Diff of /sml/trunk/HISTORY

Parent Directory | Revision Log | Patch

	revision 1049, Thu Jan 31 21:03:49 2002 UTC	revision 1171, Fri Mar 22 20:41:43 2002 UTC
#	Line 13	Line 13
13	Description:	Description:
14
15	----------------------------------------------------------------------	----------------------------------------------------------------------
16		Name: Lal George
17		Date: 2002/03/22 14:18:25 EST
18		Tag: george-20020322-cps-branch-prob
19		Description:
20
21		Implemented the Ball-Larus branch prediction-heuristics, and
22		incorporated graphical viewers for control flow graphs.
23
24		Ball-Larus Heuristics:
25		---------------------
26		See the file compiler/CodeGen/cpscompile/cpsBranchProb.sml.
27
28		By design it uses the Dempster-Shafer theory for combining
29		probabilities.  For example, in the function:
30
31		fun f(n,acc) = if n = 0 then acc else f(n-1, n*acc)
32
33		the ball-larus heuristics predicts that the n=0 is unlikely
34		(OH-heuristic), and the 'then' branch is unlikely because of the
35		RH-heuristic -- giving the 'then' branch an even lower combined
36		probability using the Dempster-Shafer theory.
37
38		Finally, John Reppy's loop analysis in MLRISC, further lowers the
39		probability of the 'then' branch because of the loop in the else
40		branch.
41
42
43		Graphical Viewing:
44		------------------
45		I merely plugged in Allen's graphical viewers into the compiler. The
46		additional code is not much. At the top level, saying:
47
48		Control.MLRISC.getFlag "cfg-graphical-view" := true;
49
50		will display the graphical view of the control flow graph just before
51		back-patching.  daVinci must be in your path for this to work. If
52		daVinci is not available, then the default viewer can be changed
53		using:
54
55		Control.MLRISC.getString "viewer"
56
57		which can be set to "dot" or "vcg" for the corresponding viewers. Of
58		course, these viewers must be in your path.
59
60		The above will display the compilation unit at the level of clusters,
61		many of which are small, boring, and un-interesting. Also setting:
62
63		Control.MLRISC.getInt "cfg-graphical-view_size"
64
65		will display clusters that are larger than the value set by the above.
66
67
68		----------------------------------------------------------------------
69		Name: Matthias Blume
70		Date: 2002/03/21 22:20:00 EST
71		Tag: blume-20020321-kmp-bugfix
72		Description:
73
74		Changed the interface to the KMP routine in PreString and fixed
75		a minor bug in one place where it was used.
76
77		----------------------------------------------------------------------
78		Name: Allen Leung
79		Date: 2002/03/21 20:30:00 EST
80		Tag: leunga-20020321-cfg
81		Description:
82
83		Fixed a potential problem in cfg edge splitting.
84
85		----------------------------------------------------------------------
86		Name: Allen Leung
87		Date: 2002/03/21 17:15:00 EST
88		Tag: leunga-20020321-x86-fp-cfg
89		Description:
90
91		1. Recoded the buggy parts of x86-fp.
92
93		a. All the block reordering code has been removed.
94		We now depend on the block placement phases to do this work.
95
96		b. Critical edge splitting code has been simplified and moved into the
97		CFG modules, as where they belong.
98
99		Both of these were quite buggy and complex.  The code is now much, much
100		simpler.
101
102		2. X86 backend.
103
104		a. Added instructions for 64-bit support.  Instruction selection for
105		64-bit has not been committed, however, since that
106		requires changes to MLTREE which haven't been approved by
107		Lal and John.
108
109		b. Added support for FUCOMI and FUCOMIP when generating code for
110		PentiumPro and above.  We only generate these instructions in
111		the fast-fp mode.
112
113		c. Added cases for JP and JNP in X86FreqProps.
114
115		3. CFG
116
117		CFG now has a bunch of methods for edge splitting and merging.
118
119		4. Machine description.
120
121		John's simplification of MLTREE_BASIS.fcond broke a few machine
122		description things:
123
124		rtl-build.{sig,sml} and hppa.mdl fixed.
125
126		NOTE: the machine description stuff in the repository is still broken.
127		Again, I can't put my fixes in because that involves
128		changes to MLTREE.
129
130		----------------------------------------------------------------------
131		Name: Matthias Blume
132		Date: 2002/03/20 15:55:00 EST
133		Tag: blume-20020320-kmp
134		Description:
135
136		Implemented Knuth-Morris-Pratt string matching in PreString and used
137		it for String.isSubstring, Substring.isSubstring, and
138		Substring.position.
139
140		(Might need some stress-testing.  Simple examples worked fine.)
141
142		----------------------------------------------------------------------
143		Name: Matthias Blume
144		Date: 2002/03/19 16:37:00 EST
145		Tag: blume-20020319-witnesses
146		Description:
147
148		Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
149
150		This implements a generic mechanism for changing constness qualifiers
151		anywhere within big C types without resorting to outright "casts".
