11 |
Date: yyyy/mm/dd |
Date: yyyy/mm/dd |
12 |
Tag: <post-commit CVS tag> |
Tag: <post-commit CVS tag> |
13 |
Description: |
Description: |
14 |
|
---------------------------------------------------------------------- |
15 |
|
Name: Matthias Blume |
16 |
|
Date: 2002/03/13 14:00:00 EST |
17 |
|
Tag: blume-20020313-overload-etc |
18 |
|
Description: |
19 |
|
|
20 |
|
1. Added _overload as a synonym for overload for backward compatibility. |
21 |
|
(Control.overloadKW must be true for either version to be accepted.) |
22 |
|
|
23 |
|
2. Fixed bug in install script that caused more things to be installed |
24 |
|
than what was requested in config/targets. |
25 |
|
|
26 |
|
3. Made CM aware of the (_)overload construct so that autoloading |
27 |
|
works. |
28 |
|
|
29 |
|
---------------------------------------------------------------------- |
30 |
|
Name: Matthias Blume |
31 |
|
Date: 2002/03/12 22:03:00 EST |
32 |
|
Tag: blume-20020312-url |
33 |
|
Description: |
34 |
|
|
35 |
|
Forgot to update BOOT and srcarchiveurl. |
36 |
|
|
37 |
|
---------------------------------------------------------------------- |
38 |
|
Name: Matthias Blume |
39 |
|
Date: 2002/03/12 17:30:00 EST |
40 |
|
Tag: blume-20020312-version110392 |
41 |
|
Description: |
42 |
|
|
43 |
|
Yet another version number bump (because of small changes to the |
44 |
|
binfile format). Version number is now 110.39.2. NEW BOOTFILES! |
45 |
|
|
46 |
|
Changes: |
47 |
|
|
48 |
|
The new pid generation scheme described a few weeks ago was overly |
49 |
|
complicated. I implemented a new mechanism that is simpler and |
50 |
|
provides a bit more "stability": Once CM has seen a compilation |
51 |
|
unit, it keeps its identity constant (as long as you do not delete |
52 |
|
those crucial CM/GUID/* files). This means that when you change |
53 |
|
and interface, compiler, then go back to the old interface, and |
54 |
|
compile again, you arrive at the original pid. |
55 |
|
|
56 |
|
There now also is a mechanism that instructs CM to use the plain |
57 |
|
environment hash as a module's pid (effectively making its GUID |
58 |
|
the empty string). For this, "noguid" must be specified as an |
59 |
|
option to the .sml file in question within its .cm file. |
60 |
|
This is most useful for code that is being generated by tools such |
61 |
|
as ml-nlffigen (because during development programmers tend to |
62 |
|
erase the tool's entire output directory tree including CM's cached |
63 |
|
GUIDs). "noguid" is somewhat dangerous (since it can be used to locally |
64 |
|
revert to the old, broken behavior of SML/NJ, but in specific cases |
65 |
|
where there is no danger of interface confusion, its use is ok |
66 |
|
(I think). |
67 |
|
|
68 |
|
ml-nlffigen by default generates "noguid" annotations. They can be |
69 |
|
turned off by specifying -guid in its command line. |
70 |
|
|
71 |
|
---------------------------------------------------------------------- |
72 |
|
Name: Lal George |
73 |
|
Date: 2002/03/12 12 14:42:36 EST |
74 |
|
Tag: george-20020312-frequency-computation |
75 |
|
Description: |
76 |
|
|
77 |
|
Integrated jump chaining and static block frequency into the |
78 |
|
compiler. More details and numbers later. |
79 |
|
|
80 |
|
---------------------------------------------------------------------- |
81 |
|
Name: Lal George |
82 |
|
Date: 2002/03/11 11 22:38:53 EST |
83 |
|
Tag: george-20020311-jump-chain-elim |
84 |
|
Description: |
85 |
|
|
86 |
|
Tested the jump chain elimination on all architectures (except the |
87 |
|
hppa). This is on by default right now and is profitable for the |
88 |
|
alpha and x86, however, it may not be profitable for the sparc and ppc |
89 |
|
when compiling the compiler. |
90 |
|
|
91 |
|
The gc test will typically jump to a label at the end of the cluster, |
92 |
|
where there is another jump to an external cluster containing the actual |
93 |
|
code to invoke gc. This is to allow factoring of common gc invocation |
94 |
|
sequences. That is to say, we generate: |
95 |
|
|
96 |
|
f: |
97 |
|
testgc |
98 |
|
ja L1 % jump if above to L1 |
99 |
|
|
100 |
|
L1: |
101 |
|
jmp L2 |
102 |
|
|
103 |
|
|
104 |
|
After jump chain elimination the 'ja L1' instructions is converted to |
105 |
|
'ja L2'. On the sparc and ppc, many of the 'ja L2' instructions may end |
106 |
|
up being implemented in their long form (if L2 is far away) using: |
107 |
|
|
108 |
|
jbe L3 % jump if below or equal to L3 |
109 |
|
jmp L2 |
110 |
|
L3: |
111 |
|
... |
112 |
|
|
113 |
|
|
114 |
|
For large compilation units L2 may be far away. |
115 |
|
|
116 |
|
|
117 |
|
---------------------------------------------------------------------- |
118 |
|
Name: Matthias Blume |
119 |
|
Date: 2002/03/11 13:30:00 EST |
120 |
|
Tag: blume-20020311-mltreeeval |
121 |
|
Description: |
122 |
|
|
123 |
|
A functor parameter was missing. |
124 |
|
|
125 |
|
---------------------------------------------------------------------- |
