Diff of /sml/trunk/HISTORY

Parent Directory | Revision Log | Patch

	revision 993, Fri Nov 23 02:37:10 2001 UTC	revision 1136, Tue Mar 12 19:44:02 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		Name: Lal George
16		Date: 2002/03/12 12 14:42:36 EST
17		Tag: george-20020312-frequency-computation
18		Description:
19
20		Integrated jump chaining and static block frequency into the
21		compiler. More details and numbers later.
22
23		----------------------------------------------------------------------
24		Name: Lal George
25		Date: 2002/03/11 11 22:38:53 EST
26		Tag: george-20020311-jump-chain-elim
27		Description:
28
29		Tested the jump chain elimination on all architectures (except the
30		hppa).  This is on by default right now and is profitable for the
31		alpha and x86, however, it may not be profitable for the sparc and ppc
32		when compiling the compiler.
33
34		The gc test will typically jump to a label at the end of the cluster,
35		where there is another jump to an external cluster containing the actual
36		code to invoke gc. This is to allow factoring of common gc invocation
37		sequences. That is to say, we generate:
38
39		f:
40		testgc
41		ja   L1      % jump if above to L1
42
43		L1:
44		jmp L2
45
46
47		After jump chain elimination the 'ja L1' instructions is converted to
48		'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
49		up being implemented in their long form (if L2 is far away) using:
50
51		jbe     L3      % jump if below or equal to L3
52		jmp     L2
53		L3:
54		...
55
56
57		For large compilation units L2  may be far away.
58
59
60		----------------------------------------------------------------------
61		Name: Matthias Blume
62		Date: 2002/03/11 13:30:00 EST
63		Tag: blume-20020311-mltreeeval
64		Description:
65
66		A functor parameter was missing.
67
68		----------------------------------------------------------------------
69		Name: Allen Leung
70		Date: 2002/03/11 10:30:00 EST
71		Tag: leunga-20020310-runtime-string0
72		Description:
73
74		The representation of the empty string now points to a
75		legal null terminated C string instead of unit.  It is now possible
76		to convert an ML string into C string with InlineT.CharVector.getData.
77		This compiles into one single machine instruction.
78
79		----------------------------------------------------------------------
80		Name: Allen Leung
81		Date: 2002/03/10 23:55:00 EST
82		Tag: leunga-20020310-x86-call
83		Description:
84
85		Added machine generation for CALL instruction (relative displacement mode)
86
87		----------------------------------------------------------------------
88		Name: Matthias Blume
89		Date: 2002/03/08 16:05:00
90		Tag: blume-20020308-entrypoints
91		Description:
92
93		Version number bumped to 110.39.1.  NEW BOOTFILES!
94
95		Entrypoints: non-zero offset into a code object where execution should begin.
96
97		- Added the notion of an entrypoint to CodeObj.
98		- Added reading/writing of entrypoint info to Binfile.
99		- Made runtime system bootloader aware of entrypoints.
100		- Use the address of the label of the first function given to mlriscGen
101		as the entrypoint.  This address is currently always 0, but it will
102		not be 0 once we turn on block placement.
103		- Removed the linkage cluster code (which was The Other Way(tm) of dealing
104		with entry points) from mlriscGen.
105
106		----------------------------------------------------------------------
107		Name: Allen Leung
108		Date: 2002/03/07 20:45:00 EST
109		Tag: leunga-20020307-x86-cmov
110		Description:
111
112		Bug fixes for CMOVcc on x86.
113
114		1. Added machine code generation for CMOVcc
115		2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
116		3. CMOVcc cannot have an immediate operand as argument.
117
118		----------------------------------------------------------------------
119		Name: Matthias Blume
120		Date: 2002/03/07 16:15:00 EST
121		Tag: blume-20020307-controls
122		Description:
123
124		This is a very large but mostly boring patch which makes (almost)
125		every tuneable compiler knob (i.e., pretty much everything under
126		Control.* plus a few other things) configurable via both the command
127		line and environment variables in the style CM did its configuration
128		until now.
129
130		Try starting sml with '-h' (or, if you are brave, '-H')
131
132		To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
133		implements the underlying generic mechanism.
134
135		The interface to some of the existing such facilities has changed somewhat.
136		For example, the MLRiscControl module now provides mkFoo instead of getFoo.
137		(The getFoo interface is still there for backward-compatibility, but its
138		use is deprecated.)
139
140		The ml-build script passes -Cxxx=yyy command-line arguments through so
141		that one can now twiddle the compiler settings when using this "batch"
142		compiler.
143
144		TODO items:
145
146		We should go through and throw out all controls that are no longer
147		connected to anything.  Moreover, we should go through and provide
148		meaningful (and correct!) documentation strings for those controls
149		that still are connected.
150
151		Currently, multiple calls to Controls.new are accepted (only the first
152		has any effect).  Eventually we should make sure that every control
153		is being made (via Controls.new) exactly once.  Future access can then
154		be done using Controls.acc.
155
156		Finally, it would probably be a good idea to use the getter-setter
157		interface to controls rather than ref cells.  For the time being, both
158		styles are provided by the Controls module, but getter-setter pairs are
159		better if thread-safety is of any concern because they can be wrapped.
