/usr/share/fpcsrc/2.6.2/compiler/optloop.pas is in fpc-source-2.6.2 2.6.2-8.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 | {
Loop optimization
Copyright (c) 2005 by Florian Klaempfl
This program is free software; you can redistribute it and/or modify
it under the terms of the GNU General Public License as published by
the Free Software Foundation; either version 2 of the License, or
(at your option) any later version.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
****************************************************************************
}
unit optloop;
{$i fpcdefs.inc}
interface
uses
node;
function unroll_loop(node : tnode) : tnode;
function OptimizeInductionVariables(node : tnode) : boolean;
implementation
uses
cutils,cclasses,
globtype,globals,constexp,
symdef,symsym,
defutil,
cpuinfo,
nutils,
nadd,nbas,nflw,ncon,ninl,ncal,nld,nmem,ncnv,
ncgmem,
pass_1,
optbase,optutils,
procinfo;
var
nodecount : aword;
function donodecount(var n: tnode; arg: pointer): foreachnoderesult;
begin
inc(nodecount);
result:=fen_false;
end;
{ rough estimation how large the tree "node" is }
function countnodes(node : tnode) : aword;
begin
nodecount:=0;
foreachnodestatic(node,@donodecount,nil);
result:=nodecount;
end;
function number_unrolls(node : tnode) : cardinal;
begin
{$ifdef i386}
{ multiply by 2 for CPUs with a long pipeline }
if current_settings.cputype in [cpu_Pentium4] then
number_unrolls:=60 div countnodes(node)
else
{$endif i386}
number_unrolls:=30 div countnodes(node);
if number_unrolls=0 then
number_unrolls:=1;
end;
function unroll_loop(node : tnode) : tnode;
var
unrolls,i : cardinal;
counts : qword;
unrollstatement,newforstatement : tstatementnode;
unrollblock : tblocknode;
begin
result:=nil;
if (cs_opt_size in current_settings.optimizerswitches) then
exit;
if not(node.nodetype in [forn]) then
exit;
unrolls:=number_unrolls(tfornode(node).t2);
if unrolls>1 then
begin
{ number of executions known? }
if (tfornode(node).right.nodetype=ordconstn) and (tfornode(node).t1.nodetype=ordconstn) then
begin
if lnf_backward in tfornode(node).loopflags then
counts:=tordconstnode(tfornode(node).right).value-tordconstnode(tfornode(node).t1).value+1
else
counts:=tordconstnode(tfornode(node).t1).value-tordconstnode(tfornode(node).right).value+1;
{ don't unroll more than we need }
if unrolls>counts then
unrolls:=counts;
{ create block statement }
unrollblock:=internalstatements(unrollstatement);
{ let's unroll (and rock of course) }
for i:=1 to unrolls do
begin
{ create and insert copy of the statement block }
addstatement(unrollstatement,tfornode(node).t2.getcopy);
{ set and insert entry label? }
if (counts mod unrolls<>0) and
((counts mod unrolls)=unrolls-i) then
begin
tfornode(node).entrylabel:=clabelnode.create(cnothingnode.create,tlabelsym.create('$optunrol'));
addstatement(unrollstatement,tfornode(node).entrylabel);
end;
{ for itself increases at the last iteration }
if i<unrolls then
begin
{ insert incr/decrementation of counter var }
if lnf_backward in tfornode(node).loopflags then
addstatement(unrollstatement,
geninlinenode(in_dec_x,false,ccallparanode.create(tfornode(node).left.getcopy,nil)))
else
addstatement(unrollstatement,
geninlinenode(in_inc_x,false,ccallparanode.create(tfornode(node).left.getcopy,nil)));
end;
end;
{ can we get rid of the for statement? }
if unrolls=counts then
begin
{ create block statement }
result:=internalstatements(newforstatement);
{ initial assignment }
addstatement(newforstatement,cassignmentnode.create(
tfornode(node).left.getcopy,tfornode(node).right.getcopy));
addstatement(newforstatement,unrollblock);
end;
end
else
begin
{ unrolling is a little bit more tricky if we don't know the
loop count at compile time, but the solution is to use a jump table
which is indexed by "loop count mod unrolls" at run time and which
jumps then at the appropriate place inside the loop. Because
a module division is expensive, we can use only unroll counts dividable
by 2 }
case unrolls of
1..2:
;
3:
unrolls:=2;
4..7:
unrolls:=4;
{ unrolls>4 already make no sense imo, but who knows (FK) }
8..15:
unrolls:=8;
16..31:
unrolls:=16;
32..63:
unrolls:=32;
64..$7fff:
unrolls:=64;
else
exit;
