Diff of /sml/trunk/HISTORY

Parent Directory | Revision Log | Patch

	revision 916, Fri Aug 24 21:13:58 2001 UTC	revision 1098, Wed Feb 27 17:29:47 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	Name: Matthias Blume
17		Date: 2002/02/27 12:29:00 EST
18		Tag: blume-20020227-cdebug
19		Description:
20
21		- made types in structure C and C_Debug to be equal
22		- got rid of code duplication (c-int.sml vs. c-int-debug.sml)
23		- there no longer is a C_Int_Debug (C_Debug is directly derived from C)
24
25		----------------------------------------------------------------------
26		Name: Matthias Blume
27		Date: 2002/02/26 12:00:00 EST
28		Tag: blume-20020226-ffi
29		Description:
30
31		1. Fixed a minor bug in CM's "noweb" tool:
32		If numbering is turned off, then truly don't number (i.e., do not
33		supply the -L option to noweb).  The previous behavior was to supply
34		-L'' -- which caused noweb to use the "default" line numbering scheme.
35		Thanks to Chris Richards for pointing this out (and supplying the fix).
36
37		2. Once again, I reworked some aspects of the FFI:
38
39		A. The incomplete/complete type business:
40
41		- Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
42		gone!
43		- ML types representing an incomplete type are now *equal* to
44		ML types representing their corresponding complete types (just like
45		in C).  This is still safe because ml-nlffigen will not generate
46		RTTI for incomplete types, nor will it generate functions that
47		require access to such RTTI.   But when ML code generated from both
48		incomplete and complete versions of the C type meet, the ML types
49		are trivially interoperable.
50
51		NOTE:  These changes restore the full generality of the translation
52		(which was previously lost when I eliminated functorization)!
53
54		B. Enum types:
55
56		- Structure C now has a type constructor "enum" that is similar to
57		how the "su" constructor works.  However, "enum" is not a phantom
58		type because each "T enum" has values (and is isomorphic to
59		MLRep.Signed.int).
60		- There are generic access operations for enum objects (using
61		MLRep.Signed.int).
62		- ml-nlffigen will generate a structure E_foo for each "enum foo".
63		* The structure contains the definition of type "mlrep" (the ML-side
64		representation type of the enum).  Normally, mlrep is the same
65		as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
66		then mlrep will be defined as a datatype -- thus facilitating
67		pattern matching on mlrep values.
68		("-ec" will be suppressed if there are duplicate values in an
69		enumeration.)
70		* Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
71		will be generated for each C enum constant xxx.
72		* Conversion functions m2i and i2m convert between mlrep and
73		MLRep.Signed.int.  (Without "-ec", these functions are identities.)
74		* Coversion functions c and ml convert between mlrep and "tag enum".
75		* Access functions (get/set) fetch and store mlrep values.
76		- By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
77		enumerations are merged into one single enumeration represented by
78		structure E_'.
79
80		----------------------------------------------------------------------
81		Name: Allen Leung
82		Date: 2002/02/25 04:45:00 EST
83		Tag: leunga-20020225-cps-spill
84
85		This is a new implementation of the CPS spill phase.
86		The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
87		In case of problems, replace it with the old file spill.sml
88
89		The current compiler runs into some serious performance problems when
90		constructing a large record.  This can happen when we try to compile a
91		structure with many items.  Even a very simple structure like the following
92		makes the compiler slow down.
93
94		structure Foo = struct
95		val x_1 = 0w1 : Word32.int
96		val x_2 = 0w2 : Word32.int
97		val x_3 = 0w3 : Word32.int
98		...
99		val x_N = 0wN : Word32.int
100		end
101
102		The following table shows the compile time, from N=1000 to N=4000,
103		with the old compiler:
104
105		N
106		1000   CPS 100 spill                           0.04u  0.00s  0.00g
107		MLRISC ra                               0.06u  0.00s  0.05g
108		(spills = 0 reloads = 0)
109		TOTAL                                   0.63u  0.07s  0.21g
110
111		1100   CPS 100 spill                           8.25u  0.32s  0.64g
112		MLRISC ra                               5.68u  0.59s  3.93g
113		(spills = 0 reloads = 0)
114		TOTAL                                   14.71u  0.99s  4.81g
115
116		1500   CPS 100 spill                           58.55u  2.34s  1.74g
117		MLRISC ra                               5.54u  0.65s  3.91g
118		(spills = 543 reloads = 1082)
119		TOTAL                                   65.40u  3.13s  6.00g
120
121		2000   CPS 100 spill                           126.69u  4.84s  3.08g
122		MLRISC ra                               0.80u  0.10s  0.55g
123		(spills = 42 reloads = 84)
124		TOTAL                                   129.42u  5.10s  4.13g
125
126		3000   CPS 100 spill                           675.59u  19.03s  11.64g
127		MLRISC ra                               2.69u  0.27s  1.38g
128		(spills = 62 reloads = 124)
129		TOTAL                                   682.48u  19.61s  13.99g
130
131		4000   CPS 100 spill                           2362.82u  56.28s  43.60g
132		MLRISC ra                               4.96u  0.27s  2.72g
133		(spills = 85 reloads = 170)
134		TOTAL                                   2375.26u  57.21s  48.00g
135
136		As you can see the old cps spill module suffers from some serious
137		performance problem but since I cannot decipher the old code fully,
138		innstead of patching the problems up, I'm reimplementing it
139		with a different algorithm.  The new code is more modular,
140		smaller when compiled, and substantially faster
141		(O(n log n) time and O(n) space).  Timing of the new spill module:
142
143		4000  CPS 100 spill                           0.02u  0.00s  0.00g
144		MLRISC ra                               0.25u  0.02s  0.15g
145		(spills=1 reloads=3)
146		TOTAL                                   7.74u  0.34s  1.62g
147
148		Implementation details:
149
150		As far as I can tell, the purpose of the CPS spill module is to make sure the
151		number of live variables at any program point (the bandwidth)
152		does not exceed a certain limit, which is determined by the
153		size of the spill area.