160
161		*****************************************
162
163		One bug fix: The function blockPlacement in three of the MLRISC
164		backpatch files used to be hard-wired to one of two possibilities at
165		link time (according to the value of the placementFlag).  But (I
166		think) it should rather sense the flag every time.
167
168		*****************************************
169
170		Other assorted changes (by other people who did not supply a HISTORY entry):
171
172		1. the cross-module inliner now works much better (Monnier)
173		2. representation of weights, frequencies, and probabilities in MLRISC
174		changed in preparation of using those for weighted block placement
175		(Reppy, George)
176
177		----------------------------------------------------------------------
178		Name: Lal George
179		Date: 2002/03/07 14:44:24 EST 2002
180		Tag: george-20020307-weighted-block-placement
181
182		Tested the weighted block placement optimization on all architectures
183		(except the hppa) using AMPL to generate the block and edge frequencies.
184		Changes were required in the machine properties to correctly
185		categorize trap instructions. There is an MLRISC flag
186		"weighted-block-placement" that can be used to enable weighted block
187		placement, but this will be ineffective without block/edge
188		frequencies (coming soon).
189
190
191		----------------------------------------------------------------------
192		Name: Lal George
193		Date: 2002/03/05 17:24:48 EST
194		Tag: george-20020305-linkage-cluster
195
196		In order to support the block placement optimization, a new cluster
197		is generated as the very first cluster (called the linkage cluster).
198		It contains a single jump to the 'real' entry point for the compilation
199		unit. Block placement has no effect on the linkage cluster itself, but
200		all the other clusters  have full freedom in the manner in which they
201		reorder blocks or functions.
202
203		On the x86 the typical linkage code that is generated is:
204		----------------------
205		.align 2
206		L0:
207		addl    $L1-L0, 72(%esp)
208		jmp     L1
209
210
211		.align  2
212		L1:
213		----------------------
214
215		72(%esp) is the memory location for the stdlink register. This
216		must contain the address of the CPS function being called. In the
217		above example, it contains the address of  L0; before
218		calling L1 (the real entry point for the compilation unit), it
219		must contain the address for L1, and hence
220
221		addl $L1-L0, 72(%esp)
222
223		I have tested this on all architectures except the hppa.The increase
224		in code size is of course negligible
225
226		----------------------------------------------------------------------
227		Name: Allen Leung
228		Date: 2002/03/03 13:20:00 EST
229		Tag: leunga-20020303-mlrisc-tools
230
231		Added #[ ... ] expressions to mlrisc tools
232
233		----------------------------------------------------------------------
234		Name: Matthias Blume
235		Date: 2002/02/27 12:29:00 EST
236		Tag: blume-20020227-cdebug
237		Description:
238
239		- made types in structure C and C_Debug to be equal
240		- got rid of code duplication (c-int.sml vs. c-int-debug.sml)
241		- there no longer is a C_Int_Debug (C_Debug is directly derived from C)
242
243		----------------------------------------------------------------------
244		Name: Matthias Blume
245		Date: 2002/02/26 12:00:00 EST
246		Tag: blume-20020226-ffi
247		Description:
248
249		1. Fixed a minor bug in CM's "noweb" tool:
250		If numbering is turned off, then truly don't number (i.e., do not
251		supply the -L option to noweb).  The previous behavior was to supply
252		-L'' -- which caused noweb to use the "default" line numbering scheme.
253		Thanks to Chris Richards for pointing this out (and supplying the fix).
254
255		2. Once again, I reworked some aspects of the FFI:
256
257		A. The incomplete/complete type business:
258
259		- Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
260		gone!
261		- ML types representing an incomplete type are now *equal* to
262		ML types representing their corresponding complete types (just like
263		in C).  This is still safe because ml-nlffigen will not generate
264		RTTI for incomplete types, nor will it generate functions that
265		require access to such RTTI.   But when ML code generated from both
266		incomplete and complete versions of the C type meet, the ML types
267		are trivially interoperable.
268
269		NOTE:  These changes restore the full generality of the translation
270		(which was previously lost when I eliminated functorization)!
271
272		B. Enum types:
273
274		- Structure C now has a type constructor "enum" that is similar to
275		how the "su" constructor works.  However, "enum" is not a phantom
276		type because each "T enum" has values (and is isomorphic to
277		MLRep.Signed.int).
278		- There are generic access operations for enum objects (using
279		MLRep.Signed.int).
280		- ml-nlffigen will generate a structure E_foo for each "enum foo".
281		* The structure contains the definition of type "mlrep" (the ML-side
282		representation type of the enum).  Normally, mlrep is the same
283		as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
284		then mlrep will be defined as a datatype -- thus facilitating
285		pattern matching on mlrep values.
286		("-ec" will be suppressed if there are duplicate values in an
287		enumeration.)
288		* Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
289		will be generated for each C enum constant xxx.
290		* Conversion functions m2i and i2m convert between mlrep and
291		MLRep.Signed.int.  (Without "-ec", these functions are identities.)
292		* Coversion functions c and ml convert between mlrep and "tag enum".
293		* Access functions (get/set) fetch and store mlrep values.
294		- By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
295		enumerations are merged into one single enumeration represented by
296		structure E_'.
