Diff of /sml/branches/primop-branch-3/NOTES/HISTORY

Parent Directory | Revision Log | Patch

	revision 890, Thu Jul 19 20:38:56 2001 UTC	revision 1127, Fri Mar 8 01:35:33 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: 2002/03/07 20:45:00 EST
18		Tag: leunga-20020307-x86-cmov
19		Description:
20
21		Bug fixes for CMOVcc on x86.
22
23		1. Added machine code generation for CMOVcc
24		2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
25		3. CMOVcc cannot have an immediate operand as argument.
26
27		----------------------------------------------------------------------
28		Name: Matthias Blume
29		Date: 2002/03/07 16:15:00 EST
30		Tag: blume-20020307-controls
31		Description:
32
33		This is a very large but mostly boring patch which makes (almost)
34		every tuneable compiler knob (i.e., pretty much everything under
35		Control.* plus a few other things) configurable via both the command
36		line and environment variables in the style CM did its configuration
37		until now.
38
39		Try starting sml with '-h' (or, if you are brave, '-H')
40
41		To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
42		implements the underlying generic mechanism.
43
44		The interface to some of the existing such facilities has changed somewhat.
45		For example, the MLRiscControl module now provides mkFoo instead of getFoo.
46		(The getFoo interface is still there for backward-compatibility, but its
47		use is deprecated.)
48
49		The ml-build script passes -Cxxx=yyy command-line arguments through so
50		that one can now twiddle the compiler settings when using this "batch"
51		compiler.
52
53		TODO items:
54
55		We should go through and throw out all controls that are no longer
56		connected to anything.  Moreover, we should go through and provide
57		meaningful (and correct!) documentation strings for those controls
58		that still are connected.
59
60		Currently, multiple calls to Controls.new are accepted (only the first
61		has any effect).  Eventually we should make sure that every control
62		is being made (via Controls.new) exactly once.  Future access can then
63		be done using Controls.acc.
64
65		Finally, it would probably be a good idea to use the getter-setter
66		interface to controls rather than ref cells.  For the time being, both
67		styles are provided by the Controls module, but getter-setter pairs are
68		better if thread-safety is of any concern because they can be wrapped.
69
70		*****************************************
71
72		One bug fix: The function blockPlacement in three of the MLRISC
73		backpatch files used to be hard-wired to one of two possibilities at
74		link time (according to the value of the placementFlag).  But (I
75		think) it should rather sense the flag every time.
76
77		*****************************************
78
79		Other assorted changes (by other people who did not supply a HISTORY entry):
80
81		1. the cross-module inliner now works much better (Monnier)
82		2. representation of weights, frequencies, and probabilities in MLRISC
83		changed in preparation of using those for weighted block placement
84		(Reppy, George)
85
86		----------------------------------------------------------------------
87		Name: Lal George
88		Date: 2002/03/07 14:44:24 EST 2002
89		Tag: george-20020307-weighted-block-placement
90
91		Tested the weighted block placement optimization on all architectures
92		(except the hppa) using AMPL to generate the block and edge frequencies.
93		Changes were required in the machine properties to correctly
94		categorize trap instructions. There is an MLRISC flag
95		"weighted-block-placement" that can be used to enable weighted block
96		placement, but this will be ineffective without block/edge
97		frequencies (coming soon).
98
99
100		----------------------------------------------------------------------
101		Name: Lal George
102		Date: 2002/03/05 17:24:48 EST
103		Tag: george-20020305-linkage-cluster
104
105		In order to support the block placement optimization, a new cluster
106		is generated as the very first cluster (called the linkage cluster).
107		It contains a single jump to the 'real' entry point for the compilation
108		unit. Block placement has no effect on the linkage cluster itself, but
109		all the other clusters  have full freedom in the manner in which they
110		reorder blocks or functions.
111
112		On the x86 the typical linkage code that is generated is:
113		----------------------
114		.align 2
115		L0:
116		addl    $L1-L0, 72(%esp)
117		jmp     L1
118
119
120		.align  2
121		L1:
122		----------------------
123
124		72(%esp) is the memory location for the stdlink register. This
125		must contain the address of the CPS function being called. In the
126		above example, it contains the address of  L0; before
127		calling L1 (the real entry point for the compilation unit), it
128		must contain the address for L1, and hence
129
130		addl $L1-L0, 72(%esp)
131
132		I have tested this on all architectures except the hppa.The increase
133		in code size is of course negligible
134
135		----------------------------------------------------------------------
136		Name: Allen Leung
137		Date: 2002/03/03 13:20:00 EST
138		Tag: leunga-20020303-mlrisc-tools
139
140		Added #[ ... ] expressions to mlrisc tools
141
142		----------------------------------------------------------------------
143		Name: Matthias Blume
144		Date: 2002/02/27 12:29:00 EST
145		Tag: blume-20020227-cdebug
146		Description:
147
148		- made types in structure C and C_Debug to be equal
149		- got rid of code duplication (c-int.sml vs. c-int-debug.sml)
150		- there no longer is a C_Int_Debug (C_Debug is directly derived from C)
151
152		----------------------------------------------------------------------
153		Name: Matthias Blume
154		Date: 2002/02/26 12:00:00 EST
155		Tag: blume-20020226-ffi
156		Description:
157
158		1. Fixed a minor bug in CM's "noweb" tool:
159		If numbering is turned off, then truly don't number (i.e., do not
160		supply the -L option to noweb).  The previous behavior was to supply
161		-L'' -- which caused noweb to use the "default" line numbering scheme.
162		Thanks to Chris Richards for pointing this out (and supplying the fix).
163
164		2. Once again, I reworked some aspects of the FFI:
165
166		A. The incomplete/complete type business:
167
168		- Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
169		gone!
170		- ML types representing an incomplete type are now *equal* to
171		ML types representing their corresponding complete types (just like
172		in C).  This is still safe because ml-nlffigen will not generate
173		RTTI for incomplete types, nor will it generate functions that
174		require access to such RTTI.   But when ML code generated from both
175		incomplete and complete versions of the C type meet, the ML types
176		are trivially interoperable.
177
178		NOTE:  These changes restore the full generality of the translation
179		(which was previously lost when I eliminated functorization)!
180
181		B. Enum types:
182
183		- Structure C now has a type constructor "enum" that is similar to
184		how the "su" constructor works.  However, "enum" is not a phantom
185		type because each "T enum" has values (and is isomorphic to
186		MLRep.Signed.int).
187		- There are generic access operations for enum objects (using
188		MLRep.Signed.int).
189		- ml-nlffigen will generate a structure E_foo for each "enum foo".
190		* The structure contains the definition of type "mlrep" (the ML-side
191		representation type of the enum).  Normally, mlrep is the same
192		as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
193		then mlrep will be defined as a datatype -- thus facilitating
194		pattern matching on mlrep values.
195		("-ec" will be suppressed if there are duplicate values in an
196		enumeration.)
197		* Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
198		will be generated for each C enum constant xxx.
199		* Conversion functions m2i and i2m convert between mlrep and
200		MLRep.Signed.int.  (Without "-ec", these functions are identities.)
201		* Coversion functions c and ml convert between mlrep and "tag enum".
202		* Access functions (get/set) fetch and store mlrep values.