154
155		When the bandwidth is too large, we decrease the register pressure by
156		packing live variables into spill records.  How we achieve this is
157		completely different than what we did in the old code.
158
159		First, there is something about the MLRiscGen code generator
160		that we should be aware of:
161
162		o MLRiscGen performs code motion!
163
164		In particular, it will move floating point computations and
165		address computations involving only the heap pointer to
166		their use sites (if there is only a single use).
167		What this means is that if we have a CPS record construction
168		statement
169
170		RECORD(k,vl,w,e)
171
172		we should never count the new record address w as live if w
173		has only one use (which is often the case).
174
175		We should do something similar to floating point, but the transformation
176		there is much more complex, so I won't deal with that.
177
178		Secondly, there are now two new cps primops at our disposal:
179
180		1. rawrecord of record_kind option
181		This pure operator allocates some uninitialized storage from the heap.
182		There are two forms:
183
184		rawrecord NONE [INT n]  allocates a tagless record of length n
185		rawrecord (SOME rk) [INT n] allocates a tagged record of length n
186		and initializes the tag.
187
188		2. rawupdate of cty
189		rawupdate cty (v,i,x)
190		Assigns to x to the ith component of record v.
191		The storelist is not updated.
192
193		We use these new primops for both spilling and increment record construction.
194
195		1. Spilling.
196
197		This is implemented with a linear scan algorithm (but generalized
198		to trees).  The algorithm will create a single spill record at the
199		beginning of the cps function and use rawupdate to spill to it,
200		and SELECT or SELp to reload from it.  So both spills and reloads
201		are fine-grain operations.  In contrast, in the old algorithm
202		"spills" have to be bundled together in records.
203
204		Ideally, we should sink the spill record construction to where
205		it is needed.  We can even split the spill record into multiple ones
206		at the places where they are needed.  But CPS is not a good
207		representation for global code motion, so I'll keep it simple and
208		am not attempting this.
209
210		2. Incremental record construction (aka record splitting).
211
212		Long records with many component values which are simulatenously live
213		(recall that single use record addresses are not considered to
214		be live) are constructed with rawrecord and rawupdate.
215		We allocate space on the heap with rawrecord first, then gradually
216		fill it in with rawupdate.  This is the technique suggested to me
217		by Matthias.
218
219		Some restrictions on when this is applicable:
220		1. It is not a VECTOR record.  The code generator currently does not handle
221		this case. VECTOR record uses double indirection like arrays.
222		2. All the record component values are defined in the same "basic block"
223		as the record constructor.  This is to prevent speculative
224		record construction.
225
226		----------------------------------------------------------------------
227		Name: Allen Leung
228		Date: 2002/02/22 01:02:00 EST
229		Tag: leunga-20020222-mlrisc-tools
230
231		Minor bug fixes in the parser and rewriter
232
233		----------------------------------------------------------------------
234		Name: Allen Leung
235		Date: 2002/02/21 20:20:00 EST
236		Tag: leunga-20020221-peephole
237
238		Regenerated the peephole files.  Some contained typos in the specification
239		and some didn't compile because of pretty printing bugs in the old version
240		of 'nowhere'.
241
242		----------------------------------------------------------------------
243		Name: Allen Leung
244		Date: 2002/02/19 20:20:00 EST
245		Tag: leunga-20020219-mlrisc-tools
246		Description:
247
248		Minor bug fixes to the mlrisc-tools library:
249
250		1.  Fixed up parsing colon suffixed keywords
251		2.  Added the ability to shut the error messages up
252		3.  Reimplemented the pretty printer and fixed up/improved
253		the pretty printing of handle and -> types.
254		4.  Fixed up generation of literal symbols in the nowhere tool.
255		5.  Added some SML keywords to to sml.sty
256
257		----------------------------------------------------------------------
258		Name: Matthias Blume
259		Date: 2002/02/19 16:20:00 EST
260		Tag: blume-20020219-cmffi
261		Description:
262
263		A wild mix of changes, some minor, some major:
264
265		* All C FFI-related libraries are now anchored under $c:
266		$/c.cm      --> $c/c.cm
267		$/c-int.cm  --> $c/internals/c-int.cm
268		$/memory.cm --> $c/memory/memory.cm
269
270		* "make" tool (in CM) now treats its argument pathname slightly
271		differently:
272		1. If the native expansion is an absolute name, then before invoking
273		the "make" command on it, CM will apply OS.Path.mkRelative
274		(with relativeTo = OS.FileSys.getDir()) to it.
275		2. The argument will be passed through to subsequent phases of CM
276		processing without "going native".  In particular, if the argument
277		was an anchored path, then "make" will not lose track of that anchor.
278
279		* Compiler backends now "know" their respective C calling conventions
280		instead of having to be told about it by ml-nlffigen.  This relieves
281		ml-nlffigen from one of its burdens.
282
283		* The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
284
285		* Export C_DEBUG and C_Debug from $c/c.cm.
286
287		* C type encoding in ml-nlffi-lib has been improved to model the conceptual
288		subtyping relationship between incomplete pointers and their complete
289		counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
290		with the convention of instantiating 'o with ('t, 'c) obj whenever
291		the pointer target type is complete.  In the incomplete case, 'o
292		will be instantiated with some "'c iobj" -- a type obtained by
293		using one of the functors PointerToIncompleteType or PointerToCompleteType.
294
295		Operations that work on both incomplete and complete pointer types are
296		typed as taking an 'o ptr while operations that require the target to
297		be known are typed as taking some ('t, 'c) obj ptr.