297
298		----------------------------------------------------------------------
299		Name: Allen Leung
300		Date: 2002/02/25 04:45:00 EST
301		Tag: leunga-20020225-cps-spill
302
303		This is a new implementation of the CPS spill phase.
304		The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
305		In case of problems, replace it with the old file spill.sml
306
307		The current compiler runs into some serious performance problems when
308		constructing a large record.  This can happen when we try to compile a
309		structure with many items.  Even a very simple structure like the following
310		makes the compiler slow down.
311
312		structure Foo = struct
313		val x_1 = 0w1 : Word32.int
314		val x_2 = 0w2 : Word32.int
315		val x_3 = 0w3 : Word32.int
316		...
317		val x_N = 0wN : Word32.int
318		end
319
320		The following table shows the compile time, from N=1000 to N=4000,
321		with the old compiler:
322
323		N
324		1000   CPS 100 spill                           0.04u  0.00s  0.00g
325		MLRISC ra                               0.06u  0.00s  0.05g
326		(spills = 0 reloads = 0)
327		TOTAL                                   0.63u  0.07s  0.21g
328
329		1100   CPS 100 spill                           8.25u  0.32s  0.64g
330		MLRISC ra                               5.68u  0.59s  3.93g
331		(spills = 0 reloads = 0)
332		TOTAL                                   14.71u  0.99s  4.81g
333
334		1500   CPS 100 spill                           58.55u  2.34s  1.74g
335		MLRISC ra                               5.54u  0.65s  3.91g
336		(spills = 543 reloads = 1082)
337		TOTAL                                   65.40u  3.13s  6.00g
338
339		2000   CPS 100 spill                           126.69u  4.84s  3.08g
340		MLRISC ra                               0.80u  0.10s  0.55g
341		(spills = 42 reloads = 84)
342		TOTAL                                   129.42u  5.10s  4.13g
343
344		3000   CPS 100 spill                           675.59u  19.03s  11.64g
345		MLRISC ra                               2.69u  0.27s  1.38g
346		(spills = 62 reloads = 124)
347		TOTAL                                   682.48u  19.61s  13.99g
348
349		4000   CPS 100 spill                           2362.82u  56.28s  43.60g
350		MLRISC ra                               4.96u  0.27s  2.72g
351		(spills = 85 reloads = 170)
352		TOTAL                                   2375.26u  57.21s  48.00g
353
354		As you can see the old cps spill module suffers from some serious
355		performance problem.  But since I cannot decipher the old code fully,
356		instead of patching the problems up, I'm reimplementing it
357		with a different algorithm.  The new code is more modular,
358		smaller when compiled, and substantially faster
359		(O(n log n) time and O(n) space).  Timing of the new spill module:
360
361		4000  CPS 100 spill                           0.02u  0.00s  0.00g
362		MLRISC ra                               0.25u  0.02s  0.15g
363		(spills=1 reloads=3)
364		TOTAL                                   7.74u  0.34s  1.62g
365
366		Implementation details:
367
368		As far as I can tell, the purpose of the CPS spill module is to make sure the
369		number of live variables at any program point (the bandwidth)
370		does not exceed a certain limit, which is determined by the
371		size of the spill area.
372
373		When the bandwidth is too large, we decrease the register pressure by
374		packing live variables into spill records.  How we achieve this is
375		completely different than what we did in the old code.
376
377		First, there is something about the MLRiscGen code generator
378		that we should be aware of:
379
380		o MLRiscGen performs code motion!
381
382		In particular, it will move floating point computations and
383		address computations involving only the heap pointer to
384		their use sites (if there is only a single use).
385		What this means is that if we have a CPS record construction
386		statement
387
388		RECORD(k,vl,w,e)
389
390		we should never count the new record address w as live if w
391		has only one use (which is often the case).
392
393		We should do something similar to floating point, but the transformation
394		there is much more complex, so I won't deal with that.
395
396		Secondly, there are now two new cps primops at our disposal:
397
398		1. rawrecord of record_kind option
399		This pure operator allocates some uninitialized storage from the heap.
400		There are two forms:
401
402		rawrecord NONE [INT n]  allocates a tagless record of length n
403		rawrecord (SOME rk) [INT n] allocates a tagged record of length n
404		and initializes the tag.
405
406		2. rawupdate of cty
407		rawupdate cty (v,i,x)
408		Assigns to x to the ith component of record v.
409		The storelist is not updated.
410
411		We use these new primops for both spilling and increment record construction.
412
413		1. Spilling.
414
415		This is implemented with a linear scan algorithm (but generalized
416		to trees).  The algorithm will create a single spill record at the
417		beginning of the cps function and use rawupdate to spill to it,
418		and SELECT or SELp to reload from it.  So both spills and reloads
419		are fine-grain operations.  In contrast, in the old algorithm
420		"spills" have to be bundled together in records.
421
422		Ideally, we should sink the spill record construction to where
423		it is needed.  We can even split the spill record into multiple ones
424		at the places where they are needed.  But CPS is not a good
425		representation for global code motion, so I'll keep it simple and
426		am not attempting this.
427
428		2. Incremental record construction (aka record splitting).