152		(So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
153		modify the constness at the outermost level.)
154		The implementation of "convert" is based on the idea of "witness"
155		values -- values that are not used by the operation but whose types
156		"testify" to their applicability.  On the implementation side, "convert"
157		is simply a projection (returning its second curried argument).  With
158		cross-module inlining, it should not result in any machine code being
159		generated.
160
161		----------------------------------------------------------------------
162		Name: Matthias Blume
163		Date: 2002/03/15 16:40:00 EST
164		Tag: blume-20020315-basis
165		Description:
166
167		Provided (preliminary?) implementations for
168
169		{String,Substring}.{concatWith,isSuffix,isSubstring}
170
171		and
172
173		Substring.full
174
175		Those are in the Basis spec but they were missing in SML/NJ.
176
177		----------------------------------------------------------------------
178		Name: Matthias Blume
179		Date: 2002/03/14 21:30:00 EST
180		Tag: blume-20020314-controls
181		Description:
182
183		Controls:
184		---------
185
186		1. Factored out the recently-added Controls : CONTROLS stuff and put
187		it into its own library $/controls-lib.cm.  The source tree for
188		this is under src/smlnj-lib/Controls.
189
190		2. Changed the names of types and functions in this interface, so they
191		make a bit more "sense":
192
193		module -> registry
194		'a registry -> 'a group
195
196		3. The interface now deals in ref cells only.  The getter/setter interface
197		is (mostly) gone.
198
199		4. Added a function that lets one register an already-existing ref cell.
200
201		5. Made the corresponding modifications to the rest of the code so that
202		everything compiles again.
203
204		6. Changed the implementation of Controls.MLRISC back to something closer
205		to the original.  In particular, this module (and therefore MLRISC)
206		does not depend on Controls.  There now is some link-time code in
207		int-sys.sml that registers the MLRISC controls with the Controls
208		module.
209
210		CM:
211		---
212
213		* One can now specify the lambda-split aggressiveness in init.cmi.
214
215		----------------------------------------------------------------------
216		Name: Allen Leung
217		Date: 2002/03/13 17:30:00 EST
218		Tag: leunga-20020313-x86-fp-unary
219		Description:
220
221		Bug fix for:
222
223		> leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
224		> Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
225		> - fun f(x,(y,z)) = Real.~ y;
226		> [autoloading]
227		> [autoloading done]
228		>       fchsl   (%eax), 184(%esp)
229		> Error: MLRisc bug: X86MCEmitter.emitInstr
230		>
231		> uncaught exception Error
232		>   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
233
234		The problem was that the code generator did not generate any fp registers
235		in this case, and the ra didn't know that it needed to run the X86FP phase to
236		translate the pseudo fp instruction.   This only happened with unary fp
237		operators in certain situations.
238
239		----------------------------------------------------------------------
240		Name: Matthias Blume
241		Date: 2002/03/13 14:00:00 EST
242		Tag: blume-20020313-overload-etc
243		Description:
244
245		1. Added _overload as a synonym for overload for backward compatibility.
246		(Control.overloadKW must be true for either version to be accepted.)
247
248		2. Fixed bug in install script that caused more things to be installed
249		than what was requested in config/targets.
250
251		3. Made CM aware of the (_)overload construct so that autoloading
252		works.
253
254		----------------------------------------------------------------------
255		Name: Matthias Blume
256		Date: 2002/03/12 22:03:00 EST
257		Tag: blume-20020312-url
258		Description:
259
260		Forgot to update BOOT and srcarchiveurl.
261
262		----------------------------------------------------------------------
263		Name: Matthias Blume
264		Date: 2002/03/12 17:30:00 EST
265		Tag: blume-20020312-version110392
266		Description:
267
268		Yet another version number bump (because of small changes to the
269		binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
270
271		Changes:
272
273		The new pid generation scheme described a few weeks ago was overly
274		complicated.  I implemented a new mechanism that is simpler and
275		provides a bit more "stability":  Once CM has seen a compilation
276		unit, it keeps its identity constant (as long as you do not delete
277		those crucial CM/GUID/* files).  This means that when you change
278		an interface, compile, then go back to the old interface, and
279		compile again, you arrive at the original pid.
280
281		There now also is a mechanism that instructs CM to use the plain
282		environment hash as a module's pid (effectively making its GUID
283		the empty string).  For this, "noguid" must be specified as an
284		option to the .sml file in question within its .cm file.
285		This is most useful for code that is being generated by tools such
286		as ml-nlffigen (because during development programmers tend to
287		erase the tool's entire output directory tree including CM's cached
288		GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
289		revert to the old, broken behavior of SML/NJ, but in specific cases
290		where there is no danger of interface confusion, its use is ok
291		(I think).