126 |
|
Name: Allen Leung |
127 |
|
Date: 2002/03/11 10:30:00 EST |
128 |
|
Tag: leunga-20020310-runtime-string0 |
129 |
|
Description: |
130 |
|
|
131 |
|
The representation of the empty string now points to a |
132 |
|
legal null terminated C string instead of unit. It is now possible |
133 |
|
to convert an ML string into C string with InlineT.CharVector.getData. |
134 |
|
This compiles into one single machine instruction. |
135 |
|
|
136 |
|
---------------------------------------------------------------------- |
137 |
|
Name: Allen Leung |
138 |
|
Date: 2002/03/10 23:55:00 EST |
139 |
|
Tag: leunga-20020310-x86-call |
140 |
|
Description: |
141 |
|
|
142 |
|
Added machine generation for CALL instruction (relative displacement mode) |
143 |
|
|
144 |
|
---------------------------------------------------------------------- |
145 |
|
Name: Matthias Blume |
146 |
|
Date: 2002/03/08 16:05:00 |
147 |
|
Tag: blume-20020308-entrypoints |
148 |
|
Description: |
149 |
|
|
150 |
|
Version number bumped to 110.39.1. NEW BOOTFILES! |
151 |
|
|
152 |
|
Entrypoints: non-zero offset into a code object where execution should begin. |
153 |
|
|
154 |
|
- Added the notion of an entrypoint to CodeObj. |
155 |
|
- Added reading/writing of entrypoint info to Binfile. |
156 |
|
- Made runtime system bootloader aware of entrypoints. |
157 |
|
- Use the address of the label of the first function given to mlriscGen |
158 |
|
as the entrypoint. This address is currently always 0, but it will |
159 |
|
not be 0 once we turn on block placement. |
160 |
|
- Removed the linkage cluster code (which was The Other Way(tm) of dealing |
161 |
|
with entry points) from mlriscGen. |
162 |
|
|
163 |
|
---------------------------------------------------------------------- |
164 |
|
Name: Allen Leung |
165 |
|
Date: 2002/03/07 20:45:00 EST |
166 |
|
Tag: leunga-20020307-x86-cmov |
167 |
|
Description: |
168 |
|
|
169 |
|
Bug fixes for CMOVcc on x86. |
170 |
|
|
171 |
|
1. Added machine code generation for CMOVcc |
172 |
|
2. CMOVcc is now generated in preference over SETcc on PentiumPro or above. |
173 |
|
3. CMOVcc cannot have an immediate operand as argument. |
174 |
|
|
175 |
|
---------------------------------------------------------------------- |
176 |
|
Name: Matthias Blume |
177 |
|
Date: 2002/03/07 16:15:00 EST |
178 |
|
Tag: blume-20020307-controls |
179 |
|
Description: |
180 |
|
|
181 |
|
This is a very large but mostly boring patch which makes (almost) |
182 |
|
every tuneable compiler knob (i.e., pretty much everything under |
183 |
|
Control.* plus a few other things) configurable via both the command |
184 |
|
line and environment variables in the style CM did its configuration |
185 |
|
until now. |
186 |
|
|
187 |
|
Try starting sml with '-h' (or, if you are brave, '-H') |
188 |
|
|
189 |
|
To this end, I added a structure Controls : CONTROLS to smlnj-lib.cm which |
190 |
|
implements the underlying generic mechanism. |
191 |
|
|
192 |
|
The interface to some of the existing such facilities has changed somewhat. |
193 |
|
For example, the MLRiscControl module now provides mkFoo instead of getFoo. |
194 |
|
(The getFoo interface is still there for backward-compatibility, but its |
195 |
|
use is deprecated.) |
196 |
|
|
197 |
|
The ml-build script passes -Cxxx=yyy command-line arguments through so |
198 |
|
that one can now twiddle the compiler settings when using this "batch" |
199 |
|
compiler. |
200 |
|
|
201 |
|
TODO items: |
202 |
|
|
203 |
|
We should go through and throw out all controls that are no longer |
204 |
|
connected to anything. Moreover, we should go through and provide |
205 |
|
meaningful (and correct!) documentation strings for those controls |
206 |
|
that still are connected. |
207 |
|
|
208 |
|
Currently, multiple calls to Controls.new are accepted (only the first |
209 |
|
has any effect). Eventually we should make sure that every control |
210 |
|
is being made (via Controls.new) exactly once. Future access can then |
211 |
|
be done using Controls.acc. |
212 |
|
|
213 |
|
Finally, it would probably be a good idea to use the getter-setter |
214 |
|
interface to controls rather than ref cells. For the time being, both |
215 |
|
styles are provided by the Controls module, but getter-setter pairs are |
216 |
|
better if thread-safety is of any concern because they can be wrapped. |
217 |
|
|
218 |
|
***************************************** |
219 |
|
|
220 |
|
One bug fix: The function blockPlacement in three of the MLRISC |
221 |
|
backpatch files used to be hard-wired to one of two possibilities at |
222 |
|
link time (according to the value of the placementFlag). But (I |
223 |
|
think) it should rather sense the flag every time. |
224 |
|
|
225 |
|
***************************************** |
226 |
|
|
227 |
|
Other assorted changes (by other people who did not supply a HISTORY entry): |
228 |
|
|
229 |
|