429
430		Long records with many component values which are simulatenously live
431		(recall that single use record addresses are not considered to
432		be live) are constructed with rawrecord and rawupdate.
433		We allocate space on the heap with rawrecord first, then gradually
434		fill it in with rawupdate.  This is the technique suggested to me
435		by Matthias.
436
437		Some restrictions on when this is applicable:
438		1. It is not a VECTOR record.  The code generator currently does not handle
439		this case. VECTOR record uses double indirection like arrays.
440		2. All the record component values are defined in the same "basic block"
441		as the record constructor.  This is to prevent speculative
442		record construction.
443
444		----------------------------------------------------------------------
445		Name: Allen Leung
446		Date: 2002/02/22 01:02:00 EST
447		Tag: leunga-20020222-mlrisc-tools
448
449		Minor bug fixes in the parser and rewriter
450
451		----------------------------------------------------------------------
452		Name: Allen Leung
453		Date: 2002/02/21 20:20:00 EST
454		Tag: leunga-20020221-peephole
455
456		Regenerated the peephole files.  Some contained typos in the specification
457		and some didn't compile because of pretty printing bugs in the old version
458		of 'nowhere'.
459
460		----------------------------------------------------------------------
461		Name: Allen Leung
462		Date: 2002/02/19 20:20:00 EST
463		Tag: leunga-20020219-mlrisc-tools
464		Description:
465
466		Minor bug fixes to the mlrisc-tools library:
467
468		1.  Fixed up parsing colon suffixed keywords
469		2.  Added the ability to shut the error messages up
470		3.  Reimplemented the pretty printer and fixed up/improved
471		the pretty printing of handle and -> types.
472		4.  Fixed up generation of literal symbols in the nowhere tool.
473		5.  Added some SML keywords to to sml.sty
474
475		----------------------------------------------------------------------
476		Name: Matthias Blume
477		Date: 2002/02/19 16:20:00 EST
478		Tag: blume-20020219-cmffi
479		Description:
480
481		A wild mix of changes, some minor, some major:
482
483		* All C FFI-related libraries are now anchored under $c:
484		$/c.cm      --> $c/c.cm
485		$/c-int.cm  --> $c/internals/c-int.cm
486		$/memory.cm --> $c/memory/memory.cm
487
488		* "make" tool (in CM) now treats its argument pathname slightly
489		differently:
490		1. If the native expansion is an absolute name, then before invoking
491		the "make" command on it, CM will apply OS.Path.mkRelative
492		(with relativeTo = OS.FileSys.getDir()) to it.
493		2. The argument will be passed through to subsequent phases of CM
494		processing without "going native".  In particular, if the argument
495		was an anchored path, then "make" will not lose track of that anchor.
496
497		* Compiler backends now "know" their respective C calling conventions
498		instead of having to be told about it by ml-nlffigen.  This relieves
499		ml-nlffigen from one of its burdens.
500
501		* The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
502
503		* Export C_DEBUG and C_Debug from $c/c.cm.
504
505		* C type encoding in ml-nlffi-lib has been improved to model the conceptual
506		subtyping relationship between incomplete pointers and their complete
507		counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
508		with the convention of instantiating 'o with ('t, 'c) obj whenever
509		the pointer target type is complete.  In the incomplete case, 'o
510		will be instantiated with some "'c iobj" -- a type obtained by
511		using one of the functors PointerToIncompleteType or PointerToCompleteType.
512
513		Operations that work on both incomplete and complete pointer types are
514		typed as taking an 'o ptr while operations that require the target to
515		be known are typed as taking some ('t, 'c) obj ptr.
516
517		voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
518		where void is an eqtype without any values.  This makes it possible
519		to work on voidptr values using functions meant to operate on light
520		incomplete pointers.
521
522		* As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
523		been vastly simplified.
524
525		----------------------------------------------------------------------
526		Name: Matthias Blume
527		Date: 2002/02/19 10:48:00 EST
528		Tag: blume-20020219-pqfix
529		Description:
530
531		Applied Chris Okasaki's bug fix for priority queues.
532
533		----------------------------------------------------------------------
534		Name: Matthias Blume
535		Date: 2002/02/15 17:05:00
536		Tag: Release_110_39
537		Description:
538
539		Last-minute retagging is becoming a tradition... :-(
540
541		This is the working release 110.39.
542
543		----------------------------------------------------------------------
544		Name: Matthias Blume
545		Date: 2002/02/15 16:00:00 EST
546		Tag: Release_110_39-orig
547		Description:
548
549		Working release 110.39.  New bootfiles.
550
551		(Update: There was a small bug in the installer so it wouldn't work
552		with all shells.  So I retagged. -Matthias)
553
554		----------------------------------------------------------------------
555		Name: Matthias Blume
556		Date: 2002/02/15 14:17:00 EST
557		Tag: blume-20020215-showbindings
558		Description:
559
560		Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
561		the latter can be useful for exploring what bindings are available at
562		the interactive prompt.  (The first function returns only the list
563		of symbols that are really bound, the second prints those but also the
564		ones that CM's autoloading mechanism knows about.)
