Diff of /sml/trunk/HISTORY

Parent Directory | Revision Log | Patch

	revision 987, Wed Nov 21 21:14:44 2001 UTC	revision 1131, Mon Mar 11 15:20:52 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: Allen Leung
17		Date: 2002/03/11 10:30:00 EST
18		Tag: leunga-20020310-runtime-string0
19		Description:
20
21		The representation of the empty string now points to a
22		legal null terminated C string instead of unit.  It is now possible
23		to convert an ML string into C string with InlineT.CharVector.getData.
24		This compiles into one single machine instruction.
25
26		----------------------------------------------------------------------
27		Name: Allen Leung
28		Date: 2002/03/10 23:55:00 EST
29		Tag: leunga-20020310-x86-call
30		Description:
31
32		Added machine generation for CALL instruction (relative displacement mode)
33
34		----------------------------------------------------------------------
35		Name: Matthias Blume
36		Date: 2002/03/08 16:05:00
37		Tag: blume-20020308-entrypoints
38		Description:
39
40		Version number bumped to 110.39.1.  NEW BOOTFILES!
41
42		Entrypoints: non-zero offset into a code object where execution should begin.
43
44		- Added the notion of an entrypoint to CodeObj.
45		- Added reading/writing of entrypoint info to Binfile.
46		- Made runtime system bootloader aware of entrypoints.
47		- Use the address of the label of the first function given to mlriscGen
48		as the entrypoint.  This address is currently always 0, but it will
49		not be 0 once we turn on block placement.
50		- Removed the linkage cluster code (which was The Other Way(tm) of dealing
51		with entry points) from mlriscGen.
52
53		----------------------------------------------------------------------
54		Name: Allen Leung
55		Date: 2002/03/07 20:45:00 EST
56		Tag: leunga-20020307-x86-cmov
57		Description:
58
59		Bug fixes for CMOVcc on x86.
60
61		1. Added machine code generation for CMOVcc
62		2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
63		3. CMOVcc cannot have an immediate operand as argument.
64
65		----------------------------------------------------------------------
66		Name: Matthias Blume
67		Date: 2002/03/07 16:15:00 EST
68		Tag: blume-20020307-controls
69		Description:
70
71		This is a very large but mostly boring patch which makes (almost)
72		every tuneable compiler knob (i.e., pretty much everything under
73		Control.* plus a few other things) configurable via both the command
74		line and environment variables in the style CM did its configuration
75		until now.
76
77		Try starting sml with '-h' (or, if you are brave, '-H')
78
79		To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
80		implements the underlying generic mechanism.
81
82		The interface to some of the existing such facilities has changed somewhat.
83		For example, the MLRiscControl module now provides mkFoo instead of getFoo.
84		(The getFoo interface is still there for backward-compatibility, but its
85		use is deprecated.)
86
87		The ml-build script passes -Cxxx=yyy command-line arguments through so
88		that one can now twiddle the compiler settings when using this "batch"
89		compiler.
90
91		TODO items:
92
93		We should go through and throw out all controls that are no longer
94		connected to anything.  Moreover, we should go through and provide
95		meaningful (and correct!) documentation strings for those controls
96		that still are connected.
97
98		Currently, multiple calls to Controls.new are accepted (only the first
99		has any effect).  Eventually we should make sure that every control
100		is being made (via Controls.new) exactly once.  Future access can then
101		be done using Controls.acc.
102
103		Finally, it would probably be a good idea to use the getter-setter
104		interface to controls rather than ref cells.  For the time being, both
105		styles are provided by the Controls module, but getter-setter pairs are
106		better if thread-safety is of any concern because they can be wrapped.
107
108		*****************************************
109
110		One bug fix: The function blockPlacement in three of the MLRISC
111		backpatch files used to be hard-wired to one of two possibilities at
112		link time (according to the value of the placementFlag).  But (I
113		think) it should rather sense the flag every time.
114
115		*****************************************
116
117		Other assorted changes (by other people who did not supply a HISTORY entry):
118
119		1. the cross-module inliner now works much better (Monnier)
120		2. representation of weights, frequencies, and probabilities in MLRISC
121		changed in preparation of using those for weighted block placement
122		(Reppy, George)
123
124		----------------------------------------------------------------------
125		Name: Lal George
126		Date: 2002/03/07 14:44:24 EST 2002
127		Tag: george-20020307-weighted-block-placement
128
129		Tested the weighted block placement optimization on all architectures
130		(except the hppa) using AMPL to generate the block and edge frequencies.
131		Changes were required in the machine properties to correctly
132		categorize trap instructions. There is an MLRISC flag
133		"weighted-block-placement" that can be used to enable weighted block
134		placement, but this will be ineffective without block/edge
135		frequencies (coming soon).
136
137
138		----------------------------------------------------------------------
139		Name: Lal George
140		Date: 2002/03/05 17:24:48 EST
141		Tag: george-20020305-linkage-cluster
142
143		In order to support the block placement optimization, a new cluster
144		is generated as the very first cluster (called the linkage cluster).
145		It contains a single jump to the 'real' entry point for the compilation
146		unit. Block placement has no effect on the linkage cluster itself, but
147		all the other clusters  have full freedom in the manner in which they
148		reorder blocks or functions.
