Diff of /sml/trunk/HISTORY

Parent Directory | Revision Log | Patch

	revision 984, Wed Nov 21 19:00:08 2001 UTC	revision 1130, Mon Mar 11 04:49:41 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: 2002/03/10 23:55:00 EST
18		Tag: leunga-20020310-x86-call
19		Description:
20
21		Added machine generation for CALL instruction (relative displacement mode)
22
23		----------------------------------------------------------------------
24		Name: Matthias Blume
25		Date: 2002/03/08 16:05:00
26		Tag: blume-20020308-entrypoints
27		Description:
28
29		Version number bumped to 110.39.1.  NEW BOOTFILES!
30
31		Entrypoints: non-zero offset into a code object where execution should begin.
32
33		- Added the notion of an entrypoint to CodeObj.
34		- Added reading/writing of entrypoint info to Binfile.
35		- Made runtime system bootloader aware of entrypoints.
36		- Use the address of the label of the first function given to mlriscGen
37		as the entrypoint.  This address is currently always 0, but it will
38		not be 0 once we turn on block placement.
39		- Removed the linkage cluster code (which was The Other Way(tm) of dealing
40		with entry points) from mlriscGen.
41
42		----------------------------------------------------------------------
43		Name: Allen Leung
44		Date: 2002/03/07 20:45:00 EST
45		Tag: leunga-20020307-x86-cmov
46		Description:
47
48		Bug fixes for CMOVcc on x86.
49
50		1. Added machine code generation for CMOVcc
51		2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
52		3. CMOVcc cannot have an immediate operand as argument.
53
54		----------------------------------------------------------------------
55		Name: Matthias Blume
56		Date: 2002/03/07 16:15:00 EST
57		Tag: blume-20020307-controls
58		Description:
59
60		This is a very large but mostly boring patch which makes (almost)
61		every tuneable compiler knob (i.e., pretty much everything under
62		Control.* plus a few other things) configurable via both the command
63		line and environment variables in the style CM did its configuration
64		until now.
65
66		Try starting sml with '-h' (or, if you are brave, '-H')
67
68		To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
69		implements the underlying generic mechanism.
70
71		The interface to some of the existing such facilities has changed somewhat.
72		For example, the MLRiscControl module now provides mkFoo instead of getFoo.
73		(The getFoo interface is still there for backward-compatibility, but its
74		use is deprecated.)
75
76		The ml-build script passes -Cxxx=yyy command-line arguments through so
77		that one can now twiddle the compiler settings when using this "batch"
78		compiler.
79
80		TODO items:
81
82		We should go through and throw out all controls that are no longer
83		connected to anything.  Moreover, we should go through and provide
84		meaningful (and correct!) documentation strings for those controls
85		that still are connected.
86
87		Currently, multiple calls to Controls.new are accepted (only the first
88		has any effect).  Eventually we should make sure that every control
89		is being made (via Controls.new) exactly once.  Future access can then
90		be done using Controls.acc.
91
92		Finally, it would probably be a good idea to use the getter-setter
93		interface to controls rather than ref cells.  For the time being, both
94		styles are provided by the Controls module, but getter-setter pairs are
95		better if thread-safety is of any concern because they can be wrapped.
96
97		*****************************************
98
99		One bug fix: The function blockPlacement in three of the MLRISC
100		backpatch files used to be hard-wired to one of two possibilities at
101		link time (according to the value of the placementFlag).  But (I
102		think) it should rather sense the flag every time.
103
104		*****************************************
105
106		Other assorted changes (by other people who did not supply a HISTORY entry):
107
108		1. the cross-module inliner now works much better (Monnier)
109		2. representation of weights, frequencies, and probabilities in MLRISC
110		changed in preparation of using those for weighted block placement
111		(Reppy, George)
112
113		----------------------------------------------------------------------
114		Name: Lal George
115		Date: 2002/03/07 14:44:24 EST 2002
116		Tag: george-20020307-weighted-block-placement
117
118		Tested the weighted block placement optimization on all architectures
119		(except the hppa) using AMPL to generate the block and edge frequencies.
120		Changes were required in the machine properties to correctly
121		categorize trap instructions. There is an MLRISC flag
122		"weighted-block-placement" that can be used to enable weighted block
123		placement, but this will be ineffective without block/edge
124		frequencies (coming soon).
125
126
127		----------------------------------------------------------------------
128		Name: Lal George
129		Date: 2002/03/05 17:24:48 EST
130		Tag: george-20020305-linkage-cluster
131
132		In order to support the block placement optimization, a new cluster
133		is generated as the very first cluster (called the linkage cluster).
134		It contains a single jump to the 'real' entry point for the compilation
135		unit. Block placement has no effect on the linkage cluster itself, but
136		all the other clusters  have full freedom in the manner in which they
137		reorder blocks or functions.