565
566		----------------------------------------------------------------------
567		Name: Matthias Blume
568		Date: 2002/02/15 12:08:00 EST
569		Tag: blume-20020215-iptrs
570		Description:
571
572		Two improvements to ml-nlffigen:
573
574		1. Write files only if they do not exist or if their current contents
575		do not coincide with what's being written.  (That is, avoid messing
576		with the time stamps unless absolutely necessary.)
577
578		2. Implement a "repository" mechanism for generated files related
579		to "incomplete pointer types".   See the README file for details.
580
581		----------------------------------------------------------------------
582		Name: Matthias Blume
583		Date: 2002/02/14 11:50:00 EST
584		Tag: blume-20020214-quote
585		Description:
586
587		Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
588		because of the new and improved tag generation scheme.  (Thanks to Allen
589		Leung for pointing it out.)
590
591		----------------------------------------------------------------------
592		Name: Lal George
593		Date: 2002/02/14 09:55:27 EST 2002
594		Tag: george-20020214-isabelle-bug
595		Description:
596
597		Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
598		of Isabelle on the x86.
599
600		From Allen:
601		-----------
602		I've found the problem:
603
604		in ra-core.sml, I use the counter "blocked" to keep track of the
605		true number of elements in the freeze queue.  When the counter goes
606		to zero, I skip examining the queue.  But I've messed up the
607		bookkeeping in combine():
608
609		else ();
610		case !ucol of
611		PSEUDO => (if !cntv > 0 then
612		(if !cntu > 0 then blocked := !blocked - 1 else ();
613		^^^^^^^^^^^^^^^^^^^^^^^
614		moveu := mergeMoveList(!movev, !moveu)
615		)
616		else ();
617
618		combine() is called to coalesce two nodes u and v.
619		I think I was thinking that if the move counts of u and v are both
620		greater than zero then after they are coalesced then one node is
621		removed from the freeze queue.  Apparently I was thinking that
622		both u and v are of low degree, but that's clearly not necessarily true.
623
624
625		02/12/2002:
626		Here's the patch.  HOL now compiles.
627
628		I don't know how this impact on performance (compile
629		time or runtime).  This bug caused the RA (especially on the x86)
630		to go thru the potential spill phase when there are still nodes on the
631		freeze queue.
632
633
634
635
636		----------------------------------------------------------------------
637		Name: Matthias Blume
638		Date: 2002/02/13 22:40:00 EST
639		Tag: blume-20020213-fptr-rtti
640		Description:
641
642		Fixed a bug in ml-nlffigen that was introduced with one of the previous
643		updates.
644
645		----------------------------------------------------------------------
646		Name: Matthias Blume
647		Date: 2002/02/13 16:41:00 EST
648		Tag: blume-20020213-cmlpq
649		Description:
650
651		Added new priority queue export symbols (which have just been added to
652		smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
653		not compile and the installer would choke.)
654
655		----------------------------------------------------------------------
656		Name: Matthias Blume
657		Date: 2002/02/13 16:15:00 EST
658		Tag: blume-20020213-various
659		Description:
660
661		1. More tweaks to ml-nlffigen:
662
663		- better internal datastructures (resulting in slight speedup)
664		- "-match" option requires exact match
665		- "localized" gensym counters (untagged structs/unions nested within
666		other structs/unions or within typedefs get a fresh counter; their
667		tag will be prefixed by a concatenation of their parents' tags)
668		- bug fixes (related to calculation of transitive closure of types
669		to be included in the output)
670
671		2. Minor Basis updates:
672
673		- added implementations for List.collate and Option.app
674
675		----------------------------------------------------------------------
676		Name: Matthias Blume
677		Date: 2002/02/11 15:55:00 EST
678		Tag: blume-20020211-gensym
679		Description:
680
681		Added a "-gensym" option to command line of ml-nlffigen.  This can be
682		used to specify a "stem" -- a string that is inserted in all "gensym'd"
683		names (ML structure names that correspond to unnamed C structs, unions,
684		and enums), so that separate runs of ml-nlffigen do not clash.
685
686		----------------------------------------------------------------------
687		Name: Matthias Blume
688		Date: 2002/02/11 12:05:00 EST
689		Tag: blume-20020211-gensml
690		Description:
691
692		A quick fix for a problem with GenSML (in the pgraph-util library):
693		Make generation of toplevel "local" optional.  (Strictly speaking,
694		signature definitions within "local" are not legal SML.)
695
696		Other than that: updates to INSTALL and cm/TODO.
697
698		----------------------------------------------------------------------
699		Name: Matthias Blume
700		Date: 2002/02/08 15:00:00 EST
701		Tag: blume-20020208-uniquepid
702		Description:
703
704		0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
705
706		1. The installer (config/install.sh) has gotten smarter:
707
708		- Configuration options are a bit easier to specify now
709		(in config/targets).
710		- Bug in recognizing .tar.bz2 files fixed.
711		- Installer automatically resolves dependencies between
712		configuration options (e.g., if you ask for eXene, you will
713		also get cml -- regardless whether you asked for it or not).
714		- Installer can run in "quieter mode" by setting the environment
715		variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
716		"completely silent", though.
717		- Build HashCons library as part of smlnj-lib.
718
719		2. A new scheme for assigning persistent identifiers to compilation
720		units (and, by extension, to types etc.) has been put into place.