203		- By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
204		enumerations are merged into one single enumeration represented by
205		structure E_'.
206
207		----------------------------------------------------------------------
208		Name: Allen Leung
209		Date: 2002/02/25 04:45:00 EST
210		Tag: leunga-20020225-cps-spill
211
212		This is a new implementation of the CPS spill phase.
213		The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
214		In case of problems, replace it with the old file spill.sml
215
216		The current compiler runs into some serious performance problems when
217		constructing a large record.  This can happen when we try to compile a
218		structure with many items.  Even a very simple structure like the following
219		makes the compiler slow down.
220
221		structure Foo = struct
222		val x_1 = 0w1 : Word32.int
223		val x_2 = 0w2 : Word32.int
224		val x_3 = 0w3 : Word32.int
225		...
226		val x_N = 0wN : Word32.int
227		end
228
229		The following table shows the compile time, from N=1000 to N=4000,
230		with the old compiler:
231
232		N
233		1000   CPS 100 spill                           0.04u  0.00s  0.00g
234		MLRISC ra                               0.06u  0.00s  0.05g
235		(spills = 0 reloads = 0)
236		TOTAL                                   0.63u  0.07s  0.21g
237
238		1100   CPS 100 spill                           8.25u  0.32s  0.64g
239		MLRISC ra                               5.68u  0.59s  3.93g
240		(spills = 0 reloads = 0)
241		TOTAL                                   14.71u  0.99s  4.81g
242
243		1500   CPS 100 spill                           58.55u  2.34s  1.74g
244		MLRISC ra                               5.54u  0.65s  3.91g
245		(spills = 543 reloads = 1082)
246		TOTAL                                   65.40u  3.13s  6.00g
247
248		2000   CPS 100 spill                           126.69u  4.84s  3.08g
249		MLRISC ra                               0.80u  0.10s  0.55g
250		(spills = 42 reloads = 84)
251		TOTAL                                   129.42u  5.10s  4.13g
252
253		3000   CPS 100 spill                           675.59u  19.03s  11.64g
254		MLRISC ra                               2.69u  0.27s  1.38g
255		(spills = 62 reloads = 124)
256		TOTAL                                   682.48u  19.61s  13.99g
257
258		4000   CPS 100 spill                           2362.82u  56.28s  43.60g
259		MLRISC ra                               4.96u  0.27s  2.72g
260		(spills = 85 reloads = 170)
261		TOTAL                                   2375.26u  57.21s  48.00g
262
263		As you can see the old cps spill module suffers from some serious
264		performance problem.  But since I cannot decipher the old code fully,
265		instead of patching the problems up, I'm reimplementing it
266		with a different algorithm.  The new code is more modular,
267		smaller when compiled, and substantially faster
268		(O(n log n) time and O(n) space).  Timing of the new spill module:
269
270		4000  CPS 100 spill                           0.02u  0.00s  0.00g
271		MLRISC ra                               0.25u  0.02s  0.15g
272		(spills=1 reloads=3)
273		TOTAL                                   7.74u  0.34s  1.62g
274
275		Implementation details:
276
277		As far as I can tell, the purpose of the CPS spill module is to make sure the
278		number of live variables at any program point (the bandwidth)
279		does not exceed a certain limit, which is determined by the
280		size of the spill area.
281
282		When the bandwidth is too large, we decrease the register pressure by
283		packing live variables into spill records.  How we achieve this is
284		completely different than what we did in the old code.
285
286		First, there is something about the MLRiscGen code generator
287		that we should be aware of:
288
289		o MLRiscGen performs code motion!
290
291		In particular, it will move floating point computations and
292		address computations involving only the heap pointer to
293		their use sites (if there is only a single use).
294		What this means is that if we have a CPS record construction
295		statement
296
297		RECORD(k,vl,w,e)
298
299		we should never count the new record address w as live if w
300		has only one use (which is often the case).
301
302		We should do something similar to floating point, but the transformation
303		there is much more complex, so I won't deal with that.
304
305		Secondly, there are now two new cps primops at our disposal:
306
307		1. rawrecord of record_kind option
308		This pure operator allocates some uninitialized storage from the heap.
309		There are two forms:
310
311		rawrecord NONE [INT n]  allocates a tagless record of length n
312		rawrecord (SOME rk) [INT n] allocates a tagged record of length n
313		and initializes the tag.
314
315		2. rawupdate of cty
316		rawupdate cty (v,i,x)
317		Assigns to x to the ith component of record v.
318		The storelist is not updated.
319
320		We use these new primops for both spilling and increment record construction.
321
322		1. Spilling.
323
324		This is implemented with a linear scan algorithm (but generalized
325		to trees).  The algorithm will create a single spill record at the
326		beginning of the cps function and use rawupdate to spill to it,
327		and SELECT or SELp to reload from it.  So both spills and reloads
328		are fine-grain operations.  In contrast, in the old algorithm
329		"spills" have to be bundled together in records.
330
331		Ideally, we should sink the spill record construction to where
332		it is needed.  We can even split the spill record into multiple ones
333		at the places where they are needed.  But CPS is not a good
334		representation for global code motion, so I'll keep it simple and
335		am not attempting this.
336
337		2. Incremental record construction (aka record splitting).
338
339		Long records with many component values which are simulatenously live
340		(recall that single use record addresses are not considered to
341		be live) are constructed with rawrecord and rawupdate.
342		We allocate space on the heap with rawrecord first, then gradually
343		fill it in with rawupdate.  This is the technique suggested to me
344		by Matthias.
345
346		Some restrictions on when this is applicable:
347		1. It is not a VECTOR record.  The code generator currently does not handle
348		this case. VECTOR record uses double indirection like arrays.
349		2. All the record component values are defined in the same "basic block"
350		as the record constructor.  This is to prevent speculative
351		record construction.
352
353		----------------------------------------------------------------------
354		Name: Allen Leung
355		Date: 2002/02/22 01:02:00 EST
356		Tag: leunga-20020222-mlrisc-tools
357
358		Minor bug fixes in the parser and rewriter
359
360		----------------------------------------------------------------------
361		Name: Allen Leung
362		Date: 2002/02/21 20:20:00 EST
363		Tag: leunga-20020221-peephole
364
365		Regenerated the peephole files.  Some contained typos in the specification
366		and some didn't compile because of pretty printing bugs in the old version
367		of 'nowhere'.
368
369		----------------------------------------------------------------------
370		Name: Allen Leung
371		Date: 2002/02/19 20:20:00 EST
372		Tag: leunga-20020219-mlrisc-tools
373		Description:
374
375		Minor bug fixes to the mlrisc-tools library:
376
377		1.  Fixed up parsing colon suffixed keywords
378		2.  Added the ability to shut the error messages up
379		3.  Reimplemented the pretty printer and fixed up/improved
380		the pretty printing of handle and -> types.
381		4.  Fixed up generation of literal symbols in the nowhere tool.
