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