at the places where they are needed.  But CPS is not a good
507		representation for global code motion, so I'll keep it simple and
508		am not attempting this.
509
510		2. Incremental record construction (aka record splitting).
511
512		Long records with many component values which are simulatenously live
513		(recall that single use record addresses are not considered to
514		be live) are constructed with rawrecord and rawupdate.
515		We allocate space on the heap with rawrecord first, then gradually
516		fill it in with rawupdate.  This is the technique suggested to me
517		by Matthias.
518
519		Some restrictions on when this is applicable:
520		1. It is not a VECTOR record.  The code generator currently does not handle
521		this case. VECTOR record uses double indirection like arrays.
522		2. All the record component values are defined in the same "basic block"
523		as the record constructor.  This is to prevent speculative
524		record construction.
525
526		----------------------------------------------------------------------
527		Name: Allen Leung
528		Date: 2002/02/22 01:02:00 EST
529		Tag: leunga-20020222-mlrisc-tools
530
531		Minor bug fixes in the parser and rewriter
532
533		----------------------------------------------------------------------
534		Name: Allen Leung
535		Date: 2002/02/21 20:20:00 EST
536		Tag: leunga-20020221-peephole
537
538		Regenerated the peephole files.  Some contained typos in the specification
539		and some didn't compile because of pretty printing bugs in the old version
540		of 'nowhere'.
541
542		----------------------------------------------------------------------
543		Name: Allen Leung
544		Date: 2002/02/19 20:20:00 EST
545		Tag: leunga-20020219-mlrisc-tools
546		Description:
547
548		Minor bug fixes to the mlrisc-tools library:
549
550		1.  Fixed up parsing colon suffixed keywords
551		2.  Added the ability to shut the error messages up
552		3.  Reimplemented the pretty printer and fixed up/improved
553		the pretty printing of handle and -> types.
554		4.  Fixed up generation of literal symbols in the nowhere tool.
555		5.  Added some SML keywords to to sml.sty
556
557		----------------------------------------------------------------------
558		Name: Matthias Blume
559		Date: 2002/02/19 16:20:00 EST
560		Tag: blume-20020219-cmffi
561		Description:
562
563		A wild mix of changes, some minor, some major:
564
565		* All C FFI-related libraries are now anchored under $c:
566		$/c.cm      --> $c/c.cm
567		$/c-int.cm  --> $c/internals/c-int.cm
568		$/memory.cm --> $c/memory/memory.cm
569
570		* "make" tool (in CM) now treats its argument pathname slightly
571		differently:
572		1. If the native expansion is an absolute name, then before invoking
573		the "make" command on it, CM will apply OS.Path.mkRelative
574		(with relativeTo = OS.FileSys.getDir()) to it.
575		2. The argument will be passed through to subsequent phases of CM
576		processing without "going native".  In particular, if the argument
577		was an anchored path, then "make" will not lose track of that anchor.
578
579		* Compiler backends now "know" their respective C calling conventions
580		instead of having to be told about it by ml-nlffigen.  This relieves
581		ml-nlffigen from one of its burdens.
582
583		* The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
584
585		* Export C_DEBUG and C_Debug from $c/c.cm.
586
587		* C type encoding in ml-nlffi-lib has been improved to model the conceptual
588		subtyping relationship between incomplete pointers and their complete
589		counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
590		with the convention of instantiating 'o with ('t, 'c) obj whenever
591		the pointer target type is complete.  In the incomplete case, 'o
592		will be instantiated with some "'c iobj" -- a type obtained by
593		using one of the functors PointerToIncompleteType or PointerToCompleteType.
594
595		Operations that work on both incomplete and complete pointer types are
596		typed as taking an 'o ptr while operations that require the target to
597		be known are typed as taking some ('t, 'c) obj ptr.
598
599		voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
600		where void is an eqtype without any values.  This makes it possible
601		to work on voidptr values using functions meant to operate on light
602		incomplete pointers.
603
604		* As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
605		been vastly simplified.