382		5.  Added some SML keywords to to sml.sty
383
384		----------------------------------------------------------------------
385		Name: Matthias Blume
386		Date: 2002/02/19 16:20:00 EST
387		Tag: blume-20020219-cmffi
388		Description:
389
390		A wild mix of changes, some minor, some major:
391
392		* All C FFI-related libraries are now anchored under $c:
393		$/c.cm      --> $c/c.cm
394		$/c-int.cm  --> $c/internals/c-int.cm
395		$/memory.cm --> $c/memory/memory.cm
396
397		* "make" tool (in CM) now treats its argument pathname slightly
398		differently:
399		1. If the native expansion is an absolute name, then before invoking
400		the "make" command on it, CM will apply OS.Path.mkRelative
401		(with relativeTo = OS.FileSys.getDir()) to it.
402		2. The argument will be passed through to subsequent phases of CM
403		processing without "going native".  In particular, if the argument
404		was an anchored path, then "make" will not lose track of that anchor.
405
406		* Compiler backends now "know" their respective C calling conventions
407		instead of having to be told about it by ml-nlffigen.  This relieves
408		ml-nlffigen from one of its burdens.
409
410		* The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
411
412		* Export C_DEBUG and C_Debug from $c/c.cm.
413
414		* C type encoding in ml-nlffi-lib has been improved to model the conceptual
415		subtyping relationship between incomplete pointers and their complete
416		counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
417		with the convention of instantiating 'o with ('t, 'c) obj whenever
418		the pointer target type is complete.  In the incomplete case, 'o
419		will be instantiated with some "'c iobj" -- a type obtained by
420		using one of the functors PointerToIncompleteType or PointerToCompleteType.
421
422		Operations that work on both incomplete and complete pointer types are
423		typed as taking an 'o ptr while operations that require the target to
424		be known are typed as taking some ('t, 'c) obj ptr.
425
426		voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
427		where void is an eqtype without any values.  This makes it possible
428		to work on voidptr values using functions meant to operate on light
429		incomplete pointers.
430
431		* As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
432		been vastly simplified.
433
434		----------------------------------------------------------------------
435		Name: Matthias Blume
436		Date: 2002/02/19 10:48:00 EST
437		Tag: blume-20020219-pqfix
438		Description:
439
440		Applied Chris Okasaki's bug fix for priority queues.
441
442		----------------------------------------------------------------------
443		Name: Matthias Blume
444		Date: 2002/02/15 17:05:00
445		Tag: Release_110_39
446		Description:
447
448		Last-minute retagging is becoming a tradition... :-(
449
450		This is the working release 110.39.
451
452		----------------------------------------------------------------------
453		Name: Matthias Blume
454		Date: 2002/02/15 16:00:00 EST
455		Tag: Release_110_39-orig
456		Description:
457
458		Working release 110.39.  New bootfiles.
459
460		(Update: There was a small bug in the installer so it wouldn't work
461		with all shells.  So I retagged. -Matthias)
462
463		----------------------------------------------------------------------
464		Name: Matthias Blume
465		Date: 2002/02/15 14:17:00 EST
466		Tag: blume-20020215-showbindings
467		Description:
468
469		Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
470		the latter can be useful for exploring what bindings are available at
471		the interactive prompt.  (The first function returns only the list
472		of symbols that are really bound, the second prints those but also the
473		ones that CM's autoloading mechanism knows about.)
474
475		----------------------------------------------------------------------
476		Name: Matthias Blume
477		Date: 2002/02/15 12:08:00 EST
478		Tag: blume-20020215-iptrs
479		Description:
480
481		Two improvements to ml-nlffigen:
482
483		1. Write files only if they do not exist or if their current contents
484		do not coincide with what's being written.  (That is, avoid messing
485		with the time stamps unless absolutely necessary.)
486
487		2. Implement a "repository" mechanism for generated files related
488		to "incomplete pointer types".   See the README file for details.
489
490		----------------------------------------------------------------------
491		Name: Matthias Blume
492		Date: 2002/02/14 11:50:00 EST
493		Tag: blume-20020214-quote
494		Description:
495
496		Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
497		because of the new and improved tag generation scheme.  (Thanks to Allen
498		Leung for pointing it out.)
499
500		----------------------------------------------------------------------
501		Name: Lal George
502		Date: 2002/02/14 09:55:27 EST 2002
503		Tag: george-20020214-isabelle-bug
504		Description:
505
506		Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
507		of Isabelle on the x86.
508
509		From Allen:
510		-----------
511		I've found the problem:
512
513		in ra-core.sml, I use the counter "blocked" to keep track of the
514		true number of elements in the freeze queue.  When the counter goes
515		to zero, I skip examining the queue.  But I've messed up the
516		bookkeeping in combine():
517
518		else ();
519		case !ucol of
520		PSEUDO => (if !cntv > 0 then
521		(if !cntu > 0 then blocked := !blocked - 1 else ();
522		^^^^^^^^^^^^^^^^^^^^^^^
523		moveu := mergeMoveList(!movev, !moveu)
524		)
525		else ();
526
527		combine() is called to coalesce two nodes u and v.
528		I think I was thinking that if the move counts of u and v are both
529		greater than zero then after they are coalesced then one node is
530		removed from the freeze queue.  Apparently I was thinking that
531		both u and v are of low degree, but that's clearly not necessarily true.
532
533
534		02/12/2002:
535		Here's the patch.  HOL now compiles.
536
537		I don't know how this impact on performance (compile
538		time or runtime).  This bug caused the RA (especially on the x86)
539		to go thru the potential spill phase when there are still nodes on the
540		freeze queue.
541
542
543
544
545		----------------------------------------------------------------------
546		Name: Matthias Blume
547		Date: 2002/02/13 22:40:00 EST
548		Tag: blume-20020213-fptr-rtti
549		Description:
550
551		Fixed a bug in ml-nlffigen that was introduced with one of the previous
552		updates.
553
554		----------------------------------------------------------------------
555		Name: Matthias Blume
556		Date: 2002/02/13 16:41:00 EST
557		Tag: blume-20020213-cmlpq
558		Description:
559
560		Added new priority queue export symbols (which have just been added to
561		smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
562		not compile and the installer would choke.)
563
564		----------------------------------------------------------------------
565		Name: Matthias Blume
566		Date: 2002/02/13 16:15:00 EST
567		Tag: blume-20020213-various
568		Description:
569
570		1. More tweaks to ml-nlffigen:
571
572		- better internal datastructures (resulting in slight speedup)
573		- "-match" option requires exact match
574		- "localized" gensym counters (untagged structs/unions nested within
575		other structs/unions or within typedefs get a fresh counter; their
576		tag will be prefixed by a concatenation of their parents' tags)
577		- bug fixes (related to calculation of transitive closure of types
578		to be included in the output)
579
580		2. Minor Basis updates:
581
582		- added implementations for List.collate and Option.app
583
584		----------------------------------------------------------------------
585		Name: Matthias Blume
586		Date: 2002/02/11 15:55:00 EST
587		Tag: blume-20020211-gensym
588		Description:
589
590		Added a "-gensym" option to command line of ml-nlffigen.  This can be
591		used to specify a "stem" -- a string that is inserted in all "gensym'd"
592		names (ML structure names that correspond to unnamed C structs, unions,
593		and enums), so that separate runs of ml-nlffigen do not clash.
594
595		----------------------------------------------------------------------
596		Name: Matthias Blume
597		Date: 2002/02/11 12:05:00 EST
598		Tag: blume-20020211-gensml
599		Description:
600
601		A quick fix for a problem with GenSML (in the pgraph-util library):
602		Make generation of toplevel "local" optional.  (Strictly speaking,
603		signature definitions within "local" are not legal SML.)
