Diff of /sml/trunk/HISTORY

Parent Directory | Revision Log | Patch

	revision 1115, Sun Mar 3 18:16:31 2002 UTC	revision 1189, Fri May 10 20:40:33 2002 UTC
#	Line 13	Line 13
13	Description:	Description:
14
15	----------------------------------------------------------------------	----------------------------------------------------------------------
16		Name: Matthias Blume
17		Date: 2002/05/10 16:40:00 EDT
18		Tag: blume-20020510-erg-textio
19		Description:
20
21		Applied the following bugfix provided by Emden Gansner:
22
23		Output is corrupted when outputSubstr is used rather than output.
24
25		The problem occurs when a substring
26
27		ss = (s, dataStart, dataLen)
28
29		where dataStart > 0, fills a stream buffer with avail bytes left.
30		avail bytes of s, starting at index dataStart, are copied into the
31		buffer, the buffer is flushed, and then the remaining dataLen-avail
32		bytes of ss are copied into the beginning of the buffer. Instead of
33		starting this copy at index dataStart+avail in s, the current code
34		starts the copy at index avail.
35
36		Fix:
37		In text-io-fn.sml, change line 695 from
38		val needsFlush = copyVec(v, avail, dataLen-avail, buf, 0)
39		to
40		val needsFlush = copyVec(v, dataStart+avail, dataLen-avail, buf, 0)
41
42		----------------------------------------------------------------------
43		Name: Matthias Blume
44		Date: 2002/04/12 13:55:00 EDT
45		Tag: blume-20020412-assyntax
46		Description:
47
48		1. Grabbed newer assyntax.h from the XFree86 project.
49		2. Fiddled with how to compile X86.prim.asm without warnings.
50		3. (Very) Minor cleanup in CM.
51
52		----------------------------------------------------------------------
53		Name: Matthias Blume
54		Date: 2002/04/01 (no joke!) 17:07:00 EST
55		Tag: blume-20020401-x86div
56		Description:
57
58		Added full support for div/mod/rem/quot on the x86, using the machine
59		instruction's two results (without clumsily recomputing the remainder)
60		directly where appropriate.
61
62		Some more extensive power-of-two support was added to the x86 instruction
63		selector (avoiding expensive divs, mods, and muls where they can be
64		replaced with cheaper shifts and masks).  However, this sort of thing
65		ought to be done earlier, e.g., within the CPS optimizer so that
66		all architectures benefit from it.
67
68		The compiler compiles to a fixed point, but changes might be somewhat
69		fragile nevertheless.  Please, report any strange things that you might
70		see wrt. div/mod/quot/rem...
71
72		----------------------------------------------------------------------
73		Name: Matthias Blume
74		Date: 2002/03/29 17:22:00
75		Tag: blume-20020329-div
76		Description:
77
78		Fixed my broken div/mod logic.  Unfortunately, this means that the
79		inline code for div/mod now has one more comparison than before.
80		Fast paths (quotient > 0 or remainder = 0) are not affected, though.
81		The problem was with quotient = 0, because that alone does not tell
82		us which way the rounding went.  One then has to look at whether
83		remainder and divisor have the same sign...  :(
84
85		Anyway, I replaced the bootfiles with fresh ones...
86
87		----------------------------------------------------------------------
88		Name: Matthias Blume
89		Date: 2002/03/29 14:10:00 EST
90		Tag: blume-20020329-inlprims
91		Description:
92
93		NEW BOOTFILES!!!    Version number bumped to 110.39.3.
94
95		Primops have changed. This means that the bin/boot-file formats have
96		changed as well.
97
98		To make sure that there is no confusion, I made a new version.
99
100
101		CHANGES:
102
103		* removed REMT from mltree (remainder should never overflow).