1. the cross-module inliner now works much better (Monnier) |
230 |
|
2. representation of weights, frequencies, and probabilities in MLRISC |
231 |
|
changed in preparation of using those for weighted block placement |
232 |
|
(Reppy, George) |
233 |
|
|
234 |
|
---------------------------------------------------------------------- |
235 |
|
Name: Lal George |
236 |
|
Date: 2002/03/07 14:44:24 EST 2002 |
237 |
|
Tag: george-20020307-weighted-block-placement |
238 |
|
|
239 |
|
Tested the weighted block placement optimization on all architectures |
240 |
|
(except the hppa) using AMPL to generate the block and edge frequencies. |
241 |
|
Changes were required in the machine properties to correctly |
242 |
|
categorize trap instructions. There is an MLRISC flag |
243 |
|
"weighted-block-placement" that can be used to enable weighted block |
244 |
|
placement, but this will be ineffective without block/edge |
245 |
|
frequencies (coming soon). |
246 |
|
|
247 |
|
|
248 |
|
---------------------------------------------------------------------- |
249 |
|
Name: Lal George |
250 |
|
Date: 2002/03/05 17:24:48 EST |
251 |
|
Tag: george-20020305-linkage-cluster |
252 |
|
|
253 |
|
In order to support the block placement optimization, a new cluster |
254 |
|
is generated as the very first cluster (called the linkage cluster). |
255 |
|
It contains a single jump to the 'real' entry point for the compilation |
256 |
|
unit. Block placement has no effect on the linkage cluster itself, but |
257 |
|
all the other clusters have full freedom in the manner in which they |
258 |
|
reorder blocks or functions. |
259 |
|
|
260 |
|
On the x86 the typical linkage code that is generated is: |
261 |
|
---------------------- |
262 |
|
.align 2 |
263 |
|
L0: |
264 |
|
addl $L1-L0, 72(%esp) |
265 |
|
jmp L1 |
266 |
|
|
267 |
|
|
268 |
|
.align 2 |
269 |
|
L1: |
270 |
|
---------------------- |
271 |
|
|
272 |
|
72(%esp) is the memory location for the stdlink register. This |
273 |
|
must contain the address of the CPS function being called. In the |
274 |
|
above example, it contains the address of L0; before |
275 |
|
calling L1 (the real entry point for the compilation unit), it |
276 |
|
must contain the address for L1, and hence |
277 |
|
|
278 |
|
addl $L1-L0, 72(%esp) |
279 |
|
|
280 |
|
I have tested this on all architectures except the hppa.The increase |
281 |
|
in code size is of course negligible |
282 |
|
|
283 |
|
---------------------------------------------------------------------- |
284 |
|
Name: Allen Leung |
285 |
|
Date: 2002/03/03 13:20:00 EST |
286 |
|
Tag: leunga-20020303-mlrisc-tools |
287 |
|
|
288 |
|
Added #[ ... ] expressions to mlrisc tools |
289 |
|
|
290 |
|
---------------------------------------------------------------------- |
291 |
|
Name: Matthias Blume |
292 |
|
Date: 2002/02/27 12:29:00 EST |
293 |
|
Tag: blume-20020227-cdebug |
294 |
|
Description: |
295 |
|
|
296 |
|
- made types in structure C and C_Debug to be equal |
297 |
|
- got rid of code duplication (c-int.sml vs. c-int-debug.sml) |
298 |
|
- there no longer is a C_Int_Debug (C_Debug is directly derived from C) |
299 |
|
|
300 |
|
---------------------------------------------------------------------- |
301 |
|
Name: Matthias Blume |
302 |
|
Date: 2002/02/26 12:00:00 EST |
303 |
|
Tag: blume-20020226-ffi |
304 |
|
Description: |
305 |
|
|
306 |
|
1. Fixed a minor bug in CM's "noweb" tool: |
307 |
|
If numbering is turned off, then truly don't number (i.e., do not |
308 |
|
supply the -L option to noweb). The previous behavior was to supply |
309 |
|
-L'' -- which caused noweb to use the "default" line numbering scheme. |
310 |
|
Thanks to Chris Richards for pointing this out (and supplying the fix). |
311 |
|
|
312 |
|
2. Once again, I reworked some aspects of the FFI: |
313 |
|
|
314 |
|
A. The incomplete/complete type business: |
315 |
|
|
316 |
|
- Signatures POINTER_TO_INCOMPLETE_TYPE and accompanying functors are |
317 |
|
gone! |
318 |
|
- ML types representing an incomplete type are now *equal* to |
319 |
|
ML types representing their corresponding complete types (just like |
320 |
|
in C). This is still safe because ml-nlffigen will not generate |
321 |
|
RTTI for incomplete types, nor will it generate functions that |
322 |
|
require access to such RTTI. But when ML code generated from both |
323 |
|
incomplete and complete versions of the C type meet, the ML types |
324 |
|
are trivially interoperable. |
325 |
|
|
326 |
|
NOTE: These changes restore the full generality of the translation |
327 |
|
(which was previously lost when I eliminated functorization)! |
328 |
|
|
329 |
|
B. Enum types: |
330 |
|
|
331 |
|
- Structure C now has a type constructor "enum" that is similar to |
332 |
|
how the "su" constructor works. However, "enum" is not a phantom |
333 |
|
type because each "T enum" has values (and is isomorphic to |
334 |
|
MLRep.Signed.int). |
335 |
|