721		This fixes a long-standing bug where types and even dynamic values
722		can get internally confused, thereby compromising type safety
723		(abstraction) and dynamic correctness.  See
724
725		http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
726
727		for an example of how things could go wrong until now.
728
729		The downside of the new scheme is that pids are not quite as
730		persistent as they used to be: CM will generate a fresh pid
731		for every compilation unit that it thinks it sees for the first
732		time.  That means that if you compile starting from a clean, fresh
733		source tree at two different times, you end up with different
734		binaries.
735
736		Cutoff recompilation, however, has not been compromised because
737		CM keeps pid information in special caches between runs.
738
739		----------------------------------------------------------------------
740		Name: Lal George
741		Date: 2002/02/07 15:34:13 EST 2002
742		Tag: <none>
743		Description:
744
745		Compilers that generate assembly code may produce  global labels
746		whose value is resolved at link time. The various peephole optimization
747		modules did not take this in account.
748
749		TODO. The Labels.addrOf function should really return an option
750		type so that clients are forced to deal with this issue, rather
751		than an exception being raised.
752
753		----------------------------------------------------------------------
754		Name: Lal George
755		Date: 2002/02/06 13:55:02 EST
756		Tag: george-20020206-ra-breakup
757		Description:
758
759		1. A bug fix from Allen.
760
761		A typo causes extra fstp %st(0)'s to be generated at compensation
762		edges, which might cause stack underflow traps at runtime.  This
763		occurs in fft where there are extraneous fstps right before the 'into'
764		trap instruction (in this case they are harmless since none of the
765		integers overflow.)
766
767		2. Pulled out various utility modules that were embedded in the modules
768		of the register allocator. I need these modules for other purposes, but
769		they are not complete enough to put into a library (just yet).
770		----------------------------------------------------------------------
771		Name: Matthias Blume
772		Date: 2002/01/31 16:05:00 EST
773		Tag: blume-20020131-sparc-ccalls
774		Description:
775
776		1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
777		of extra stack space by mistake.  Fixed.
778
779		2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
780
781		----------------------------------------------------------------------
782		Name: Allen Leung
783		Date: 2002/01/30
784		Tag: leunga-20020130-nowhere-bug-fix
785		Description:
786
787		MLRISC bug fixes:
788		1. Fixed a bindings computation bug in the 'nowhere' program generator tool.
789		2. MachineInt.fromString was negating its value.
790
791		----------------------------------------------------------------------
792		Name: Matthias Blume
793		Date: 2002/01/29
794		Tag: blume-20020129-INSTALL
795		Description:
796
797		- Added somewhat detailed installation instructions (file INSTALL).
798		- Fixed curl-detection bug in config/install.sh.
799		- It is now possible to select the URL getter using the URLGETTER
800		environment variable:
801
802		not set / "unknown"      --> automatic detection (script tries wget,
803		curl, and lynx)
804		"wget" / "curl" / "lynx" --> use the specified program (script "knows"
805		how to properly invoke them)
806		other                    --> use $URLGETTER directly, it must take
807		precisely two command-line arguments
808		(source URL and destination file name)
809
810		----------------------------------------------------------------------
811		Name: Matthias Blume
812		Date: 2002/01/28
813		Tag: blume-20020128-sparc-ccalls
814		Description:
815
816		- Fixed problem with calculation of "used" registers in sparc-c-calls.
817		- Make use of the allocParam argument in sparc-c-calls.
818
819		----------------------------------------------------------------------
820		Name: Matthias Blume
821		Date: 2002/01/28
822		Tag: blume-20020128-allocParam
823		Description:
824
825		John Reppy:  Changes c-calls API to accept client-callback for
826		allocating extra stack space.
827		me: Corresponding changes to mlriscGen (using a dummy argument that
828		does not change the current behavior).
829
830		----------------------------------------------------------------------
831		Name: Matthias Blume
832		Date: 2002/01/28 12:00:00
833		Tag: Release_110_38
834		Description:
835
836		This time for real!!!
837
838		----------------------------------------------------------------------
839		Name: Matthias Blume
840		Date: 2002/01/28 10:56:00 EST
841		Tag: blume-20020128-retraction
842		Description:
843
844		0. Retracted earlier 110.38.  (The Release_110_38 tag has been replaced
845		with blume-Release_110_38-retracted.)
846
847		1. Fixed a problem with incorrect rounding modes in real64.sml.
848		(Thanks to Andrew Mccreight <andrew.mccreight@yale.edu>.)
849
850		2. A bug in ml-nlffigen related to the handling of unnamed structs, unions,
851		and enums fixed.  The naming of corresponding ML identifiers should
852		now be consistent again.
853
854		----------------------------------------------------------------------
855		Name: Allen Leung
856		Date: 2002/01/27
857		Tag: leunga-20020127-nowhere
858		Description:
859
860		Added a target called nowhere in the configuration scripts.
861		Enabling this will build the MLRISC 'nowhere' tool (for translating
862		programs with where-clauses into legal SML code) during installation.