606
607		----------------------------------------------------------------------
608		Name: Matthias Blume
609		Date: 2002/02/19 10:48:00 EST
610		Tag: blume-20020219-pqfix
611		Description:
612
613		Applied Chris Okasaki's bug fix for priority queues.
614
615		----------------------------------------------------------------------
616		Name: Matthias Blume
617		Date: 2002/02/15 17:05:00
618		Tag: Release_110_39
619		Description:
620
621		Last-minute retagging is becoming a tradition... :-(
622
623		This is the working release 110.39.
624
625		----------------------------------------------------------------------
626		Name: Matthias Blume
627		Date: 2002/02/15 16:00:00 EST
628		Tag: Release_110_39-orig
629		Description:
630
631		Working release 110.39.  New bootfiles.
632
633		(Update: There was a small bug in the installer so it wouldn't work
634		with all shells.  So I retagged. -Matthias)
635
636		----------------------------------------------------------------------
637		Name: Matthias Blume
638		Date: 2002/02/15 14:17:00 EST
639		Tag: blume-20020215-showbindings
640		Description:
641
642		Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
643		the latter can be useful for exploring what bindings are available at
644		the interactive prompt.  (The first function returns only the list
645		of symbols that are really bound, the second prints those but also the
646		ones that CM's autoloading mechanism knows about.)
647
648		----------------------------------------------------------------------
649		Name: Matthias Blume
650		Date: 2002/02/15 12:08:00 EST
651		Tag: blume-20020215-iptrs
652		Description:
653
654		Two improvements to ml-nlffigen:
655
656		1. Write files only if they do not exist or if their current contents
657		do not coincide with what's being written.  (That is, avoid messing
658		with the time stamps unless absolutely necessary.)
659
660		2. Implement a "repository" mechanism for generated files related
661		to "incomplete pointer types".   See the README file for details.
662
663		----------------------------------------------------------------------
664		Name: Matthias Blume
665		Date: 2002/02/14 11:50:00 EST
666		Tag: blume-20020214-quote
667		Description:
668
669		Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
670		because of the new and improved tag generation scheme.  (Thanks to Allen
671		Leung for pointing it out.)
672
673		----------------------------------------------------------------------
674		Name: Lal George
675		Date: 2002/02/14 09:55:27 EST 2002
676		Tag: george-20020214-isabelle-bug
677		Description:
678
679		Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
680		of Isabelle on the x86.
681
682		From Allen:
683		-----------
684		I've found the problem:
685
686		in ra-core.sml, I use the counter "blocked" to keep track of the
687		true number of elements in the freeze queue.  When the counter goes
688		to zero, I skip examining the queue.  But I've messed up the
689		bookkeeping in combine():
690
691		else ();
692		case !ucol of
693		PSEUDO => (if !cntv > 0 then
694		(if !cntu > 0 then blocked := !blocked - 1 else ();
695		^^^^^^^^^^^^^^^^^^^^^^^
696		moveu := mergeMoveList(!movev, !moveu)
697		)
698		else ();
699
700		combine() is called to coalesce two nodes u and v.
701		I think I was thinking that if the move counts of u and v are both
702		greater than zero then after they are coalesced then one node is
703		removed from the freeze queue.  Apparently I was thinking that
704		both u and v are of low degree, but that's clearly not necessarily true.
705
706
707		02/12/2002:
708		Here's the patch.  HOL now compiles.
709
710		I don't know how this impact on performance (compile
711		time or runtime).  This bug caused the RA (especially on the x86)
712		to go thru the potential spill phase when there are still nodes on the
713		freeze queue.
714
715
716
717
718		----------------------------------------------------------------------
719		Name: Matthias Blume
720		Date: 2002/02/13 22:40:00 EST
721		Tag: blume-20020213-fptr-rtti
722		Description:
723
724		Fixed a bug in ml-nlffigen that was introduced with one of the previous
725		updates.
726
727		----------------------------------------------------------------------
728		Name: Matthias Blume
729		Date: 2002/02/13 16:41:00 EST
730		Tag: blume-20020213-cmlpq
731		Description:
732
733		Added new priority queue export symbols (which have just been added to
734		smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
735		not compile and the installer would choke.)
736
737		----------------------------------------------------------------------
738		Name: Matthias Blume
739		Date: 2002/02/13 16:15:00 EST
740		Tag: blume-20020213-various
741		Description:
742
743		1. More tweaks to ml-nlffigen:
744
745		- better internal datastructures (resulting in slight speedup)
746		- "-match" option requires exact match
747		- "localized" gensym counters (untagged structs/unions nested within
748		other structs/unions or within typedefs get a fresh counter; their
749		tag will be prefixed by a concatenation of their parents' tags)
750		- bug fixes (related to calculation of transitive closure of types
751		to be included in the output)
752
753		2. Minor Basis updates:
754
755		- added implementations for List.collate and Option.app
756
757		----------------------------------------------------------------------
758		Name: Matthias Blume
759		Date: 2002/02/11 15:55:00 EST
760		Tag: blume-20020211-gensym
761		Description:
762
763		Added a "-gensym" option to command line of ml-nlffigen.  This can be
764		used to specify a "stem" -- a string that is inserted in all "gensym'd"
765		names (ML structure names that correspond to unnamed C structs, unions,
766		and enums), so that separate runs of ml-nlffigen do not clash.