138
139		On the x86 the typical linkage code that is generated is:
140		----------------------
141		.align 2
142		L0:
143		addl    $L1-L0, 72(%esp)
144		jmp     L1
145
146
147		.align  2
148		L1:
149		----------------------
150
151		72(%esp) is the memory location for the stdlink register. This
152		must contain the address of the CPS function being called. In the
153		above example, it contains the address of  L0; before
154		calling L1 (the real entry point for the compilation unit), it
155		must contain the address for L1, and hence
156
157		addl $L1-L0, 72(%esp)
158
159		I have tested this on all architectures except the hppa.The increase
160		in code size is of course negligible
161
162		----------------------------------------------------------------------
163		Name: Allen Leung
164		Date: 2002/03/03 13:20:00 EST
165		Tag: leunga-20020303-mlrisc-tools
166
167		Added #[ ... ] expressions to mlrisc tools
168
169		----------------------------------------------------------------------
170		Name: Matthias Blume
171		Date: 2002/02/27 12:29:00 EST
172		Tag: blume-20020227-cdebug
173		Description:
174
175		- made types in structure C and C_Debug to be equal
176		- got rid of code duplication (c-int.sml vs. c-int-debug.sml)
177		- there no longer is a C_Int_Debug (C_Debug is directly derived from C)
178
179		----------------------------------------------------------------------
180		Name: Matthias Blume
181		Date: 2002/02/26 12:00:00 EST
182		Tag: blume-20020226-ffi
183		Description:
184
185		1. Fixed a minor bug in CM's "noweb" tool:
186		If numbering is turned off, then truly don't number (i.e., do not
187		supply the -L option to noweb).  The previous behavior was to supply
188		-L'' -- which caused noweb to use the "default" line numbering scheme.
189		Thanks to Chris Richards for pointing this out (and supplying the fix).
190
191		2. Once again, I reworked some aspects of the FFI:
192
193		A. The incomplete/complete type business:
194
195		- Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
196		gone!
197		- ML types representing an incomplete type are now *equal* to
198		ML types representing their corresponding complete types (just like
199		in C).  This is still safe because ml-nlffigen will not generate
200		RTTI for incomplete types, nor will it generate functions that
201		require access to such RTTI.   But when ML code generated from both
202		incomplete and complete versions of the C type meet, the ML types
203		are trivially interoperable.
204
205		NOTE:  These changes restore the full generality of the translation
206		(which was previously lost when I eliminated functorization)!
207
208		B. Enum types:
209
210		- Structure C now has a type constructor "enum" that is similar to
211		how the "su" constructor works.  However, "enum" is not a phantom
212		type because each "T enum" has values (and is isomorphic to
213		MLRep.Signed.int).
214		- There are generic access operations for enum objects (using
215		MLRep.Signed.int).
216		- ml-nlffigen will generate a structure E_foo for each "enum foo".
217		* The structure contains the definition of type "mlrep" (the ML-side
218		representation type of the enum).  Normally, mlrep is the same
219		as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
220		then mlrep will be defined as a datatype -- thus facilitating
221		pattern matching on mlrep values.
222		("-ec" will be suppressed if there are duplicate values in an
223		enumeration.)
224		* Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
225		will be generated for each C enum constant xxx.
226		* Conversion functions m2i and i2m convert between mlrep and
227		MLRep.Signed.int.  (Without "-ec", these functions are identities.)
228		* Coversion functions c and ml convert between mlrep and "tag enum".
229		* Access functions (get/set) fetch and store mlrep values.
230		- By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
231		enumerations are merged into one single enumeration represented by
232		structure E_'.
233
234		----------------------------------------------------------------------
235		Name: Allen Leung
236		Date: 2002/02/25 04:45:00 EST
237		Tag: leunga-20020225-cps-spill
238
239		This is a new implementation of the CPS spill phase.
240		The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
241		In case of problems, replace it with the old file spill.sml
242
243		The current compiler runs into some serious performance problems when
244		constructing a large record.  This can happen when we try to compile a
245		structure with many items.  Even a very simple structure like the following
246		makes the compiler slow down.
247
248		structure Foo = struct
249		val x_1 = 0w1 : Word32.int
250		val x_2 = 0w2 : Word32.int
251		val x_3 = 0w3 : Word32.int
252		...
253		val x_N = 0wN : Word32.int
254		end
255
256		The following table shows the compile time, from N=1000 to N=4000,
257		with the old compiler:
258
259		N
260		1000   CPS 100 spill                           0.04u  0.00s  0.00g
261		MLRISC ra                               0.06u  0.00s  0.05g
262		(spills = 0 reloads = 0)
263		TOTAL                                   0.63u  0.07s  0.21g
264
265		1100   CPS 100 spill                           8.25u  0.32s  0.64g
266		MLRISC ra                               5.68u  0.59s  3.93g
267		(spills = 0 reloads = 0)
268		TOTAL                                   14.71u  0.99s  4.81g
269
270		1500   CPS 100 spill                           58.55u  2.34s  1.74g
271		MLRISC ra                               5.54u  0.65s  3.91g
272		(spills = 543 reloads = 1082)
273		TOTAL                                   65.40u  3.13s  6.00g
274
275		2000   CPS 100 spill                           126.69u  4.84s  3.08g
276		MLRISC ra                               0.80u  0.10s  0.55g
277		(spills = 42 reloads = 84)
278		TOTAL                                   129.42u  5.10s  4.13g
279
280		3000   CPS 100 spill                           675.59u  19.03s  11.64g
281		MLRISC ra                               2.69u  0.27s  1.38g
282		(spills = 62 reloads = 124)
283		TOTAL                                   682.48u  19.61s  13.99g
284
285		4000   CPS 100 spill                           2362.82u  56.28s  43.60g
286		MLRISC ra                               4.96u  0.27s  2.72g
287		(spills = 85 reloads = 170)
288		TOTAL                                   2375.26u  57.21s  48.00g
289
290		As you can see the old cps spill module suffers from some serious
291		performance problem.  But since I cannot decipher the old code fully,
292		instead of patching the problems up, I'm reimplementing it
293		with a different algorithm.  The new code is more modular,
294		smaller when compiled, and substantially faster
295		(O(n log n) time and O(n) space).  Timing of the new spill module:
296
297		4000  CPS 100 spill                           0.02u  0.00s  0.00g
298		MLRISC ra                               0.25u  0.02s  0.15g
299		(spills=1 reloads=3)
300		TOTAL                                   7.74u  0.34s  1.62g
301
302		Implementation details:
303
304		As far as I can tell, the purpose of the CPS spill module is to make sure the
305		number of live variables at any program point (the bandwidth)
306		does not exceed a certain limit, which is determined by the
307		size of the spill area.