149
150		On the x86 the typical linkage code that is generated is:
151		----------------------
152		.align 2
153		L0:
154		addl    $L1-L0, 72(%esp)
155		jmp     L1
156
157
158		.align  2
159		L1:
160		----------------------
161
162		72(%esp) is the memory location for the stdlink register. This
163		must contain the address of the CPS function being called. In the
164		above example, it contains the address of  L0; before
165		calling L1 (the real entry point for the compilation unit), it
166		must contain the address for L1, and hence
167
168		addl $L1-L0, 72(%esp)
169
170		I have tested this on all architectures except the hppa.The increase
171		in code size is of course negligible
172
173		----------------------------------------------------------------------
174		Name: Allen Leung
175		Date: 2002/03/03 13:20:00 EST
176		Tag: leunga-20020303-mlrisc-tools
177
178		Added #[ ... ] expressions to mlrisc tools
179
180		----------------------------------------------------------------------
181		Name: Matthias Blume
182		Date: 2002/02/27 12:29:00 EST
183		Tag: blume-20020227-cdebug
184		Description:
185
186		- made types in structure C and C_Debug to be equal
187		- got rid of code duplication (c-int.sml vs. c-int-debug.sml)
188		- there no longer is a C_Int_Debug (C_Debug is directly derived from C)
189
190		----------------------------------------------------------------------
191		Name: Matthias Blume
192		Date: 2002/02/26 12:00:00 EST
193		Tag: blume-20020226-ffi
194		Description:
195
196		1. Fixed a minor bug in CM's "noweb" tool:
197		If numbering is turned off, then truly don't number (i.e., do not
198		supply the -L option to noweb).  The previous behavior was to supply
199		-L'' -- which caused noweb to use the "default" line numbering scheme.
200		Thanks to Chris Richards for pointing this out (and supplying the fix).
201
202		2. Once again, I reworked some aspects of the FFI:
203
204		A. The incomplete/complete type business:
205
206		- Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
207		gone!
208		- ML types representing an incomplete type are now *equal* to
209		ML types representing their corresponding complete types (just like
210		in C).  This is still safe because ml-nlffigen will not generate
211		RTTI for incomplete types, nor will it generate functions that
212		require access to such RTTI.   But when ML code generated from both
213		incomplete and complete versions of the C type meet, the ML types
214		are trivially interoperable.
215
216		NOTE:  These changes restore the full generality of the translation
217		(which was previously lost when I eliminated functorization)!
218
219		B. Enum types:
220
221		- Structure C now has a type constructor "enum" that is similar to
222		how the "su" constructor works.  However, "enum" is not a phantom
223		type because each "T enum" has values (and is isomorphic to
224		MLRep.Signed.int).
225		- There are generic access operations for enum objects (using
226		MLRep.Signed.int).
227		- ml-nlffigen will generate a structure E_foo for each "enum foo".
228		* The structure contains the definition of type "mlrep" (the ML-side
229		representation type of the enum).  Normally, mlrep is the same
230		as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
231		then mlrep will be defined as a datatype -- thus facilitating
232		pattern matching on mlrep values.
233		("-ec" will be suppressed if there are duplicate values in an
234		enumeration.)
235		* Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
236		will be generated for each C enum constant xxx.
237		* Conversion functions m2i and i2m convert between mlrep and
238		MLRep.Signed.int.  (Without "-ec", these functions are identities.)
239		* Coversion functions c and ml convert between mlrep and "tag enum".
240		* Access functions (get/set) fetch and store mlrep values.
241		- By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
242		enumerations are merged into one single enumeration represented by
243		structure E_'.
244
245		----------------------------------------------------------------------
246		Name: Allen Leung
247		Date: 2002/02/25 04:45:00 EST
248		Tag: leunga-20020225-cps-spill
249
250		This is a new implementation of the CPS spill phase.
251		The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
252		In case of problems, replace it with the old file spill.sml
253
254		The current compiler runs into some serious performance problems when
255		constructing a large record.  This can happen when we try to compile a
256		structure with many items.  Even a very simple structure like the following
257		makes the compiler slow down.
258
259		structure Foo = struct
260		val x_1 = 0w1 : Word32.int
261		val x_2 = 0w2 : Word32.int
262		val x_3 = 0w3 : Word32.int
263		...