292
293		ml-nlffigen by default generates "noguid" annotations.  They can be
294		turned off by specifying -guid in its command line.
295
296		----------------------------------------------------------------------
297		Name: Lal George
298		Date: 2002/03/12 12 14:42:36 EST
299		Tag: george-20020312-frequency-computation
300		Description:
301
302		Integrated jump chaining and static block frequency into the
303		compiler. More details and numbers later.
304
305		----------------------------------------------------------------------
306		Name: Lal George
307		Date: 2002/03/11 11 22:38:53 EST
308		Tag: george-20020311-jump-chain-elim
309		Description:
310
311		Tested the jump chain elimination on all architectures (except the
312		hppa).  This is on by default right now and is profitable for the
313		alpha and x86, however, it may not be profitable for the sparc and ppc
314		when compiling the compiler.
315
316		The gc test will typically jump to a label at the end of the cluster,
317		where there is another jump to an external cluster containing the actual
318		code to invoke gc. This is to allow factoring of common gc invocation
319		sequences. That is to say, we generate:
320
321		f:
322		testgc
323		ja   L1      % jump if above to L1
324
325		L1:
326		jmp L2
327
328
329		After jump chain elimination the 'ja L1' instructions is converted to
330		'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
331		up being implemented in their long form (if L2 is far away) using:
332
333		jbe     L3      % jump if below or equal to L3
334		jmp     L2
335		L3:
336		...
337
338
339		For large compilation units L2  may be far away.
340
341
342		----------------------------------------------------------------------
343		Name: Matthias Blume
344		Date: 2002/03/11 13:30:00 EST
345		Tag: blume-20020311-mltreeeval
346		Description:
347
348		A functor parameter was missing.
349
350		----------------------------------------------------------------------
351		Name: Allen Leung
352		Date: 2002/03/11 10:30:00 EST
353		Tag: leunga-20020311-runtime-string0
354		Description:
355
356		The representation of the empty string now points to a
357		legal null terminated C string instead of unit.  It is now possible
358		to convert an ML string into C string with InlineT.CharVector.getData.
359		This compiles into one single machine instruction.
360
361		----------------------------------------------------------------------
362		Name: Allen Leung
363		Date: 2002/03/10 23:55:00 EST
364		Tag: leunga-20020310-x86-call
365		Description:
366
367		Added machine generation for CALL instruction (relative displacement mode)
368
369		----------------------------------------------------------------------
370		Name: Matthias Blume
371		Date: 2002/03/08 16:05:00
372		Tag: blume-20020308-entrypoints
373		Description:
374
375		Version number bumped to 110.39.1.  NEW BOOTFILES!
376
377		Entrypoints: non-zero offset into a code object where execution should begin.
378
379		- Added the notion of an entrypoint to CodeObj.
380		- Added reading/writing of entrypoint info to Binfile.
381		- Made runtime system bootloader aware of entrypoints.
382		- Use the address of the label of the first function given to mlriscGen
383		as the entrypoint.  This address is currently always 0, but it will
384		not be 0 once we turn on block placement.
385		- Removed the linkage cluster code (which was The Other Way(tm) of dealing
386		with entry points) from mlriscGen.
387
388		----------------------------------------------------------------------
389		Name: Allen Leung
390		Date: 2002/03/07 20:45:00 EST
391		Tag: leunga-20020307-x86-cmov
392		Description:
393
394		Bug fixes for CMOVcc on x86.
395
396		1. Added machine code generation for CMOVcc
397		2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
398		3. CMOVcc cannot have an immediate operand as argument.
399
400		----------------------------------------------------------------------
401		Name: Matthias Blume
402		Date: 2002/03/07 16:15:00 EST
403		Tag: blume-20020307-controls
404		Description:
405
406		This is a very large but mostly boring patch which makes (almost)
407		every tuneable compiler knob (i.e., pretty much everything under
408		Control.* plus a few other things) configurable via both the command
409		line and environment variables in the style CM did its configuration
410		until now.
411
412		Try starting sml with '-h' (or, if you are brave, '-H')
413
414		To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
415		implements the underlying generic mechanism.
416
417		The interface to some of the existing such facilities has changed somewhat.
418		For example, the MLRiscControl module now provides mkFoo instead of getFoo.
419		(The getFoo interface is still there for backward-compatibility, but its
420		use is deprecated.)
421
422		The ml-build script passes -Cxxx=yyy command-line arguments through so
423		that one can now twiddle the compiler settings when using this "batch"
424		compiler.
425
426		TODO items:
427
428		We should go through and throw out all controls that are no longer
429		connected to anything.  Moreover, we should go through and provide
430		meaningful (and correct!) documentation strings for those controls
431		that still are connected.
432
433		Currently, multiple calls to Controls.new are accepted (only the first
434		has any effect).  Eventually we should make sure that every control
435		is being made (via Controls.new) exactly once.  Future access can then
436		be done using Controls.acc.