604
605		Other than that: updates to INSTALL and cm/TODO.
606
607		----------------------------------------------------------------------
608		Name: Matthias Blume
609		Date: 2002/02/08 15:00:00 EST
610		Tag: blume-20020208-uniquepid
611		Description:
612
613		0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
614
615		1. The installer (config/install.sh) has gotten smarter:
616
617		- Configuration options are a bit easier to specify now
618		(in config/targets).
619		- Bug in recognizing .tar.bz2 files fixed.
620		- Installer automatically resolves dependencies between
621		configuration options (e.g., if you ask for eXene, you will
622		also get cml -- regardless whether you asked for it or not).
623		- Installer can run in "quieter mode" by setting the environment
624		variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
625		"completely silent", though.
626		- Build HashCons library as part of smlnj-lib.
627
628		2. A new scheme for assigning persistent identifiers to compilation
629		units (and, by extension, to types etc.) has been put into place.
630		This fixes a long-standing bug where types and even dynamic values
631		can get internally confused, thereby compromising type safety
632		(abstraction) and dynamic correctness.  See
633
634		http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
635
636		for an example of how things could go wrong until now.
637
638		The downside of the new scheme is that pids are not quite as
639		persistent as they used to be: CM will generate a fresh pid
640		for every compilation unit that it thinks it sees for the first
641		time.  That means that if you compile starting from a clean, fresh
642		source tree at two different times, you end up with different
643		binaries.
644
645		Cutoff recompilation, however, has not been compromised because
646		CM keeps pid information in special caches between runs.
647
648		----------------------------------------------------------------------
649		Name: Lal George
650		Date: 2002/02/07 15:34:13 EST 2002
651		Tag: <none>
652		Description:
653
654		Compilers that generate assembly code may produce  global labels
655		whose value is resolved at link time. The various peephole optimization
656		modules did not take this in account.
657
658		TODO. The Labels.addrOf function should really return an option
659		type so that clients are forced to deal with this issue, rather
660		than an exception being raised.
661
662		----------------------------------------------------------------------
663		Name: Lal George
664		Date: 2002/02/06 13:55:02 EST
665		Tag: george-20020206-ra-breakup
666		Description:
667
668		1. A bug fix from Allen.
669
670		A typo causes extra fstp %st(0)'s to be generated at compensation
671		edges, which might cause stack underflow traps at runtime.  This
672		occurs in fft where there are extraneous fstps right before the 'into'
673		trap instruction (in this case they are harmless since none of the
674		integers overflow.)
675
676		2. Pulled out various utility modules that were embedded in the modules
677		of the register allocator. I need these modules for other purposes, but
678		they are not complete enough to put into a library (just yet).
679		----------------------------------------------------------------------
680		Name: Matthias Blume
681		Date: 2002/01/31 16:05:00 EST
682		Tag: blume-20020131-sparc-ccalls
683		Description:
684
685		1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
686		of extra stack space by mistake.  Fixed.
687
688		2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
689
690		----------------------------------------------------------------------
691		Name: Allen Leung
692		Date: 2002/01/30
693		Tag: leunga-20020130-nowhere-bug-fix
694		Description:
695
696		MLRISC bug fixes:
697		1. Fixed a bindings computation bug in the 'nowhere' program generator tool.
698		2. MachineInt.fromString was negating its value.
699
700		----------------------------------------------------------------------
701		Name: Matthias Blume
702		Date: 2002/01/29
703		Tag: blume-20020129-INSTALL
704		Description:
705
706		- Added somewhat detailed installation instructions (file INSTALL).
707		- Fixed curl-detection bug in config/install.sh.
708		- It is now possible to select the URL getter using the URLGETTER
709		environment variable:
710
711		not set / "unknown"      --> automatic detection (script tries wget,
712		curl, and lynx)
713		"wget" / "curl" / "lynx" --> use the specified program (script "knows"
714		how to properly invoke them)
715		other                    --> use $URLGETTER directly, it must take
716		precisely two command-line arguments
717		(source URL and destination file name)
718
719		----------------------------------------------------------------------
720		Name: Matthias Blume
721		Date: 2002/01/28
722		Tag: blume-20020128-sparc-ccalls
723		Description:
724
725		- Fixed problem with calculation of "used" registers in sparc-c-calls.
726		- Make use of the allocParam argument in sparc-c-calls.
727
728		----------------------------------------------------------------------
729		Name: Matthias Blume
730		Date: 2002/01/28
731		Tag: blume-20020128-allocParam
732		Description:
733
734		John Reppy:  Changes c-calls API to accept client-callback for
735		allocating extra stack space.
736		me: Corresponding changes to mlriscGen (using a dummy argument that
737		does not change the current behavior).
738
739		----------------------------------------------------------------------
740		Name: Matthias Blume
741		Date: 2002/01/28 12:00:00
742		Tag: Release_110_38
743		Description:
744
745		This time for real!!!
746
747		----------------------------------------------------------------------
748		Name: Matthias Blume
749		Date: 2002/01/28 10:56:00 EST
750		Tag: blume-20020128-retraction
751		Description:
752
753		0. Retracted earlier 110.38.  (The Release_110_38 tag has been replaced
754		with blume-Release_110_38-retracted.)
755
756		1. Fixed a problem with incorrect rounding modes in real64.sml.
757		(Thanks to Andrew Mccreight <andrew.mccreight@yale.edu>.)
758
759		2. A bug in ml-nlffigen related to the handling of unnamed structs, unions,
760		and enums fixed.  The naming of corresponding ML identifiers should
761		now be consistent again.
762
763		----------------------------------------------------------------------
764		Name: Allen Leung
765		Date: 2002/01/27
766		Tag: leunga-20020127-nowhere
767		Description:
768
769		Added a target called nowhere in the configuration scripts.
770		Enabling this will build the MLRISC 'nowhere' tool (for translating
771		programs with where-clauses into legal SML code) during installation.
772
773		----------------------------------------------------------------------
774		Name: Matthias Blume
775		Date: 2002/01/25 21:27:00 EST
776		Tag: blume-Release_110_38-retracted
777		Description:
778
779		Call it a (working) release!  Version is 110.38. Bootfiles are ready.
780
781		README will be added later.
782
783		!!! NOTE:  Re-tagged as blume-Release_110_38-retracted. Original tag
784		(Release_110_38) removed.  Reason: Last-minute bug fixes.
785
786		----------------------------------------------------------------------
787		Name: Matthias Blume
788		Date: 2002/01/25
789		Tag: blume-20020125-ffi
790		Description:
791
792		A large number of tweaks and improvements to ml-nlffi-lib and
793		ml-nlffigen:
794
795		- ML represenation types have been streamlined
796		- getter and setter functions work with concrete values, not abstract
797		ones where possible
798		- ml-nlffigen command line more flexible (see README file there)
799		- some bugs have been fixed (hopefully)
800
801		----------------------------------------------------------------------
802		Name: Lal George
803		Date: 2002/01/24
804		Tag: george-20020124-risc-ra-interface
805		Description:
806
807		There is a dramatic simplification in the interface to the
808		register allocator for RISC architectures as a result of making
809		parallel copy instructions explicit.