- There are generic access operations for enum objects (using |
336 |
|
MLRep.Signed.int). |
337 |
|
- ml-nlffigen will generate a structure E_foo for each "enum foo". |
338 |
|
* The structure contains the definition of type "mlrep" (the ML-side |
339 |
|
representation type of the enum). Normally, mlrep is the same |
340 |
|
as "MLRep.Signed.int", but if ml-nlffigen was invoked with "-ec", |
341 |
|
then mlrep will be defined as a datatype -- thus facilitating |
342 |
|
pattern matching on mlrep values. |
343 |
|
("-ec" will be suppressed if there are duplicate values in an |
344 |
|
enumeration.) |
345 |
|
* Constructors ("-ec") or values (no "-ec") e_xxx of type mlrep |
346 |
|
will be generated for each C enum constant xxx. |
347 |
|
* Conversion functions m2i and i2m convert between mlrep and |
348 |
|
MLRep.Signed.int. (Without "-ec", these functions are identities.) |
349 |
|
* Coversion functions c and ml convert between mlrep and "tag enum". |
350 |
|
* Access functions (get/set) fetch and store mlrep values. |
351 |
|
- By default (unless ml-nlffigen was invoked with "-nocollect"), unnamed |
352 |
|
enumerations are merged into one single enumeration represented by |
353 |
|
structure E_'. |
354 |
|
|
355 |
|
---------------------------------------------------------------------- |
356 |
|
Name: Allen Leung |
357 |
|
Date: 2002/02/25 04:45:00 EST |
358 |
|
Tag: leunga-20020225-cps-spill |
359 |
|
|
360 |
|
This is a new implementation of the CPS spill phase. |
361 |
|
The new phase is in the new file compiler/CodeGen/cpscompile/spill-new.sml |
362 |
|
In case of problems, replace it with the old file spill.sml |
363 |
|
|
364 |
|
The current compiler runs into some serious performance problems when |
365 |
|
constructing a large record. This can happen when we try to compile a |
366 |
|
structure with many items. Even a very simple structure like the following |
367 |
|
makes the compiler slow down. |
368 |
|
|
369 |
|
structure Foo = struct |
370 |
|
val x_1 = 0w1 : Word32.int |
371 |
|
val x_2 = 0w2 : Word32.int |
372 |
|
val x_3 = 0w3 : Word32.int |
373 |
|
... |
374 |
|
val x_N = 0wN : Word32.int |
375 |
|
end |
376 |
|
|
377 |
|
The following table shows the compile time, from N=1000 to N=4000, |
378 |
|
with the old compiler: |
379 |
|
|
380 |
|
N |
381 |
|
1000 CPS 100 spill 0.04u 0.00s 0.00g |
382 |
|
MLRISC ra 0.06u 0.00s 0.05g |
383 |
|
(spills = 0 reloads = 0) |
384 |
|
TOTAL 0.63u 0.07s 0.21g |
385 |
|
|
386 |
|
1100 CPS 100 spill 8.25u 0.32s 0.64g |
387 |
|
MLRISC ra 5.68u 0.59s 3.93g |
388 |
|
(spills = 0 reloads = 0) |
389 |
|
TOTAL 14.71u 0.99s 4.81g |
390 |
|
|
391 |
|
1500 CPS 100 spill 58.55u 2.34s 1.74g |
392 |
|
MLRISC ra 5.54u 0.65s 3.91g |
393 |
|
(spills = 543 reloads = 1082) |
394 |
|
TOTAL 65.40u 3.13s 6.00g |
395 |
|
|
396 |
|
2000 CPS 100 spill 126.69u 4.84s 3.08g |
397 |
|
MLRISC ra 0.80u 0.10s 0.55g |
398 |
|
(spills = 42 reloads = 84) |
399 |
|
TOTAL 129.42u 5.10s 4.13g |
400 |
|
|
401 |
|
3000 CPS 100 spill 675.59u 19.03s 11.64g |
402 |
|
MLRISC ra 2.69u 0.27s 1.38g |
403 |
|
(spills = 62 reloads = 124) |
404 |
|
TOTAL 682.48u 19.61s 13.99g |
405 |
|
|
406 |
|
4000 CPS 100 spill 2362.82u 56.28s 43.60g |
407 |
|
MLRISC ra 4.96u 0.27s 2.72g |
408 |
|
(spills = 85 reloads = 170) |
409 |
|
TOTAL 2375.26u 57.21s 48.00g |
410 |
|
|
411 |
|
As you can see the old cps spill module suffers from some serious |
412 |
|
performance problem. But since I cannot decipher the old code fully, |
413 |
|
instead of patching the problems up, I'm reimplementing it |
414 |
|
with a different algorithm. The new code is more modular, |
415 |
|
smaller when compiled, and substantially faster |
416 |
|
(O(n log n) time and O(n) space). Timing of the new spill module: |
417 |
|
|
418 |
|
4000 CPS 100 spill 0.02u 0.00s 0.00g |
419 |
|
MLRISC ra 0.25u 0.02s 0.15g |
420 |
|
(spills=1 reloads=3) |
421 |
|
TOTAL 7.74u 0.34s 1.62g |
422 |
|
|
423 |
|
Implementation details: |
424 |
|
|
425 |
|
As far as I can tell, the purpose of the CPS spill module is to make sure the |
426 |
|
number of live variables at any program point (the bandwidth) |
427 |
|
does not exceed a certain limit, which is determined by the |
428 |
|
size of the spill area. |
429 |
|
|
430 |
|
When the bandwidth is too large, we decrease the register pressure by |
431 |
|
packing live variables into spill records. How we achieve this is |
432 |
|
completely different than what we did in the old code. |
433 |
|
|
434 |
|
First, there is something about the MLRiscGen code generator |
435 |
|
that we should be aware of: |
436 |
|
|
437 |
|
o MLRiscGen performs code motion! |
438 |
|
|
439 |
|
In particular, it will move floating point computations and |
440 |
|
address computations involving only the heap pointer to |
441 |
|
their use sites (if there is only a single use). |
442 |
|
What this means is that if we have a CPS record construction |
443 |
|
statement |
444 |
|
|
445 |