308
309		When the bandwidth is too large, we decrease the register pressure by
310		packing live variables into spill records.  How we achieve this is
311		completely different than what we did in the old code.
312
313		First, there is something about the MLRiscGen code generator
314		that we should be aware of:
315
316		o MLRiscGen performs code motion!
317
318		In particular, it will move floating point computations and
319		address computations involving only the heap pointer to
320		their use sites (if there is only a single use).
321		What this means is that if we have a CPS record construction
322		statement
323
324		RECORD(k,vl,w,e)
325
326		we should never count the new record address w as live if w
327		has only one use (which is often the case).
328
329		We should do something similar to floating point, but the transformation
330		there is much more complex, so I won't deal with that.
331
332		Secondly, there are now two new cps primops at our disposal:
333
334		1. rawrecord of record_kind option
335		This pure operator allocates some uninitialized storage from the heap.
336		There are two forms:
337
338		rawrecord NONE [INT n]  allocates a tagless record of length n
339		rawrecord (SOME rk) [INT n] allocates a tagged record of length n
340		and initializes the tag.
341
342		2. rawupdate of cty
343		rawupdate cty (v,i,x)
344		Assigns to x to the ith component of record v.
345		The storelist is not updated.
346
347		We use these new primops for both spilling and increment record construction.
348
349		1. Spilling.
350
351		This is implemented with a linear scan algorithm (but generalized
352		to trees).  The algorithm will create a single spill record at the
353		beginning of the cps function and use rawupdate to spill to it,
354		and SELECT or SELp to reload from it.  So both spills and reloads
355		are fine-grain operations.  In contrast, in the old algorithm
356		"spills" have to be bundled together in records.
357
358		Ideally, we should sink the spill record construction to where
359		it is needed.  We can even split the spill record into multiple ones
360		at the places where they are needed.  But CPS is not a good
361		representation for global code motion, so I'll keep it simple and
362		am not attempting this.
363
364		2. Incremental record construction (aka record splitting).
365
366		Long records with many component values which are simulatenously live
367		(recall that single use record addresses are not considered to
368		be live) are constructed with rawrecord and rawupdate.
369		We allocate space on the heap with rawrecord first, then gradually
370		fill it in with rawupdate.  This is the technique suggested to me
371		by Matthias.
372
373		Some restrictions on when this is applicable:
374		1. It is not a VECTOR record.  The code generator currently does not handle
375		this case. VECTOR record uses double indirection like arrays.
376		2. All the record component values are defined in the same "basic block"
377		as the record constructor.  This is to prevent speculative
378		record construction.
379
380		----------------------------------------------------------------------
381		Name: Allen Leung
382		Date: 2002/02/22 01:02:00 EST
383		Tag: leunga-20020222-mlrisc-tools
384
385		Minor bug fixes in the parser and rewriter
386
387		----------------------------------------------------------------------
388		Name: Allen Leung
389		Date: 2002/02/21 20:20:00 EST
390		Tag: leunga-20020221-peephole
391
392		Regenerated the peephole files.  Some contained typos in the specification
393		and some didn't compile because of pretty printing bugs in the old version
394		of 'nowhere'.
395
396		----------------------------------------------------------------------
397		Name: Allen Leung
398		Date: 2002/02/19 20:20:00 EST
399		Tag: leunga-20020219-mlrisc-tools
400		Description:
401
402		Minor bug fixes to the mlrisc-tools library:
403
404		1.  Fixed up parsing colon suffixed keywords
405		2.  Added the ability to shut the error messages up
406		3.  Reimplemented the pretty printer and fixed up/improved
407		the pretty printing of handle and -> types.
408		4.  Fixed up generation of literal symbols in the nowhere tool.
409		5.  Added some SML keywords to to sml.sty
410
411		----------------------------------------------------------------------
412		Name: Matthias Blume
413		Date: 2002/02/19 16:20:00 EST
414		Tag: blume-20020219-cmffi
415		Description:
416
417		A wild mix of changes, some minor, some major:
418
419		* All C FFI-related libraries are now anchored under $c:
420		$/c.cm      --> $c/c.cm
421		$/c-int.cm  --> $c/internals/c-int.cm
422		$/memory.cm --> $c/memory/memory.cm
423
424		* "make" tool (in CM) now treats its argument pathname slightly
425		differently:
426		1. If the native expansion is an absolute name, then before invoking
427		the "make" command on it, CM will apply OS.Path.mkRelative
428		(with relativeTo = OS.FileSys.getDir()) to it.