437
438		Finally, it would probably be a good idea to use the getter-setter
439		interface to controls rather than ref cells.  For the time being, both
440		styles are provided by the Controls module, but getter-setter pairs are
441		better if thread-safety is of any concern because they can be wrapped.
442
443		*****************************************
444
445		One bug fix: The function blockPlacement in three of the MLRISC
446		backpatch files used to be hard-wired to one of two possibilities at
447		link time (according to the value of the placementFlag).  But (I
448		think) it should rather sense the flag every time.
449
450		*****************************************
451
452		Other assorted changes (by other people who did not supply a HISTORY entry):
453
454		1. the cross-module inliner now works much better (Monnier)
455		2. representation of weights, frequencies, and probabilities in MLRISC
456		changed in preparation of using those for weighted block placement
457		(Reppy, George)
458
459		----------------------------------------------------------------------
460		Name: Lal George
461		Date: 2002/03/07 14:44:24 EST 2002
462		Tag: george-20020307-weighted-block-placement
463
464		Tested the weighted block placement optimization on all architectures
465		(except the hppa) using AMPL to generate the block and edge frequencies.
466		Changes were required in the machine properties to correctly
467		categorize trap instructions. There is an MLRISC flag
468		"weighted-block-placement" that can be used to enable weighted block
469		placement, but this will be ineffective without block/edge
470		frequencies (coming soon).
471
472
473		----------------------------------------------------------------------
474		Name: Lal George
475		Date: 2002/03/05 17:24:48 EST
476		Tag: george-20020305-linkage-cluster
477
478		In order to support the block placement optimization, a new cluster
479		is generated as the very first cluster (called the linkage cluster).
480		It contains a single jump to the 'real' entry point for the compilation
481		unit. Block placement has no effect on the linkage cluster itself, but
482		all the other clusters  have full freedom in the manner in which they
483		reorder blocks or functions.
484
485		On the x86 the typical linkage code that is generated is:
486		----------------------
487		.align 2
488		L0:
489		addl    $L1-L0, 72(%esp)
490		jmp     L1
491
492
493		.align  2
494		L1:
495		----------------------
496
497		72(%esp) is the memory location for the stdlink register. This
498		must contain the address of the CPS function being called. In the
499		above example, it contains the address of  L0; before
500		calling L1 (the real entry point for the compilation unit), it
501		must contain the address for L1, and hence
502
503		addl $L1-L0, 72(%esp)
504
505		I have tested this on all architectures except the hppa.The increase
506		in code size is of course negligible
507
508		----------------------------------------------------------------------
509		Name: Allen Leung
510		Date: 2002/03/03 13:20:00 EST
511		Tag: leunga-20020303-mlrisc-tools
512
513		Added #[ ... ] expressions to mlrisc tools
514
515		----------------------------------------------------------------------
516		Name: Matthias Blume
517		Date: 2002/02/27 12:29:00 EST
518		Tag: blume-20020227-cdebug
519		Description:
520
521		- made types in structure C and C_Debug to be equal
522		- got rid of code duplication (c-int.sml vs. c-int-debug.sml)
523		- there no longer is a C_Int_Debug (C_Debug is directly derived from C)
524
525		----------------------------------------------------------------------
526		Name: Matthias Blume
527		Date: 2002/02/26 12:00:00 EST
528		Tag: blume-20020226-ffi
529		Description:
530
531		1. Fixed a minor bug in CM's "noweb" tool:
532		If numbering is turned off, then truly don't number (i.e., do not
533		supply the -L option to noweb).  The previous behavior was to supply
534		-L'' -- which caused noweb to use the "default" line numbering scheme.
535		Thanks to Chris Richards for pointing this out (and supplying the fix).
536
537		2. Once again, I reworked some aspects of the FFI:
538
539		A. The incomplete/complete type business:
540
541		- Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are
542		gone!
543		- ML types representing an incomplete type are now *equal* to
544		ML types representing their corresponding complete types (just like
545		in C).  This is still safe because ml-nlffigen will not generate
546		RTTI for incomplete types, nor will it generate functions that
547		require access to such RTTI.   But when ML code generated from both
548		incomplete and complete versions of the C type meet, the ML types
549		are trivially interoperable.
550
551		NOTE:  These changes restore the full generality of the translation
552		(which was previously lost when I eliminated functorization)!
553
554		B. Enum types:
555
556		- Structure C now has a type constructor "enum" that is similar to
557		how the "su" constructor works.  However, "enum" is not a phantom
558		type because each "T enum" has values (and is isomorphic to
559		MLRep.Signed.int).
560		- There are generic access operations for enum objects (using
561		MLRep.Signed.int).
562		- ml-nlffigen will generate a structure E_foo for each "enum foo".
563		* The structure contains the definition of type "mlrep" (the ML-side
564		representation type of the enum).  Normally, mlrep is the same
565		as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec",
566		then mlrep will be defined as a datatype -- thus facilitating
567		pattern matching on mlrep values.