|
RECORD(k,vl,w,e) |
446 |
|
|
447 |
|
we should never count the new record address w as live if w |
448 |
|
has only one use (which is often the case). |
449 |
|
|
450 |
|
We should do something similar to floating point, but the transformation |
451 |
|
there is much more complex, so I won't deal with that. |
452 |
|
|
453 |
|
Secondly, there are now two new cps primops at our disposal: |
454 |
|
|
455 |
|
1. rawrecord of record_kind option |
456 |
|
This pure operator allocates some uninitialized storage from the heap. |
457 |
|
There are two forms: |
458 |
|
|
459 |
|
rawrecord NONE [INT n] allocates a tagless record of length n |
460 |
|
rawrecord (SOME rk) [INT n] allocates a tagged record of length n |
461 |
|
and initializes the tag. |
462 |
|
|
463 |
|
2. rawupdate of cty |
464 |
|
rawupdate cty (v,i,x) |
465 |
|
Assigns to x to the ith component of record v. |
466 |
|
The storelist is not updated. |
467 |
|
|
468 |
|
We use these new primops for both spilling and increment record construction. |
469 |
|
|
470 |
|
1. Spilling. |
471 |
|
|
472 |
|
This is implemented with a linear scan algorithm (but generalized |
473 |
|
to trees). The algorithm will create a single spill record at the |
474 |
|
beginning of the cps function and use rawupdate to spill to it, |
475 |
|
and SELECT or SELp to reload from it. So both spills and reloads |
476 |
|
are fine-grain operations. In contrast, in the old algorithm |
477 |
|
"spills" have to be bundled together in records. |
478 |
|
|
479 |
|
Ideally, we should sink the spill record construction to where |
480 |
|
it is needed. We can even split the spill record into multiple ones |
481 |
|
at the places where they are needed. But CPS is not a good |
482 |
|
representation for global code motion, so I'll keep it simple and |
483 |
|
am not attempting this. |
484 |
|
|
485 |
|
2. Incremental record construction (aka record splitting). |
486 |
|
|
487 |
|
Long records with many component values which are simulatenously live |
488 |
|
(recall that single use record addresses are not considered to |
489 |
|
be live) are constructed with rawrecord and rawupdate. |
490 |
|
We allocate space on the heap with rawrecord first, then gradually |
491 |
|
fill it in with rawupdate. This is the technique suggested to me |
492 |
|
by Matthias. |
493 |
|
|
494 |
|
Some restrictions on when this is applicable: |
495 |
|
1. It is not a VECTOR record. The code generator currently does not handle |
496 |
|
this case. VECTOR record uses double indirection like arrays. |
497 |
|
2. All the record component values are defined in the same "basic block" |
498 |
|
as the record constructor. This is to prevent speculative |
499 |
|
record construction. |
500 |
|
|
501 |
|
---------------------------------------------------------------------- |
502 |
|
Name: Allen Leung |
503 |
|
Date: 2002/02/22 01:02:00 EST |
504 |
|
Tag: leunga-20020222-mlrisc-tools |
505 |
|
|
506 |
|
Minor bug fixes in the parser and rewriter |
507 |
|
|
508 |
|
---------------------------------------------------------------------- |
509 |
|
Name: Allen Leung |
510 |
|
Date: 2002/02/21 20:20:00 EST |
511 |
|
Tag: leunga-20020221-peephole |
512 |
|
|
513 |
|
Regenerated the peephole files. Some contained typos in the specification |
514 |
|
and some didn't compile because of pretty printing bugs in the old version |
515 |
|
of 'nowhere'. |
516 |
|
|
517 |
|
---------------------------------------------------------------------- |
518 |
|
Name: Allen Leung |
519 |
|
Date: 2002/02/19 20:20:00 EST |
520 |
|
Tag: leunga-20020219-mlrisc-tools |
521 |
|
Description: |
522 |
|
|
523 |
|
Minor bug fixes to the mlrisc-tools library: |
524 |
|
|
525 |
|
1. Fixed up parsing colon suffixed keywords |
526 |
|
2. Added the ability to shut the error messages up |
527 |
|
3. Reimplemented the pretty printer and fixed up/improved |
528 |
|
the pretty printing of handle and -> types. |
529 |
|
4. Fixed up generation of literal symbols in the nowhere tool. |
530 |
|
5. Added some SML keywords to to sml.sty |
531 |
|
|
532 |
|
---------------------------------------------------------------------- |
533 |
|
Name: Matthias Blume |
534 |
|
Date: 2002/02/19 16:20:00 EST |
535 |
|
Tag: blume-20020219-cmffi |
536 |
|
Description: |
537 |
|
|
538 |
|
A wild mix of changes, some minor, some major: |
539 |
|
|
540 |
|
* All C FFI-related libraries are now anchored under $c: |
541 |
|
$/c.cm --> $c/c.cm |
542 |
|
$/c-int.cm --> $c/internals/c-int.cm |
543 |
|
$/memory.cm --> $c/memory/memory.cm |
544 |
|
|
545 |
|
* "make" tool (in CM) now treats its argument pathname slightly |
546 |
|
differently: |
547 |
|
1. If the native expansion is an absolute name, then before invoking |
548 |
|
the "make" command on it, CM will apply OS.Path.mkRelative |
549 |
|
(with relativeTo = OS.FileSys.getDir()) to it. |
550 |
|
2. The argument will be passed through to subsequent phases of CM |
551 |