863
864		----------------------------------------------------------------------
865		Name: Matthias Blume
866		Date: 2002/01/25 21:27:00 EST
867		Tag: blume-Release_110_38-retracted
868		Description:
869
870		Call it a (working) release!  Version is 110.38. Bootfiles are ready.
871
872		README will be added later.
873
874		!!! NOTE:  Re-tagged as blume-Release_110_38-retracted. Original tag
875		(Release_110_38) removed.  Reason: Last-minute bug fixes.
876
877		----------------------------------------------------------------------
878		Name: Matthias Blume
879		Date: 2002/01/25
880		Tag: blume-20020125-ffi
881		Description:
882
883		A large number of tweaks and improvements to ml-nlffi-lib and
884		ml-nlffigen:
885
886		- ML represenation types have been streamlined
887		- getter and setter functions work with concrete values, not abstract
888		ones where possible
889		- ml-nlffigen command line more flexible (see README file there)
890		- some bugs have been fixed (hopefully)
891
892		----------------------------------------------------------------------
893		Name: Lal George
894		Date: 2002/01/24
895		Tag: george-20020124-risc-ra-interface
896		Description:
897
898		There is a dramatic simplification in the interface to the
899		register allocator for RISC architectures as a result of making
900		parallel copy instructions explicit.
901
902		----------------------------------------------------------------------
903		Name: Matthias Blume
904		Date: 2002/01/22
905		Tag: blume-20020122-x86-ccalls
906		Description:
907
908		Bug fix for c-calls on x86 (having to do with how char- and
909		short-arguments are being handled).
910
911		----------------------------------------------------------------------
912		Name: Matthias Blume
913		Date: 2002/01/21
914		Tag: blume-20020121-ff
915		Description:
916
917		Another day of fiddling with the FFI...
918
919		1. Bug fix/workaround:  CKIT does not complain about negative array
920		dimensions, so ml-nlffigen has to guard itself against this possibility.
921		(Otherwise a negative dimension would send it into an infinite loop.)
922
923		2. Some of the abstract types (light objects, light pointers, most "base"
924		types) in structure C are now eqtypes.
925
926		3. Added constructors and test functions for NULL function pointers.
927
928		----------------------------------------------------------------------
929		Name: Matthias Blume
930		Date: 2002/01/18
931		Tag: blume-20020118-ready-for-new-release
932		Description:
933
934		Made config/srcarchiveurl point to a new place.  (Will provide boot
935		files shortly.)
936
937		Maybe we christen this to be 110.38?
938
939		----------------------------------------------------------------------
940		Name: Matthias Blume
941		Date: 2002/01/18
942		Tag: blume-20020118-more-ffifiddle
943		Description:
944
945		Today's FFI fiddling:
946
947		- Provided a structure CGetSet with "convenient" versions of C.Get.* and
948		C.Set.* that use concrete (MLRep.*) arguments and results instead
949		of abstract ones.
950
951		- Provided word-style bit operations etc. for "int" representation
952		types in MLRep.S<Foo>Bitops where <Foo> ranges over Char, Int, Short,
953		and Long.
954
955		----------------------------------------------------------------------
956		Name: Matthias Blume
957		Date: 2002/01/18
958		Tag: blume-20020118-use-x86-fp
959		Description:
960
961		Now that x86-fast-fp seems to be working, I turned it back on again
962		by default.  (Seems to work fine now, even with the FFI.)
963
964		Other than that, I added some documentation about the FFI to
965		src/ml-nlffigen/README and updated the FFI test examples in
966		src/ml-nlffi-lib/Tests/*.
967
968		----------------------------------------------------------------------
969		Name: Allen Leung
970		Date: 2002/01/17
971		Tag: leunga-20020117-x86-fast-fp-call
972		Description:
973
974		1. Fixed a problem with handling return fp values when x86's fast fp
975		mode is turned on.
976
977		2. Minor pretty printing fix for cellset.  Print %st(0) as %st(0) instead
978		of %f32.
979
980		3. Added a constructor INT32lit to the ast of MLRISC tools.
981
982		----------------------------------------------------------------------
983		Name: Matthias Blume
984		Date: 2002/01/16
985		Tag: blume-20020116-ffifiddle
986		Description:
987
988		More fiddling with the FFI interface:
989
990		- Make constness 'c instead of rw wherever possible.  This eliminates
991		the need for certain explicit coercions.  (However, due to ML's
992		value polymorphism, there will still be many cases where explicit
993		coercions are necessary.  Phantom types are not the whole answer
994		to modeling a subtyping relationship in ML.)
995
996		- ro/rw coersions for pointers added.  (Avoids the detour through */&.)
997
998		- "printf" test example added to src/ml-nlffi-lib/Tests.  (Demonstrates
999		clumsy workaround for varargs problem.)
1000
1001		----------------------------------------------------------------------
1002		Name: Lal George
1003		Date: 2002/01/15
1004		Tag: <none>
1005		Description:
1006
1007		1. Since COPY instructions are no longer native to the architecture,
1008		a generic functor can be used to implement the expandCopies function.
1009
1010		2. Allowed EXPORT and IMPORT pseudo-op declarations to appear inside a
1011		TEXT segment.
1012
1013		----------------------------------------------------------------------
1014		Name: Matthias Blume
1015		Date: 2002/01/15
1016		Tag: blume-20020115-ffiupdates
1017		Description:
1018
1019		1. Fix for bug resulting in single-precision float values being returned
1020		incorrectly from FFI calls.