104
105		* added primops to deal with divisions of all flavors to the frontend
106
107		* handled these primops all the way through so they map to their respective
108		MLRISC support
109
110		* used these primops in the implementation of Int, Int32, Word, Word32
111
112		* removed INLDIV, INLMOD, and INLREM as they are no longer necessary
113
114		* parameterized INLMIN, INLMAX, and INLABS by a numkind
115
116		* translate.sml now deals with all flavors of INL{MIN,MAX,ABS}, including
117		floating point
118
119		* used INL{MIN,MAX,ABS} in the implementation of Int, Int32, Word, Word32,
120		and Real (but Real.abs maps to a separate floating-point-only primop)
121
122
123		TODO items:
124
125		* Hacked Alpha32 instruction selection, disabling the selection of REMx
126		instructions because the machine instruction encoder cannot handle
127		them.  (Hppa, PPC, and Sparc instruction selection did not handle
128		REM in the first place, and REM is supported by the x86 machine coder.)
129
130		* Handle DIV and MOD with DIV_TO_NEGINF directly in the x86 instruction
131		selection phase.  (The two can be streamlined because the hardware
132		delivers both quotient and remainder at the same time anyway.)
133
134		* Think about what to do with "valOf(Int32.minInt) div ~1" and friends.
135		(Currently the behavior is inconsistent both across architectures and
136		wrt. the draft Basis spec.)
137
138		* Word8 should eventually be handled natively, too.
139
140		* There seems to be one serious bug in mltree-gen.sml.  It appears, though,
141		as if there currently is no execution path that could trigger it in
142		SML/NJ.  (The assumptions underlying functions arith and promotable do not
143		hold for things like multiplication and division.)
144
145		----------------------------------------------------------------------
146		Name: Matthias Blume
147		Date: 2002/03/27 16:27:00 EST
148		Tag: blume-20020327-mlrisc-divisions
149		Description:
150
151		Added support for all four division operations (ML's div, mod, quot,
152		and rem) to MLRISC.  In the course of doing so, I also rationalized
153		the naming (no more annoying switch-around of DIV and QUOT), by
154		parameterizing the operation by div_rounding_mode (which can be either
155		DIV_TO_ZERO or DIV_TO_NEGINF).
156
157		The generic MLTreeGen functor takes care of compiling all four
158		operations down to only round-to-zero div.
159
160		Missing pieces:
161
162		* Doing something smarter than relying on MLTreeGen on architectures
163		like, e.g., the x86 where hardware division delivers both quotient and
164		remainder at the same time.  With this, the implementation of the
165		round-to-neginf operations could be further streamlined.
166
167		* Remove inlining support for div/mod/rem from the frontend and replace it
168		with primops that get carried through to the backend.  Do this for all
169		int and word types.
170
171		----------------------------------------------------------------------
172		Name: Matthias Blume
173		Date: 2002/03/25 17:25:00 EST
174		Tag: blume-20020325-divmod
175		Description:
176
177		I improved (hopefully without breaking them) the implementation of Int.div,
178		Int.mod, and Int.rem.   For this, the code in translate.sml now takes
179		advantage of the following observations:
180
181		Let  q = x quot y      r = x rem y
182		d = x div  y      m = x mod y
183
184		where "quot" is the round-to-zero version of integer division that
185		hardware usually provides.  Then we have:
186
187		r = x - q * y        where neither the * nor the - will overflow
188		d = if q >= 0 orelse x = q * y then q else q - 1
189		where neither the * nor the - will overflow
190		m = if q >= 0 orelse r = 0 then r else r + y
191		where the + will not overflow
192
193		This results in substantial simplification of the generated code.
194		The following table shows the number of CFG nodes and edges generated
195		for
196		fun f (x, y) = x OPER y
197		(* with OPER \in div, mod, quot, rem *)
198
199
200		OPER | nodes(old) | edges(old) | nodes(new) | edges(new)
201		--------------------------------------------------------
202		div |         24 |         39 |         12 |         16
203		mod |         41 |         71 |         12 |         16
204		quot |          8 |         10 |          8 |         10
205		rem |         10 |         14 |          8 |         10
206
207
208		----------------------------------------------------------------------
209		Name: Matthias Blume
210		Date: 2002/03/25 22:06:00 EST
211		Tag: blume-20020325-cprotobug
212		Description:
213
214		Fixed a bug in cproto (c prototype decoder).
215
216		----------------------------------------------------------------------
217		Name: Matthias Blume
218		Date: 2002/03/25 16:00:00 EST
219		Tag: blume-20020325-raw-primops
220		Description:
221
222		I did some cleanup to Allen's new primop code and
223		replaced yesterday's bootfiles with new ones.