568		("-ec" will be suppressed if there are duplicate values in an
569		enumeration.)
570		* Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep
571		will be generated for each C enum constant xxx.
572		* Conversion functions m2i and i2m convert between mlrep and
573		MLRep.Signed.int.  (Without "-ec", these functions are identities.)
574		* Coversion functions c and ml convert between mlrep and "tag enum".
575		* Access functions (get/set) fetch and store mlrep values.
576		- By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed
577		enumerations are merged into one single enumeration represented by
578		structure E_'.
579
580		----------------------------------------------------------------------
581		Name: Allen Leung
582		Date: 2002/02/25 04:45:00 EST
583		Tag: leunga-20020225-cps-spill
584
585		This is a new implementation of the CPS spill phase.
586		The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml
587		In case of problems, replace it with the old file spill.sml
588
589		The current compiler runs into some serious performance problems when
590		constructing a large record.  This can happen when we try to compile a
591		structure with many items.  Even a very simple structure like the following
592		makes the compiler slow down.
593
594		structure Foo = struct
595		val x_1 = 0w1 : Word32.int
596		val x_2 = 0w2 : Word32.int
597		val x_3 = 0w3 : Word32.int
598		...
599		val x_N = 0wN : Word32.int
600		end
601
602		The following table shows the compile time, from N=1000 to N=4000,
603		with the old compiler:
604
605		N
606		1000   CPS 100 spill                           0.04u  0.00s  0.00g
607		MLRISC ra                               0.06u  0.00s  0.05g
608		(spills = 0 reloads = 0)
609		TOTAL                                   0.63u  0.07s  0.21g
610
611		1100   CPS 100 spill                           8.25u  0.32s  0.64g
612		MLRISC ra                               5.68u  0.59s  3.93g
613		(spills = 0 reloads = 0)
614		TOTAL                                   14.71u  0.99s  4.81g
615
616		1500   CPS 100 spill                           58.55u  2.34s  1.74g
617		MLRISC ra                               5.54u  0.65s  3.91g
618		(spills = 543 reloads = 1082)
619		TOTAL                                   65.40u  3.13s  6.00g
620
621		2000   CPS 100 spill                           126.69u  4.84s  3.08g
622		MLRISC ra                               0.80u  0.10s  0.55g
623		(spills = 42 reloads = 84)
624		TOTAL                                   129.42u  5.10s  4.13g
625
626		3000   CPS 100 spill                           675.59u  19.03s  11.64g
627		MLRISC ra                               2.69u  0.27s  1.38g
628		(spills = 62 reloads = 124)
629		TOTAL                                   682.48u  19.61s  13.99g
630
631		4000   CPS 100 spill                           2362.82u  56.28s  43.60g
632		MLRISC ra                               4.96u  0.27s  2.72g
633		(spills = 85 reloads = 170)
634		TOTAL                                   2375.26u  57.21s  48.00g
635
636		As you can see the old cps spill module suffers from some serious
637		performance problem.  But since I cannot decipher the old code fully,
638		instead of patching the problems up, I'm reimplementing it
639		with a different algorithm.  The new code is more modular,
640		smaller when compiled, and substantially faster
641		(O(n log n) time and O(n) space).  Timing of the new spill module:
642
643		4000  CPS 100 spill                           0.02u  0.00s  0.00g
644		MLRISC ra                               0.25u  0.02s  0.15g
645		(spills=1 reloads=3)
646		TOTAL                                   7.74u  0.34s  1.62g
647
648		Implementation details:
649
650		As far as I can tell, the purpose of the CPS spill module is to make sure the
651		number of live variables at any program point (the bandwidth)
652		does not exceed a certain limit, which is determined by the
653		size of the spill area.
654
655		When the bandwidth is too large, we decrease the register pressure by
656		packing live variables into spill records.  How we achieve this is
657		completely different than what we did in the old code.
658
659		First, there is something about the MLRiscGen code generator
660		that we should be aware of:
661
662		o MLRiscGen performs code motion!
663
664		In particular, it will move floating point computations and
665		address computations involving only the heap pointer to
666		their use sites (if there is only a single use).
667		What this means is that if we have a CPS record construction
668		statement
669
670		RECORD(k,vl,w,e)
671
672		we should never count the new record address w as live if w
673		has only one use (which is often the case).
674
675		We should do something similar to floating point, but the transformation
676		there is much more complex, so I won't deal with that.
677
678		Secondly, there are now two new cps primops at our disposal:
679
680		1. rawrecord of record_kind option
681		This pure operator allocates some uninitialized storage from the heap.
682		There are two forms:
683
684		rawrecord NONE [INT n]  allocates a tagless record of length n
685		rawrecord (SOME rk) [INT n] allocates a tagged record of length n
686		and initializes the tag.
