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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499
|
(*
* X86 specific register allocator.
* This module abstracts out all the nasty RA business on the x86.
* So you should only have to write the callbacks.
*)
functor X86RA
( structure I : X86INSTR
structure InsnProps : INSN_PROPERTIES where I = I
structure F : FLOWGRAPH where I = I
structure Asm : INSTRUCTION_EMITTER where I = I
(* Spilling heuristics determines which node should be spilled.
* You can use Chaitin, ChowHenessey, or one of your own.
*)
structure SpillHeur : RA_SPILL_HEURISTICS
(* The Spill module figures out the strategies for inserting
* spill code. You can use RASpill, or RASpillWithRenaming,
* or write your own if you are feeling adventurous.
*)
structure Spill : RA_SPILL where I = I
sharing F.P = Asm.P
(* Should we use allocate register on the floating point stack?
* Note that this flag must match the one passed to the code generator
* module.
*)
val fast_floating_point : bool ref
datatype raPhase = SPILL_PROPAGATION
| SPILL_COLORING
(* Called before register allocation; perform your initialization here. *)
val beforeRA : unit -> unit
(* Integer register allocation parameters *)
structure Int :
sig
val avail : I.C.cell list
val dedicated : I.C.cell list
val memRegs : I.C.cell list
val phases : raPhase list
val spillLoc : Annotations.annotations ref *
RAGraph.logical_spill_id -> I.operand
(* This function is called once before spilling begins *)
val spillInit : RAGraph.interferenceGraph -> unit
end
(* Floating point register allocation parameters *)
structure Float :
sig
(* Sethi-Ullman mode *)
val avail : I.C.cell list
val dedicated : I.C.cell list
val memRegs : I.C.cell list
val phases : raPhase list
val spillLoc : Annotations.annotations ref *
RAGraph.logical_spill_id -> I.operand
(* This function is called once before spilling begins *)
val spillInit : RAGraph.interferenceGraph -> unit
(* When fast_floating_point is on, use these instead: *)
val fastMemRegs : I.C.cell list
val fastPhases : raPhase list
end
) : CLUSTER_OPTIMIZATION =
struct
structure F = F
structure I = I
structure C = I.C
val name = "X86RA"
type flowgraph = F.cluster
val intSpillCnt = MLRiscControl.getCounter "ra-int-spills"
val floatSpillCnt = MLRiscControl.getCounter "ra-float-spills"
val intReloadCnt = MLRiscControl.getCounter "ra-int-reloads"
val floatReloadCnt = MLRiscControl.getCounter "ra-float-reloads"
val intRenameCnt = MLRiscControl.getCounter "ra-int-renames"
val floatRenameCnt = MLRiscControl.getCounter "ra-float-renames"
val x86CfgDebugFlg = MLRiscControl.getFlag "x86-cfg-debug"
(*
val deadcode = MLRiscControl.getCounter "x86-dead-code"
val deadblocks = MLRiscControl.getCounter "x86-dead-blocks"
*)
structure PrintFlowGraph=
PrintCluster(structure Flowgraph=F
structure Asm = Asm)
structure X86FP =
X86FP(structure X86Instr = I
structure X86Props = InsnProps
structure Flowgraph = F
structure Liveness = Liveness(F)
structure Asm = Asm
)
structure X86Spill = X86Spill(structure Instr=I structure Props=InsnProps)
(*
* Dead code elimination
*)
exception X86DeadCode
val affectedBlocks =
IntHashTable.mkTable(32,X86DeadCode) : bool IntHashTable.hash_table
val deadRegs =
IntHashTable.mkTable(32,X86DeadCode) : bool IntHashTable.hash_table
fun removeDeadCode(F.CLUSTER{blocks, ...}) =
let val find = IntHashTable.find deadRegs
fun isDead r =
case find (C.registerId r) of
SOME _ => true
| NONE => false
fun isAffected i = getOpt (IntHashTable.find affectedBlocks i, false)
fun isDeadInstr(I.ANNOTATION{i, ...}) = isDeadInstr i
| isDeadInstr(I.MOVE{dst=I.Direct rd, ...}) = isDead rd
| isDeadInstr(I.MOVE{dst=I.MemReg rd, ...}) = isDead rd
| isDeadInstr(I.COPY{dst=[rd], ...}) = isDead rd
| isDeadInstr _ = false
fun scan [] = ()
| scan(F.BBLOCK{blknum, insns, ...}::rest) =
(if isAffected blknum then
((* deadblocks := !deadblocks + 1; *)
insns := elim(!insns, [])
) else ();
scan rest)
| scan(_::rest) = scan rest
and elim([], code) = rev code
| elim(i::instrs, code) =
if isDeadInstr i then
((* deadcode := !deadcode + 1; *) elim(instrs, code))
else elim(instrs, i::code)
in if IntHashTable.numItems affectedBlocks > 0 then
(scan blocks;
IntHashTable.clear deadRegs;
IntHashTable.clear affectedBlocks)
else ()
end
(* This function finds out which pseudo memory registers are unused.