224		(But they are stored in the same place.)
225
226		----------------------------------------------------------------------
227		Name: Matthias Blume
228		Date: 2002/03/24 22:40:00 EST
229		Tag: blume-20020324-bootfiles
230		Description:
231
232		Made the bootfiles that Allen asked for.
233
234		----------------------------------------------------------------------
235		Name: Allen Leung
236		Date: 2002/03/23 15:50:00 EST
237		Tag: leunga-20020323-flint-cps-rcc-primops
238		Description:
239
240		1. Changes to FLINT primops:
241
242		(* make a call to a C-function;
243		* The primop carries C function prototype information and specifies
244		* which of its (ML-) arguments are floating point. C prototype
245		* information is for use by the backend, ML information is for
246		* use by the CPS converter. *)
247		| RAW_CCALL of { c_proto: CTypes.c_proto,
248		ml_args: ccall_type list,
249		ml_res_opt: ccall_type option,
250		reentrant : bool
251		} option
252		(* Allocate uninitialized storage on the heap.
253		* The record is meant to hold short-lived C objects, i.e., they
254		* are not ML pointers.  With the tag, the representation is
255		* the same as RECORD with tag tag_raw32 (sz=4), or tag_fblock (sz=8)
256		*)
257		| RAW_RECORD of {tag:bool,sz:int}
258		and ccall_type = CCALL_INT32 | CCALL_REAL64 | CCALL_ML_PTR
259
260		2.  These CPS primops are now overloaded:
261
262		rawload of {kind:numkind}
263		rawstore of {kind:numkind}
264
265		The one argument form is:
266
267		rawload {kind} address
268
269		The two argument form is:
270
271		rawload {kind} [ml object, byte-offset]
272
273		3. RAW_CCALL/RCC now takes two extra arguments:
274
275		a. The first is whether the C call is reentrant, i.e., whether
276		ML state should be saved and restored.
277		b. The second argument is a string argument specifying the name of
278		library and the C function.
279
280		These things are currently not handled in the code generator, yet.
281
282		4. In CProto,
283
284		An encoding type of "bool" means "ml object" and is mapped into
285		C prototype of PTR.  Note that "bool" is different than "string",
286		even though "string" is also mapped into PTR, because "bool"
287		is assigned an CPS type of BOGt, while "string" is assigned INT32t.
288
289		5. Pickler/unpicker
290
291		Changed to handle RAW_RECORD and newest RAW_CCALL
292
293		6. MLRiscGen,
294
295		1. Changed to handle the new rawload/rawstore/rawrecord operators.
296		2. Code for handling C Calls has been moved to a new module CPSCCalls,
297		in the file CodeGen/cpscompile/cps-c-calls.sml
298
299		7. Added the conditional move operator
300
301		condmove of branch
302
303		to cps.  Generation of this is still buggy so it is currently
304		disabled.
305
306		----------------------------------------------------------------------
307		Name: Lal George
308		Date: 2002/03/22 14:18:25 EST
309		Tag: george-20020322-cps-branch-prob
310		Description:
311
312		Implemented the Ball-Larus branch prediction-heuristics, and
313		incorporated graphical viewers for control flow graphs.
314
315		Ball-Larus Heuristics:
316		---------------------
317		See the file compiler/CodeGen/cpscompile/cpsBranchProb.sml.
318
319		By design it uses the Dempster-Shafer theory for combining
320		probabilities.  For example, in the function:
321
322		fun f(n,acc) = if n = 0 then acc else f(n-1, n*acc)
323
324		the ball-larus heuristics predicts that the n=0 is unlikely
325		(OH-heuristic), and the 'then' branch is unlikely because of the
326		RH-heuristic -- giving the 'then' branch an even lower combined
327		probability using the Dempster-Shafer theory.
328
329		Finally, John Reppy's loop analysis in MLRISC, further lowers the
330		probability of the 'then' branch because of the loop in the else
331		branch.