687
688		2. rawupdate of cty
689		rawupdate cty (v,i,x)
690		Assigns to x to the ith component of record v.
691		The storelist is not updated.
692
693		We use these new primops for both spilling and increment record construction.
694
695		1. Spilling.
696
697		This is implemented with a linear scan algorithm (but generalized
698		to trees).  The algorithm will create a single spill record at the
699		beginning of the cps function and use rawupdate to spill to it,
700		and SELECT or SELp to reload from it.  So both spills and reloads
701		are fine-grain operations.  In contrast, in the old algorithm
702		"spills" have to be bundled together in records.
703
704		Ideally, we should sink the spill record construction to where
705		it is needed.  We can even split the spill record into multiple ones
706		at the places where they are needed.  But CPS is not a good
707		representation for global code motion, so I'll keep it simple and
708		am not attempting this.
709
710		2. Incremental record construction (aka record splitting).
711
712		Long records with many component values which are simulatenously live
713		(recall that single use record addresses are not considered to
714		be live) are constructed with rawrecord and rawupdate.
715		We allocate space on the heap with rawrecord first, then gradually
716		fill it in with rawupdate.  This is the technique suggested to me
717		by Matthias.
718
719		Some restrictions on when this is applicable:
720		1. It is not a VECTOR record.  The code generator currently does not handle
721		this case. VECTOR record uses double indirection like arrays.
722		2. All the record component values are defined in the same "basic block"
723		as the record constructor.  This is to prevent speculative
724		record construction.
725
726		----------------------------------------------------------------------
727		Name: Allen Leung
728		Date: 2002/02/22 01:02:00 EST
729		Tag: leunga-20020222-mlrisc-tools
730
731		Minor bug fixes in the parser and rewriter
732
733		----------------------------------------------------------------------
734		Name: Allen Leung
735		Date: 2002/02/21 20:20:00 EST
736		Tag: leunga-20020221-peephole
737
738		Regenerated the peephole files.  Some contained typos in the specification
739		and some didn't compile because of pretty printing bugs in the old version
740		of 'nowhere'.
741
742		----------------------------------------------------------------------
743		Name: Allen Leung
744		Date: 2002/02/19 20:20:00 EST
745		Tag: leunga-20020219-mlrisc-tools
746		Description:
747
748		Minor bug fixes to the mlrisc-tools library:
749
750		1.  Fixed up parsing colon suffixed keywords
751		2.  Added the ability to shut the error messages up
752		3.  Reimplemented the pretty printer and fixed up/improved
753		the pretty printing of handle and -> types.
754		4.  Fixed up generation of literal symbols in the nowhere tool.
755		5.  Added some SML keywords to to sml.sty
756
757		----------------------------------------------------------------------
758		Name: Matthias Blume
759		Date: 2002/02/19 16:20:00 EST
760		Tag: blume-20020219-cmffi
761		Description:
762
763		A wild mix of changes, some minor, some major:
764
765		* All C FFI-related libraries are now anchored under $c:
766		$/c.cm      --> $c/c.cm
767		$/c-int.cm  --> $c/internals/c-int.cm
768		$/memory.cm --> $c/memory/memory.cm
769
770		* "make" tool (in CM) now treats its argument pathname slightly
771		differently:
772		1. If the native expansion is an absolute name, then before invoking
773		the "make" command on it, CM will apply OS.Path.mkRelative
774		(with relativeTo = OS.FileSys.getDir()) to it.
775		2. The argument will be passed through to subsequent phases of CM
776		processing without "going native".  In particular, if the argument
777		was an anchored path, then "make" will not lose track of that anchor.
778
779		* Compiler backends now "know" their respective C calling conventions
780		instead of having to be told about it by ml-nlffigen.  This relieves
781		ml-nlffigen from one of its burdens.
782
783		* The X86Backend has been split into X86CCallBackend and X86StdCallBackend.
784
785		* Export C_DEBUG and C_Debug from $c/c.cm.
786
787		* C type encoding in ml-nlffi-lib has been improved to model the conceptual
788		subtyping relationship between incomplete pointers and their complete
789		counterparts.  For this, ('t, 'c) ptr has been changed to 'o ptr --
790		with the convention of instantiating 'o with ('t, 'c) obj whenever
791		the pointer target type is complete.  In the incomplete case, 'o
792		will be instantiated with some "'c iobj" -- a type obtained by
793		using one of the functors PointerToIncompleteType or PointerToCompleteType.
794
795		Operations that work on both incomplete and complete pointer types are
796		typed as taking an 'o ptr while operations that require the target to
797		be known are typed as taking some ('t, 'c) obj ptr.
798
799		voidptr is now a bit "more concrete", namely "type voidptr = void ptr'"
800		where void is an eqtype without any values.  This makes it possible
801		to work on voidptr values using functions meant to operate on light
802		incomplete pointers.