|
processing without "going native". In particular, if the argument |
552 |
|
was an anchored path, then "make" will not lose track of that anchor. |
553 |
|
|
554 |
|
* Compiler backends now "know" their respective C calling conventions |
555 |
|
instead of having to be told about it by ml-nlffigen. This relieves |
556 |
|
ml-nlffigen from one of its burdens. |
557 |
|
|
558 |
|
* The X86Backend has been split into X86CCallBackend and X86StdCallBackend. |
559 |
|
|
560 |
|
* Export C_DEBUG and C_Debug from $c/c.cm. |
561 |
|
|
562 |
|
* C type encoding in ml-nlffi-lib has been improved to model the conceptual |
563 |
|
subtyping relationship between incomplete pointers and their complete |
564 |
|
counterparts. For this, ('t, 'c) ptr has been changed to 'o ptr -- |
565 |
|
with the convention of instantiating 'o with ('t, 'c) obj whenever |
566 |
|
the pointer target type is complete. In the incomplete case, 'o |
567 |
|
will be instantiated with some "'c iobj" -- a type obtained by |
568 |
|
using one of the functors PointerToIncompleteType or PointerToCompleteType. |
569 |
|
|
570 |
|
Operations that work on both incomplete and complete pointer types are |
571 |
|
typed as taking an 'o ptr while operations that require the target to |
572 |
|
be known are typed as taking some ('t, 'c) obj ptr. |
573 |
|
|
574 |
|
voidptr is now a bit "more concrete", namely "type voidptr = void ptr'" |
575 |
|
where void is an eqtype without any values. This makes it possible |
576 |
|
to work on voidptr values using functions meant to operate on light |
577 |
|
incomplete pointers. |
578 |
|
|
579 |
|
* As a result of the above, signature POINTER_TO_INCOMPLETE_TYPE has |
580 |
|
been vastly simplified. |
581 |
|
|
582 |
|
---------------------------------------------------------------------- |
583 |
|
Name: Matthias Blume |
584 |
|
Date: 2002/02/19 10:48:00 EST |
585 |
|
Tag: blume-20020219-pqfix |
586 |
|
Description: |
587 |
|
|
588 |
|
Applied Chris Okasaki's bug fix for priority queues. |
589 |
|
|
590 |
|
---------------------------------------------------------------------- |
591 |
|
Name: Matthias Blume |
592 |
|
Date: 2002/02/15 17:05:00 |
593 |
|
Tag: Release_110_39 |
594 |
|
Description: |
595 |
|
|
596 |
|
Last-minute retagging is becoming a tradition... :-( |
597 |
|
|
598 |
|
This is the working release 110.39. |
599 |
|
|
600 |
|
---------------------------------------------------------------------- |
601 |
|
Name: Matthias Blume |
602 |
|
Date: 2002/02/15 16:00:00 EST |
603 |
|
Tag: Release_110_39-orig |
604 |
|
Description: |
605 |
|
|
606 |
|
Working release 110.39. New bootfiles. |
607 |
|
|
608 |
|
(Update: There was a small bug in the installer so it wouldn't work |
609 |
|
with all shells. So I retagged. -Matthias) |
610 |
|
|
611 |
|
---------------------------------------------------------------------- |
612 |
|
Name: Matthias Blume |
613 |
|
Date: 2002/02/15 14:17:00 EST |
614 |
|
Tag: blume-20020215-showbindings |
615 |
|
Description: |
616 |
|
|
617 |
|
Added EnvRef.listBoundSymbols and CM.State.showBindings. Especially |
618 |
|
the latter can be useful for exploring what bindings are available at |
619 |
|
the interactive prompt. (The first function returns only the list |
620 |
|
of symbols that are really bound, the second prints those but also the |
621 |
|
ones that CM's autoloading mechanism knows about.) |
622 |
|
|
623 |
|
---------------------------------------------------------------------- |
624 |
|
Name: Matthias Blume |
625 |
|
Date: 2002/02/15 12:08:00 EST |
626 |
|
Tag: blume-20020215-iptrs |
627 |
|
Description: |
628 |
|
|
629 |
|
Two improvements to ml-nlffigen: |
630 |
|
|
631 |
|
1. Write files only if they do not exist or if their current contents |
632 |
|
do not coincide with what's being written. (That is, avoid messing |
633 |
|
with the time stamps unless absolutely necessary.) |
634 |
|
|
635 |
|
2. Implement a "repository" mechanism for generated files related |
636 |
|
to "incomplete pointer types". See the README file for details. |
637 |
|
|
638 |
|
---------------------------------------------------------------------- |
639 |
|
Name: Matthias Blume |
640 |
|
Date: 2002/02/14 11:50:00 EST |
641 |
|
Tag: blume-20020214-quote |
642 |
|
Description: |
643 |
|
|
644 |
|
Added a type 't t_' to tag.sml (in ml-nlffi-lib.cm). This is required |
645 |
|
because of the new and improved tag generation scheme. (Thanks to Allen |
646 |
|
Leung for pointing it out.) |
647 |
|
|
648 |
|
---------------------------------------------------------------------- |
649 |
|
Name: Lal George |
650 |
|
Date: 2002/02/14 09:55:27 EST 2002 |
651 |
|
Tag: george-20020214-isabelle-bug |
652 |
|
Description: |
653 |
|
|
654 |
|
Fixed the MLRISC bug sent by Markus Wenzel regarding the compilation |
655 |
|
of Isabelle on the x86. |
656 |
|
|
657 |
|
From Allen: |
658 |
|
----------- |
659 |
|
I've found the problem: |
660 |
|
|