264		val x_N = 0wN : Word32.int
265		end
266
267		The following table shows the compile time, from N=1000 to N=4000,
268		with the old compiler:
269
270		N
271		1000   CPS 100 spill                           0.04u  0.00s  0.00g
272		MLRISC ra                               0.06u  0.00s  0.05g
273		(spills = 0 reloads = 0)
274		TOTAL                                   0.63u  0.07s  0.21g
275
276		1100   CPS 100 spill                           8.25u  0.32s  0.64g
277		MLRISC ra                               5.68u  0.59s  3.93g
278		(spills = 0 reloads = 0)
279		TOTAL                                   14.71u  0.99s  4.81g
280
281		1500   CPS 100 spill                           58.55u  2.34s  1.74g
282		MLRISC ra                               5.54u  0.65s  3.91g
283		(spills = 543 reloads = 1082)
284		TOTAL                                   65.40u  3.13s  6.00g
285
286		2000   CPS 100 spill                           126.69u  4.84s  3.08g
287		MLRISC ra                               0.80u  0.10s  0.55g
288		(spills = 42 reloads = 84)
289		TOTAL                                   129.42u  5.10s  4.13g
290
291		3000   CPS 100 spill                           675.59u  19.03s  11.64g
292		MLRISC ra                               2.69u  0.27s  1.38g
293		(spills = 62 reloads = 124)
294		TOTAL                                   682.48u  19.61s  13.99g
295
296		4000   CPS 100 spill                           2362.82u  56.28s  43.60g
297		MLRISC ra                               4.96u  0.27s  2.72g
298		(spills = 85 reloads = 170)
299		TOTAL                                   2375.26u  57.21s  48.00g
300
301		As you can see the old cps spill module suffers from some serious
302		performance problem.  But since I cannot decipher the old code fully,
303		instead of patching the problems up, I'm reimplementing it
304		with a different algorithm.  The new code is more modular,
305		smaller when compiled, and substantially faster
306		(O(n log n) time and O(n) space).  Timing of the new spill module:
307
308		4000  CPS 100 spill                           0.02u  0.00s  0.00g
309		MLRISC ra                               0.25u  0.02s  0.15g
310		(spills=1 reloads=3)
311		TOTAL                                   7.74u  0.34s  1.62g
312
313		Implementation details:
314
315		As far as I can tell, the purpose of the CPS spill module is to make sure the
316		number of live variables at any program point (the bandwidth)
317		does not exceed a certain limit, which is determined by the
318		size of the spill area.
319
320		When the bandwidth is too large, we decrease the register pressure by
321		packing live variables into spill records.  How we achieve this is
322		completely different than what we did in the old code.
323
324		First, there is something about the MLRiscGen code generator
325		that we should be aware of:
326
327		o MLRiscGen performs code motion!
328
329		In particular, it will move floating point computations and
330		address computations involving only the heap pointer to
331		their use sites (if there is only a single use).
332		What this means is that if we have a CPS record construction
333		statement
334
335		RECORD(k,vl,w,e)
336
337		we should never count the new record address w as live if w
338		has only one use (which is often the case).
339
340		We should do something similar to floating point, but the transformation
341		there is much more complex, so I won't deal with that.
342
343		Secondly, there are now two new cps primops at our disposal:
344
345		1. rawrecord of record_kind option
346		This pure operator allocates some uninitialized storage from the heap.
347		There are two forms:
348
349		rawrecord NONE [INT n]  allocates a tagless record of length n
350		rawrecord (SOME rk) [INT n] allocates a tagged record of length n
351		and initializes the tag.
352
353		2. rawupdate of cty
354		rawupdate cty (v,i,x)
355		Assigns to x to the ith component of record v.
356		The storelist is not updated.
357
358		We use these new primops for both spilling and increment record construction.
359
360		1. Spilling.
361
362		This is implemented with a linear scan algorithm (but generalized
363		to trees).  The algorithm will create a single spill record at the
364		beginning of the cps function and use rawupdate to spill to it,
365		and SELECT or SELp to reload from it.  So both spills and reloads
366		are fine-grain operations.  In contrast, in the old algorithm
367		"spills" have to be bundled together in records.
368
369		Ideally, we should sink the spill record construction to where
370		it is needed.  We can even split the spill record into multiple ones
371		at the places where they are needed.  But CPS is not a good
372		representation for global code motion, so I'll keep it simple and
373		am not attempting this.
374
375		2. Incremental record construction (aka record splitting).
376
377		Long records with many component values which are simulatenously live
378		(recall that single use record addresses are not considered to
379		be live) are constructed with rawrecord and rawupdate.
380		We allocate space on the heap with rawrecord first, then gradually
381		fill it in with rawupdate.  This is the technique suggested to me
382		by Matthias.
383
384		Some restrictions on when this is applicable:
385		1. It is not a VECTOR record.  The code generator currently does not handle
386		this case. VECTOR record uses double indirection like arrays.
387		2. All the record component values are defined in the same "basic block"
388		as the record constructor.  This is to prevent speculative
389		record construction.
390
391		----------------------------------------------------------------------
392		Name: Allen Leung
393		Date: 2002/02/22 01:02:00 EST
394		Tag: leunga-20020222-mlrisc-tools
395
396		Minor bug fixes in the parser and rewriter
397
398		----------------------------------------------------------------------
399		Name: Allen Leung
400		Date: 2002/02/21 20:20:00 EST
401		Tag: leunga-20020221-peephole
402
403		Regenerated the peephole files.  Some contained typos in the specification
404		and some didn't compile because of pretty printing bugs in the old version
405		of 'nowhere'.
406
407		----------------------------------------------------------------------
408		Name: Allen Leung
409		Date: 2002/02/19 20:20:00 EST
410		Tag: leunga-20020219-mlrisc-tools
411		Description:
412
413		Minor bug fixes to the mlrisc-tools library:
414
415		1.  Fixed up parsing colon suffixed keywords
416		2.  Added the ability to shut the error messages up
417		3.  Reimplemented the pretty printer and fixed up/improved
418		the pretty printing of handle and -> types.