* Those that are unused are made available for spilling.
* The register allocator calls this function right before spilling
* a set of nodes.
*)
val firstSpill = ref true
val firstFPSpill = ref true
fun spillInit(graph, I.C.GP) =
if !firstSpill then (* only do this once! *)
(Int.spillInit graph;
firstSpill := false
)
else ()
| spillInit(graph, I.C.FP) =
if !firstFPSpill then
(Float.spillInit graph;
firstFPSpill := false
)
else ()
(* This is the generic register allocator *)
structure Ra =
RegisterAllocator
(SpillHeur)
(MemoryRA (* for memory coalescing *)
(RADeadCodeElim (* do the funky dead code elimination stuff *)
(ClusterRA
(structure Flowgraph = F
structure Asm = Asm
structure InsnProps = InsnProps
structure Spill = Spill
)
)
(fun cellkind I.C.GP = true | cellkind _ = false
val deadRegs = deadRegs
val affectedBlocks = affectedBlocks
val spillInit = spillInit
)
)
)
(* -------------------------------------------------------------------
* Floating point stuff
* -------------------------------------------------------------------*)
val KF32 = length Float.avail
structure FR32 = GetReg(val nRegs=KF32
val available=map C.registerId Float.avail
val first=C.registerId(I.C.ST 8))
val availF8 = C.Regs C.FP {from=0, to=6, step=1}
val KF8 = length availF8
structure FR8 = GetReg(val nRegs=KF8
val available=map C.registerId availF8
val first=C.registerId(I.C.ST 0))
(* -------------------------------------------------------------------
* Callbacks for floating point K=32
* -------------------------------------------------------------------*)
fun copyInstrF((rds as [_], rss as [_]), _) =
I.FCOPY{dst=rds, src=rss, tmp=NONE}
| copyInstrF((rds, rss), I.FCOPY{tmp, ...}) =
I.FCOPY{dst=rds, src=rss, tmp=tmp}
| copyInstrF(x, I.ANNOTATION{i,a}) =
I.ANNOTATION{i=copyInstrF(x, i), a=a}
val copyInstrF = fn x => [copyInstrF x]
fun getFregLoc(an, Ra.FRAME loc) = Float.spillLoc(an, loc)
| getFregLoc(an, Ra.MEM_REG r) = I.FDirect r
(* spill floating point *)
fun spillF{instr, reg, spillLoc, kill, annotations=an} =
(floatSpillCnt := !floatSpillCnt + 1;
X86Spill.fspill(instr, reg, getFregLoc(an, spillLoc))
)
fun spillFreg{src, reg, spillLoc, annotations=an} =
(floatSpillCnt := !floatSpillCnt + 1;
[I.FLDL(I.FDirect(src)), I.FSTPL(getFregLoc(an, spillLoc))]
)
fun spillFcopyTmp{copy=I.FCOPY{dst, src, ...}, spillLoc,
annotations=an} =
(floatSpillCnt := !floatSpillCnt + 1;
I.FCOPY{dst=dst, src=src, tmp=SOME(getFregLoc(an, spillLoc))}
)
| spillFcopyTmp{copy=I.ANNOTATION{i,a}, spillLoc, annotations} =
let val i = spillFcopyTmp{copy=i, spillLoc=spillLoc,
annotations=annotations}
in I.ANNOTATION{i=i, a=a} end
(* rename floating point *)
fun renameF{instr, fromSrc, toSrc} =
(floatRenameCnt := !floatRenameCnt + 1;
X86Spill.freload(instr, fromSrc, I.FDirect toSrc)
)
(* reload floating point *)
fun reloadF{instr, reg, spillLoc, annotations=an} =
(floatReloadCnt := !floatReloadCnt + 1;
X86Spill.freload(instr, reg, getFregLoc(an, spillLoc))
)
fun reloadFreg{dst, reg, spillLoc, annotations=an} =
(floatReloadCnt := !floatReloadCnt + 1;
[I.FLDL(getFregLoc(an, spillLoc)), I.FSTPL(I.FDirect dst)]
)
(* -------------------------------------------------------------------
* Callbacks for floating point K=7