298
299		voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
300		where void is an eqtype without any values.  This makes it possible
301		to work on voidptr values using functions meant to operate on light
302		incomplete pointers.
303
304		* As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
305		been vastly simplified.
306
307		----------------------------------------------------------------------
308		Name: Matthias Blume
309		Date: 2002/02/19 10:48:00 EST
310		Tag: blume-20020219-pqfix
311		Description:
312
313		Applied Chris Okasaki's bug fix for priority queues.
314
315		----------------------------------------------------------------------
316		Name: Matthias Blume
317		Date: 2002/02/15 17:05:00
318		Tag: Release_110_39
319		Description:
320
321		Last-minute retagging is becoming a tradition... :-(
322
323		This is the working release 110.39.
324
325		----------------------------------------------------------------------
326		Name: Matthias Blume
327		Date: 2002/02/15 16:00:00 EST
328		Tag: Release_110_39-orig
329		Description:
330
331		Working release 110.39.  New bootfiles.
332
333		(Update: There was a small bug in the installer so it wouldn't work
334		with all shells.  So I retagged. -Matthias)
335
336		----------------------------------------------------------------------
337		Name: Matthias Blume
338		Date: 2002/02/15 14:17:00 EST
339		Tag: blume-20020215-showbindings
340		Description:
341
342		Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
343		the latter can be useful for exploring what bindings are available at
344		the interactive prompt.  (The first function returns only the list
345		of symbols that are really bound, the second prints those but also the
346		ones that CM's autoloading mechanism knows about.)
347
348		----------------------------------------------------------------------
349		Name: Matthias Blume
350		Date: 2002/02/15 12:08:00 EST
351		Tag: blume-20020215-iptrs
352		Description:
353
354		Two improvements to ml-nlffigen:
355
356		1. Write files only if they do not exist or if their current contents
357		do not coincide with what's being written.  (That is, avoid messing
358		with the time stamps unless absolutely necessary.)
359
360		2. Implement a "repository" mechanism for generated files related
361		to "incomplete pointer types".   See the README file for details.
362
363		----------------------------------------------------------------------
364		Name: Matthias Blume
365		Date: 2002/02/14 11:50:00 EST
366		Tag: blume-20020214-quote
367		Description:
368
369		Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
370		because of the new and improved tag generation scheme.  (Thanks to Allen
371		Leung for pointing it out.)
372
373		----------------------------------------------------------------------
374		Name: Lal George
375		Date: 2002/02/14 09:55:27 EST 2002
376		Tag: george-20020214-isabelle-bug
377		Description:
378
379		Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
380		of Isabelle on the x86.
381
382		From Allen:
383		-----------
384		I've found the problem:
385
386		in ra-core.sml, I use the counter "blocked" to keep track of the
387		true number of elements in the freeze queue.  When the counter goes
388		to zero, I skip examining the queue.  But I've messed up the
389		bookkeeping in combine():
390
391		else ();
392		case !ucol of
393		PSEUDO => (if !cntv > 0 then
394		(if !cntu > 0 then blocked := !blocked - 1 else ();
395		^^^^^^^^^^^^^^^^^^^^^^^
396		moveu := mergeMoveList(!movev, !moveu)
397		)
398		else ();
399
400		combine() is called to coalesce two nodes u and v.
401		I think I was thinking that if the move counts of u and v are both
402		greater than zero then after they are coalesced then one node is
403		removed from the freeze queue.  Apparently I was thinking that
404		both u and v are of low degree, but that's clearly not necessarily true.
405
406
407		02/12/2002:
408		Here's the patch.  HOL now compiles.
409
410		I don't know how this impact on performance (compile
411		time or runtime).  This bug caused the RA (especially on the x86)
412		to go thru the potential spill phase when there are still nodes on the
413		freeze queue.
414
415
416
417
418		----------------------------------------------------------------------
419		Name: Matthias Blume
420		Date: 2002/02/13 22:40:00 EST
421		Tag: blume-20020213-fptr-rtti
422		Description:
423
424		Fixed a bug in ml-nlffigen that was introduced with one of the previous
425		updates.
426
427		----------------------------------------------------------------------
428		Name: Matthias Blume
429		Date: 2002/02/13 16:41:00 EST
430		Tag: blume-20020213-cmlpq
431		Description:
432
433		Added new priority queue export symbols (which have just been added to
434		smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
435		not compile and the installer would choke.)
436
437		----------------------------------------------------------------------
438		Name: Matthias Blume
439		Date: 2002/02/13 16:15:00 EST
440		Tag: blume-20020213-various
441		Description:
442
443		1. More tweaks to ml-nlffigen:
444
445		- better internal datastructures (resulting in slight speedup)
446		- "-match" option requires exact match
447		- "localized" gensym counters (untagged structs/unions nested within
448		other structs/unions or within typedefs get a fresh counter; their
449		tag will be prefixed by a concatenation of their parents' tags)
450		- bug fixes (related to calculation of transitive closure of types
451		to be included in the output)
452
453		2. Minor Basis updates:
454
455		- added implementations for List.collate and Option.app
456
457		----------------------------------------------------------------------
458		Name: Matthias Blume
459		Date: 2002/02/11 15:55:00 EST
460		Tag: blume-20020211-gensym
461		Description:
462
463		Added a "-gensym" option to command line of ml-nlffigen.  This can be
464		used to specify a "stem" -- a string that is inserted in all "gensym'd"
465		names (ML structure names that correspond to unnamed C structs, unions,
466		and enums), so that separate runs of ml-nlffigen do not clash.
467
468		----------------------------------------------------------------------
469		Name: Matthias Blume
470		Date: 2002/02/11 12:05:00 EST
471		Tag: blume-20020211-gensml
472		Description:
473
474		A quick fix for a problem with GenSML (in the pgraph-util library):
475		Make generation of toplevel "local" optional.  (Strictly speaking,
476		signature definitions within "local" are not legal SML.)