419		4.  Fixed up generation of literal symbols in the nowhere tool.
420		5.  Added some SML keywords to to sml.sty
421
422		----------------------------------------------------------------------
423		Name: Matthias Blume
424		Date: 2002/02/19 16:20:00 EST
425		Tag: blume-20020219-cmffi
426		Description:
427
428		A wild mix of changes, some minor, some major:
429
430		* All C FFI-related libraries are now anchored under $c:
431		$/c.cm      --> $c/c.cm
432		$/c-int.cm  --> $c/internals/c-int.cm
433		$/memory.cm --> $c/memory/memory.cm
434
435		* "make" tool (in CM) now treats its argument pathname slightly
436		differently:
437		1. If the native expansion is an absolute name, then before invoking
438		the "make" command on it, CM will apply OS.Path.mkRelative
439		(with relativeTo = OS.FileSys.getDir()) to it.
440		2. The argument will be passed through to subsequent phases of CM
441		processing without "going native".  In particular, if the argument
442		was an anchored path, then "make" will not lose track of that anchor.
443
444		* Compiler backends now "know" their respective C calling conventions
445		instead of having to be told about it by ml-nlffigen.  This relieves
446		ml-nlffigen from one of its burdens.
447
448		* The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
449
450		* Export C_DEBUG and C_Debug from $c/c.cm.
451
452		* C type encoding in ml-nlffi-lib has been improved to model the conceptual
453		subtyping relationship between incomplete pointers and their complete
454		counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
455		with the convention of instantiating 'o with ('t, 'c) obj whenever
456		the pointer target type is complete.  In the incomplete case, 'o
457		will be instantiated with some "'c iobj" -- a type obtained by
458		using one of the functors PointerToIncompleteType or PointerToCompleteType.
459
460		Operations that work on both incomplete and complete pointer types are
461		typed as taking an 'o ptr while operations that require the target to
462		be known are typed as taking some ('t, 'c) obj ptr.
463
464		voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
465		where void is an eqtype without any values.  This makes it possible
466		to work on voidptr values using functions meant to operate on light
467		incomplete pointers.
468
469		* As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
470		been vastly simplified.
471
472		----------------------------------------------------------------------
473		Name: Matthias Blume
474		Date: 2002/02/19 10:48:00 EST
475		Tag: blume-20020219-pqfix
476		Description:
477
478		Applied Chris Okasaki's bug fix for priority queues.
479
480		----------------------------------------------------------------------
481		Name: Matthias Blume
482		Date: 2002/02/15 17:05:00
483		Tag: Release_110_39
484		Description:
485
486		Last-minute retagging is becoming a tradition... :-(
487
488		This is the working release 110.39.
489
490		----------------------------------------------------------------------
491		Name: Matthias Blume
492		Date: 2002/02/15 16:00:00 EST
493		Tag: Release_110_39-orig
494		Description:
495
496		Working release 110.39.  New bootfiles.
497
498		(Update: There was a small bug in the installer so it wouldn't work
499		with all shells.  So I retagged. -Matthias)
500
501		----------------------------------------------------------------------
502		Name: Matthias Blume
503		Date: 2002/02/15 14:17:00 EST
504		Tag: blume-20020215-showbindings
505		Description:
506
507		Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
508		the latter can be useful for exploring what bindings are available at
509		the interactive prompt.  (The first function returns only the list
510		of symbols that are really bound, the second prints those but also the
511		ones that CM's autoloading mechanism knows about.)
512
513		----------------------------------------------------------------------
514		Name: Matthias Blume
515		Date: 2002/02/15 12:08:00 EST
516		Tag: blume-20020215-iptrs
517		Description:
518
519		Two improvements to ml-nlffigen:
520
521		1. Write files only if they do not exist or if their current contents
522		do not coincide with what's being written.  (That is, avoid messing
523		with the time stamps unless absolutely necessary.)
524
525		2. Implement a "repository" mechanism for generated files related
526		to "incomplete pointer types".   See the README file for details.
527
528		----------------------------------------------------------------------
529		Name: Matthias Blume
530		Date: 2002/02/14 11:50:00 EST
531		Tag: blume-20020214-quote
532		Description:
533
534		Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
535		because of the new and improved tag generation scheme.  (Thanks to Allen
536		Leung for pointing it out.)
537
538		----------------------------------------------------------------------
539		Name: Lal George
540		Date: 2002/02/14 09:55:27 EST 2002
541		Tag: george-20020214-isabelle-bug
542		Description:
543
544		Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
545		of Isabelle on the x86.
546
547		From Allen:
548		-----------
549		I've found the problem:
550
551		in ra-core.sml, I use the counter "blocked" to keep track of the
552		true number of elements in the freeze queue.  When the counter goes
553		to zero, I skip examining the queue.  But I've messed up the
554		bookkeeping in combine():
555
556		else ();
557		case !ucol of
558		PSEUDO => (if !cntv > 0 then
559		(if !cntu > 0 then blocked := !blocked - 1 else ();
560		^^^^^^^^^^^^^^^^^^^^^^^
561		moveu := mergeMoveList(!movev, !moveu)
562		)
563		else ();
564
565		combine() is called to coalesce two nodes u and v.