810
811		----------------------------------------------------------------------
812		Name: Matthias Blume
813		Date: 2002/01/22
814		Tag: blume-20020122-x86-ccalls
815		Description:
816
817		Bug fix for c-calls on x86 (having to do with how char- and
818		short-arguments are being handled).
819
820		----------------------------------------------------------------------
821		Name: Matthias Blume
822		Date: 2002/01/21
823		Tag: blume-20020121-ff
824		Description:
825
826		Another day of fiddling with the FFI...
827
828		1. Bug fix/workaround:  CKIT does not complain about negative array
829		dimensions, so ml-nlffigen has to guard itself against this possibility.
830		(Otherwise a negative dimension would send it into an infinite loop.)
831
832		2. Some of the abstract types (light objects, light pointers, most "base"
833		types) in structure C are now eqtypes.
834
835		3. Added constructors and test functions for NULL function pointers.
836
837		----------------------------------------------------------------------
838		Name: Matthias Blume
839		Date: 2002/01/18
840		Tag: blume-20020118-ready-for-new-release
841		Description:
842
843		Made config/srcarchiveurl point to a new place.  (Will provide boot
844		files shortly.)
845
846		Maybe we christen this to be 110.38?
847
848		----------------------------------------------------------------------
849		Name: Matthias Blume
850		Date: 2002/01/18
851		Tag: blume-20020118-more-ffifiddle
852		Description:
853
854		Today's FFI fiddling:
855
856		- Provided a structure CGetSet with "convenient" versions of C.Get.* and
857		C.Set.* that use concrete (MLRep.*) arguments and results instead
858		of abstract ones.
859
860		- Provided word-style bit operations etc. for "int" representation
861		types in MLRep.S<Foo>Bitops where <Foo> ranges over Char, Int, Short,
862		and Long.
863
864		----------------------------------------------------------------------
865		Name: Matthias Blume
866		Date: 2002/01/18
867		Tag: blume-20020118-use-x86-fp
868		Description:
869
870		Now that x86-fast-fp seems to be working, I turned it back on again
871		by default.  (Seems to work fine now, even with the FFI.)
872
873		Other than that, I added some documentation about the FFI to
874		src/ml-nlffigen/README and updated the FFI test examples in
875		src/ml-nlffi-lib/Tests/*.
876
877		----------------------------------------------------------------------
878		Name: Allen Leung
879		Date: 2002/01/17
880		Tag: leunga-20020117-x86-fast-fp-call
881		Description:
882
883		1. Fixed a problem with handling return fp values when x86's fast fp
884		mode is turned on.
885
886		2. Minor pretty printing fix for cellset.  Print %st(0) as %st(0) instead
887		of %f32.
888
889		3. Added a constructor INT32lit to the ast of MLRISC tools.
890
891		----------------------------------------------------------------------
892		Name: Matthias Blume
893		Date: 2002/01/16
894		Tag: blume-20020116-ffifiddle
895		Description:
896
897		More fiddling with the FFI interface:
898
899		- Make constness 'c instead of rw wherever possible.  This eliminates
900		the need for certain explicit coercions.  (However, due to ML's
901		value polymorphism, there will still be many cases where explicit
902		coercions are necessary.  Phantom types are not the whole answer
903		to modeling a subtyping relationship in ML.)
904
905		- ro/rw coersions for pointers added.  (Avoids the detour through */&.)
906
907		- "printf" test example added to src/ml-nlffi-lib/Tests.  (Demonstrates
908		clumsy workaround for varargs problem.)
909
910		----------------------------------------------------------------------
911		Name: Lal George
912		Date: 2002/01/15
913		Tag: <none>
914		Description:
915
916		1. Since COPY instructions are no longer native to the architecture,
917		a generic functor can be used to implement the expandCopies function.
918
919		2. Allowed EXPORT and IMPORT pseudo-op declarations to appear inside a
920		TEXT segment.
921
922		----------------------------------------------------------------------
923		Name: Matthias Blume
924		Date: 2002/01/15
925		Tag: blume-20020115-ffiupdates
926		Description:
927
928		1. Fix for bug resulting in single-precision float values being returned
929		incorrectly from FFI calls.
930
931		2. Small modifications to C FFI API:
932
933		- memory-allocation routines return straight objects (no options)
934		and raise an exception in out-of-memory situations
935		- unsafe extensions to cast between function pointers and pointers
936		from/to ints
937		- added structure C_Debug as an alternative to structure C where
938		pointer-dereferencing (|*| and |*!) always check for null-pointers
939		- added open_lib' to DynLinkage;  open_lib' works like open_lib
940		but also takes a (possibly empty) list of existing library handles
941		that the current library depends on
942
943		----------------------------------------------------------------------
944		Name: Matthias Blume
945		Date: 2002/01/10
946		Tag: blume-20020110-newffigen
947		Description:
948
949		1. Updates to portable graph code.
950
951		2. Major update to ml-nlffigen and ml-nlffi-lib.  Things are much
952		more scalable now so that even huge interfaces such as the one
953		for GTK compile in finite time and space. :-)
954		See src/ml-nlffigen/README for details on what's new.
955
956		----------------------------------------------------------------------
957		Name: Lal George
958		Date: 2001/01/09 14:31:35 EST 2002
959		Tag: george-20011206-rm-native-copy
960		Description:
961
962		Removed the native COPY and FCOPY instructions
963		from all the architectures and replaced it with the
964		explicit COPY instruction from the previous commit.
965
966		It is now possible to simplify many of the optimizations
967		modules that manipulate copies. This has not been
968		done in this change.
969
970		----------------------------------------------------------------------
971		Name: Lal George
972		Date: 2001/12/06 16:50:13 EST 2001
973		Tag: george-20011206-mlrisc-instruction
974		Description:
975
976		Changed the representation of instructions from being fully abstract
977		to being partially concrete. That is to say:
978
979		from
980		type instruction
981
982		to
983		type instr                              (* machine instruction *)
984
985		datatype instruction =
986		LIVE of {regs: C.cellset, spilled: C.cellset}
987		| KILL of {regs: C.cellset, spilled: C.cellset}
988		| COPYXXX of {k: CB.cellkind, dst: CB.cell list, src: CB.cell list}
989		| ANNOTATION of {i: instruction, a: Annotations.annotation}
990		| INSTR of instr
991
992		This makes the handling of certain special instructions that appear on
993		all architectures easier and uniform.
994
995		LIVE and KILL say that a list of registers are live or killed at the
996		program point where they appear. No spill code is generated when an
997		element of the 'regs' field is spilled, but the register is moved to
998		the 'spilled' (which is present, more for debugging than anything else).
999
1000		LIVE replaces the (now deprecated) DEFFREG instruction on the alpha.
1001		We used to generate:
1002
1003		DEFFREG f1
1004		f1 := f2 + f3
1005		trapb
1006
1007		but now generate:
1008
1009		f1 := f2 + f3
1010		trapb
1011		LIVE {regs=[f1,f2,f3], spilled=[]}
1012
1013		Furthermore, the DEFFREG (hack) required that all floating point instruction
1014		use all registers mentioned in the instruction. Therefore f1 := f2 + f3,
1015		defines f1 and uses [f1,f2,f3]! This hack is no longer required resulting
1016		in a cleaner alpha implementation. (Hopefully, intel will not get rid of
1017		this architecture).