803
804		* As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has
805		been vastly simplified.
806
807		----------------------------------------------------------------------
808		Name: Matthias Blume
809		Date: 2002/02/19 10:48:00 EST
810		Tag: blume-20020219-pqfix
811		Description:
812
813		Applied Chris Okasaki's bug fix for priority queues.
814
815		----------------------------------------------------------------------
816		Name: Matthias Blume
817		Date: 2002/02/15 17:05:00
818		Tag: Release_110_39
819		Description:
820
821		Last-minute retagging is becoming a tradition... :-(
822
823		This is the working release 110.39.
824
825		----------------------------------------------------------------------
826		Name: Matthias Blume
827		Date: 2002/02/15 16:00:00 EST
828		Tag: Release_110_39-orig
829		Description:
830
831		Working release 110.39.  New bootfiles.
832
833		(Update: There was a small bug in the installer so it wouldn't work
834		with all shells.  So I retagged. -Matthias)
835
836		----------------------------------------------------------------------
837		Name: Matthias Blume
838		Date: 2002/02/15 14:17:00 EST
839		Tag: blume-20020215-showbindings
840		Description:
841
842		Added EnvRef.listBoundSymbols and CM.State.showBindings.  Especially
843		the latter can be useful for exploring what bindings are available at
844		the interactive prompt.  (The first function returns only the list
845		of symbols that are really bound, the second prints those but also the
846		ones that CM's autoloading mechanism knows about.)
847
848		----------------------------------------------------------------------
849		Name: Matthias Blume
850		Date: 2002/02/15 12:08:00 EST
851		Tag: blume-20020215-iptrs
852		Description:
853
854		Two improvements to ml-nlffigen:
855
856		1. Write files only if they do not exist or if their current contents
857		do not coincide with what's being written.  (That is, avoid messing
858		with the time stamps unless absolutely necessary.)
859
860		2. Implement a "repository" mechanism for generated files related
861		to "incomplete pointer types".   See the README file for details.
862
863		----------------------------------------------------------------------
864		Name: Matthias Blume
865		Date: 2002/02/14 11:50:00 EST
866		Tag: blume-20020214-quote
867		Description:
868
869		Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm).  This is required
870		because of the new and improved tag generation scheme.  (Thanks to Allen
871		Leung for pointing it out.)
872
873		----------------------------------------------------------------------
874		Name: Lal George
875		Date: 2002/02/14 09:55:27 EST 2002
876		Tag: george-20020214-isabelle-bug
877		Description:
878
879		Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation
880		of Isabelle on the x86.
881
882		From Allen:
883		-----------
884		I've found the problem:
885
886		in ra-core.sml, I use the counter "blocked" to keep track of the
887		true number of elements in the freeze queue.  When the counter goes
888		to zero, I skip examining the queue.  But I've messed up the
889		bookkeeping in combine():
890
891		else ();
892		case !ucol of
893		PSEUDO => (if !cntv > 0 then
894		(if !cntu > 0 then blocked := !blocked - 1 else ();
895		^^^^^^^^^^^^^^^^^^^^^^^
896		moveu := mergeMoveList(!movev, !moveu)
897		)
898		else ();
899
900		combine() is called to coalesce two nodes u and v.
901		I think I was thinking that if the move counts of u and v are both
902		greater than zero then after they are coalesced then one node is
903		removed from the freeze queue.  Apparently I was thinking that
904		both u and v are of low degree, but that's clearly not necessarily true.
905
906
907		02/12/2002:
908		Here's the patch.  HOL now compiles.
909
910		I don't know how this impact on performance (compile
911		time or runtime).  This bug caused the RA (especially on the x86)
912		to go thru the potential spill phase when there are still nodes on the
913		freeze queue.
914
915
916
917
918		----------------------------------------------------------------------
919		Name: Matthias Blume
920		Date: 2002/02/13 22:40:00 EST
921		Tag: blume-20020213-fptr-rtti
922		Description:
923
924		Fixed a bug in ml-nlffigen that was introduced with one of the previous
925		updates.
926
927		----------------------------------------------------------------------
928		Name: Matthias Blume
929		Date: 2002/02/13 16:41:00 EST
930		Tag: blume-20020213-cmlpq
931		Description:
932
933		Added new priority queue export symbols (which have just been added to
934		smlnj-lib.cm) to CML's version of smlnj-lib.cm.  (Otherwise CML would
935		not compile and the installer would choke.)