566		I think I was thinking that if the move counts of u and v are both
567		greater than zero then after they are coalesced then one node is
568		removed from the freeze queue.  Apparently I was thinking that
569		both u and v are of low degree, but that's clearly not necessarily true.
570
571
572		02/12/2002:
573		Here's the patch.  HOL now compiles.
574
575		I don't know how this impact on performance (compile
576		time or runtime).  This bug caused the RA (especially on the x86)
577		to go thru the potential spill phase when there are still nodes on the
578		freeze queue.
579
580
581
582
583		----------------------------------------------------------------------
584		Name: Matthias Blume
585		Date: 2002/02/13 22:40:00 EST
586		Tag: blume-20020213-fptr-rtti
587		Description:
588
589		Fixed a bug in ml-nlffigen that was introduced with one of the previous
590		updates.
591
592		----------------------------------------------------------------------
593		Name: Matthias Blume
594		Date: 2002/02/13 16:41:00 EST
595		Tag: blume-20020213-cmlpq
596		Description:
597
598		Added new priority queue export symbols (which have just been added to
599		smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
600		not compile and the installer would choke.)
601
602		----------------------------------------------------------------------
603		Name: Matthias Blume
604		Date: 2002/02/13 16:15:00 EST
605		Tag: blume-20020213-various
606		Description:
607
608		1. More tweaks to ml-nlffigen:
609
610		- better internal datastructures (resulting in slight speedup)
611		- "-match" option requires exact match
612		- "localized" gensym counters (untagged structs/unions nested within
613		other structs/unions or within typedefs get a fresh counter; their
614		tag will be prefixed by a concatenation of their parents' tags)
615		- bug fixes (related to calculation of transitive closure of types
616		to be included in the output)
617
618		2. Minor Basis updates:
619
620		- added implementations for List.collate and Option.app
621
622		----------------------------------------------------------------------
623		Name: Matthias Blume
624		Date: 2002/02/11 15:55:00 EST
625		Tag: blume-20020211-gensym
626		Description:
627
628		Added a "-gensym" option to command line of ml-nlffigen.  This can be
629		used to specify a "stem" -- a string that is inserted in all "gensym'd"
630		names (ML structure names that correspond to unnamed C structs, unions,
631		and enums), so that separate runs of ml-nlffigen do not clash.
632
633		----------------------------------------------------------------------
634		Name: Matthias Blume
635		Date: 2002/02/11 12:05:00 EST
636		Tag: blume-20020211-gensml
637		Description:
638
639		A quick fix for a problem with GenSML (in the pgraph-util library):
640		Make generation of toplevel "local" optional.  (Strictly speaking,
641		signature definitions within "local" are not legal SML.)
642
643		Other than that: updates to INSTALL and cm/TODO.
644
645		----------------------------------------------------------------------
646		Name: Matthias Blume
647		Date: 2002/02/08 15:00:00 EST
648		Tag: blume-20020208-uniquepid
649		Description:
650
651		0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
652
653		1. The installer (config/install.sh) has gotten smarter:
654
655		- Configuration options are a bit easier to specify now
656		(in config/targets).
657		- Bug in recognizing .tar.bz2 files fixed.
658		- Installer automatically resolves dependencies between
659		configuration options (e.g., if you ask for eXene, you will
660		also get cml -- regardless whether you asked for it or not).
661		- Installer can run in "quieter mode" by setting the environment
662		variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
663		"completely silent", though.
664		- Build HashCons library as part of smlnj-lib.
665
666		2. A new scheme for assigning persistent identifiers to compilation
667		units (and, by extension, to types etc.) has been put into place.
668		This fixes a long-standing bug where types and even dynamic values
669		can get internally confused, thereby compromising type safety
670		(abstraction) and dynamic correctness.  See
671
672		http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
673
674		for an example of how things could go wrong until now.
675
676		The downside of the new scheme is that pids are not quite as
677		persistent as they used to be: CM will generate a fresh pid
678		for every compilation unit that it thinks it sees for the first
679		time.  That means that if you compile starting from a clean, fresh
680		source tree at two different times, you end up with different
681		binaries.
682
683		Cutoff recompilation, however, has not been compromised because
684		CM keeps pid information in special caches between runs.
685
686		----------------------------------------------------------------------
687		Name: Lal George
688		Date: 2002/02/07 15:34:13 EST 2002
689		Tag: <none>
690		Description:
691
692		Compilers that generate assembly code may produce  global labels
693		whose value is resolved at link time. The various peephole optimization
694		modules did not take this in account.
695
696		TODO. The Labels.addrOf function should really return an option
697		type so that clients are forced to deal with this issue, rather
698		than an exception being raised.
699
700		----------------------------------------------------------------------
701		Name: Lal George
702		Date: 2002/02/06 13:55:02 EST
703		Tag: george-20020206-ra-breakup
704		Description:
705
706		1. A bug fix from Allen.
707
708		A typo causes extra fstp %st(0)'s to be generated at compensation
709		edges, which might cause stack underflow traps at runtime.  This
710		occurs in fft where there are extraneous fstps right before the 'into'
711		trap instruction (in this case they are harmless since none of the
712		integers overflow.)