477
478		Other than that: updates to INSTALL and cm/TODO.
479
480		----------------------------------------------------------------------
481		Name: Matthias Blume
482		Date: 2002/02/08 15:00:00 EST
483		Tag: blume-20020208-uniquepid
484		Description:
485
486		0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
487
488		1. The installer (config/install.sh) has gotten smarter:
489
490		- Configuration options are a bit easier to specify now
491		(in config/targets).
492		- Bug in recognizing .tar.bz2 files fixed.
493		- Installer automatically resolves dependencies between
494		configuration options (e.g., if you ask for eXene, you will
495		also get cml -- regardless whether you asked for it or not).
496		- Installer can run in "quieter mode" by setting the environment
497		variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
498		"completely silent", though.
499		- Build HashCons library as part of smlnj-lib.
500
501		2. A new scheme for assigning persistent identifiers to compilation
502		units (and, by extension, to types etc.) has been put into place.
503		This fixes a long-standing bug where types and even dynamic values
504		can get internally confused, thereby compromising type safety
505		(abstraction) and dynamic correctness.  See
506
507		http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
508
509		for an example of how things could go wrong until now.
510
511		The downside of the new scheme is that pids are not quite as
512		persistent as they used to be: CM will generate a fresh pid
513		for every compilation unit that it thinks it sees for the first
514		time.  That means that if you compile starting from a clean, fresh
515		source tree at two different times, you end up with different
516		binaries.
517
518		Cutoff recompilation, however, has not been compromised because
519		CM keeps pid information in special caches between runs.
520
521		----------------------------------------------------------------------
522		Name: Lal George
523		Date: 2002/02/07 15:34:13 EST 2002
524		Tag: <none>
525		Description:
526
527		Compilers that generate assembly code may produce  global labels
528		whose value is resolved at link time. The various peephole optimization
529		modules did not take this in account.
530
531		TODO. The Labels.addrOf function should really return an option
532		type so that clients are forced to deal with this issue, rather
533		than an exception being raised.
534
535		----------------------------------------------------------------------
536		Name: Lal George
537		Date: 2002/02/06 13:55:02 EST
538		Tag: george-20020206-ra-breakup
539		Description:
540
541		1. A bug fix from Allen.
542
543		A typo causes extra fstp %st(0)'s to be generated at compensation
544		edges, which might cause stack underflow traps at runtime.  This
545		occurs in fft where there are extraneous fstps right before the 'into'
546		trap instruction (in this case they are harmless since none of the
547		integers overflow.)
548
549		2. Pulled out various utility modules that were embedded in the modules
550		of the register allocator. I need these modules for other purposes, but
551		they are not complete enough to put into a library (just yet).
552		----------------------------------------------------------------------
553		Name: Matthias Blume
554		Date: 2002/01/31 16:05:00 EST
555		Tag: blume-20020131-sparc-ccalls
556		Description:
557
558		1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
559		of extra stack space by mistake.  Fixed.
560
561		2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
562
563		----------------------------------------------------------------------
564		Name: Allen Leung
565		Date: 2002/01/30
566		Tag: leunga-20020130-nowhere-bug-fix
567		Description:
568
569		MLRISC bug fixes:
570		1. Fixed a bindings computation bug in the 'nowhere' program generator tool.
571		2. MachineInt.fromString was negating its value.
572
573		----------------------------------------------------------------------
574		Name: Matthias Blume
575		Date: 2002/01/29
576		Tag: blume-20020129-INSTALL
577		Description:
578
579		- Added somewhat detailed installation instructions (file INSTALL).
580		- Fixed curl-detection bug in config/install.sh.
581		- It is now possible to select the URL getter using the URLGETTER
582		environment variable:
583
584		not set / "unknown"      --> automatic detection (script tries wget,
585		curl, and lynx)
586		"wget" / "curl" / "lynx" --> use the specified program (script "knows"
587		how to properly invoke them)
588		other                    --> use $URLGETTER directly, it must take
589		precisely two command-line arguments
590		(source URL and destination file name)
591
592		----------------------------------------------------------------------
593		Name: Matthias Blume
594		Date: 2002/01/28
595		Tag: blume-20020128-sparc-ccalls
596		Description:
597
598		- Fixed problem with calculation of "used" registers in sparc-c-calls.
599		- Make use of the allocParam argument in sparc-c-calls.
600
601		----------------------------------------------------------------------
602		Name: Matthias Blume
603		Date: 2002/01/28
604		Tag: blume-20020128-allocParam
605		Description:
606
607		John Reppy:  Changes c-calls API to accept client-callback for
608		allocating extra stack space.
609		me: Corresponding changes to mlriscGen (using a dummy argument that
610		does not change the current behavior).
611
612		----------------------------------------------------------------------
613		Name: Matthias Blume
614		Date: 2002/01/28 12:00:00
615		Tag: Release_110_38
616		Description:
617
618		This time for real!!!
619
620		----------------------------------------------------------------------
621		Name: Matthias Blume
622		Date: 2002/01/28 10:56:00 EST
623		Tag: blume-20020128-retraction
624		Description:
625
626		0. Retracted earlier 110.38.  (The Release_110_38 tag has been replaced
627		with blume-Release_110_38-retracted.)
628
629		1. Fixed a problem with incorrect rounding modes in real64.sml.
630		(Thanks to Andrew Mccreight <andrew.mccreight@yale.edu>.)
631
632		2. A bug in ml-nlffigen related to the handling of unnamed structs, unions,
633		and enums fixed.  The naming of corresponding ML identifiers should
634		now be consistent again.