429		2. The argument will be passed through to subsequent phases of CM
430		processing without "going native".  In particular, if the argument
431		was an anchored path, then "make" will not lose track of that anchor.
432
433		* Compiler backends now "know" their respective C calling conventions
434		instead of having to be told about it by ml-nlffigen.  This relieves
435		ml-nlffigen from one of its burdens.
436
437		* The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
438
439		* Export C_DEBUG and C_Debug from $c/c.cm.
440
441		* C type encoding in ml-nlffi-lib has been improved to model the conceptual
442		subtyping relationship between incomplete pointers and their complete
443		counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
444		with the convention of instantiating 'o with ('t, 'c) obj whenever
445		the pointer target type is complete.  In the incomplete case, 'o
446		will be instantiated with some "'c iobj" -- a type obtained by
447		using one of the functors PointerToIncompleteType or PointerToCompleteType.
448
449		Operations that work on both incomplete and complete pointer types are
450		typed as taking an 'o ptr while operations that require the target to
451		be known are typed as taking some ('t, 'c) obj ptr.
452
453		voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
454		where void is an eqtype without any values.  This makes it possible
455		to work on voidptr values using functions meant to operate on light
456		incomplete pointers.
457
458		* As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
459		been vastly simplified.
460
461		----------------------------------------------------------------------
462		Name: Matthias Blume
463		Date: 2002/02/19 10:48:00 EST
464		Tag: blume-20020219-pqfix
465		Description:
466
467		Applied Chris Okasaki's bug fix for priority queues.
468
469		----------------------------------------------------------------------
470		Name: Matthias Blume
471		Date: 2002/02/15 17:05:00
472		Tag: Release_110_39
473		Description:
474
475		Last-minute retagging is becoming a tradition... :-(
476
477		This is the working release 110.39.
478
479		----------------------------------------------------------------------
480		Name: Matthias Blume
481		Date: 2002/02/15 16:00:00 EST
482		Tag: Release_110_39-orig
483		Description:
484
485		Working release 110.39.  New bootfiles.
486
487		(Update: There was a small bug in the installer so it wouldn't work
488		with all shells.  So I retagged. -Matthias)
489
490		----------------------------------------------------------------------
491		Name: Matthias Blume
492		Date: 2002/02/15 14:17:00 EST
493		Tag: blume-20020215-showbindings
494		Description:
495
496		Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
497		the latter can be useful for exploring what bindings are available at
498		the interactive prompt.  (The first function returns only the list
499		of symbols that are really bound, the second prints those but also the
500		ones that CM's autoloading mechanism knows about.)
501
502		----------------------------------------------------------------------
503		Name: Matthias Blume
504		Date: 2002/02/15 12:08:00 EST
505		Tag: blume-20020215-iptrs
506		Description:
507
508		Two improvements to ml-nlffigen:
509
510		1. Write files only if they do not exist or if their current contents
511		do not coincide with what's being written.  (That is, avoid messing
512		with the time stamps unless absolutely necessary.)
513
514		2. Implement a "repository" mechanism for generated files related
515		to "incomplete pointer types".   See the README file for details.
516
517		----------------------------------------------------------------------
518		Name: Matthias Blume
519		Date: 2002/02/14 11:50:00 EST
520		Tag: blume-20020214-quote
521		Description:
522
523		Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
524		because of the new and improved tag generation scheme.  (Thanks to Allen
525		Leung for pointing it out.)
526
527		----------------------------------------------------------------------
528		Name: Lal George
529		Date: 2002/02/14 09:55:27 EST 2002
530		Tag: george-20020214-isabelle-bug
531		Description:
532
533		Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
534		of Isabelle on the x86.
535
536		From Allen:
537		-----------
538		I've found the problem:
539
540		in ra-core.sml, I use the counter "blocked" to keep track of the
541		true number of elements in the freeze queue.  When the counter goes
542		to zero, I skip examining the queue.  But I've messed up the
543		bookkeeping in combine():
544
545		else ();
546		case !ucol of
547		PSEUDO => (if !cntv > 0 then
548		(if !cntu > 0 then blocked := !blocked - 1 else ();
549		^^^^^^^^^^^^^^^^^^^^^^^
550		moveu := mergeMoveList(!movev, !moveu)
551		)
552		else ();
553
554		combine() is called to coalesce two nodes u and v.
555		I think I was thinking that if the move counts of u and v are both
556		greater than zero then after they are coalesced then one node is
557		removed from the freeze queue.  Apparently I was thinking that
558		both u and v are of low degree, but that's clearly not necessarily true.
559
560
561		02/12/2002:
562		Here's the patch.  HOL now compiles.
563
564		I don't know how this impact on performance (compile
565		time or runtime).  This bug caused the RA (especially on the x86)
566		to go thru the potential spill phase when there are still nodes on the
567		freeze queue.