* -------------------------------------------------------------------*)
fun FMemReg f = let val fx = C.registerNum f
in if fx >= 8 andalso fx < 32
then I.FDirect f else I.FPR f
end
fun copyInstrF'((rds as [d], rss as [s]), _) =
I.FMOVE{fsize=I.FP64,src=FMemReg s,dst=FMemReg d}
| copyInstrF'((rds, rss), I.FCOPY{tmp, ...}) =
I.FCOPY{dst=rds, src=rss, tmp=tmp}
| copyInstrF'(x, I.ANNOTATION{i, a}) =
I.ANNOTATION{i=copyInstrF'(x,i), a=a}
val copyInstrF' = fn x => [copyInstrF' x]
fun spillFreg'{src, reg, spillLoc, annotations=an} =
(floatSpillCnt := !floatSpillCnt + 1;
[I.FMOVE{fsize=I.FP64, src=FMemReg src, dst=getFregLoc(an,spillLoc)}]
)
fun renameF'{instr, fromSrc, toSrc} =
(floatRenameCnt := !floatRenameCnt + 1;
X86Spill.freload(instr, fromSrc, I.FPR toSrc)
)
fun reloadFreg'{dst, reg, spillLoc, annotations=an} =
(floatReloadCnt := !floatReloadCnt + 1;
[I.FMOVE{fsize=I.FP64, dst=FMemReg dst, src=getFregLoc(an,spillLoc)}]
)
(* -------------------------------------------------------------------
* Integer 8 stuff
* -------------------------------------------------------------------*)
fun memToMemMove{dst, src} =
let val tmp = I.C.newReg()
in [I.MOVE{mvOp=I.MOVL,src=src,dst=I.Direct tmp},
I.MOVE{mvOp=I.MOVL,src=I.Direct tmp,dst=dst}
]
end
fun copyInstrR((rds as [d], rss as [s]), _) =
if C.sameColor(d,s) then [] else
let val dx = C.registerNum d and sx = C.registerNum s
in case (dx >= 8 andalso dx < 32, sx >= 8 andalso sx < 32) of
(false, false) => [I.COPY{dst=rds, src=rss, tmp=NONE}]
| (true, false) => [I.MOVE{mvOp=I.MOVL,src=I.Direct s,
dst=I.MemReg d}]
| (false, true) => [I.MOVE{mvOp=I.MOVL,src=I.MemReg s,
dst=I.Direct d}]
| (true, true) => memToMemMove{src=I.MemReg s, dst=I.MemReg d}
end
| copyInstrR((rds, rss), I.COPY{tmp, ...}) =
[I.COPY{dst=rds, src=rss, tmp=tmp}]
| copyInstrR(x, I.ANNOTATION{i, a}) =
copyInstrR(x, i) (* XXX *)
fun getRegLoc(an, Ra.FRAME loc) = Int.spillLoc(an, loc)
| getRegLoc(an, Ra.MEM_REG r) = I.MemReg r
(* No, logical spill locations... *)
structure GR8 = GetReg(val nRegs=8
val available=map C.registerId Int.avail
val first=0)
val K8 = length Int.avail
(* register allocation for general purpose registers *)
fun spillR8{instr, reg, spillLoc, kill, annotations=an} =
(intSpillCnt := !intSpillCnt + 1;
X86Spill.spill(instr, reg, getRegLoc(an, spillLoc))
)
fun isMemReg r = let val x = C.registerNum r
in x >= 8 andalso x < 32 end
fun spillReg{src, reg, spillLoc, annotations=an} =
let val _ = intSpillCnt := !intSpillCnt + 1;
val dstLoc = getRegLoc(an,spillLoc)
val isMemReg = isMemReg src
val srcLoc = if isMemReg then I.MemReg src else I.Direct src
in if InsnProps.eqOpn(srcLoc, dstLoc) then []
else if isMemReg then memToMemMove{dst=dstLoc, src=srcLoc}
else [I.MOVE{mvOp=I.MOVL, src=srcLoc, dst=dstLoc}]
end
fun spillCopyTmp{copy=I.COPY{src, dst,...}, spillLoc, annotations=an} =
(intSpillCnt := !intSpillCnt + 1;
I.COPY{dst=dst, src=src, tmp=SOME(getRegLoc(an, spillLoc))}
)
fun renameR8{instr, fromSrc, toSrc} =
(intRenameCnt := !intRenameCnt + 1;
X86Spill.reload(instr, fromSrc, I.Direct toSrc)
)
fun reloadR8{instr, reg, spillLoc, annotations=an} =
(intReloadCnt := !intReloadCnt + 1;
X86Spill.reload(instr, reg, getRegLoc(an,spillLoc))
)