end;
{ we don't handle this yet }
exit;
end;
if not(assigned(result)) then
begin
tfornode(node).t2.free;
tfornode(node).t2:=unrollblock;
end;
end;
end;
var
initcode,
calccode,
deletecode : tblocknode;
initcodestatements,
calccodestatements,
deletecodestatements: tstatementnode;
templist : tfplist;
inductionexprs : tfplist;
changedforloop,
containsnestedforloop : boolean;
function is_loop_invariant(loop : tnode;expr : tnode) : boolean;
begin
result:=is_constnode(expr);
case expr.nodetype of
loadn:
begin
if (pi_dfaavailable in current_procinfo.flags) and
assigned(loop.optinfo) and
assigned(expr.optinfo) then
{ no aliasing? }
result:=not(tabstractvarsym(tloadnode(expr).symtableentry).addr_taken) and
{ no definition in the loop? }
not(DFASetIn(loop.optinfo^.defsum,expr.optinfo^.index));
end;
vecn:
begin
result:=((tvecnode(expr).left.nodetype=loadn) or is_loop_invariant(loop,tvecnode(expr).left)) and
is_loop_invariant(loop,tvecnode(expr).right);
end;
typeconvn:
result:=is_loop_invariant(loop,ttypeconvnode(expr).left);
end;
end;
{ checks if the strength of n can be recuded, arg is the tforloop being considered }
function dostrengthreductiontest(var n: tnode; arg: pointer): foreachnoderesult;
function findpreviousstrengthreduction : boolean;
var
i : longint;
begin
result:=false;
for i:=0 to inductionexprs.count-1 do
begin
{ do we already maintain one expression? }
if tnode(inductionexprs[i]).isequal(n) then
begin
n.free;
case n.nodetype of
muln:
n:=ctemprefnode.create(ttempcreatenode(templist[i]));
vecn:
n:=ctypeconvnode.create_internal(cderefnode.create(ctemprefnode.create(
ttempcreatenode(templist[i]))),n.resultdef);
else
internalerror(200809211);
end;
result:=true;
exit;
end;
end;
end;
procedure CreateNodes;
begin
if not assigned(initcode) then
begin
initcode:=internalstatements(initcodestatements);
calccode:=internalstatements(calccodestatements);
deletecode:=internalstatements(deletecodestatements);
end;
end;
var
tempnode : ttempcreatenode;
dummy : longint;
begin
result:=fen_false;
case n.nodetype of
forn:
{ inform for loop search routine, that it needs to search more deeply }
containsnestedforloop:=true;
muln:
begin
if (taddnode(n).right.nodetype=loadn) and
taddnode(n).right.isequal(tfornode(arg).left) and
{ plain read of the loop variable? }
not(nf_write in taddnode(n).right.flags) and
not(nf_modify in taddnode(n).right.flags) and
is_loop_invariant(tfornode(arg),taddnode(n).left) and
{ for now, we can handle only constant lower borders }
is_constnode(tfornode(arg).right) then
taddnode(n).swapleftright;
if (taddnode(n).left.nodetype=loadn) and
taddnode(n).left.isequal(tfornode(arg).left) and
{ plain read of the loop variable? }
not(nf_write in taddnode(n).left.flags) and
not(nf_modify in taddnode(n).left.flags) and
is_loop_invariant(tfornode(arg),taddnode(n).right) and
{ for now, we can handle only constant lower borders }
is_constnode(tfornode(arg).right) then
begin
changedforloop:=true;
{ did we use the same expression before already? }
if not(findpreviousstrengthreduction) then
begin
tempnode:=ctempcreatenode.create(n.resultdef,n.resultdef.size,tt_persistent,
tstoreddef(n.resultdef).is_intregable or tstoreddef(n.resultdef).is_fpuregable);
templist.Add(tempnode);
inductionexprs.Add(n);
CreateNodes;
if lnf_backward in tfornode(arg).loopflags then
addstatement(calccodestatements,
geninlinenode(in_dec_x,false,
ccallparanode.create(ctemprefnode.create(tempnode),ccallparanode.create(taddnode(n).right.getcopy,nil))))
else
addstatement(calccodestatements,
geninlinenode(in_inc_x,false,
ccallparanode.create(ctemprefnode.create(tempnode),ccallparanode.create(taddnode(n).right.getcopy,nil))));
addstatement(initcodestatements,tempnode);
addstatement(initcodestatements,cassignmentnode.create(ctemprefnode.create(tempnode),
caddnode.create(muln,
caddnode.create(subn,tfornode(arg).right.getcopy,cordconstnode.create(1,tfornode(arg).right.resultdef,false)),
taddnode(n).right.getcopy)
)
);
{ finally replace the node by a temp. ref }
n:=ctemprefnode.create(tempnode);
{ ... and add a temp. release node }
addstatement(deletecodestatements,ctempdeletenode.create(tempnode));
end;
{ set types }
do_firstpass(n);