568
569
570
571
572		----------------------------------------------------------------------
573		Name: Matthias Blume
574		Date: 2002/02/13 22:40:00 EST
575		Tag: blume-20020213-fptr-rtti
576		Description:
577
578		Fixed a bug in ml-nlffigen that was introduced with one of the previous
579		updates.
580
581		----------------------------------------------------------------------
582		Name: Matthias Blume
583		Date: 2002/02/13 16:41:00 EST
584		Tag: blume-20020213-cmlpq
585		Description:
586
587		Added new priority queue export symbols (which have just been added to
588		smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
589		not compile and the installer would choke.)
590
591		----------------------------------------------------------------------
592		Name: Matthias Blume
593		Date: 2002/02/13 16:15:00 EST
594		Tag: blume-20020213-various
595		Description:
596
597		1. More tweaks to ml-nlffigen:
598
599		- better internal datastructures (resulting in slight speedup)
600		- "-match" option requires exact match
601		- "localized" gensym counters (untagged structs/unions nested within
602		other structs/unions or within typedefs get a fresh counter; their
603		tag will be prefixed by a concatenation of their parents' tags)
604		- bug fixes (related to calculation of transitive closure of types
605		to be included in the output)
606
607		2. Minor Basis updates:
608
609		- added implementations for List.collate and Option.app
610
611		----------------------------------------------------------------------
612		Name: Matthias Blume
613		Date: 2002/02/11 15:55:00 EST
614		Tag: blume-20020211-gensym
615		Description:
616
617		Added a "-gensym" option to command line of ml-nlffigen.  This can be
618		used to specify a "stem" -- a string that is inserted in all "gensym'd"
619		names (ML structure names that correspond to unnamed C structs, unions,
620		and enums), so that separate runs of ml-nlffigen do not clash.
621
622		----------------------------------------------------------------------
623		Name: Matthias Blume
624		Date: 2002/02/11 12:05:00 EST
625		Tag: blume-20020211-gensml
626		Description:
627
628		A quick fix for a problem with GenSML (in the pgraph-util library):
629		Make generation of toplevel "local" optional.  (Strictly speaking,
630		signature definitions within "local" are not legal SML.)
631
632		Other than that: updates to INSTALL and cm/TODO.
633
634		----------------------------------------------------------------------
635		Name: Matthias Blume
636		Date: 2002/02/08 15:00:00 EST
637		Tag: blume-20020208-uniquepid
638		Description:
639
640		0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
641
642		1. The installer (config/install.sh) has gotten smarter:
643
644		- Configuration options are a bit easier to specify now
645		(in config/targets).
646		- Bug in recognizing .tar.bz2 files fixed.
647		- Installer automatically resolves dependencies between
648		configuration options (e.g., if you ask for eXene, you will
649		also get cml -- regardless whether you asked for it or not).
650		- Installer can run in "quieter mode" by setting the environment
651		variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
652		"completely silent", though.
653		- Build HashCons library as part of smlnj-lib.
654
655		2. A new scheme for assigning persistent identifiers to compilation
656		units (and, by extension, to types etc.) has been put into place.
657		This fixes a long-standing bug where types and even dynamic values
658		can get internally confused, thereby compromising type safety
659		(abstraction) and dynamic correctness.  See
660
661		http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
662
663		for an example of how things could go wrong until now.
664
665		The downside of the new scheme is that pids are not quite as
666		persistent as they used to be: CM will generate a fresh pid
667		for every compilation unit that it thinks it sees for the first
668		time.  That means that if you compile starting from a clean, fresh
669		source tree at two different times, you end up with different
670		binaries.
671
672		Cutoff recompilation, however, has not been compromised because
673		CM keeps pid information in special caches between runs.
674
675		----------------------------------------------------------------------
676		Name: Lal George
677		Date: 2002/02/07 15:34:13 EST 2002
678		Tag: <none>
679		Description:
680
681		Compilers that generate assembly code may produce  global labels
682		whose value is resolved at link time. The various peephole optimization
683		modules did not take this in account.
684
685		TODO. The Labels.addrOf function should really return an option
686		type so that clients are forced to deal with this issue, rather
687		than an exception being raised.
688
689		----------------------------------------------------------------------
690		Name: Lal George
691		Date: 2002/02/06 13:55:02 EST
692		Tag: george-20020206-ra-breakup
693		Description:
694
695		1. A bug fix from Allen.
696
697		A typo causes extra fstp %st(0)'s to be generated at compensation
698		edges, which might cause stack underflow traps at runtime.  This
699		occurs in fft where there are extraneous fstps right before the 'into'
700		trap instruction (in this case they are harmless since none of the
701		integers overflow.)
702
703		2. Pulled out various utility modules that were embedded in the modules
704		of the register allocator. I need these modules for other purposes, but
705		they are not complete enough to put into a library (just yet).
706		----------------------------------------------------------------------
707		Name: Matthias Blume
708		Date: 2002/01/31 16:05:00 EST
709		Tag: blume-20020131-sparc-ccalls
710		Description:
711
712		1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes)
713		of extra stack space by mistake.  Fixed.
714
715		2. Bug in logic of handling of command-line options in ml-nlffigen fixed.
716
717		----------------------------------------------------------------------
718		Name: Allen Leung
719		Date: 2002/01/30
720		Tag: leunga-20020130-nowhere-bug-fix
721		Description:
722
723		MLRISC bug fixes:
724		1. Fixed a bindings computation bug in the 'nowhere' program generator tool.