661 |
|
in ra-core.sml, I use the counter "blocked" to keep track of the |
662 |
|
true number of elements in the freeze queue. When the counter goes |
663 |
|
to zero, I skip examining the queue. But I've messed up the |
664 |
|
bookkeeping in combine(): |
665 |
|
|
666 |
|
else (); |
667 |
|
case !ucol of |
668 |
|
PSEUDO => (if !cntv > 0 then |
669 |
|
(if !cntu > 0 then blocked := !blocked - 1 else (); |
670 |
|
^^^^^^^^^^^^^^^^^^^^^^^ |
671 |
|
moveu := mergeMoveList(!movev, !moveu) |
672 |
|
) |
673 |
|
else (); |
674 |
|
|
675 |
|
combine() is called to coalesce two nodes u and v. |
676 |
|
I think I was thinking that if the move counts of u and v are both |
677 |
|
greater than zero then after they are coalesced then one node is |
678 |
|
removed from the freeze queue. Apparently I was thinking that |
679 |
|
both u and v are of low degree, but that's clearly not necessarily true. |
680 |
|
|
681 |
|
|
682 |
|
02/12/2002: |
683 |
|
Here's the patch. HOL now compiles. |
684 |
|
|
685 |
|
I don't know how this impact on performance (compile |
686 |
|
time or runtime). This bug caused the RA (especially on the x86) |
687 |
|
to go thru the potential spill phase when there are still nodes on the |
688 |
|
freeze queue. |
689 |
|
|
690 |
|
|
691 |
|
|
692 |
|
|
693 |
|
---------------------------------------------------------------------- |
694 |
|
Name: Matthias Blume |
695 |
|
Date: 2002/02/13 22:40:00 EST |
696 |
|
Tag: blume-20020213-fptr-rtti |
697 |
|
Description: |
698 |
|
|
699 |
|
Fixed a bug in ml-nlffigen that was introduced with one of the previous |
700 |
|
updates. |
701 |
|
|
702 |
|
---------------------------------------------------------------------- |
703 |
|
Name: Matthias Blume |
704 |
|
Date: 2002/02/13 16:41:00 EST |
705 |
|
Tag: blume-20020213-cmlpq |
706 |
|
Description: |
707 |
|
|
708 |
|
Added new priority queue export symbols (which have just been added to |
709 |
|
smlnj-lib.cm) to CML's version of smlnj-lib.cm. (Otherwise CML would |
710 |
|
not compile and the installer would choke.) |
711 |
|
|
712 |
|
---------------------------------------------------------------------- |
713 |
|
Name: Matthias Blume |
714 |
|
Date: 2002/02/13 16:15:00 EST |
715 |
|
Tag: blume-20020213-various |
716 |
|
Description: |
717 |
|
|
718 |
|
1. More tweaks to ml-nlffigen: |
719 |
|
|
720 |
|
- better internal datastructures (resulting in slight speedup) |
721 |
|
- "-match" option requires exact match |
722 |
|
- "localized" gensym counters (untagged structs/unions nested within |
723 |
|
other structs/unions or within typedefs get a fresh counter; their |
724 |
|
tag will be prefixed by a concatenation of their parents' tags) |
725 |
|
- bug fixes (related to calculation of transitive closure of types |
726 |
|
to be included in the output) |
727 |
|
|
728 |
|
2. Minor Basis updates: |
729 |
|
|
730 |
|
- added implementations for List.collate and Option.app |
731 |
|
|
732 |
|
---------------------------------------------------------------------- |
733 |
|
Name: Matthias Blume |
734 |
|
Date: 2002/02/11 15:55:00 EST |
735 |
|
Tag: blume-20020211-gensym |
736 |
|
Description: |
737 |
|
|
738 |
|
Added a "-gensym" option to command line of ml-nlffigen. This can be |
739 |
|
used to specify a "stem" -- a string that is inserted in all "gensym'd" |
740 |
|
names (ML structure names that correspond to unnamed C structs, unions, |
741 |
|
and enums), so that separate runs of ml-nlffigen do not clash. |
742 |
|
|
743 |
|
---------------------------------------------------------------------- |
744 |
|
Name: Matthias Blume |
745 |
|
Date: 2002/02/11 12:05:00 EST |
746 |
|
Tag: blume-20020211-gensml |
747 |
|
Description: |
748 |
|
|
749 |
|
A quick fix for a problem with GenSML (in the pgraph-util library): |
750 |
|
Make generation of toplevel "local" optional. (Strictly speaking, |
751 |
|
signature definitions within "local" are not legal SML.) |
752 |
|
|
753 |
|
Other than that: updates to INSTALL and cm/TODO. |
754 |
|
|
755 |
|
---------------------------------------------------------------------- |
756 |
|
Name: Matthias Blume |
757 |
|
Date: 2002/02/08 15:00:00 EST |
758 |
|
Tag: blume-20020208-uniquepid |
759 |
|
Description: |
760 |
|
|
761 |
|
0. Version number has been bumped to 110.38.1. NEW BOOTFILES!!! |
762 |
|
|
763 |
|
1. The installer (config/install.sh) has gotten smarter: |
764 |
|
|
765 |
|
- Configuration options are a bit easier to specify now |
766 |
|
(in config/targets). |
767 |
|
- Bug in recognizing .tar.bz2 files fixed. |
768 |
|
- Installer automatically resolves dependencies between |
769 |
|
configuration options (e.g., if you ask for eXene, you will |
770 |
|
also get cml -- regardless whether you asked for it or not). |
771 |
|
- Installer can run in "quieter mode" by setting the environment |
772 |
|
variable INSTALL_QUIETLY to "true". "Quieter" does not mean |
773 |
|
"completely silent", though. |