332
333
334		Graphical Viewing:
335		------------------
336		I merely plugged in Allen's graphical viewers into the compiler. The
337		additional code is not much. At the top level, saying:
338
339		Control.MLRISC.getFlag "cfg-graphical-view" := true;
340
341		will display the graphical view of the control flow graph just before
342		back-patching.  daVinci must be in your path for this to work. If
343		daVinci is not available, then the default viewer can be changed
344		using:
345
346		Control.MLRISC.getString "viewer"
347
348		which can be set to "dot" or "vcg" for the corresponding viewers. Of
349		course, these viewers must be in your path.
350
351		The above will display the compilation unit at the level of clusters,
352		many of which are small, boring, and un-interesting. Also setting:
353
354		Control.MLRISC.getInt "cfg-graphical-view_size"
355
356		will display clusters that are larger than the value set by the above.
357
358
359		----------------------------------------------------------------------
360		Name: Matthias Blume
361		Date: 2002/03/21 22:20:00 EST
362		Tag: blume-20020321-kmp-bugfix
363		Description:
364
365		Changed the interface to the KMP routine in PreString and fixed
366		a minor bug in one place where it was used.
367
368		----------------------------------------------------------------------
369		Name: Allen Leung
370		Date: 2002/03/21 20:30:00 EST
371		Tag: leunga-20020321-cfg
372		Description:
373
374		Fixed a potential problem in cfg edge splitting.
375
376		----------------------------------------------------------------------
377		Name: Allen Leung
378		Date: 2002/03/21 17:15:00 EST
379		Tag: leunga-20020321-x86-fp-cfg
380		Description:
381
382		1. Recoded the buggy parts of x86-fp.
383
384		a. All the block reordering code has been removed.
385		We now depend on the block placement phases to do this work.
386
387		b. Critical edge splitting code has been simplified and moved into the
388		CFG modules, as where they belong.
389
390		Both of these were quite buggy and complex.  The code is now much, much
391		simpler.
392
393		2. X86 backend.
394
395		a. Added instructions for 64-bit support.  Instruction selection for
396		64-bit has not been committed, however, since that
397		requires changes to MLTREE which haven't been approved by
398		Lal and John.
399
400		b. Added support for FUCOMI and FUCOMIP when generating code for
401		PentiumPro and above.  We only generate these instructions in
402		the fast-fp mode.
403
404		c. Added cases for JP and JNP in X86FreqProps.
405
406		3. CFG
407
408		CFG now has a bunch of methods for edge splitting and merging.
409
410		4. Machine description.
411
412		John's simplification of MLTREE_BASIS.fcond broke a few machine
413		description things:
414
415		rtl-build.{sig,sml} and hppa.mdl fixed.
416
417		NOTE: the machine description stuff in the repository is still broken.
418		Again, I can't put my fixes in because that involves
419		changes to MLTREE.
420
421		----------------------------------------------------------------------
422		Name: Matthias Blume
423		Date: 2002/03/20 15:55:00 EST
424		Tag: blume-20020320-kmp
425		Description:
426
427		Implemented Knuth-Morris-Pratt string matching in PreString and used
428		it for String.isSubstring, Substring.isSubstring, and
429		Substring.position.
430
431		(Might need some stress-testing.  Simple examples worked fine.)
432
433		----------------------------------------------------------------------
434		Name: Matthias Blume
435		Date: 2002/03/19 16:37:00 EST
436		Tag: blume-20020319-witnesses
437		Description:
438
439		Added a structure C.W and functions convert/Ptr.convert to ml-nlffi-lib.
440
441		This implements a generic mechanism for changing constness qualifiers
442		anywhere within big C types without resorting to outright "casts".
443		(So far, functions such as C.rw/C.ro or C.Ptr.rw/C.Ptr.ro only let you
444		modify the constness at the outermost level.)
445		The implementation of "convert" is based on the idea of "witness"
446		values -- values that are not used by the operation but whose types
447		"testify" to their applicability.  On the implementation side, "convert"
448		is simply a projection (returning its second curried argument).  With
449		cross-module inlining, it should not result in any machine code being
450		generated.