635
636		----------------------------------------------------------------------
637		Name: Allen Leung
638		Date: 2002/01/27
639		Tag: leunga-20020127-nowhere
640		Description:
641
642		Added a target called nowhere in the configuration scripts.
643		Enabling this will build the MLRISC 'nowhere' tool (for translating
644		programs with where-clauses into legal SML code) during installation.
645
646		----------------------------------------------------------------------
647		Name: Matthias Blume
648		Date: 2002/01/25 21:27:00 EST
649		Tag: blume-Release_110_38-retracted
650		Description:
651
652		Call it a (working) release!  Version is 110.38. Bootfiles are ready.
653
654		README will be added later.
655
656		!!! NOTE:  Re-tagged as blume-Release_110_38-retracted. Original tag
657		(Release_110_38) removed.  Reason: Last-minute bug fixes.
658
659		----------------------------------------------------------------------
660		Name: Matthias Blume
661		Date: 2002/01/25
662		Tag: blume-20020125-ffi
663		Description:
664
665		A large number of tweaks and improvements to ml-nlffi-lib and
666		ml-nlffigen:
667
668		- ML represenation types have been streamlined
669		- getter and setter functions work with concrete values, not abstract
670		ones where possible
671		- ml-nlffigen command line more flexible (see README file there)
672		- some bugs have been fixed (hopefully)
673
674		----------------------------------------------------------------------
675		Name: Lal George
676		Date: 2002/01/24
677		Tag: george-20020124-risc-ra-interface
678		Description:
679
680		There is a dramatic simplification in the interface to the
681		register allocator for RISC architectures as a result of making
682		parallel copy instructions explicit.
683
684		----------------------------------------------------------------------
685		Name: Matthias Blume
686		Date: 2002/01/22
687		Tag: blume-20020122-x86-ccalls
688		Description:
689
690		Bug fix for c-calls on x86 (having to do with how char- and
691		short-arguments are being handled).
692
693		----------------------------------------------------------------------
694		Name: Matthias Blume
695		Date: 2002/01/21
696		Tag: blume-20020121-ff
697		Description:
698
699		Another day of fiddling with the FFI...
700
701		1. Bug fix/workaround:  CKIT does not complain about negative array
702		dimensions, so ml-nlffigen has to guard itself against this possibility.
703		(Otherwise a negative dimension would send it into an infinite loop.)
704
705		2. Some of the abstract types (light objects, light pointers, most "base"
706		types) in structure C are now eqtypes.
707
708		3. Added constructors and test functions for NULL function pointers.
709
710		----------------------------------------------------------------------
711		Name: Matthias Blume
712		Date: 2002/01/18
713		Tag: blume-20020118-ready-for-new-release
714		Description:
715
716		Made config/srcarchiveurl point to a new place.  (Will provide boot
717		files shortly.)
718
719		Maybe we christen this to be 110.38?
720
721		----------------------------------------------------------------------
722		Name: Matthias Blume
723		Date: 2002/01/18
724		Tag: blume-20020118-more-ffifiddle
725		Description:
726
727		Today's FFI fiddling:
728
729		- Provided a structure CGetSet with "convenient" versions of C.Get.* and
730		C.Set.* that use concrete (MLRep.*) arguments and results instead
731		of abstract ones.
732
733		- Provided word-style bit operations etc. for "int" representation
734		types in MLRep.S<Foo>Bitops where <Foo> ranges over Char, Int, Short,
735		and Long.
736
737		----------------------------------------------------------------------
738		Name: Matthias Blume
739		Date: 2002/01/18
740		Tag: blume-20020118-use-x86-fp
741		Description:
742
743		Now that x86-fast-fp seems to be working, I turned it back on again
744		by default.  (Seems to work fine now, even with the FFI.)
745
746		Other than that, I added some documentation about the FFI to
747		src/ml-nlffigen/README and updated the FFI test examples in
748		src/ml-nlffi-lib/Tests/*.
749
750		----------------------------------------------------------------------
751		Name: Allen Leung
752		Date: 2002/01/17
753		Tag: leunga-20020117-x86-fast-fp-call
754		Description:
755
756		1. Fixed a problem with handling return fp values when x86's fast fp
757		mode is turned on.
758
759		2. Minor pretty printing fix for cellset.  Print %st(0) as %st(0) instead
760		of %f32.
761
762		3. Added a constructor INT32lit to the ast of MLRISC tools.
763
764		----------------------------------------------------------------------
765		Name: Matthias Blume
766		Date: 2002/01/16
767		Tag: blume-20020116-ffifiddle
768		Description:
769
770		More fiddling with the FFI interface:
771
772		- Make constness 'c instead of rw wherever possible.  This eliminates
773		the need for certain explicit coercions.  (However, due to ML's
774		value polymorphism, there will still be many cases where explicit
775		coercions are necessary.  Phantom types are not the whole answer
776		to modeling a subtyping relationship in ML.)
777
778		- ro/rw coersions for pointers added.  (Avoids the detour through */&.)
779
780		- "printf" test example added to src/ml-nlffi-lib/Tests.  (Demonstrates
781		clumsy workaround for varargs problem.)
782
783		----------------------------------------------------------------------
784		Name: Lal George
785		Date: 2002/01/15
786		Tag: <none>
787		Description:
788
789		1. Since COPY instructions are no longer native to the architecture,
790		a generic functor can be used to implement the expandCopies function.
791
792		2. Allowed EXPORT and IMPORT pseudo-op declarations to appear inside a
793		TEXT segment.
794
795		----------------------------------------------------------------------
796		Name: Matthias Blume
797		Date: 2002/01/15
798		Tag: blume-20020115-ffiupdates
799		Description:
800
801		1. Fix for bug resulting in single-precision float values being returned
802		incorrectly from FFI calls.