774 |
|
- Build HashCons library as part of smlnj-lib. |
775 |
|
|
776 |
|
2. A new scheme for assigning persistent identifiers to compilation |
777 |
|
units (and, by extension, to types etc.) has been put into place. |
778 |
|
This fixes a long-standing bug where types and even dynamic values |
779 |
|
can get internally confused, thereby compromising type safety |
780 |
|
(abstraction) and dynamic correctness. See |
781 |
|
|
782 |
|
http://cm.bell-labs.com/cm/cs/who/blume/pid-confusion.tgz |
783 |
|
|
784 |
|
for an example of how things could go wrong until now. |
785 |
|
|
786 |
|
The downside of the new scheme is that pids are not quite as |
787 |
|
persistent as they used to be: CM will generate a fresh pid |
788 |
|
for every compilation unit that it thinks it sees for the first |
789 |
|
time. That means that if you compile starting from a clean, fresh |
790 |
|
source tree at two different times, you end up with different |
791 |
|
binaries. |
792 |
|
|
793 |
|
Cutoff recompilation, however, has not been compromised because |
794 |
|
CM keeps pid information in special caches between runs. |
795 |
|
|
796 |
|
---------------------------------------------------------------------- |
797 |
|
Name: Lal George |
798 |
|
Date: 2002/02/07 15:34:13 EST 2002 |
799 |
|
Tag: <none> |
800 |
|
Description: |
801 |
|
|
802 |
|
Compilers that generate assembly code may produce global labels |
803 |
|
whose value is resolved at link time. The various peephole optimization |
804 |
|
modules did not take this in account. |
805 |
|
|
806 |
|
TODO. The Labels.addrOf function should really return an option |
807 |
|
type so that clients are forced to deal with this issue, rather |
808 |
|
than an exception being raised. |
809 |
|
|
810 |
|
---------------------------------------------------------------------- |
811 |
|
Name: Lal George |
812 |
|
Date: 2002/02/06 13:55:02 EST |
813 |
|
Tag: george-20020206-ra-breakup |
814 |
|
Description: |
815 |
|
|
816 |
|
1. A bug fix from Allen. |
817 |
|
|
818 |
|
A typo causes extra fstp %st(0)'s to be generated at compensation |
819 |
|
edges, which might cause stack underflow traps at runtime. This |
820 |
|
occurs in fft where there are extraneous fstps right before the 'into' |
821 |
|
trap instruction (in this case they are harmless since none of the |
822 |
|
integers overflow.) |
823 |
|
|
824 |
|
2. Pulled out various utility modules that were embedded in the modules |
825 |
|
of the register allocator. I need these modules for other purposes, but |
826 |
|
they are not complete enough to put into a library (just yet). |
827 |
|
---------------------------------------------------------------------- |
828 |
|
Name: Matthias Blume |
829 |
|
Date: 2002/01/31 16:05:00 EST |
830 |
|
Tag: blume-20020131-sparc-ccalls |
831 |
|
Description: |
832 |
|
|
833 |
|
1. C-calls on Sparc needlessly allocated a huge chunk (96 bytes) |
834 |
|
of extra stack space by mistake. Fixed. |
835 |
|
|
836 |
|
2. Bug in logic of handling of command-line options in ml-nlffigen fixed. |
837 |
|
|
838 |
|
---------------------------------------------------------------------- |
839 |
|
Name: Allen Leung |
840 |
|
Date: 2002/01/30 |
841 |
|
Tag: leunga-20020130-nowhere-bug-fix |
842 |
|
Description: |
843 |
|
|
844 |
|
MLRISC bug fixes: |
845 |
|
1. Fixed a bindings computation bug in the 'nowhere' program generator tool. |
846 |
|
2. MachineInt.fromString was negating its value. |
847 |
|
|
848 |
|
---------------------------------------------------------------------- |
849 |
|
Name: Matthias Blume |
850 |
|
Date: 2002/01/29 |
851 |
|
Tag: blume-20020129-INSTALL |
852 |
|
Description: |
853 |
|
|
854 |
|
- Added somewhat detailed installation instructions (file INSTALL). |
855 |
|
- Fixed curl-detection bug in config/install.sh. |
856 |
|
- It is now possible to select the URL getter using the URLGETTER |
857 |
|
environment variable: |
858 |
|
|
859 |
|
not set / "unknown" --> automatic detection (script tries wget, |
860 |
|
curl, and lynx) |
861 |
|
"wget" / "curl" / "lynx" --> use the specified program (script "knows" |
862 |
|
how to properly invoke them) |
863 |
|
other --> use $URLGETTER directly, it must take |
864 |
|
precisely two command-line arguments |
865 |
|
(source URL and destination file name) |
866 |
|
|
867 |
---------------------------------------------------------------------- |
---------------------------------------------------------------------- |
868 |
Name: Matthias Blume |
Name: Matthias Blume |
1445 |
|
|
1446 |
---------------------------------------------------------------------- |
---------------------------------------------------------------------- |
1447 |
Name: Matthias Blume |
Name: Matthias Blume |
|
>>>>>>> 1.169 |
|
1448 |
Date: 2001/09/18 15:35:00 EDT |
Date: 2001/09/18 15:35:00 EDT |
1449 |
Tag: blume-20010918-readme11036 |
Tag: blume-20010918-readme11036 |
1450 |
Description: |
Description: |