767
768		----------------------------------------------------------------------
769		Name: Matthias Blume
770		Date: 2002/02/11 12:05:00 EST
771		Tag: blume-20020211-gensml
772		Description:
773
774		A quick fix for a problem with GenSML (in the pgraph-util library):
775		Make generation of toplevel "local" optional.  (Strictly speaking,
776		signature definitions within "local" are not legal SML.)
777
778		Other than that: updates to INSTALL and cm/TODO.
779
780		----------------------------------------------------------------------
781		Name: Matthias Blume
782		Date: 2002/02/08 15:00:00 EST
783		Tag: blume-20020208-uniquepid
784		Description:
785
786		0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
787
788		1. The installer (config/install.sh) has gotten smarter:
789
790		- Configuration options are a bit easier to specify now
791		(in config/targets).
792		- Bug in recognizing .tar.bz2 files fixed.
793		- Installer automatically resolves dependencies between
794		configuration options (e.g., if you ask for eXene, you will
795		also get cml -- regardless whether you asked for it or not).
796		- Installer can run in "quieter mode" by setting the environment
797		variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
798		"completely silent", though.
799		- Build HashCons library as part of smlnj-lib.
800
801		2. A new scheme for assigning persistent identifiers to compilation
802		units (and, by extension, to types etc.) has been put into place.
803		This fixes a long-standing bug where types and even dynamic values
804		can get internally confused, thereby compromising type safety
805		(abstraction) and dynamic correctness.  See
806
807		http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
808
809		for an example of how things could go wrong until now.
810
811		The downside of the new scheme is that pids are not quite as
812		persistent as they used to be: CM will generate a fresh pid
813		for every compilation unit that it thinks it sees for the first
814		time.  That means that if you compile starting from a clean, fresh
815		source tree at two different times, you end up with different
816		binaries.
817
818		Cutoff recompilation, however, has not been compromised because
819		CM keeps pid information in special caches between runs.
820
821		----------------------------------------------------------------------
822		Name: Lal George
823		Date: 2002/02/07 15:34:13 EST 2002
824		Tag: <none>
825		Description:
826
827		Compilers that generate assembly code may produce  global labels
828		whose value is resolved at link time. The various peephole optimization
829		modules did not take this in account.
830
831		TODO. The Labels.addrOf function should really return an option
832		type so that clients are forced to deal with this issue, rather
833		than an exception being raised.
834
835		----------------------------------------------------------------------
836		Name: Lal George
837		Date: 2002/02/06 13:55:02 EST
838		Tag: george-20020206-ra-breakup
839		Description:
840
841		1. A bug fix from Allen.
842
843		A typo causes extra fstp %st(0)'s to be generated at compensation
844		edges, which might cause stack underflow traps at runtime.  This
845		occurs in fft where there are extraneous fstps right before the 'into'
846		trap instruction (in this case they are harmless since none of the
847		integers overflow.)
848
849		2. Pulled out various utility modules that were embedded in the modules
850		of the register allocator. I need these modules for other purposes, but
851		they are not complete enough to put into a library (just yet).
852		----------------------------------------------------------------------
853		Name: Matthias Blume
854		Date: 2002/01/31 16:05:00 EST
855		Tag: blume-20020131-sparc-ccalls
856		Description:
857
858		1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
859		of extra stack space by mistake.  Fixed.
860
861		2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
862
863		----------------------------------------------------------------------
864		Name: Allen Leung
865	Date: 2002/01/30	Date: 2002/01/30
866	Tag: leunga-20020130-nowhere-bug-fix	Tag: leunga-20020130-nowhere-bug-fix
867	Description:	Description:
#	Line 622	Line 1470
1470
1471	----------------------------------------------------------------------	----------------------------------------------------------------------
1472	Name: Matthias Blume	Name: Matthias Blume
	>>>>>>> 1.169
1473	Date: 2001/09/18 15:35:00 EDT	Date: 2001/09/18 15:35:00 EDT
1474	Tag: blume-20010918-readme11036	Tag: blume-20010918-readme11036
1475	Description:	Description:

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.1048   changed lines   Added in v.1142

---

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