725		2. MachineInt.fromString was negating its value.
726
727		----------------------------------------------------------------------
728		Name: Matthias Blume
729		Date: 2002/01/29
730		Tag: blume-20020129-INSTALL
731		Description:
732
733		- Added somewhat detailed installation instructions (file INSTALL).
734		- Fixed curl-detection bug in config/install.sh.
735		- It is now possible to select the URL getter using the URLGETTER
736		environment variable:
737
738		not set / "unknown"      --> automatic detection (script tries wget,
739		curl, and lynx)
740		"wget" / "curl" / "lynx" --> use the specified program (script "knows"
741		how to properly invoke them)
742		other                    --> use $URLGETTER directly, it must take
743		precisely two command-line arguments
744		(source URL and destination file name)
745
746		----------------------------------------------------------------------
747		Name: Matthias Blume
748		Date: 2002/01/28
749		Tag: blume-20020128-sparc-ccalls
750		Description:
751
752		- Fixed problem with calculation of "used" registers in sparc-c-calls.
753		- Make use of the allocParam argument in sparc-c-calls.
754
755		----------------------------------------------------------------------
756		Name: Matthias Blume
757		Date: 2002/01/28
758		Tag: blume-20020128-allocParam
759		Description:
760
761		John Reppy:  Changes c-calls API to accept client-callback for
762		allocating extra stack space.
763		me: Corresponding changes to mlriscGen (using a dummy argument that
764		does not change the current behavior).
765
766		----------------------------------------------------------------------
767		Name: Matthias Blume
768		Date: 2002/01/28 12:00:00
769		Tag: Release_110_38
770		Description:
771
772		This time for real!!!
773
774		----------------------------------------------------------------------
775		Name: Matthias Blume
776		Date: 2002/01/28 10:56:00 EST
777		Tag: blume-20020128-retraction
778		Description:
779
780		0. Retracted earlier 110.38.  (The Release_110_38 tag has been replaced
781		with blume-Release_110_38-retracted.)
782
783		1. Fixed a problem with incorrect rounding modes in real64.sml.
784		(Thanks to Andrew Mccreight <andrew.mccreight@yale.edu>.)
785
786		2. A bug in ml-nlffigen related to the handling of unnamed structs, unions,
787		and enums fixed.  The naming of corresponding ML identifiers should
788		now be consistent again.
789
790		----------------------------------------------------------------------
791		Name: Allen Leung
792		Date: 2002/01/27
793		Tag: leunga-20020127-nowhere
794		Description:
795
796		Added a target called nowhere in the configuration scripts.
797		Enabling this will build the MLRISC 'nowhere' tool (for translating
798		programs with where-clauses into legal SML code) during installation.
799
800		----------------------------------------------------------------------
801		Name: Matthias Blume
802		Date: 2002/01/25 21:27:00 EST
803		Tag: blume-Release_110_38-retracted
804		Description:
805
806		Call it a (working) release!  Version is 110.38. Bootfiles are ready.
807
808		README will be added later.
809
810		!!! NOTE:  Re-tagged as blume-Release_110_38-retracted. Original tag
811		(Release_110_38) removed.  Reason: Last-minute bug fixes.
812
813		----------------------------------------------------------------------
814		Name: Matthias Blume
815		Date: 2002/01/25
816		Tag: blume-20020125-ffi
817		Description:
818
819		A large number of tweaks and improvements to ml-nlffi-lib and
820		ml-nlffigen:
821
822		- ML represenation types have been streamlined
823		- getter and setter functions work with concrete values, not abstract
824		ones where possible
825		- ml-nlffigen command line more flexible (see README file there)
826		- some bugs have been fixed (hopefully)
827
828		----------------------------------------------------------------------
829		Name: Lal George
830		Date: 2002/01/24
831		Tag: george-20020124-risc-ra-interface
832		Description:
833
834		There is a dramatic simplification in the interface to the
835		register allocator for RISC architectures as a result of making
836		parallel copy instructions explicit.
837
838		----------------------------------------------------------------------
839		Name: Matthias Blume
840		Date: 2002/01/22
841		Tag: blume-20020122-x86-ccalls
842		Description:
843
844		Bug fix for c-calls on x86 (having to do with how char- and
845		short-arguments are being handled).
846
847		----------------------------------------------------------------------
848		Name: Matthias Blume
849		Date: 2002/01/21
850		Tag: blume-20020121-ff
851		Description:
852
853		Another day of fiddling with the FFI...
854
855		1. Bug fix/workaround:  CKIT does not complain about negative array
856		dimensions, so ml-nlffigen has to guard itself against this possibility.
857		(Otherwise a negative dimension would send it into an infinite loop.)
858
859		2. Some of the abstract types (light objects, light pointers, most "base"
860		types) in structure C are now eqtypes.
861
862		3. Added constructors and test functions for NULL function pointers.
863
864		----------------------------------------------------------------------
865		Name: Matthias Blume
866		Date: 2002/01/18
867		Tag: blume-20020118-ready-for-new-release
868		Description:
869
870		Made config/srcarchiveurl point to a new place.  (Will provide boot
871		files shortly.)