713
714		2. Pulled out various utility modules that were embedded in the modules
715		of the register allocator. I need these modules for other purposes, but
716		they are not complete enough to put into a library (just yet).
717		----------------------------------------------------------------------
718		Name: Matthias Blume
719		Date: 2002/01/31 16:05:00 EST
720		Tag: blume-20020131-sparc-ccalls
721		Description:
722
723		1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
724		of extra stack space by mistake.  Fixed.
725
726		2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
727
728		----------------------------------------------------------------------
729		Name: Allen Leung
730		Date: 2002/01/30
731		Tag: leunga-20020130-nowhere-bug-fix
732		Description:
733
734		MLRISC bug fixes:
735		1. Fixed a bindings computation bug in the 'nowhere' program generator tool.
736		2. MachineInt.fromString was negating its value.
737
738		----------------------------------------------------------------------
739		Name: Matthias Blume
740		Date: 2002/01/29
741		Tag: blume-20020129-INSTALL
742		Description:
743
744		- Added somewhat detailed installation instructions (file INSTALL).
745		- Fixed curl-detection bug in config/install.sh.
746		- It is now possible to select the URL getter using the URLGETTER
747		environment variable:
748
749		not set / "unknown"      --> automatic detection (script tries wget,
750		curl, and lynx)
751		"wget" / "curl" / "lynx" --> use the specified program (script "knows"
752		how to properly invoke them)
753		other                    --> use $URLGETTER directly, it must take
754		precisely two command-line arguments
755		(source URL and destination file name)
756
757		----------------------------------------------------------------------
758		Name: Matthias Blume
759		Date: 2002/01/28
760		Tag: blume-20020128-sparc-ccalls
761		Description:
762
763		- Fixed problem with calculation of "used" registers in sparc-c-calls.
764		- Make use of the allocParam argument in sparc-c-calls.
765
766		----------------------------------------------------------------------
767		Name: Matthias Blume
768		Date: 2002/01/28
769		Tag: blume-20020128-allocParam
770		Description:
771
772		John Reppy:  Changes c-calls API to accept client-callback for
773		allocating extra stack space.
774		me: Corresponding changes to mlriscGen (using a dummy argument that
775		does not change the current behavior).
776
777		----------------------------------------------------------------------
778		Name: Matthias Blume
779		Date: 2002/01/28 12:00:00
780		Tag: Release_110_38
781		Description:
782
783		This time for real!!!
784
785		----------------------------------------------------------------------
786		Name: Matthias Blume
787		Date: 2002/01/28 10:56:00 EST
788		Tag: blume-20020128-retraction
789		Description:
790
791		0. Retracted earlier 110.38.  (The Release_110_38 tag has been replaced
792		with blume-Release_110_38-retracted.)
793
794		1. Fixed a problem with incorrect rounding modes in real64.sml.
795		(Thanks to Andrew Mccreight <andrew.mccreight@yale.edu>.)
796
797		2. A bug in ml-nlffigen related to the handling of unnamed structs, unions,
798		and enums fixed.  The naming of corresponding ML identifiers should
799		now be consistent again.
800
801		----------------------------------------------------------------------
802		Name: Allen Leung
803		Date: 2002/01/27
804		Tag: leunga-20020127-nowhere
805		Description:
806
807		Added a target called nowhere in the configuration scripts.
808		Enabling this will build the MLRISC 'nowhere' tool (for translating
809		programs with where-clauses into legal SML code) during installation.
810
811		----------------------------------------------------------------------
812		Name: Matthias Blume
813		Date: 2002/01/25 21:27:00 EST
814		Tag: blume-Release_110_38-retracted
815		Description:
816
817		Call it a (working) release!  Version is 110.38. Bootfiles are ready.
818
819		README will be added later.
820
821		!!! NOTE:  Re-tagged as blume-Release_110_38-retracted. Original tag
822		(Release_110_38) removed.  Reason: Last-minute bug fixes.
823
824		----------------------------------------------------------------------
825		Name: Matthias Blume
826		Date: 2002/01/25
827		Tag: blume-20020125-ffi
828		Description:
829
830		A large number of tweaks and improvements to ml-nlffi-lib and
831		ml-nlffigen:
832
833		- ML represenation types have been streamlined
834		- getter and setter functions work with concrete values, not abstract
835		ones where possible
836		- ml-nlffigen command line more flexible (see README file there)
837		- some bugs have been fixed (hopefully)
838
839		----------------------------------------------------------------------
840		Name: Lal George
841		Date: 2002/01/24
842		Tag: george-20020124-risc-ra-interface
843		Description:
844
845		There is a dramatic simplification in the interface to the
846		register allocator for RISC architectures as a result of making
847		parallel copy instructions explicit.
848
849		----------------------------------------------------------------------
850		Name: Matthias Blume
851		Date: 2002/01/22
852		Tag: blume-20020122-x86-ccalls
853		Description:
854
855		Bug fix for c-calls on x86 (having to do with how char- and
856		short-arguments are being handled).