451
452		----------------------------------------------------------------------
453		Name: Matthias Blume
454		Date: 2002/03/15 16:40:00 EST
455		Tag: blume-20020315-basis
456		Description:
457
458		Provided (preliminary?) implementations for
459
460		{String,Substring}.{concatWith,isSuffix,isSubstring}
461
462		and
463
464		Substring.full
465
466		Those are in the Basis spec but they were missing in SML/NJ.
467
468		----------------------------------------------------------------------
469		Name: Matthias Blume
470		Date: 2002/03/14 21:30:00 EST
471		Tag: blume-20020314-controls
472		Description:
473
474		Controls:
475		---------
476
477		1. Factored out the recently-added Controls : CONTROLS stuff and put
478		it into its own library $/controls-lib.cm.  The source tree for
479		this is under src/smlnj-lib/Controls.
480
481		2. Changed the names of types and functions in this interface, so they
482		make a bit more "sense":
483
484		module -> registry
485		'a registry -> 'a group
486
487		3. The interface now deals in ref cells only.  The getter/setter interface
488		is (mostly) gone.
489
490		4. Added a function that lets one register an already-existing ref cell.
491
492		5. Made the corresponding modifications to the rest of the code so that
493		everything compiles again.
494
495		6. Changed the implementation of Controls.MLRISC back to something closer
496		to the original.  In particular, this module (and therefore MLRISC)
497		does not depend on Controls.  There now is some link-time code in
498		int-sys.sml that registers the MLRISC controls with the Controls
499		module.
500
501		CM:
502		---
503
504		* One can now specify the lambda-split aggressiveness in init.cmi.
505
506		----------------------------------------------------------------------
507		Name: Allen Leung
508		Date: 2002/03/13 17:30:00 EST
509		Tag: leunga-20020313-x86-fp-unary
510		Description:
511
512		Bug fix for:
513
514		> leunga@weaselbane:~/Yale/tmp/sml-dist{21} bin/sml
515		> Standard ML of New Jersey v110.39.1 [FLINT v1.5], March 08, 2002
516		> - fun f(x,(y,z)) = Real.~ y;
517		> [autoloading]
518		> [autoloading done]
519		>       fchsl   (%eax), 184(%esp)
520		> Error: MLRisc bug: X86MCEmitter.emitInstr
521		>
522		> uncaught exception Error
523		>   raised at: ../MLRISC/control/mlriscErrormsg.sml:16.14-16.19
524
525		The problem was that the code generator did not generate any fp registers
526		in this case, and the ra didn't know that it needed to run the X86FP phase to
527		translate the pseudo fp instruction.   This only happened with unary fp
528		operators in certain situations.
529
530		----------------------------------------------------------------------
531		Name: Matthias Blume
532		Date: 2002/03/13 14:00:00 EST
533		Tag: blume-20020313-overload-etc
534		Description:
535
536		1. Added _overload as a synonym for overload for backward compatibility.
537		(Control.overloadKW must be true for either version to be accepted.)
538
539		2. Fixed bug in install script that caused more things to be installed
540		than what was requested in config/targets.
541
542		3. Made CM aware of the (_)overload construct so that autoloading
543		works.
544
545		----------------------------------------------------------------------
546		Name: Matthias Blume
547		Date: 2002/03/12 22:03:00 EST
548		Tag: blume-20020312-url
549		Description:
550
551		Forgot to update BOOT and srcarchiveurl.
552
553		----------------------------------------------------------------------
554		Name: Matthias Blume
555		Date: 2002/03/12 17:30:00 EST
556		Tag: blume-20020312-version110392
557		Description:
558
559		Yet another version number bump (because of small changes to the
560		binfile format).  Version number is now 110.39.2.  NEW BOOTFILES!
561
562		Changes:
563
564		The new pid generation scheme described a few weeks ago was overly
565		complicated.  I implemented a new mechanism that is simpler and
566		provides a bit more "stability":  Once CM has seen a compilation
567		unit, it keeps its identity constant (as long as you do not delete
568		those crucial CM/GUID/* files).  This means that when you change
569		an interface, compile, then go back to the old interface, and
570		compile again, you arrive at the original pid.
571
572		There now also is a mechanism that instructs CM to use the plain
573		environment hash as a module's pid (effectively making its GUID
574		the empty string).  For this, "noguid" must be specified as an
575		option to the .sml file in question within its .cm file.