803
804		2. Small modifications to C FFI API:
805
806		- memory-allocation routines return straight objects (no options)
807		and raise an exception in out-of-memory situations
808		- unsafe extensions to cast between function pointers and pointers
809		from/to ints
810		- added structure C_Debug as an alternative to structure C where
811		pointer-dereferencing (|*| and |*!) always check for null-pointers
812		- added open_lib' to DynLinkage;  open_lib' works like open_lib
813		but also takes a (possibly empty) list of existing library handles
814		that the current library depends on
815
816		----------------------------------------------------------------------
817		Name: Matthias Blume
818		Date: 2002/01/10
819		Tag: blume-20020110-newffigen
820		Description:
821
822		1. Updates to portable graph code.
823
824		2. Major update to ml-nlffigen and ml-nlffi-lib.  Things are much
825		more scalable now so that even huge interfaces such as the one
826		for GTK compile in finite time and space. :-)
827		See src/ml-nlffigen/README for details on what's new.
828
829		----------------------------------------------------------------------
830		Name: Lal George
831		Date: 2001/01/09 14:31:35 EST 2002
832		Tag: george-20011206-rm-native-copy
833		Description:
834
835		Removed the native COPY and FCOPY instructions
836		from all the architectures and replaced it with the
837		explicit COPY instruction from the previous commit.
838
839		It is now possible to simplify many of the optimizations
840		modules that manipulate copies. This has not been
841		done in this change.
842
843		----------------------------------------------------------------------
844		Name: Lal George
845		Date: 2001/12/06 16:50:13 EST 2001
846		Tag: george-20011206-mlrisc-instruction
847		Description:
848
849		Changed the representation of instructions from being fully abstract
850		to being partially concrete. That is to say:
851
852		from
853		type instruction
854
855		to
856		type instr                              (* machine instruction *)
857
858		datatype instruction =
859		LIVE of {regs: C.cellset, spilled: C.cellset}
860		| KILL of {regs: C.cellset, spilled: C.cellset}
861		| COPYXXX of {k: CB.cellkind, dst: CB.cell list, src: CB.cell list}
862		| ANNOTATION of {i: instruction, a: Annotations.annotation}
863		| INSTR of instr
864
865		This makes the handling of certain special instructions that appear on
866		all architectures easier and uniform.
867
868		LIVE and KILL say that a list of registers are live or killed at the
869		program point where they appear. No spill code is generated when an
870		element of the 'regs' field is spilled, but the register is moved to
871		the 'spilled' (which is present, more for debugging than anything else).
872
873		LIVE replaces the (now deprecated) DEFFREG instruction on the alpha.
874		We used to generate:
875
876		DEFFREG f1
877		f1 := f2 + f3
878		trapb
879
880		but now generate:
881
882		f1 := f2 + f3
883		trapb
884		LIVE {regs=[f1,f2,f3], spilled=[]}
885
886		Furthermore, the DEFFREG (hack) required that all floating point instruction
887		use all registers mentioned in the instruction. Therefore f1 := f2 + f3,
888		defines f1 and uses [f1,f2,f3]! This hack is no longer required resulting
889		in a cleaner alpha implementation. (Hopefully, intel will not get rid of
890		this architecture).
891
892		COPYXXX is intended to replace the parallel COPY and FCOPY  available on
893		all the architectures. This will result in further simplification of the
894		register allocator that must be aware of them for coalescing purposes, and
895		will also simplify certain aspects of the machine description that provides
896		callbacks related to parallel copies.
897
898		ANNOTATION should be obvious, and now INSTR represents the honest to God
899		machine instruction set!
900
901		The <arch>/instructions/<arch>Instr.sml files define certain utility
902		functions for making porting easier -- essentially converting upper case
903		to lower case. All machine instructions (of type instr) are in upper case,
904		and the lower case form generates an MLRISC instruction. For example on
905		the alpha we have:
906
907		datatype instr =
908		LDA of {r:cell, b:cell, d:operand}
909		| ...
910
911		val lda : {r:cell, b:cell, d:operand} -> instruction
912		...
913
914		where lda is just (INSTR o LDA), etc.
915
916		----------------------------------------------------------------------
917		Name: Matthias Blume
918		Date: 2001/11/22 21:40:00 EST
919		Tag: Release_110_37
920		Description:
921
922		Release 110.37.  This time for real.
923
924		----------------------------------------------------------------------
925		Name: Matthias Blume
926		Date: 2001/11/21 16:35:00 EST
927		Tag: blume-20011121-foot-in-mouth
928		Description:
929
930		Removed the "Release_110_37" tag because of a serious bug.
931		This will be re-tagged once the bug is fixed.
932
933		----------------------------------------------------------------------
934		Name: Matthias Blume
935		Date: 2001/11/21 16:14:00 EST
936		Tag: blume-20011121-forgottenfile
937		Description:
938
939		Forgot to add a file.  (Just a .tex-file -- part of
940		the CM manual source.)
941
942		----------------------------------------------------------------------
943		Name: Matthias Blume
944		Date: 2001/11/21 16:10:00 EST
945		Tag: blume-20011121-invalid_110_37
946		Description:
947
948		Note: I removed the original tag "Release_110_37" from this commit
949		because we found a serious bug in all non-x86 backends.
950		- Matthias
951
952		1. Modifications to the SML/NJ code generator and to the runtime system
953		so that code object name strings are directly inserted into code
954		objects at code generation time.  The only business the runtime system
955		has with this is now to read the name strings on occasions.
956		(The encoding of the name string has also changed somewhat.)
957
958		2. CM now implements a simple "set calculus" for specifying export lists.
959		In particular, it is now possible to refer to the export lists of
960		other libraries/groups/sources and form unions as well as differences.
