Diff of /sml/trunk/HISTORY

Parent Directory | Revision Log | Patch

	revision 1158, Fri Mar 22 01:21:25 2002 UTC	revision 1192, Wed May 15 14:02:06 2002 UTC
#	Line 13	Line 13
13	Description:	Description:
14
15	----------------------------------------------------------------------	----------------------------------------------------------------------
16		Name: Lal George
17		Date: 2002/05/15 09:20:10 EDT
18		Tag: george-20020515-pseudo-op-decls
19		Description:
20
21		All pseudo-ops emitted before the first segment declaration
22		such as TEXT, DATA, and BSS directives are assumed to be global
23		declarations and are emitted first in the assembly file. This is
24		useful in a number of situations where one has pseudo-ops that are not
25		specific to any segment, and also works around the constraint that one
26		cannot have client pseudo-ops in the TEXT segment.
27
28		Because no segment is associated with these declarations it is
29		an error to allocate any space or objects before the first segment
30		directive and an exception will be raised. However, we cannot make
31		this check for client pseudo-ops.
32
33		These top level declarations are a field in the CFG graph_info.
34		In theory you can continue to add to this field after the CFG has been
35		built -- provided you know what you are doing;-)
36
37		----------------------------------------------------------------------
38		Name: Matthias Blume
39		Date: 2002/05/13 16:40:00 EDT
40		Tag: blume-20020513-pp-etc
41		Description:
42
43		A few minor bugfixes:
44
45		- Stopgap measure for bug recently reported by Elsa Gunter (ppDec).
46		(Bogus printouts for redefined bindings still occur.  Compiler
47		bug should no longer occur now.  We need to redo the prettyprinter
48		from scratch.)
49
50		- CM pathname printer now also adds escape sequences for ( and )
51
52		- commend and docu fixes for ml-nlffi
53
54		----------------------------------------------------------------------
55		Name: Matthias Blume
56		Date: 2002/05/10 16:40:00 EDT
57		Tag: blume-20020510-erg-textio
58		Description:
59
60		Applied the following bugfix provided by Emden Gansner:
61
62		Output is corrupted when outputSubstr is used rather than output.
63
64		The problem occurs when a substring
65
66		ss = (s, dataStart, dataLen)
67
68		where dataStart > 0, fills a stream buffer with avail bytes left.
69		avail bytes of s, starting at index dataStart, are copied into the
70		buffer, the buffer is flushed, and then the remaining dataLen-avail
71		bytes of ss are copied into the beginning of the buffer. Instead of
72		starting this copy at index dataStart+avail in s, the current code
73		starts the copy at index avail.
74
75		Fix:
76		In text-io-fn.sml, change line 695 from
77		val needsFlush = copyVec(v, avail, dataLen-avail, buf, 0)
78		to
79		val needsFlush = copyVec(v, dataStart+avail, dataLen-avail, buf, 0)
80
81		----------------------------------------------------------------------
82		Name: Matthias Blume
83		Date: 2002/04/12 13:55:00 EDT
84		Tag: blume-20020412-assyntax
85		Description:
86
87		1. Grabbed newer assyntax.h from the XFree86 project.
88		2. Fiddled with how to compile X86.prim.asm without warnings.
89		3. (Very) Minor cleanup in CM.
90
91		----------------------------------------------------------------------
92		Name: Matthias Blume
93		Date: 2002/04/01 (no joke!) 17:07:00 EST
94		Tag: blume-20020401-x86div
95		Description:
96
97		Added full support for div/mod/rem/quot on the x86, using the machine
98		instruction's two results (without clumsily recomputing the remainder)
99		directly where appropriate.
100
101		Some more extensive power-of-two support was added to the x86 instruction
102		selector (avoiding expensive divs, mods, and muls where they can be
103		replaced with cheaper shifts and masks).  However, this sort of thing
104		ought to be done earlier, e.g., within the CPS optimizer so that
105		all architectures benefit from it.
106
107		The compiler compiles to a fixed point, but changes might be somewhat
108		fragile nevertheless.  Please, report any strange things that you might
109		see wrt. div/mod/quot/rem...
110
111		----------------------------------------------------------------------
112		Name: Matthias Blume
113		Date: 2002/03/29 17:22:00
114		Tag: blume-20020329-div
115		Description:
116
117		Fixed my broken div/mod logic.  Unfortunately, this means that the
118		inline code for div/mod now has one more comparison than before.
119		Fast paths (quotient > 0 or remainder = 0) are not affected, though.
120		The problem was with quotient = 0, because that alone does not tell
121		us which way the rounding went.  One then has to look at whether
122		remainder and divisor have the same sign...  :(
123
124		Anyway, I replaced the bootfiles with fresh ones...
125
126		----------------------------------------------------------------------
127		Name: Matthias Blume
128		Date: 2002/03/29 14:10:00 EST
129		Tag: blume-20020329-inlprims
130		Description:
131
132		NEW BOOTFILES!!!    Version number bumped to 110.39.3.
133
134		Primops have changed. This means that the bin/boot-file formats have
135		changed as well.
136
137		To make sure that there is no confusion, I made a new version.
138
139
140		CHANGES:
141
142		* removed REMT from mltree (remainder should never overflow).
143
144		* added primops to deal with divisions of all flavors to the frontend
145
146		* handled these primops all the way through so they map to their respective
147		MLRISC support
148
149		* used these primops in the implementation of Int, Int32, Word, Word32
150
151		* removed INLDIV, INLMOD, and INLREM as they are no longer necessary
152
153		* parameterized INLMIN, INLMAX, and INLABS by a numkind
154
155		* translate.sml now deals with all flavors of INL{MIN,MAX,ABS}, including
156		floating point
157
158		* used INL{MIN,MAX,ABS} in the implementation of Int, Int32, Word, Word32,
159		and Real (but Real.abs maps to a separate floating-point-only primop)
160
161
162		TODO items:
163
164		* Hacked Alpha32 instruction selection, disabling the selection of REMx
165		instructions because the machine instruction encoder cannot handle
166		them.  (Hppa, PPC, and Sparc instruction selection did not handle
167		REM in the first place, and REM is supported by the x86 machine coder.)
168
169		* Handle DIV and MOD with DIV_TO_NEGINF directly in the x86 instruction
170		selection phase.  (The two can be streamlined because the hardware
171		delivers both quotient and remainder at the same time anyway.)
172
173		* Think about what to do with "valOf(Int32.minInt) div ~1" and friends.
174		(Currently the behavior is inconsistent both across architectures and
175		wrt. the draft Basis spec.)
176
177		* Word8 should eventually be handled natively, too.
178
179		* There seems to be one serious bug in mltree-gen.sml.  It appears, though,
180		as if there currently is no execution path that could trigger it in
181		SML/NJ.  (The assumptions underlying functions arith and promotable do not
182		hold for things like multiplication and division.)
183
184		----------------------------------------------------------------------
185		Name: Matthias Blume
186		Date: 2002/03/27 16:27:00 EST
187		Tag: blume-20020327-mlrisc-divisions
188		Description:
189
190		Added support for all four division operations (ML's div, mod, quot,
191		and rem) to MLRISC.  In the course of doing so, I also rationalized
192		the naming (no more annoying switch-around of DIV and QUOT), by
193		parameterizing the operation by div_rounding_mode (which can be either
194		DIV_TO_ZERO or DIV_TO_NEGINF).
195
196		The generic MLTreeGen functor takes care of compiling all four
197		operations down to only round-to-zero div.
198
199		Missing pieces:
200
201		* Doing something smarter than relying on MLTreeGen on architectures
202		like, e.g., the x86 where hardware division delivers both quotient and
203		remainder at the same time.  With this, the implementation of the
204		round-to-neginf operations could be further streamlined.
205
206		* Remove inlining support for div/mod/rem from the frontend and replace it
207		with primops that get carried through to the backend.  Do this for all
208		int and word types.
209
210		----------------------------------------------------------------------
211		Name: Matthias Blume
212		Date: 2002/03/25 17:25:00 EST
213		Tag: blume-20020325-divmod
214		Description:
215
216		I improved (hopefully without breaking them) the implementation of Int.div,
217		Int.mod, and Int.rem.   For this, the code in translate.sml now takes
218		advantage of the following observations:
219
220		Let  q = x quot y      r = x rem y
221		d = x div  y      m = x mod y
222
223		where "quot" is the round-to-zero version of integer division that
224		hardware usually provides.  Then we have:
225
226		r = x - q * y        where neither the * nor the - will overflow
227		d = if q >= 0 orelse x = q * y then q else q - 1
228		where neither the * nor the - will overflow
229		m = if q >= 0 orelse r = 0 then r else r + y
230		where the + will not overflow
231
232		This results in substantial simplification of the generated code.
233		The following table shows the number of CFG nodes and edges generated
234		for
235		fun f (x, y) = x OPER y
236		(* with OPER \in div, mod, quot, rem *)
237
238
239		OPER | nodes(old) | edges(old) | nodes(new) | edges(new)
240		--------------------------------------------------------
241		div |         24 |         39 |         12 |         16
242		mod |         41 |         71 |         12 |         16
243		quot |          8 |         10 |          8 |         10
244		rem |         10 |         14 |          8 |         10
245
246
247		----------------------------------------------------------------------
248		Name: Matthias Blume
249		Date: 2002/03/25 22:06:00 EST
250		Tag: blume-20020325-cprotobug
251		Description:
252
253		Fixed a bug in cproto (c prototype decoder).
254
255		----------------------------------------------------------------------
256		Name: Matthias Blume
257		Date: 2002/03/25 16:00:00 EST
258		Tag: blume-20020325-raw-primops
259		Description:
260
261		I did some cleanup to Allen's new primop code and
262		replaced yesterday's bootfiles with new ones.
263		(But they are stored in the same place.)
264
265		----------------------------------------------------------------------
266		Name: Matthias Blume
267		Date: 2002/03/24 22:40:00 EST
268		Tag: blume-20020324-bootfiles
269		Description:
270
271		Made the bootfiles that Allen asked for.
272
273		----------------------------------------------------------------------
274		Name: Allen Leung
275		Date: 2002/03/23 15:50:00 EST
276		Tag: leunga-20020323-flint-cps-rcc-primops
277		Description:
278
279		1. Changes to FLINT primops:
280
281		(* make a call to a C-function;
282		* The primop carries C function prototype information and specifies
283		* which of its (ML-) arguments are floating point. C prototype
284		* information is for use by the backend, ML information is for
285		* use by the CPS converter. *)
286		| RAW_CCALL of { c_proto: CTypes.c_proto,
287		ml_args: ccall_type list,
288		ml_res_opt: ccall_type option,
289		reentrant : bool
290		} option
291		(* Allocate uninitialized storage on the heap.
292		* The record is meant to hold short-lived C objects, i.e., they
293		* are not ML pointers.  With the tag, the representation is
294		* the same as RECORD with tag tag_raw32 (sz=4), or tag_fblock (sz=8)
295		*)
296		| RAW_RECORD of {tag:bool,sz:int}
297		and ccall_type = CCALL_INT32 | CCALL_REAL64 | CCALL_ML_PTR
298
299		2.  These CPS primops are now overloaded:
300
301		rawload of {kind:numkind}
302		rawstore of {kind:numkind}
303
304		The one argument form is:
305
306		rawload {kind} address
307
308		The two argument form is:
309
310		rawload {kind} [ml object, byte-offset]
311
312		3. RAW_CCALL/RCC now takes two extra arguments:
313
314		a. The first is whether the C call is reentrant, i.e., whether
315		ML state should be saved and restored.
316		b. The second argument is a string argument specifying the name of
317		library and the C function.
318
319		These things are currently not handled in the code generator, yet.
320
321		4. In CProto,
322
323		An encoding type of "bool" means "ml object" and is mapped into
324		C prototype of PTR.  Note that "bool" is different than "string",
325		even though "string" is also mapped into PTR, because "bool"
326		is assigned an CPS type of BOGt, while "string" is assigned INT32t.
327
328		5. Pickler/unpicker
329
330		Changed to handle RAW_RECORD and newest RAW_CCALL
331
332		6. MLRiscGen,
333
334		1. Changed to handle the new rawload/rawstore/rawrecord operators.
335		2. Code for handling C Calls has been moved to a new module CPSCCalls,
336		in the file CodeGen/cpscompile/cps-c-calls.sml
337
338		7. Added the conditional move operator
339
340		condmove of branch
341
342		to cps.  Generation of this is still buggy so it is currently
343		disabled.
344
345		----------------------------------------------------------------------
346		Name: Lal George
347		Date: 2002/03/22 14:18:25 EST
348		Tag: george-20020322-cps-branch-prob
349		Description:
350
351		Implemented the Ball-Larus branch prediction-heuristics, and
352		incorporated graphical viewers for control flow graphs.
353
354		Ball-Larus Heuristics:
355		---------------------
356		See the file compiler/CodeGen/cpscompile/cpsBranchProb.sml.
357
358		By design it uses the Dempster-Shafer theory for combining
359		probabilities.  For example, in the function:
360
361		fun f(n,acc) = if n = 0 then acc else f(n-1, n*acc)
362
363		the ball-larus heuristics predicts that the n=0 is unlikely
364		(OH-heuristic), and the 'then' branch is unlikely because of the
365		RH-heuristic -- giving the 'then' branch an even lower combined
366		probability using the Dempster-Shafer theory.
367
368		Finally, John Reppy's loop analysis in MLRISC, further lowers the
369		probability of the 'then' branch because of the loop in the else
370		branch.
371
372
373		Graphical Viewing:
374		------------------
375		I merely plugged in Allen's graphical viewers into the compiler. The
376		additional code is not much. At the top level, saying:
377
378		Control.MLRISC.getFlag "cfg-graphical-view" := true;
379
380		will display the graphical view of the control flow graph just before
381		back-patching.  daVinci must be in your path for this to work. If
382		daVinci is not available, then the default viewer can be changed
383		using:
384
385		Control.MLRISC.getString "viewer"
386
387		which can be set to "dot" or "vcg" for the corresponding viewers. Of
388		course, these viewers must be in your path.
389
390		The above will display the compilation unit at the level of clusters,
391		many of which are small, boring, and un-interesting. Also setting:
392
393		Control.MLRISC.getInt "cfg-graphical-view_size"
394
395		will display clusters that are larger than the value set by the above.
396
397
398		----------------------------------------------------------------------
399		Name: Matthias Blume
400		Date: 2002/03/21 22:20:00 EST
401		Tag: blume-20020321-kmp-bugfix
402		Description:
403
404		Changed the interface to the KMP routine in PreString and fixed
405		a minor bug in one place where it was used.
406
407		----------------------------------------------------------------------
408	Name: Allen Leung	Name: Allen Leung
409	Date: 2002/03/21 20:30:00 EST	Date: 2002/03/21 20:30:00 EST
410	Tag: leunga-20020321-cfg	Tag: leunga-20020321-cfg

---

Colored Diff Long Colored Diff Unidiff Context Diff Side by Side

Legend: Removed from v.1158   changed lines   Added in v.1192

---

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