576		This is most useful for code that is being generated by tools such
577		as ml-nlffigen (because during development programmers tend to
578		erase the tool's entire output directory tree including CM's cached
579		GUIDs).  "noguid" is somewhat dangerous (since it can be used to locally
580		revert to the old, broken behavior of SML/NJ, but in specific cases
581		where there is no danger of interface confusion, its use is ok
582		(I think).
583
584		ml-nlffigen by default generates "noguid" annotations.  They can be
585		turned off by specifying -guid in its command line.
586
587		----------------------------------------------------------------------
588		Name: Lal George
589		Date: 2002/03/12 12 14:42:36 EST
590		Tag: george-20020312-frequency-computation
591		Description:
592
593		Integrated jump chaining and static block frequency into the
594		compiler. More details and numbers later.
595
596		----------------------------------------------------------------------
597		Name: Lal George
598		Date: 2002/03/11 11 22:38:53 EST
599		Tag: george-20020311-jump-chain-elim
600		Description:
601
602		Tested the jump chain elimination on all architectures (except the
603		hppa).  This is on by default right now and is profitable for the
604		alpha and x86, however, it may not be profitable for the sparc and ppc
605		when compiling the compiler.
606
607		The gc test will typically jump to a label at the end of the cluster,
608		where there is another jump to an external cluster containing the actual
609		code to invoke gc. This is to allow factoring of common gc invocation
610		sequences. That is to say, we generate:
611
612		f:
613		testgc
614		ja   L1      % jump if above to L1
615
616		L1:
617		jmp L2
618
619
620		After jump chain elimination the 'ja L1' instructions is converted to
621		'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end
622		up being implemented in their long form (if L2 is far away) using:
623
624		jbe     L3      % jump if below or equal to L3
625		jmp     L2
626		L3:
627		...
628
629
630		For large compilation units L2  may be far away.
631
632
633		----------------------------------------------------------------------
634		Name: Matthias Blume
635		Date: 2002/03/11 13:30:00 EST
636		Tag: blume-20020311-mltreeeval
637		Description:
638
639		A functor parameter was missing.
640
641		----------------------------------------------------------------------
642		Name: Allen Leung
643		Date: 2002/03/11 10:30:00 EST
644		Tag: leunga-20020311-runtime-string0
645		Description:
646
647		The representation of the empty string now points to a
648		legal null terminated C string instead of unit.  It is now possible
649		to convert an ML string into C string with InlineT.CharVector.getData.
650		This compiles into one single machine instruction.
651
652		----------------------------------------------------------------------
653		Name: Allen Leung
654		Date: 2002/03/10 23:55:00 EST
655		Tag: leunga-20020310-x86-call
656		Description:
657
658		Added machine generation for CALL instruction (relative displacement mode)
659
660		----------------------------------------------------------------------
661		Name: Matthias Blume
662		Date: 2002/03/08 16:05:00
663		Tag: blume-20020308-entrypoints
664		Description:
665
666		Version number bumped to 110.39.1.  NEW BOOTFILES!
667
668		Entrypoints: non-zero offset into a code object where execution should begin.
669
670		- Added the notion of an entrypoint to CodeObj.
671		- Added reading/writing of entrypoint info to Binfile.
672		- Made runtime system bootloader aware of entrypoints.
673		- Use the address of the label of the first function given to mlriscGen
674		as the entrypoint.  This address is currently always 0, but it will
675		not be 0 once we turn on block placement.
676		- Removed the linkage cluster code (which was The Other Way(tm) of dealing
677		with entry points) from mlriscGen.
678
679		----------------------------------------------------------------------
680		Name: Allen Leung
681		Date: 2002/03/07 20:45:00 EST
682		Tag: leunga-20020307-x86-cmov
683		Description:
684
685		Bug fixes for CMOVcc on x86.
686
687		1. Added machine code generation for CMOVcc
688		2. CMOVcc is now generated in preference over SETcc on PentiumPro or above.
689		3. CMOVcc cannot have an immediate operand as argument.