961		See the latest CM manual for details.
962
963		3. An separate notion of "proxy" libraries has again be eliminated from
964		CM's model.  (Proxy libraries are now simply a special case of using
965		the export list calculus.)
966
967		4. Some of the existing libraries now take advantage of the new set
968		calculus.
969		(Notice that not all libraries have been converted because some
970		of the existing .cm-files are supposed to be backward compatible
971		with 110.0.x.)
972
973		5. Some cleanup in stand-alone programs.  (Don't use "exnMessage" -- use
974		"General.exnMessage"!  The former relies on a certain hook to be
975		initialized, and that often does not happen in the stand-alone case.)
976
977		----------------------------------------------------------------------
978		Name: Lal George
979		Date: 2001/11/21  13:56:18 EST
980		Tag: george-2001121-pseudo-ops
981		Description:
982
983		Implemented a complete redesign of MLRISC pseudo-ops. Now there
984		ought to never be any question of incompatabilities with
985		pseudo-op syntax expected by host assemblers.
986
987		For now, only modules supporting GAS syntax are implemented
988		but more should follow, such as MASM, and vendor assembler
989		syntax, e.g. IBM as, Sun as, etc.
990
991		----------------------------------------------------------------------
992		Name: Matthias Blume
993		Date: 2001/11/14 11:52:00 EST
994		Tag: blume-20011114-srcname
995		Description:
996
997		1. Routed the name of the current source file to mlriscgen where it
998		should be directly emitted into the code object.  (This last part
999		is yet to be done.)
1000
1001		2. Some cleanup of the pgraph code to make it match the proposal that
1002		I put out the other day.  (The proposal notwithstanding, things are
1003		still in flux here.)
1004
1005		----------------------------------------------------------------------
1006		Name: Lal George
1007		Date: 2001/11/14 09:44:04 EST
1008		Tag:
1009		Description:
1010
1011		Fix for a backpatching bug reported by Allen.
1012
1013		Because the boundary between short and long span-dependent
1014		instructions is +/- 128, there are an astounding number of
1015		span-dependent instructions whose size is over estimated.
1016
1017		Allen came up with the idea of letting the size of span
1018		dependent instructions be non-monotonic, for a maxIter
1019		number of times, after which the size must be monotonically
1020		increasing.
1021
1022		This table shows the number of span-dependent instructions
1023		whose size was over-estimated as a function of maxIter, for the
1024		file Parse/parse/ml.grm.sml:
1025
1026		maxIter            # of instructions:
1027		10                      687
1028		20                      438
1029		30                      198
1030		40                        0
1031
1032		In compiling the compiler, there is no significant difference in
1033		compilation speed between maxIter=10 and maxIter=40. Actually,
1034		my measurements showed that maxIter=40 was a tad faster than
1035		maxIter=10! Also 96% of the  files in the compiler reach a fix
1036		point within 13 iterations, so fixing maxIter at 40, while high,
1037		is okay.
1038
1039		----------------------------------------------------------------------
1040		Name: Matthias Blume
1041		Date: 2001/10/31 15:25:00 EST
1042		Tag: blume-20011031-pgraph
1043		Description:
1044
1045		CKIT:
1046		* Changed the "Function" constructor of type Ast.ctype to carry optional
1047		argument identifiers.
1048		* Changed the return type of TypeUtil.getFunction accordingly.
1049		* Type equality ignores the argument names.
1050		* TypeUtil.composite tries to preserve argument names but gives up quickly
1051		if there is a mismatch.
1052
1053		installation script:
1054		* attempts to use "curl" if available (unless "wget" is available as well)
1055
1056		CM:
1057		* has an experimental implementation of "portable graphs" which I will
1058		soon propose as an implementation-independent library format
1059		* there are also new libraries $/pgraph.cm and $/pgraph-util.cm
1060
1061		NLFFI-LIB:
1062		* some cleanup (all cosmetic)
1063
1064		NLFFIGEN:
1065		* temporarily disabled the mechanism that suppresses ML output for
1066		C definitions whose identifiers start with an underscore character
1067		* generate val bindings for enum constants
1068		* user can request that only one style (light or heavy) is being used;
1069		default is to use both (command-line arguments: -heavy and -light)
1070		* fixed bug in handling of function types involving incomplete pointers
1071		* generate ML entry points that take record arguments (i.e., using
1072		named arguments) for C functions that have a prototype with named
1073		arguments
1074		(see changes to CKIT)
1075
1076		----------------------------------------------------------------------
1077		Name: Allen Leung
1078		Date: 2001/10/27 20:34:00 EDT
1079		Tag: leunga-20011027-x86-fast-fp-call
1080		Description:
1081
1082		Fixed the bug described in blume-20010920-slowfp.
1083
1084		The fix involves
1085		1. generating FCOPYs in FSTP in ia32-svid
1086		2. marking a CALL with the appropriate annotation
1087
1088		----------------------------------------------------------------------
1089		Name: Matthias Blume
1090		Date: 2001/10/16 11:32:00 EDT
1091		Tag: blume-20011016-netbsd
1092		Description:
1093
1094		Underscore patch from Chris Richards (fixing problem with compiling
1095		runtime system under recent NetBSD).
1096
1097		----------------------------------------------------------------------
1098		Name: Allen Leung
1099		Date: 2001/10/12 17:18:32 EDT 2001
1100		Tag: leung-20011012-x86-printflowgraph
1101		Description:
1102
1103		X86RA now uses a valid (instead of dummy) PrintFlowgraph module.
1104
1105		----------------------------------------------------------------------
1106		Name: Lal George
1107		Date: 2001/10/11 23:51:34 EDT
1108		Tag: george-20011011-too-many-instrs
1109		Description:
1110
1111		The representation of a program point never expected to see more
1112		than 65536 instructions in a basic block!