1018
1019		COPYXXX is intended to replace the parallel COPY and FCOPY  available on
1020		all the architectures. This will result in further simplification of the
1021		register allocator that must be aware of them for coalescing purposes, and
1022		will also simplify certain aspects of the machine description that provides
1023		callbacks related to parallel copies.
1024
1025		ANNOTATION should be obvious, and now INSTR represents the honest to God
1026		machine instruction set!
1027
1028		The <arch>/instructions/<arch>Instr.sml files define certain utility
1029		functions for making porting easier -- essentially converting upper case
1030		to lower case. All machine instructions (of type instr) are in upper case,
1031		and the lower case form generates an MLRISC instruction. For example on
1032		the alpha we have:
1033
1034		datatype instr =
1035		LDA of {r:cell, b:cell, d:operand}
1036		| ...
1037
1038		val lda : {r:cell, b:cell, d:operand} -> instruction
1039		...
1040
1041		where lda is just (INSTR o LDA), etc.
1042
1043		----------------------------------------------------------------------
1044		Name: Matthias Blume
1045		Date: 2001/11/22 21:40:00 EST
1046		Tag: Release_110_37
1047		Description:
1048
1049		Release 110.37.  This time for real.
1050
1051		----------------------------------------------------------------------
1052		Name: Matthias Blume
1053		Date: 2001/11/21 16:35:00 EST
1054		Tag: blume-20011121-foot-in-mouth
1055		Description:
1056
1057		Removed the "Release_110_37" tag because of a serious bug.
1058		This will be re-tagged once the bug is fixed.
1059
1060		----------------------------------------------------------------------
1061		Name: Matthias Blume
1062		Date: 2001/11/21 16:14:00 EST
1063		Tag: blume-20011121-forgottenfile
1064		Description:
1065
1066		Forgot to add a file.  (Just a .tex-file -- part of
1067		the CM manual source.)
1068
1069		----------------------------------------------------------------------
1070		Name: Matthias Blume
1071		Date: 2001/11/21 16:10:00 EST
1072		Tag: blume-20011121-invalid_110_37
1073		Description:
1074
1075		Note: I removed the original tag "Release_110_37" from this commit
1076		because we found a serious bug in all non-x86 backends.
1077		- Matthias
1078
1079		1. Modifications to the SML/NJ code generator and to the runtime system
1080		so that code object name strings are directly inserted into code
1081		objects at code generation time.  The only business the runtime system
1082		has with this is now to read the name strings on occasions.
1083		(The encoding of the name string has also changed somewhat.)
1084
1085		2. CM now implements a simple "set calculus" for specifying export lists.
1086		In particular, it is now possible to refer to the export lists of
1087		other libraries/groups/sources and form unions as well as differences.
1088		See the latest CM manual for details.
1089
1090		3. An separate notion of "proxy" libraries has again be eliminated from
1091		CM's model.  (Proxy libraries are now simply a special case of using
1092		the export list calculus.)
1093
1094		4. Some of the existing libraries now take advantage of the new set
1095		calculus.
1096		(Notice that not all libraries have been converted because some
1097		of the existing .cm-files are supposed to be backward compatible
1098		with 110.0.x.)
1099
1100		5. Some cleanup in stand-alone programs.  (Don't use "exnMessage" -- use
1101		"General.exnMessage"!  The former relies on a certain hook to be
1102		initialized, and that often does not happen in the stand-alone case.)
1103
1104		----------------------------------------------------------------------
1105		Name: Lal George
1106		Date: 2001/11/21  13:56:18 EST
1107		Tag: george-2001121-pseudo-ops
1108		Description:
1109
1110		Implemented a complete redesign of MLRISC pseudo-ops. Now there
1111		ought to never be any question of incompatabilities with
1112		pseudo-op syntax expected by host assemblers.
1113
1114		For now, only modules supporting GAS syntax are implemented
1115		but more should follow, such as MASM, and vendor assembler
1116		syntax, e.g. IBM as, Sun as, etc.
1117
1118		----------------------------------------------------------------------
1119		Name: Matthias Blume
1120		Date: 2001/11/14 11:52:00 EST
1121		Tag: blume-20011114-srcname
1122		Description:
1123
1124		1. Routed the name of the current source file to mlriscgen where it
1125		should be directly emitted into the code object.  (This last part
1126		is yet to be done.)
1127
1128		2. Some cleanup of the pgraph code to make it match the proposal that
1129		I put out the other day.  (The proposal notwithstanding, things are
1130		still in flux here.)
1131
1132		----------------------------------------------------------------------
1133		Name: Lal George
1134		Date: 2001/11/14 09:44:04 EST
1135		Tag:
1136		Description:
1137
1138		Fix for a backpatching bug reported by Allen.
1139
1140		Because the boundary between short and long span-dependent
1141		instructions is +/- 128, there are an astounding number of
1142		span-dependent instructions whose size is over estimated.
1143
1144		Allen came up with the idea of letting the size of span
1145		dependent instructions be non-monotonic, for a maxIter
1146		number of times, after which the size must be monotonically
1147		increasing.
1148
1149		This table shows the number of span-dependent instructions
1150		whose size was over-estimated as a function of maxIter, for the
1151		file Parse/parse/ml.grm.sml:
1152
1153		maxIter            # of instructions:
1154		10                      687
1155		20                      438
1156		30                      198
1157		40                        0
1158
1159		In compiling the compiler, there is no significant difference in
1160		compilation speed between maxIter=10 and maxIter=40. Actually,
1161		my measurements showed that maxIter=40 was a tad faster than
1162		maxIter=10! Also 96% of the  files in the compiler reach a fix
1163		point within 13 iterations, so fixing maxIter at 40, while high,
1164		is okay.
1165
1166		----------------------------------------------------------------------
1167		Name: Matthias Blume
1168		Date: 2001/10/31 15:25:00 EST
1169		Tag: blume-20011031-pgraph
1170		Description:
1171
1172		CKIT:
1173		* Changed the "Function" constructor of type Ast.ctype to carry optional
1174		argument identifiers.
1175		* Changed the return type of TypeUtil.getFunction accordingly.
1176		* Type equality ignores the argument names.
1177		* TypeUtil.composite tries to preserve argument names but gives up quickly
1178		if there is a mismatch.
1179
1180		installation script:
1181		* attempts to use "curl" if available (unless "wget" is available as well)
1182
1183		CM:
1184		* has an experimental implementation of "portable graphs" which I will
1185		soon propose as an implementation-independent library format
1186		* there are also new libraries $/pgraph.cm and $/pgraph-util.cm
1187
1188		NLFFI-LIB:
1189		* some cleanup (all cosmetic)
1190
1191		NLFFIGEN:
1192		* temporarily disabled the mechanism that suppresses ML output for
1193		C definitions whose identifiers start with an underscore character
1194		* generate val bindings for enum constants
1195		* user can request that only one style (light or heavy) is being used;
1196		default is to use both (command-line arguments: -heavy and -light)
1197		* fixed bug in handling of function types involving incomplete pointers
1198		* generate ML entry points that take record arguments (i.e., using
1199		named arguments) for C functions that have a prototype with named
1200		arguments
1201		(see changes to CKIT)
1202
1203		----------------------------------------------------------------------
1204		Name: Allen Leung
1205		Date: 2001/10/27 20:34:00 EDT
1206		Tag: leunga-20011027-x86-fast-fp-call
1207		Description:
1208
1209		Fixed the bug described in blume-20010920-slowfp.
1210
1211		The fix involves
1212		1. generating FCOPYs in FSTP in ia32-svid
1213		2. marking a CALL with the appropriate annotation
1214
1215		----------------------------------------------------------------------
1216		Name: Matthias Blume
1217		Date: 2001/10/16 11:32:00 EDT
1218		Tag: blume-20011016-netbsd
1219		Description:
1220
1221		Underscore patch from Chris Richards (fixing problem with compiling
1222		runtime system under recent NetBSD).