857
858		----------------------------------------------------------------------
859		Name: Matthias Blume
860		Date: 2002/01/21
861		Tag: blume-20020121-ff
862		Description:
863
864		Another day of fiddling with the FFI...
865
866		1. Bug fix/workaround:  CKIT does not complain about negative array
867		dimensions, so ml-nlffigen has to guard itself against this possibility.
868		(Otherwise a negative dimension would send it into an infinite loop.)
869
870		2. Some of the abstract types (light objects, light pointers, most "base"
871		types) in structure C are now eqtypes.
872
873		3. Added constructors and test functions for NULL function pointers.
874
875		----------------------------------------------------------------------
876		Name: Matthias Blume
877		Date: 2002/01/18
878		Tag: blume-20020118-ready-for-new-release
879		Description:
880
881		Made config/srcarchiveurl point to a new place.  (Will provide boot
882		files shortly.)
883
884		Maybe we christen this to be 110.38?
885
886		----------------------------------------------------------------------
887		Name: Matthias Blume
888		Date: 2002/01/18
889		Tag: blume-20020118-more-ffifiddle
890		Description:
891
892		Today's FFI fiddling:
893
894		- Provided a structure CGetSet with "convenient" versions of C.Get.* and
895		C.Set.* that use concrete (MLRep.*) arguments and results instead
896		of abstract ones.
897
898		- Provided word-style bit operations etc. for "int" representation
899		types in MLRep.S<Foo>Bitops where <Foo> ranges over Char, Int, Short,
900		and Long.
901
902		----------------------------------------------------------------------
903		Name: Matthias Blume
904		Date: 2002/01/18
905		Tag: blume-20020118-use-x86-fp
906		Description:
907
908		Now that x86-fast-fp seems to be working, I turned it back on again
909		by default.  (Seems to work fine now, even with the FFI.)
910
911		Other than that, I added some documentation about the FFI to
912		src/ml-nlffigen/README and updated the FFI test examples in
913		src/ml-nlffi-lib/Tests/*.
914
915		----------------------------------------------------------------------
916		Name: Allen Leung
917		Date: 2002/01/17
918		Tag: leunga-20020117-x86-fast-fp-call
919		Description:
920
921		1. Fixed a problem with handling return fp values when x86's fast fp
922		mode is turned on.
923
924		2. Minor pretty printing fix for cellset.  Print %st(0) as %st(0) instead
925		of %f32.
926
927		3. Added a constructor INT32lit to the ast of MLRISC tools.
928
929		----------------------------------------------------------------------
930		Name: Matthias Blume
931		Date: 2002/01/16
932		Tag: blume-20020116-ffifiddle
933		Description:
934
935		More fiddling with the FFI interface:
936
937		- Make constness 'c instead of rw wherever possible.  This eliminates
938		the need for certain explicit coercions.  (However, due to ML's
939		value polymorphism, there will still be many cases where explicit
940		coercions are necessary.  Phantom types are not the whole answer
941		to modeling a subtyping relationship in ML.)
942
943		- ro/rw coersions for pointers added.  (Avoids the detour through */&.)
944
945		- "printf" test example added to src/ml-nlffi-lib/Tests.  (Demonstrates
946		clumsy workaround for varargs problem.)
947
948		----------------------------------------------------------------------
949		Name: Lal George
950		Date: 2002/01/15
951		Tag: <none>
952		Description:
953
954		1. Since COPY instructions are no longer native to the architecture,
955		a generic functor can be used to implement the expandCopies function.
956
957		2. Allowed EXPORT and IMPORT pseudo-op declarations to appear inside a
958		TEXT segment.
959
960		----------------------------------------------------------------------
961		Name: Matthias Blume
962		Date: 2002/01/15
963		Tag: blume-20020115-ffiupdates
964		Description:
965
966		1. Fix for bug resulting in single-precision float values being returned
967		incorrectly from FFI calls.
968
969		2. Small modifications to C FFI API:
970
971		- memory-allocation routines return straight objects (no options)
972		and raise an exception in out-of-memory situations
973		- unsafe extensions to cast between function pointers and pointers
974		from/to ints
975		- added structure C_Debug as an alternative to structure C where
976		pointer-dereferencing (|*| and |*!) always check for null-pointers
977		- added open_lib' to DynLinkage;  open_lib' works like open_lib
978		but also takes a (possibly empty) list of existing library handles
979		that the current library depends on
980
981		----------------------------------------------------------------------
982		Name: Matthias Blume
983		Date: 2002/01/10
984		Tag: blume-20020110-newffigen
985		Description:
986
987		1. Updates to portable graph code.
988
989		2. Major update to ml-nlffigen and ml-nlffi-lib.  Things are much
990		more scalable now so that even huge interfaces such as the one
991		for GTK compile in finite time and space. :-)
992		See src/ml-nlffigen/README for details on what's new.
993
994		----------------------------------------------------------------------
995		Name: Lal George
996		Date: 2001/01/09 14:31:35 EST 2002
997		Tag: george-20011206-rm-native-copy
998		Description:
999
1000		Removed the native COPY and FCOPY instructions
1001		from all the architectures and replaced it with the
1002		explicit COPY instruction from the previous commit.
1003
1004		It is now possible to simplify many of the optimizations
1005		modules that manipulate copies. This has not been
1006		done in this change.