1021
1022		2. Small modifications to C FFI API:
1023
1024		- memory-allocation routines return straight objects (no options)
1025		and raise an exception in out-of-memory situations
1026		- unsafe extensions to cast between function pointers and pointers
1027		from/to ints
1028		- added structure C_Debug as an alternative to structure C where
1029		pointer-dereferencing (|*| and |*!) always check for null-pointers
1030		- added open_lib' to DynLinkage;  open_lib' works like open_lib
1031		but also takes a (possibly empty) list of existing library handles
1032		that the current library depends on
1033
1034		----------------------------------------------------------------------
1035		Name: Matthias Blume
1036		Date: 2002/01/10
1037		Tag: blume-20020110-newffigen
1038		Description:
1039
1040		1. Updates to portable graph code.
1041
1042		2. Major update to ml-nlffigen and ml-nlffi-lib.  Things are much
1043		more scalable now so that even huge interfaces such as the one
1044		for GTK compile in finite time and space. :-)
1045		See src/ml-nlffigen/README for details on what's new.
1046
1047		----------------------------------------------------------------------
1048		Name: Lal George
1049		Date: 2001/01/09 14:31:35 EST 2002
1050		Tag: george-20011206-rm-native-copy
1051		Description:
1052
1053		Removed the native COPY and FCOPY instructions
1054		from all the architectures and replaced it with the
1055		explicit COPY instruction from the previous commit.
1056
1057		It is now possible to simplify many of the optimizations
1058		modules that manipulate copies. This has not been
1059		done in this change.
1060
1061		----------------------------------------------------------------------
1062		Name: Lal George
1063		Date: 2001/12/06 16:50:13 EST 2001
1064		Tag: george-20011206-mlrisc-instruction
1065		Description:
1066
1067		Changed the representation of instructions from being fully abstract
1068		to being partially concrete. That is to say:
1069
1070		from
1071		type instruction
1072
1073		to
1074		type instr                              (* machine instruction *)
1075
1076		datatype instruction =
1077		LIVE of {regs: C.cellset, spilled: C.cellset}
1078		| KILL of {regs: C.cellset, spilled: C.cellset}
1079		| COPYXXX of {k: CB.cellkind, dst: CB.cell list, src: CB.cell list}
1080		| ANNOTATION of {i: instruction, a: Annotations.annotation}
1081		| INSTR of instr
1082
1083		This makes the handling of certain special instructions that appear on
1084		all architectures easier and uniform.
1085
1086		LIVE and KILL say that a list of registers are live or killed at the
1087		program point where they appear. No spill code is generated when an
1088		element of the 'regs' field is spilled, but the register is moved to
1089		the 'spilled' (which is present, more for debugging than anything else).
1090
1091		LIVE replaces the (now deprecated) DEFFREG instruction on the alpha.
1092		We used to generate:
1093
1094		DEFFREG f1
1095		f1 := f2 + f3
1096		trapb
1097
1098		but now generate:
1099
1100		f1 := f2 + f3
1101		trapb
1102		LIVE {regs=[f1,f2,f3], spilled=[]}
1103
1104		Furthermore, the DEFFREG (hack) required that all floating point instruction
1105		use all registers mentioned in the instruction. Therefore f1 := f2 + f3,
1106		defines f1 and uses [f1,f2,f3]! This hack is no longer required resulting
1107		in a cleaner alpha implementation. (Hopefully, intel will not get rid of
1108		this architecture).
1109
1110		COPYXXX is intended to replace the parallel COPY and FCOPY  available on
1111		all the architectures. This will result in further simplification of the
1112		register allocator that must be aware of them for coalescing purposes, and
1113		will also simplify certain aspects of the machine description that provides
1114		callbacks related to parallel copies.
1115
1116		ANNOTATION should be obvious, and now INSTR represents the honest to God
1117		machine instruction set!
1118
1119		The <arch>/instructions/<arch>Instr.sml files define certain utility
1120		functions for making porting easier -- essentially converting upper case
1121		to lower case. All machine instructions (of type instr) are in upper case,
1122		and the lower case form generates an MLRISC instruction. For example on
1123		the alpha we have:
1124
1125		datatype instr =
1126		LDA of {r:cell, b:cell, d:operand}
1127		| ...
1128
1129		val lda : {r:cell, b:cell, d:operand} -> instruction
1130		...
1131
1132		where lda is just (INSTR o LDA), etc.
1133
1134	----------------------------------------------------------------------	----------------------------------------------------------------------
1135	Name: Matthias Blume	Name: Matthias Blume
#	Line 269	Line 1388
1388
1389	----------------------------------------------------------------------	----------------------------------------------------------------------
1390	Name: Matthias Blume	Name: Matthias Blume
	>>>>>>> 1.169
1391	Date: 2001/09/18 15:35:00 EDT	Date: 2001/09/18 15:35:00 EDT
1392	Tag: blume-20010918-readme11036	Tag: blume-20010918-readme11036
1393	Description:	Description:

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.993   changed lines   Added in v.1136

---

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