690
691		----------------------------------------------------------------------
692		Name: Matthias Blume
693		Date: 2002/03/07 16:15:00 EST
694		Tag: blume-20020307-controls
695		Description:
696
697		This is a very large but mostly boring patch which makes (almost)
698		every tuneable compiler knob (i.e., pretty much everything under
699		Control.* plus a few other things) configurable via both the command
700		line and environment variables in the style CM did its configuration
701		until now.
702
703		Try starting sml with '-h' (or, if you are brave, '-H')
704
705		To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which
706		implements the underlying generic mechanism.
707
708		The interface to some of the existing such facilities has changed somewhat.
709		For example, the MLRiscControl module now provides mkFoo instead of getFoo.
710		(The getFoo interface is still there for backward-compatibility, but its
711		use is deprecated.)
712
713		The ml-build script passes -Cxxx=yyy command-line arguments through so
714		that one can now twiddle the compiler settings when using this "batch"
715		compiler.
716
717		TODO items:
718
719		We should go through and throw out all controls that are no longer
720		connected to anything.  Moreover, we should go through and provide
721		meaningful (and correct!) documentation strings for those controls
722		that still are connected.
723
724		Currently, multiple calls to Controls.new are accepted (only the first
725		has any effect).  Eventually we should make sure that every control
726		is being made (via Controls.new) exactly once.  Future access can then
727		be done using Controls.acc.
728
729		Finally, it would probably be a good idea to use the getter-setter
730		interface to controls rather than ref cells.  For the time being, both
731		styles are provided by the Controls module, but getter-setter pairs are
732		better if thread-safety is of any concern because they can be wrapped.
733
734		*****************************************
735
736		One bug fix: The function blockPlacement in three of the MLRISC
737		backpatch files used to be hard-wired to one of two possibilities at
738		link time (according to the value of the placementFlag).  But (I
739		think) it should rather sense the flag every time.
740
741		*****************************************
742
743		Other assorted changes (by other people who did not supply a HISTORY entry):
744
745		1. the cross-module inliner now works much better (Monnier)
746		2. representation of weights, frequencies, and probabilities in MLRISC
747		changed in preparation of using those for weighted block placement
748		(Reppy, George)
749
750		----------------------------------------------------------------------
751		Name: Lal George
752		Date: 2002/03/07 14:44:24 EST 2002
753		Tag: george-20020307-weighted-block-placement
754
755		Tested the weighted block placement optimization on all architectures
756		(except the hppa) using AMPL to generate the block and edge frequencies.
757		Changes were required in the machine properties to correctly
758		categorize trap instructions. There is an MLRISC flag
759		"weighted-block-placement" that can be used to enable weighted block
760		placement, but this will be ineffective without block/edge
761		frequencies (coming soon).
762
763
764		----------------------------------------------------------------------
765		Name: Lal George
766		Date: 2002/03/05 17:24:48 EST
767		Tag: george-20020305-linkage-cluster
768
769		In order to support the block placement optimization, a new cluster
770		is generated as the very first cluster (called the linkage cluster).
771		It contains a single jump to the 'real' entry point for the compilation
772		unit. Block placement has no effect on the linkage cluster itself, but
773		all the other clusters  have full freedom in the manner in which they
774		reorder blocks or functions.
775
776		On the x86 the typical linkage code that is generated is:
777		----------------------
778		.align 2
779		L0:
780		addl    $L1-L0, 72(%esp)
781		jmp     L1
782
783
784		.align  2
785		L1:
786		----------------------
787
788		72(%esp) is the memory location for the stdlink register. This
789		must contain the address of the CPS function being called. In the
790		above example, it contains the address of  L0; before
791		calling L1 (the real entry point for the compilation unit), it
792		must contain the address for L1, and hence
793
794		addl $L1-L0, 72(%esp)
795
796		I have tested this on all architectures except the hppa.The increase
797		in code size is of course negligible
798
799		----------------------------------------------------------------------
800	Name: Allen Leung	Name: Allen Leung
801	Date: 2002/03/03 13:20:00 EST	Date: 2002/03/03 13:20:00 EST
802	Tag: leunga-20020303-mlrisc-tools	Tag: leunga-20020303-mlrisc-tools

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.1115   changed lines   Added in v.1189

---

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