1113
1114		----------------------------------------------------------------------
1115		Name: Lal George
1116		Date: 2001/10/09 09:41:37 EDT
1117		Tag: george-20011008-mlrisc-labels
1118		Description:
1119
1120		Changed the machine description files to support printing of
1121		local and global labels in assembly code, based on host assembler
1122		conventions.
1123
1124		----------------------------------------------------------------------
1125		Name: Matthias Blume
1126		Date: 2001/09/25 15:25:00 EDT
1127		Tag: blume-20010925-exninfo
1128		Description:
1129
1130		I provided a non-hook implementation of exnName (at the toplevel) and
1131		made the "dummy" implementation of exnMessage (at the toplevel) more
1132		useful: if nothing gets "hooked in", then at least you are going to
1133		see the exception name and a message indicating why you don't see more.
1134
1135		[For the time being, programs that need exnMessage and want to use
1136		ml-build should either use General.exnMessage (strongly recommended) or
1137		refer to structure General at some other point so that CM sees a
1138		static dependency.]
1139
1140		[Similar remarks go for "print" and "use":  If you want to use their
1141		functionality in stand-alone programs generated by ml-build, then use
1142		TextIO.output and Backend.Interact.useFile (from $smlnj/compiler.cm).]
1143
1144		----------------------------------------------------------------------
1145		Name: Matthias Blume
1146		Date: 2001/09/20 17:28:00 EDT
1147		Tag: blume-20010920-slowfp
1148		Description:
1149
1150		Allen says that x86-fast-fp is not safe yet, so I turned it off again...
1151
1152		----------------------------------------------------------------------
1153		Name: Matthias Blume
1154		Date: 2001/09/20 17:20:00 EDT
1155		Tag: blume-20010920-canonicalpaths
1156		Description:
1157
1158		0. Updated the BOOT file (something that I forgot to do earlier).
1159
1160		1. Small internal change to CM so that it avoids "/../" in filenames
1161		as much as possible (but only where it is safe).
1162
1163		2. Changed config/_run-sml (resulting in a changed bin/.run-sml) so
1164		that arguments that contain delimiters are passed through correctly.
1165		This change also means that all "special" arguments of the form
1166		@SMLxxx... must come first.
1167
1168		3. Changed install script to put relative anchor names for tool commands
1169		into pathconfig.
1170
1171		----------------------------------------------------------------------
1172		Name: Matthias Blume
1173		Date: 2001/09/18 15:35:00 EDT
1174		Tag: blume-20010918-readme11036
1175		Description:
1176
1177		Added README files.
1178
1179		----------------------------------------------------------------------
1180		Name: Matthias Blume
1181		Date: 2001/09/18 11:45:00 EDT
1182		Tag: Release_110_36 (retag)
1183		Description:
1184
1185		Fixed mistake in config/preloads. Retagged as 110.36.
1186
1187		----------------------------------------------------------------------
1188		Name: Matthias Blume
1189		Date: 2001/09/18 09:40:00 EDT
1190		Tag: Release_110_36_orig (tag changed)
1191		Description:
1192
1193		New version (110.36).  New bootfiles.
1194
1195		----------------------------------------------------------------------
1196		Name: Matthias Blume
1197		Date: 2001/09/14 16:15:00 EDT
1198		Tag: blume-20010914-x86fastfp
1199		Description:
1200
1201		John committed some changes that Allen made, in particular a (hopefully)
1202		correctly working version of the x86-fp module.
1203
1204		I changed the default setting of the Control.MLRISC.getFlag "x86-fast-fp"
1205		flag to "true".  Everything seems to compile to a fixpoint ok, and
1206		"mandelbrot" speeds up by about 15%.
1207
1208		----------------------------------------------------------------------
1209		Name: Matthias Blume
1210		Date: 2001/09/13 11:20:00 EDT
1211		Tag: blume-20010913-minimal
1212		Description:
1213
1214		1. Stefan Monnier's patch to fix a miscompilation problem that
1215		was brought to light by John Reppy's work on Moby.
1216
1217		2. Implemented a minimal "structure Compiler" that contains just
1218		"version" and "architecture".  The minimal version will be
1219		available when the full version is not.  This is for backward-
1220		compatibility with code that wants to test Compiler.version.
1221
1222		----------------------------------------------------------------------
1223		Name: Matthias Blume
1224		Date: 2001/08/28 14:03:00 EDT
1225		Tag: blume-20010828-ml-lex
1226		Description:
1227
1228		Fix for bug 1581, received from Neophytos Michael.
1229
1230		----------------------------------------------------------------------
1231		Name: Matthias Blume
1232		Date: 2001/08/27 11:20:00 EDT
1233		Tag: blume-20010827-readme11035
1234		Description:
1235
1236		Fleshed out the README file for 110.35.
1237
1238		----------------------------------------------------------------------
1239		Name: Matthias Blume
1240	Date: 2001/08/24 17:10:00 EDT	Date: 2001/08/24 17:10:00 EDT
1241	Tag: Release_110_35	Tag: Release_110_35
1242	Description:	Description:
#	Line 2923	Line 4146
4146	elaborator). There were a lot of changes during my "linkpath" trials	elaborator). There were a lot of changes during my "linkpath" trials
4147	that could have been reverted to their original state but weren't.	that could have been reverted to their original state but weren't.
4148	Please, don't be too harsh on me for messing with this code a bit more	Please, don't be too harsh on me for messing with this code a bit more
4149	than what was strictly necessary...  (I _did_ resist the tempation	than what was strictly necessary...  (I _did_ resist the temptation
4150	of doing any "global reformatting" to avoid an untimely death at	of doing any "global reformatting" to avoid an untimely death at
4151	Dave's hands. :)	Dave's hands. :)
4152

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.916   changed lines   Added in v.1098

---

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