936
937		----------------------------------------------------------------------
938		Name: Matthias Blume
939		Date: 2002/02/13 16:15:00 EST
940		Tag: blume-20020213-various
941		Description:
942
943		1. More tweaks to ml-nlffigen:
944
945		- better internal datastructures (resulting in slight speedup)
946		- "-match" option requires exact match
947		- "localized" gensym counters (untagged structs/unions nested within
948		other structs/unions or within typedefs get a fresh counter; their
949		tag will be prefixed by a concatenation of their parents' tags)
950		- bug fixes (related to calculation of transitive closure of types
951		to be included in the output)
952
953		2. Minor Basis updates:
954
955		- added implementations for List.collate and Option.app
956
957		----------------------------------------------------------------------
958		Name: Matthias Blume
959		Date: 2002/02/11 15:55:00 EST
960		Tag: blume-20020211-gensym
961		Description:
962
963		Added a "-gensym" option to command line of ml-nlffigen.  This can be
964		used to specify a "stem" -- a string that is inserted in all "gensym'd"
965		names (ML structure names that correspond to unnamed C structs, unions,
966		and enums), so that separate runs of ml-nlffigen do not clash.
967
968		----------------------------------------------------------------------
969		Name: Matthias Blume
970		Date: 2002/02/11 12:05:00 EST
971		Tag: blume-20020211-gensml
972		Description:
973
974		A quick fix for a problem with GenSML (in the pgraph-util library):
975		Make generation of toplevel "local" optional.  (Strictly speaking,
976		signature definitions within "local" are not legal SML.)
977
978		Other than that: updates to INSTALL and cm/TODO.
979
980		----------------------------------------------------------------------
981		Name: Matthias Blume
982		Date: 2002/02/08 15:00:00 EST
983		Tag: blume-20020208-uniquepid
984		Description:
985
986		0. Version number has been bumped to 110.38.1.  NEW BOOTFILES!!!
987
988		1. The installer (config/install.sh) has gotten smarter:
989
990		- Configuration options are a bit easier to specify now
991		(in config/targets).
992		- Bug in recognizing .tar.bz2 files fixed.
993		- Installer automatically resolves dependencies between
994		configuration options (e.g., if you ask for eXene, you will
995		also get cml -- regardless whether you asked for it or not).
996		- Installer can run in "quieter mode" by setting the environment
997		variable INSTALL_QUIETLY to "true".  "Quieter" does not mean
998		"completely silent", though.
999		- Build HashCons library as part of smlnj-lib.
1000
1001		2. A new scheme for assigning persistent identifiers to compilation
1002		units (and, by extension, to types etc.) has been put into place.
1003		This fixes a long-standing bug where types and even dynamic values
1004		can get internally confused, thereby compromising type safety
1005		(abstraction) and dynamic correctness.  See
1006
1007		http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz
1008
1009		for an example of how things could go wrong until now.
1010
1011		The downside of the new scheme is that pids are not quite as
1012		persistent as they used to be: CM will generate a fresh pid
1013		for every compilation unit that it thinks it sees for the first
1014		time.  That means that if you compile starting from a clean, fresh
1015		source tree at two different times, you end up with different
1016		binaries.
1017
1018		Cutoff recompilation, however, has not been compromised because
1019		CM keeps pid information in special caches between runs.
1020
1021		----------------------------------------------------------------------
1022		Name: Lal George
1023		Date: 2002/02/07 15:34:13 EST 2002
1024		Tag: <none>
1025		Description:
1026
1027		Compilers that generate assembly code may produce  global labels
1028		whose value is resolved at link time. The various peephole optimization
1029		modules did not take this in account.
1030
1031		TODO. The Labels.addrOf function should really return an option
1032		type so that clients are forced to deal with this issue, rather
1033		than an exception being raised.
1034
1035		----------------------------------------------------------------------
1036		Name: Lal George
1037		Date: 2002/02/06 13:55:02 EST
1038		Tag: george-20020206-ra-breakup
1039		Description:
1040
1041		1. A bug fix from Allen.
1042
1043		A typo causes extra fstp %st(0)'s to be generated at compensation
1044		edges, which might cause stack underflow traps at runtime.  This
1045		occurs in fft where there are extraneous fstps right before the 'into'
1046		trap instruction (in this case they are harmless since none of the
1047		integers overflow.)
1048
1049		2. Pulled out various utility modules that were embedded in the modules
1050		of the register allocator. I need these modules for other purposes, but
1051		they are not complete enough to put into a library (just yet).
1052		----------------------------------------------------------------------
1053	Name: Matthias Blume	Name: Matthias Blume
1054	Date: 2002/01/31 16:05:00 EST	Date: 2002/01/31 16:05:00 EST
1055	Tag: blume-20020131-sparc-ccalls	Tag: blume-20020131-sparc-ccalls
#	Line 633	Line 1670
1670
1671	----------------------------------------------------------------------	----------------------------------------------------------------------
1672	Name: Matthias Blume	Name: Matthias Blume
	>>>>>>> 1.169
1673	Date: 2001/09/18 15:35:00 EDT	Date: 2001/09/18 15:35:00 EDT
1674	Tag: blume-20010918-readme11036	Tag: blume-20010918-readme11036
1675	Description:	Description:

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.1049   changed lines   Added in v.1171

---

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