1223
1224		----------------------------------------------------------------------
1225		Name: Allen Leung
1226		Date: 2001/10/12 17:18:32 EDT 2001
1227		Tag: leung-20011012-x86-printflowgraph
1228		Description:
1229
1230		X86RA now uses a valid (instead of dummy) PrintFlowgraph module.
1231
1232		----------------------------------------------------------------------
1233		Name: Lal George
1234		Date: 2001/10/11 23:51:34 EDT
1235		Tag: george-20011011-too-many-instrs
1236		Description:
1237
1238		The representation of a program point never expected to see more
1239		than 65536 instructions in a basic block!
1240
1241		----------------------------------------------------------------------
1242		Name: Lal George
1243		Date: 2001/10/09 09:41:37 EDT
1244		Tag: george-20011008-mlrisc-labels
1245		Description:
1246
1247		Changed the machine description files to support printing of
1248		local and global labels in assembly code, based on host assembler
1249		conventions.
1250
1251		----------------------------------------------------------------------
1252		Name: Matthias Blume
1253		Date: 2001/09/25 15:25:00 EDT
1254		Tag: blume-20010925-exninfo
1255		Description:
1256
1257		I provided a non-hook implementation of exnName (at the toplevel) and
1258		made the "dummy" implementation of exnMessage (at the toplevel) more
1259		useful: if nothing gets "hooked in", then at least you are going to
1260		see the exception name and a message indicating why you don't see more.
1261
1262		[For the time being, programs that need exnMessage and want to use
1263		ml-build should either use General.exnMessage (strongly recommended) or
1264		refer to structure General at some other point so that CM sees a
1265		static dependency.]
1266
1267		[Similar remarks go for "print" and "use":  If you want to use their
1268		functionality in stand-alone programs generated by ml-build, then use
1269		TextIO.output and Backend.Interact.useFile (from $smlnj/compiler.cm).]
1270
1271		----------------------------------------------------------------------
1272		Name: Matthias Blume
1273		Date: 2001/09/20 17:28:00 EDT
1274		Tag: blume-20010920-slowfp
1275		Description:
1276
1277		Allen says that x86-fast-fp is not safe yet, so I turned it off again...
1278
1279		----------------------------------------------------------------------
1280		Name: Matthias Blume
1281		Date: 2001/09/20 17:20:00 EDT
1282		Tag: blume-20010920-canonicalpaths
1283		Description:
1284
1285		0. Updated the BOOT file (something that I forgot to do earlier).
1286
1287		1. Small internal change to CM so that it avoids "/../" in filenames
1288		as much as possible (but only where it is safe).
1289
1290		2. Changed config/_run-sml (resulting in a changed bin/.run-sml) so
1291		that arguments that contain delimiters are passed through correctly.
1292		This change also means that all "special" arguments of the form
1293		@SMLxxx... must come first.
1294
1295		3. Changed install script to put relative anchor names for tool commands
1296		into pathconfig.
1297
1298		----------------------------------------------------------------------
1299		Name: Matthias Blume
1300		Date: 2001/09/18 15:35:00 EDT
1301		Tag: blume-20010918-readme11036
1302		Description:
1303
1304		Added README files.
1305
1306		----------------------------------------------------------------------
1307		Name: Matthias Blume
1308		Date: 2001/09/18 11:45:00 EDT
1309		Tag: Release_110_36 (retag)
1310		Description:
1311
1312		Fixed mistake in config/preloads. Retagged as 110.36.
1313
1314		----------------------------------------------------------------------
1315		Name: Matthias Blume
1316		Date: 2001/09/18 09:40:00 EDT
1317		Tag: Release_110_36_orig (tag changed)
1318		Description:
1319
1320		New version (110.36).  New bootfiles.
1321
1322		----------------------------------------------------------------------
1323		Name: Matthias Blume
1324		Date: 2001/09/14 16:15:00 EDT
1325		Tag: blume-20010914-x86fastfp
1326		Description:
1327
1328		John committed some changes that Allen made, in particular a (hopefully)
1329		correctly working version of the x86-fp module.
1330
1331		I changed the default setting of the Control.MLRISC.getFlag "x86-fast-fp"
1332		flag to "true".  Everything seems to compile to a fixpoint ok, and
1333		"mandelbrot" speeds up by about 15%.
1334
1335		----------------------------------------------------------------------
1336		Name: Matthias Blume
1337		Date: 2001/09/13 11:20:00 EDT
1338		Tag: blume-20010913-minimal
1339		Description:
1340
1341		1. Stefan Monnier's patch to fix a miscompilation problem that
1342		was brought to light by John Reppy's work on Moby.
1343
1344		2. Implemented a minimal "structure Compiler" that contains just
1345		"version" and "architecture".  The minimal version will be
1346		available when the full version is not.  This is for backward-
1347		compatibility with code that wants to test Compiler.version.
1348
1349		----------------------------------------------------------------------
1350		Name: Matthias Blume
1351		Date: 2001/08/28 14:03:00 EDT
1352		Tag: blume-20010828-ml-lex
1353		Description:
1354
1355		Fix for bug 1581, received from Neophytos Michael.
1356
1357		----------------------------------------------------------------------
1358		Name: Matthias Blume
1359		Date: 2001/08/27 11:20:00 EDT
1360		Tag: blume-20010827-readme11035
1361		Description:
1362
1363		Fleshed out the README file for 110.35.
1364
1365		----------------------------------------------------------------------
1366		Name: Matthias Blume
1367		Date: 2001/08/24 17:10:00 EDT
1368		Tag: Release_110_35
1369		Description:
1370
1371		New version number (110.35).  New bootfiles.
1372
1373		----------------------------------------------------------------------
1374		Name: Lal George
1375		Date: 2001/08/24 13:47:18 EDT 2001
1376		Tag: george-20010824-MLRISC-graphs
1377		Description:
1378
1379		removed clusters from MLRISC completely and replaced with graphs.
1380
1381		----------------------------------------------------------------------
1382		Name: Matthias Blume
1383		Date: 2001/08/23 17:50:00 EDT
1384		Tag: blume-20010823-toplevel
1385		Description:
1386
1387		- some reorganization of the code that implements various kinds of
1388		environments in the compiler (static, dynamic, symbolic, combined)
1389		- re-implemented the EnvRef module so that evalStream works properly
1390		(if the stream contains references to "use", "CM.make", etc.)