872
873		Maybe we christen this to be 110.38?
874
875		----------------------------------------------------------------------
876		Name: Matthias Blume
877		Date: 2002/01/18
878		Tag: blume-20020118-more-ffifiddle
879		Description:
880
881		Today's FFI fiddling:
882
883		- Provided a structure CGetSet with "convenient" versions of C.Get.* and
884		C.Set.* that use concrete (MLRep.*) arguments and results instead
885		of abstract ones.
886
887		- Provided word-style bit operations etc. for "int" representation
888		types in MLRep.S<Foo>Bitops where <Foo> ranges over Char, Int, Short,
889		and Long.
890
891		----------------------------------------------------------------------
892		Name: Matthias Blume
893		Date: 2002/01/18
894		Tag: blume-20020118-use-x86-fp
895		Description:
896
897		Now that x86-fast-fp seems to be working, I turned it back on again
898		by default.  (Seems to work fine now, even with the FFI.)
899
900		Other than that, I added some documentation about the FFI to
901		src/ml-nlffigen/README and updated the FFI test examples in
902		src/ml-nlffi-lib/Tests/*.
903
904		----------------------------------------------------------------------
905		Name: Allen Leung
906		Date: 2002/01/17
907		Tag: leunga-20020117-x86-fast-fp-call
908		Description:
909
910		1. Fixed a problem with handling return fp values when x86's fast fp
911		mode is turned on.
912
913		2. Minor pretty printing fix for cellset.  Print %st(0) as %st(0) instead
914		of %f32.
915
916		3. Added a constructor INT32lit to the ast of MLRISC tools.
917
918		----------------------------------------------------------------------
919		Name: Matthias Blume
920		Date: 2002/01/16
921		Tag: blume-20020116-ffifiddle
922		Description:
923
924		More fiddling with the FFI interface:
925
926		- Make constness 'c instead of rw wherever possible.  This eliminates
927		the need for certain explicit coercions.  (However, due to ML's
928		value polymorphism, there will still be many cases where explicit
929		coercions are necessary.  Phantom types are not the whole answer
930		to modeling a subtyping relationship in ML.)
931
932		- ro/rw coersions for pointers added.  (Avoids the detour through */&.)
933
934		- "printf" test example added to src/ml-nlffi-lib/Tests.  (Demonstrates
935		clumsy workaround for varargs problem.)
936
937		----------------------------------------------------------------------
938		Name: Lal George
939		Date: 2002/01/15
940		Tag: <none>
941		Description:
942
943		1. Since COPY instructions are no longer native to the architecture,
944		a generic functor can be used to implement the expandCopies function.
945
946		2. Allowed EXPORT and IMPORT pseudo-op declarations to appear inside a
947		TEXT segment.
948
949		----------------------------------------------------------------------
950		Name: Matthias Blume
951		Date: 2002/01/15
952		Tag: blume-20020115-ffiupdates
953		Description:
954
955		1. Fix for bug resulting in single-precision float values being returned
956		incorrectly from FFI calls.
957
958		2. Small modifications to C FFI API:
959
960		- memory-allocation routines return straight objects (no options)
961		and raise an exception in out-of-memory situations
962		- unsafe extensions to cast between function pointers and pointers
963		from/to ints
964		- added structure C_Debug as an alternative to structure C where
965		pointer-dereferencing (|*| and |*!) always check for null-pointers
966		- added open_lib' to DynLinkage;  open_lib' works like open_lib
967		but also takes a (possibly empty) list of existing library handles
968		that the current library depends on
969
970		----------------------------------------------------------------------
971		Name: Matthias Blume
972		Date: 2002/01/10
973		Tag: blume-20020110-newffigen
974		Description:
975
976		1. Updates to portable graph code.
977
978		2. Major update to ml-nlffigen and ml-nlffi-lib.  Things are much
979		more scalable now so that even huge interfaces such as the one
980		for GTK compile in finite time and space. :-)
981		See src/ml-nlffigen/README for details on what's new.
982
983		----------------------------------------------------------------------
984		Name: Lal George
985		Date: 2001/01/09 14:31:35 EST 2002
986		Tag: george-20011206-rm-native-copy
987		Description:
988
989		Removed the native COPY and FCOPY instructions
990		from all the architectures and replaced it with the
991		explicit COPY instruction from the previous commit.
992
993		It is now possible to simplify many of the optimizations
994		modules that manipulate copies. This has not been
995		done in this change.
996
997		----------------------------------------------------------------------
998		Name: Lal George
999		Date: 2001/12/06 16:50:13 EST 2001
1000		Tag: george-20011206-mlrisc-instruction
1001		Description:
1002
1003		Changed the representation of instructions from being fully abstract
1004		to being partially concrete. That is to say:
1005
1006		from
1007		type instruction
1008
1009		to
1010		type instr                              (* machine instruction *)
1011
1012		datatype instruction =
1013		LIVE of {regs: C.cellset, spilled: C.cellset}
1014		| KILL of {regs: C.cellset, spilled: C.cellset}
1015		| COPYXXX of {k: CB.cellkind, dst: CB.cell list, src: CB.cell list}
1016		| ANNOTATION of {i: instruction, a: Annotations.annotation}
1017		| INSTR of instr
1018
1019		This makes the handling of certain special instructions that appear on
1020		all architectures easier and uniform.