1007
1008		----------------------------------------------------------------------
1009		Name: Lal George
1010		Date: 2001/12/06 16:50:13 EST 2001
1011		Tag: george-20011206-mlrisc-instruction
1012		Description:
1013
1014		Changed the representation of instructions from being fully abstract
1015		to being partially concrete. That is to say:
1016
1017		from
1018		type instruction
1019
1020		to
1021		type instr                              (* machine instruction *)
1022
1023		datatype instruction =
1024		LIVE of {regs: C.cellset, spilled: C.cellset}
1025		| KILL of {regs: C.cellset, spilled: C.cellset}
1026		| COPYXXX of {k: CB.cellkind, dst: CB.cell list, src: CB.cell list}
1027		| ANNOTATION of {i: instruction, a: Annotations.annotation}
1028		| INSTR of instr
1029
1030		This makes the handling of certain special instructions that appear on
1031		all architectures easier and uniform.
1032
1033		LIVE and KILL say that a list of registers are live or killed at the
1034		program point where they appear. No spill code is generated when an
1035		element of the 'regs' field is spilled, but the register is moved to
1036		the 'spilled' (which is present, more for debugging than anything else).
1037
1038		LIVE replaces the (now deprecated) DEFFREG instruction on the alpha.
1039		We used to generate:
1040
1041		DEFFREG f1
1042		f1 := f2 + f3
1043		trapb
1044
1045		but now generate:
1046
1047		f1 := f2 + f3
1048		trapb
1049		LIVE {regs=[f1,f2,f3], spilled=[]}
1050
1051		Furthermore, the DEFFREG (hack) required that all floating point instruction
1052		use all registers mentioned in the instruction. Therefore f1 := f2 + f3,
1053		defines f1 and uses [f1,f2,f3]! This hack is no longer required resulting
1054		in a cleaner alpha implementation. (Hopefully, intel will not get rid of
1055		this architecture).
1056
1057		COPYXXX is intended to replace the parallel COPY and FCOPY  available on
1058		all the architectures. This will result in further simplification of the
1059		register allocator that must be aware of them for coalescing purposes, and
1060		will also simplify certain aspects of the machine description that provides
1061		callbacks related to parallel copies.
1062
1063		ANNOTATION should be obvious, and now INSTR represents the honest to God
1064		machine instruction set!
1065
1066		The <arch>/instructions/<arch>Instr.sml files define certain utility
1067		functions for making porting easier -- essentially converting upper case
1068		to lower case. All machine instructions (of type instr) are in upper case,
1069		and the lower case form generates an MLRISC instruction. For example on
1070		the alpha we have:
1071
1072		datatype instr =
1073		LDA of {r:cell, b:cell, d:operand}
1074		| ...
1075
1076		val lda : {r:cell, b:cell, d:operand} -> instruction
1077		...
1078
1079		where lda is just (INSTR o LDA), etc.
1080
1081		----------------------------------------------------------------------
1082		Name: Matthias Blume
1083		Date: 2001/11/22 21:40:00 EST
1084		Tag: Release_110_37
1085		Description:
1086
1087		Release 110.37.  This time for real.
1088
1089		----------------------------------------------------------------------
1090		Name: Matthias Blume
1091		Date: 2001/11/21 16:35:00 EST
1092		Tag: blume-20011121-foot-in-mouth
1093		Description:
1094
1095		Removed the "Release_110_37" tag because of a serious bug.
1096		This will be re-tagged once the bug is fixed.
1097
1098		----------------------------------------------------------------------
1099	Name: Matthias Blume	Name: Matthias Blume
1100	Date: 2001/11/21 16:14:00 EST	Date: 2001/11/21 16:14:00 EST
1101	Tag: blume-20011121-forgottenfile	Tag: blume-20011121-forgottenfile
#	Line 24	Line 1107
1107	----------------------------------------------------------------------	----------------------------------------------------------------------
1108	Name: Matthias Blume	Name: Matthias Blume
1109	Date: 2001/11/21 16:10:00 EST	Date: 2001/11/21 16:10:00 EST
1110	Tag: Release_110_37	Tag: blume-20011121-invalid_110_37
1111	Description:	Description:
1112
1113		Note: I removed the original tag "Release_110_37" from this commit
1114		because we found a serious bug in all non-x86 backends.
1115		- Matthias
1116
1117	1. Modifications to the SML/NJ code generator and to the runtime system	1. Modifications to the SML/NJ code generator and to the runtime system
1118	so that code object name strings are directly inserted into code	so that code object name strings are directly inserted into code
1119	objects at code generation time.  The only business the runtime system	objects at code generation time.  The only business the runtime system
#	Line 248	Line 1335
1335
1336	----------------------------------------------------------------------	----------------------------------------------------------------------
1337	Name: Matthias Blume	Name: Matthias Blume
	>>>>>>> 1.169
1338	Date: 2001/09/18 15:35:00 EDT	Date: 2001/09/18 15:35:00 EDT
1339	Tag: blume-20010918-readme11036	Tag: blume-20010918-readme11036
1340	Description:	Description:

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.987   changed lines   Added in v.1131

---

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