1391		- cleaned up evalloop.sml and interact.sml (but they need more cleaning)
1392
1393		----------------------------------------------------------------------
1394		Name: Matthias Blume
1395		Date: 2001/08/20 15:50 EDT
1396		Tag: blume20010820-slipup
1397		Description:
1398
1399		I forgot to commit a few files.  Here they are...
1400
1401		----------------------------------------------------------------------
1402		Name: Matthias Blume
1403		Date: 2001/08/20 15:35:00 EDT
1404		Tag: blume-20010820-debugprof
1405		Description:
1406
1407		!!!! NEW BOOTFILES !!!!
1408
1409		This is another round of reorganizing the compiler sources.  This
1410		time the main goal was to factor out all the "instrumentation"
1411		passes (for profiling and backtracing) into their own library.
1412		The difficulty was to do it in such a way that it does not depend
1413		on elaborate.cm but only on elabdata.cm.
1414
1415		Therefore there have been further changes to both elaborate.cm and
1416		elabdata.cm -- more "generic" things have been moved from the former
1417		to the latter.  As a result, I was forced to split the assignment
1418		of numbers indicating "primtyc"s into two portions: SML-generic and
1419		SML/NJ-specific.  Since it would have been awkward to maintain,
1420		I bit the bullet and actually _changed_ the mapping between these
1421		numbers and primtycs.  The bottom line of this is that you need
1422		a new set of bin- and bootfiles.
1423
1424		I have built new bootfiles for all architectures, so doing a fresh
1425		checkout and config/install.sh should be all you need.
1426
1427		The newly created library's name is
1428
1429		$smlnj/viscomp/debugprof.cm
1430
1431		and its sources live under
1432
1433		src/compiler/DebugProf
1434
1435		----------------------------------------------------------------------
1436		Name: Matthias Blume
1437		Date: 2001/08/15 17:15:00 EDT
1438		Tag: blume-20010815-compreorg
1439		Description:
1440
1441		This is a first cut at reorganizing the CM libraries that make up the
1442		core of the compiler.  The idea is to separate out pieces that could
1443		be used independently by tools, e.g., the parser, the typechecker, etc.
1444
1445		The current status is a step in this direction, but it is not quite
1446		satisfactory yet.  Expect more changes in the future.
1447
1448		Here is the current (new) organization...
1449
1450		What used to be $smlnj/viscomp/core.cm is now divided into
1451		six CM libraries:
1452
1453		$smlnj/viscomp/basics.cm
1454		/parser.cm
1455		/elabdata.cm
1456		/elaborate.cm
1457		/execute.cm
1458		/core.cm
1459
1460		The CM files for these libraries live under src/system/smlnj/viscomp.
1461		All these libraries are proxy libraries that contain precisely
1462		one CM library component.  Here are the locations of the components
1463		(all within the src/compiler tree):
1464
1465		Basics/basics.cm
1466		Parse/parser.cm
1467		ElabData/elabdata.cm
1468		Elaborator/elaborate.cm
1469		Execution/execute.cm
1470		core.cm
1471
1472		[This organization is the same that has been used already
1473		for a while for the architecture-specific parts of the visible
1474		compiler and for the old version of core.cm.]
1475
1476		As you will notice, many source files have been moved from their
1477		respective original locations to a new home in one of the above
1478		subtrees.
1479
1480		The division of labor between the new libraries is the following:
1481
1482		basics.cm:
1483		- Simple, basic definitions that pertain to many (or all) of
1484		the other libraries.
1485		parser.cm:
1486		- The SML parser, producing output of type Ast.dec.
1487		- The type family for Ast is also defined and exported here.
1488		elabdata.cm:
1489		- The datatypes that describe input and output of the elaborator.
1490		This includes types, absyn, and static environments.
1491		elaborator.cm:
1492		- The SML/NJ type checker and elaborator.
1493		This maps an Ast.dec (with a given static environment) to
1494		an Absyn.dec (with a new static environment).
1495		- This libraries implements certain modules that used to be
1496		structures as functors (to remove dependencies on FLINT).
1497		execute.cm:
1498		- Everything having to do with executing binary code objects.
1499		- Dynamic environments.
1500		core.cm:
1501		- SML/NJ-specific instantiations of the elaborator and MLRISC.
1502		- Top-level modules.
1503		- FLINT (this should eventually become its own library)
1504
1505		Notes:
1506
1507		I am not 100% happy with the way I separated the elaborator (and its
1508		data structures) from FLINT.  Two instances of the same problem:
1509
1510		1. Data structures contain certain fields that carry FLINT-specific
1511		information.  I hacked around this using exn and the property list
1512		module from smlnj-lib.  But the fact that there are middle-end
1513		specific fields around at all is a bit annoying.
1514
1515		2. The elaborator calculates certain FLINT-related information.  I tried
1516		to make this as abstract as I could using functorization, but, again,
1517		the fact that the elaborator has to perform calculations on behalf
1518		of the middle-end at all is not nice.
1519
1520		3. Having to used exn and property lists is unfortunate because it
1521		weakens type checking.  The other alternative (parameterizing
1522		nearly *everything*) is not appealing, though.
1523
1524		I removed the "rebinding =" warning hack because due to the new organization
1525		it was awkward to maintain it.  As a result, the compiler now issues some of
1526		these warnings when compiling init.cmi during bootstrap compilation. On
1527		the plus side, you also get a warning when you do, for example:
1528		val op = = Int32.+
1529		which was not the case up to now.
1530
1531		I placed "assign" and "deref" into the _Core structure so that the
1532		code that deals with the "lazy" keyword can find them there.  This
1533		removes the need for having access to the primitive environment
1534		during elaboration.
1535
1536		----------------------------------------------------------------------
1537		Name: Matthias Blume
1538		Date: 2001/08/13
1539		Tag: blume-20010813-closures
1540		Description:
1541
1542		This fix was sent to us by Zhong Shao.  It is supposed to improve the
1543		performance of certain loops by avoiding needless closure allocation.
1544
1545		----------------------------------------------------------------------
1546		Name: Lal George
1547		Date: 2001/07/31 10:03:23 EDT 2001
1548		Tag: george-20010731-x86-fmalloc
1549		Description: Fixed bug in x86 calls
1550
1551		There was a bug where call instructions would mysteriously
1552		vanish. The call instruction had to be one that returned
1553		a floating point value.
1554
1555		----------------------------------------------------------------------
1556	Name: Lal George	Name: Lal George
1557	Date: 2001/07/19 16:36:29 EDT 2001	Date: 2001/07/19 16:36:29 EDT 2001
1558	Tag: george-20010719-simple-cells	Tag: george-20010719-simple-cells
#	Line 2733	Line 4273
4273	elaborator). There were a lot of changes during my "linkpath" trials	elaborator). There were a lot of changes during my "linkpath" trials
4274	that could have been reverted to their original state but weren't.	that could have been reverted to their original state but weren't.
4275	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
4276	than what was strictly necessary...  (I _did_ resist the tempation	than what was strictly necessary...  (I _did_ resist the temptation
4277	of doing any "global reformatting" to avoid an untimely death at	of doing any "global reformatting" to avoid an untimely death at
4278	Dave's hands. :)	Dave's hands. :)
4279

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.890   changed lines   Added in v.1127

---

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