fun reloadReg{dst, reg, spillLoc, annotations=an} =
let val _ = intReloadCnt := !intReloadCnt + 1
val srcLoc = getRegLoc(an, spillLoc)
val isMemReg = isMemReg dst
val dstLoc = if isMemReg then I.MemReg dst else I.Direct dst
in if InsnProps.eqOpn(srcLoc,dstLoc) then []
else if isMemReg then memToMemMove{dst=dstLoc, src=srcLoc}
else [I.MOVE{mvOp=I.MOVL, src=srcLoc, dst=dstLoc}]
end
fun resetRA() =
(firstSpill := true;
firstFPSpill := true;
IntHashTable.clear affectedBlocks;
IntHashTable.clear deadRegs;
if !fast_floating_point then FR8.reset() else FR32.reset();
GR8.reset()
)
(* Dedicated + available registers *)
fun mark(a, l) = app (fn r => Array.update(a, C.registerId r, true)) l
val dedicatedR = Array.array(32,false)
val dedicatedF32 = Array.array(64,false)
val dedicatedF8 = Array.array(64,false)
val _ = mark(dedicatedR, Int.dedicated)
val _ = mark(dedicatedF32, Float.dedicated)
fun phases ps =
let fun f([], m) = m
| f(SPILL_PROPAGATION::ps, m) = f(ps, Ra.SPILL_PROPAGATION+m)
| f(SPILL_COLORING::ps, m) = f(ps, Ra.SPILL_COLORING+m)
in f(ps, Ra.NO_OPTIMIZATION)
end
(* RA parameters *)
(* How to allocate integer registers:
* Perform register alocation + memory allocation
*)
val RAInt = {spill = spillR8,
spillSrc = spillReg,
spillCopyTmp= spillCopyTmp,
reload = reloadR8,
reloadDst = reloadReg,
renameSrc = renameR8,
copyInstr = copyInstrR,
K = K8,
getreg = GR8.getreg,
cellkind = I.C.GP,
dedicated = dedicatedR,
spillProh = [],
memRegs = Int.memRegs,
mode = phases(Int.phases)
} : Ra.raClient
(* How to allocate floating point registers:
* Allocate all fp registers on the stack. This is the easy way.
*)
val RAFP32 ={spill = spillF,
spillSrc = spillFreg,
spillCopyTmp= spillFcopyTmp,
reload = reloadF,
reloadDst = reloadFreg,
renameSrc = renameF,
copyInstr = copyInstrF,
K = KF32,
getreg = FR32.getreg,
cellkind = I.C.FP,
dedicated = dedicatedF32,
spillProh = [],
memRegs = Float.memRegs,
mode = phases(Float.phases)
} : Ra.raClient
(* How to allocate floating point registers:
* Allocate fp registers on the %st stack. Also perform
* memory allcoation.
*)
val RAFP8 ={spill = spillF,
spillSrc = spillFreg',
spillCopyTmp= spillFcopyTmp,
reload = reloadF,
reloadDst = reloadFreg',
renameSrc = renameF',
copyInstr = copyInstrF',
K = KF8,
getreg = FR8.getreg,
cellkind = I.C.FP,
dedicated = dedicatedF8,
spillProh = [],
memRegs = Float.fastMemRegs,
mode = phases(Float.fastPhases)
} : Ra.raClient
(* Two RA modes, fast and normal *)
val fast_fp = [RAInt, RAFP8]
val normal_fp = [RAInt, RAFP32]
(* The main ra routine *)
fun run cluster =
let val printGraph =
if !x86CfgDebugFlg then
PrintFlowGraph.printCluster(!MLRiscControl.debug_stream)
else fn msg => fn _ => ()
val _ = beforeRA()
val _ = resetRA()
(* generic register allocator *)
val cluster = Ra.ra
(if !fast_floating_point then fast_fp else normal_fp)
cluster
val _ = removeDeadCode cluster
val _ = printGraph "\t---After register allocation K=8---\n" cluster
(* Run the FP translation phase when fast floating point has
* been enabled
*)
val cluster =
if !fast_floating_point andalso I.C.numCell I.C.FP () > 0 then
let val cluster = X86FP.run cluster
in printGraph "\t---After X86 FP translation ---\n" cluster;
cluster
end
else cluster
in cluster
end
end
|