1021
1022		LIVE and KILL say that a list of registers are live or killed at the
1023		program point where they appear. No spill code is generated when an
1024		element of the 'regs' field is spilled, but the register is moved to
1025		the 'spilled' (which is present, more for debugging than anything else).
1026
1027		LIVE replaces the (now deprecated) DEFFREG instruction on the alpha.
1028		We used to generate:
1029
1030		DEFFREG f1
1031		f1 := f2 + f3
1032		trapb
1033
1034		but now generate:
1035
1036		f1 := f2 + f3
1037		trapb
1038		LIVE {regs=[f1,f2,f3], spilled=[]}
1039
1040		Furthermore, the DEFFREG (hack) required that all floating point instruction
1041		use all registers mentioned in the instruction. Therefore f1 := f2 + f3,
1042		defines f1 and uses [f1,f2,f3]! This hack is no longer required resulting
1043		in a cleaner alpha implementation. (Hopefully, intel will not get rid of
1044		this architecture).
1045
1046		COPYXXX is intended to replace the parallel COPY and FCOPY  available on
1047		all the architectures. This will result in further simplification of the
1048		register allocator that must be aware of them for coalescing purposes, and
1049		will also simplify certain aspects of the machine description that provides
1050		callbacks related to parallel copies.
1051
1052		ANNOTATION should be obvious, and now INSTR represents the honest to God
1053		machine instruction set!
1054
1055		The <arch>/instructions/<arch>Instr.sml files define certain utility
1056		functions for making porting easier -- essentially converting upper case
1057		to lower case. All machine instructions (of type instr) are in upper case,
1058		and the lower case form generates an MLRISC instruction. For example on
1059		the alpha we have:
1060
1061		datatype instr =
1062		LDA of {r:cell, b:cell, d:operand}
1063		| ...
1064
1065		val lda : {r:cell, b:cell, d:operand} -> instruction
1066		...
1067
1068		where lda is just (INSTR o LDA), etc.
1069
1070		----------------------------------------------------------------------
1071		Name: Matthias Blume
1072		Date: 2001/11/22 21:40:00 EST
1073		Tag: Release_110_37
1074		Description:
1075
1076		Release 110.37.  This time for real.
1077
1078		----------------------------------------------------------------------
1079		Name: Matthias Blume
1080		Date: 2001/11/21 16:35:00 EST
1081		Tag: blume-20011121-foot-in-mouth
1082		Description:
1083
1084		Removed the "Release_110_37" tag because of a serious bug.
1085		This will be re-tagged once the bug is fixed.
1086
1087		----------------------------------------------------------------------
1088		Name: Matthias Blume
1089		Date: 2001/11/21 16:14:00 EST
1090		Tag: blume-20011121-forgottenfile
1091		Description:
1092
1093		Forgot to add a file.  (Just a .tex-file -- part of
1094		the CM manual source.)
1095
1096		----------------------------------------------------------------------
1097		Name: Matthias Blume
1098		Date: 2001/11/21 16:10:00 EST
1099		Tag: blume-20011121-invalid_110_37
1100		Description:
1101
1102		Note: I removed the original tag "Release_110_37" from this commit
1103		because we found a serious bug in all non-x86 backends.
1104		- Matthias
1105
1106		1. Modifications to the SML/NJ code generator and to the runtime system
1107		so that code object name strings are directly inserted into code
1108		objects at code generation time.  The only business the runtime system
1109		has with this is now to read the name strings on occasions.
1110		(The encoding of the name string has also changed somewhat.)
1111
1112		2. CM now implements a simple "set calculus" for specifying export lists.
1113		In particular, it is now possible to refer to the export lists of
1114		other libraries/groups/sources and form unions as well as differences.
1115		See the latest CM manual for details.
1116
1117		3. An separate notion of "proxy" libraries has again be eliminated from
1118		CM's model.  (Proxy libraries are now simply a special case of using
1119		the export list calculus.)
1120
1121		4. Some of the existing libraries now take advantage of the new set
1122		calculus.
1123		(Notice that not all libraries have been converted because some
1124		of the existing .cm-files are supposed to be backward compatible
1125		with 110.0.x.)
1126
1127		5. Some cleanup in stand-alone programs.  (Don't use "exnMessage" -- use
1128		"General.exnMessage"!  The former relies on a certain hook to be
1129		initialized, and that often does not happen in the stand-alone case.)
1130
1131		----------------------------------------------------------------------
1132	Name: Lal George	Name: Lal George
1133	Date: 2001/11/21  13:56:18 EST	Date: 2001/11/21  13:56:18 EST
1134	Tag: george-2001121-pseudo-ops	Tag: george-2001121-pseudo-ops
#	Line 208	Line 1324
1324
1325	----------------------------------------------------------------------	----------------------------------------------------------------------
1326	Name: Matthias Blume	Name: Matthias Blume
	>>>>>>> 1.169
1327	Date: 2001/09/18 15:35:00 EDT	Date: 2001/09/18 15:35:00 EDT
1328	Tag: blume-20010918-readme11036	Tag: blume-20010918-readme11036
1329	Description:	Description:

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.984   changed lines   Added in v.1130

---

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