result:=fen_norecurse_false;
end;
end;
vecn:
begin
{ is the index the counter variable? }
if not(is_special_array(tvecnode(n).left.resultdef)) and
(tvecnode(n).right.isequal(tfornode(arg).left) or
{ fpc usually creates a type cast to access an array }
((tvecnode(n).right.nodetype=typeconvn) and
ttypeconvnode(tvecnode(n).right).left.isequal(tfornode(arg).left)
)
) and
{ plain read of the loop variable? }
not(nf_write in tvecnode(n).right.flags) and
not(nf_modify in tvecnode(n).right.flags) and
{ direct array access? }
((tvecnode(n).left.nodetype=loadn) or
{ ... or loop invariant expression? }
is_loop_invariant(tfornode(arg),tvecnode(n).left)) and
{ removing the multiplication is only worth the
effort if it's not a simple shift }
not(ispowerof2(tcgvecnode(n).get_mul_size,dummy)) then
begin
changedforloop:=true;
{ did we use the same expression before already? }
if not(findpreviousstrengthreduction) then
begin
tempnode:=ctempcreatenode.create(voidpointertype,voidpointertype.size,tt_persistent,true);
templist.Add(tempnode);
inductionexprs.Add(n);
CreateNodes;
if lnf_backward in tfornode(arg).loopflags then
addstatement(calccodestatements,
geninlinenode(in_dec_x,false,
ccallparanode.create(ctemprefnode.create(tempnode),ccallparanode.create(
cordconstnode.create(tcgvecnode(n).get_mul_size,tfornode(arg).right.resultdef,false),nil))))
else
addstatement(calccodestatements,
geninlinenode(in_inc_x,false,
ccallparanode.create(ctemprefnode.create(tempnode),ccallparanode.create(
cordconstnode.create(tcgvecnode(n).get_mul_size,tfornode(arg).right.resultdef,false),nil))));
addstatement(initcodestatements,tempnode);
addstatement(initcodestatements,cassignmentnode.create(ctemprefnode.create(tempnode),
caddrnode.create(
cvecnode.create(tvecnode(n).left.getcopy,tfornode(arg).right.getcopy)
)
));
{ finally replace the node by a temp. ref }
n:=ctypeconvnode.create_internal(cderefnode.create(ctemprefnode.create(tempnode)),n.resultdef);
{ ... and add a temp. release node }
addstatement(deletecodestatements,ctempdeletenode.create(tempnode));
end;
{ set types }
do_firstpass(n);
result:=fen_norecurse_false;
end;
end;
end;
end;
function OptimizeInductionVariablesSingleForLoop(node : tnode) : tnode;
var
loopcode : tblocknode;
loopcodestatements,
newcodestatements : tstatementnode;
fornode : tfornode;
begin
result:=nil;
if node.nodetype<>forn then
exit;
templist:=TFPList.Create;
inductionexprs:=TFPList.Create;
initcode:=nil;
calccode:=nil;
deletecode:=nil;
initcodestatements:=nil;
calccodestatements:=nil;
deletecodestatements:=nil;
{ find all expressions being candidates for strength reduction
and replace them }
foreachnodestatic(pm_postprocess,node,@dostrengthreductiontest,node);
{ clue everything together }
if assigned(initcode) then
begin
do_firstpass(tnode(initcode));
do_firstpass(tnode(calccode));
do_firstpass(tnode(deletecode));
{ create a new for node, the old one will be released by the compiler }
with tfornode(node) do
begin
fornode:=cfornode.create(left,right,t1,t2,lnf_backward in loopflags);
left:=nil;
right:=nil;
t1:=nil;
t2:=nil;
end;
node:=fornode;
loopcode:=internalstatements(loopcodestatements);
addstatement(loopcodestatements,calccode);
addstatement(loopcodestatements,tfornode(node).t2);
tfornode(node).t2:=loopcode;
do_firstpass(node);
result:=internalstatements(newcodestatements);
addstatement(newcodestatements,initcode);
addstatement(newcodestatements,node);
addstatement(newcodestatements,deletecode);
end;
templist.Free;
inductionexprs.Free;
end;
function iterforloops(var n: tnode; arg: pointer): foreachnoderesult;
var
hp : tnode;
begin
Result:=fen_false;
if n.nodetype=forn then
begin
{ do we have DFA available? }
if pi_dfaavailable in current_procinfo.flags then
begin
CalcDefSum(n);
end;
containsnestedforloop:=false;
hp:=OptimizeInductionVariablesSingleForLoop(n);
if assigned(hp) then
begin
n.Free;
n:=hp;
end;
{ can we avoid further searching? }
if not(containsnestedforloop) then
Result:=fen_norecurse_false;
end;
end;
function OptimizeInductionVariables(node : tnode) : boolean;
begin
changedforloop:=false;
foreachnodestatic(pm_postprocess,node,@iterforloops,nil);
Result:=changedforloop;
end;
end.
|