1 //===--------- PPCPreEmitPeephole.cpp - Late peephole optimizations -------===//
2 //
3 // Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
4 // See https://llvm.org/LICENSE.txt for license information.
5 // SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
6 //
7 //===----------------------------------------------------------------------===//
8 //
9 // A pre-emit peephole for catching opportunities introduced by late passes such
10 // as MachineBlockPlacement.
11 //
12 //===----------------------------------------------------------------------===//
13
14 #include "PPC.h"
15 #include "PPCInstrInfo.h"
16 #include "PPCSubtarget.h"
17 #include "llvm/ADT/Statistic.h"
18 #include "llvm/CodeGen/LivePhysRegs.h"
19 #include "llvm/CodeGen/MachineBasicBlock.h"
20 #include "llvm/CodeGen/MachineFunctionPass.h"
21 #include "llvm/CodeGen/MachineInstrBuilder.h"
22 #include "llvm/CodeGen/RegisterScavenging.h"
23 #include "llvm/MC/MCContext.h"
24 #include "llvm/Support/CommandLine.h"
25 #include "llvm/Support/Debug.h"
26
27 using namespace llvm;
28
29 #define DEBUG_TYPE "ppc-pre-emit-peephole"
30
31 STATISTIC(NumRRConvertedInPreEmit,
32 "Number of r+r instructions converted to r+i in pre-emit peephole");
33 STATISTIC(NumRemovedInPreEmit,
34 "Number of instructions deleted in pre-emit peephole");
35 STATISTIC(NumberOfSelfCopies,
36 "Number of self copy instructions eliminated");
37 STATISTIC(NumFrameOffFoldInPreEmit,
38 "Number of folding frame offset by using r+r in pre-emit peephole");
39 STATISTIC(NumCmpsInPreEmit,
40 "Number of compares eliminated in pre-emit peephole");
41
42 static cl::opt<bool>
43 EnablePCRelLinkerOpt("ppc-pcrel-linker-opt", cl::Hidden, cl::init(true),
44 cl::desc("enable PC Relative linker optimization"));
45
46 static cl::opt<bool>
47 RunPreEmitPeephole("ppc-late-peephole", cl::Hidden, cl::init(true),
48 cl::desc("Run pre-emit peephole optimizations."));
49
50 static cl::opt<uint64_t>
51 DSCRValue("ppc-set-dscr", cl::Hidden,
52 cl::desc("Set the Data Stream Control Register."));
53
54 namespace {
55
hasPCRelativeForm(MachineInstr & Use)56 static bool hasPCRelativeForm(MachineInstr &Use) {
57 switch (Use.getOpcode()) {
58 default:
59 return false;
60 case PPC::LBZ:
61 case PPC::LBZ8:
62 case PPC::LHA:
63 case PPC::LHA8:
64 case PPC::LHZ:
65 case PPC::LHZ8:
66 case PPC::LWZ:
67 case PPC::LWZ8:
68 case PPC::STB:
69 case PPC::STB8:
70 case PPC::STH:
71 case PPC::STH8:
72 case PPC::STW:
73 case PPC::STW8:
74 case PPC::LD:
75 case PPC::STD:
76 case PPC::LWA:
77 case PPC::LXSD:
78 case PPC::LXSSP:
79 case PPC::LXV:
80 case PPC::STXSD:
81 case PPC::STXSSP:
82 case PPC::STXV:
83 case PPC::LFD:
84 case PPC::LFS:
85 case PPC::STFD:
86 case PPC::STFS:
87 case PPC::DFLOADf32:
88 case PPC::DFLOADf64:
89 case PPC::DFSTOREf32:
90 case PPC::DFSTOREf64:
91 return true;
92 }
93 }
94
95 class PPCPreEmitPeephole : public MachineFunctionPass {
96 public:
97 static char ID;
PPCPreEmitPeephole()98 PPCPreEmitPeephole() : MachineFunctionPass(ID) {}
99
getAnalysisUsage(AnalysisUsage & AU) const100 void getAnalysisUsage(AnalysisUsage &AU) const override {
101 MachineFunctionPass::getAnalysisUsage(AU);
102 }
103
getRequiredProperties() const104 MachineFunctionProperties getRequiredProperties() const override {
105 return MachineFunctionProperties().setNoVRegs();
106 }
107
108 // This function removes any redundant load immediates. It has two level
109 // loops - The outer loop finds the load immediates BBI that could be used
110 // to replace following redundancy. The inner loop scans instructions that
111 // after BBI to find redundancy and update kill/dead flags accordingly. If
112 // AfterBBI is the same as BBI, it is redundant, otherwise any instructions
113 // that modify the def register of BBI would break the scanning.
114 // DeadOrKillToUnset is a pointer to the previous operand that had the
115 // kill/dead flag set. It keeps track of the def register of BBI, the use
116 // registers of AfterBBIs and the def registers of AfterBBIs.
removeRedundantLIs(MachineBasicBlock & MBB,const TargetRegisterInfo * TRI)117 bool removeRedundantLIs(MachineBasicBlock &MBB,
118 const TargetRegisterInfo *TRI) {
119 LLVM_DEBUG(dbgs() << "Remove redundant load immediates from MBB:\n";
120 MBB.dump(); dbgs() << "\n");
121
122 DenseSet<MachineInstr *> InstrsToErase;
123 for (auto BBI = MBB.instr_begin(); BBI != MBB.instr_end(); ++BBI) {
124 // Skip load immediate that is marked to be erased later because it
125 // cannot be used to replace any other instructions.
126 if (InstrsToErase.contains(&*BBI))
127 continue;
128 // Skip non-load immediate.
129 unsigned Opc = BBI->getOpcode();
130 if (Opc != PPC::LI && Opc != PPC::LI8 && Opc != PPC::LIS &&
131 Opc != PPC::LIS8)
132 continue;
133 // Skip load immediate, where the operand is a relocation (e.g., $r3 =
134 // LI target-flags(ppc-lo) %const.0).
135 if (!BBI->getOperand(1).isImm())
136 continue;
137 assert(BBI->getOperand(0).isReg() &&
138 "Expected a register for the first operand");
139
140 LLVM_DEBUG(dbgs() << "Scanning after load immediate: "; BBI->dump(););
141
142 Register Reg = BBI->getOperand(0).getReg();
143 int64_t Imm = BBI->getOperand(1).getImm();
144 MachineOperand *DeadOrKillToUnset = nullptr;
145 if (BBI->getOperand(0).isDead()) {
146 DeadOrKillToUnset = &BBI->getOperand(0);
147 LLVM_DEBUG(dbgs() << " Kill flag of " << *DeadOrKillToUnset
148 << " from load immediate " << *BBI
149 << " is a unsetting candidate\n");
150 }
151 // This loop scans instructions after BBI to see if there is any
152 // redundant load immediate.
153 for (auto AfterBBI = std::next(BBI); AfterBBI != MBB.instr_end();
154 ++AfterBBI) {
155 // Track the operand that kill Reg. We would unset the kill flag of
156 // the operand if there is a following redundant load immediate.
157 int KillIdx = AfterBBI->findRegisterUseOperandIdx(Reg, TRI, true);
158
159 // We can't just clear implicit kills, so if we encounter one, stop
160 // looking further.
161 if (KillIdx != -1 && AfterBBI->getOperand(KillIdx).isImplicit()) {
162 LLVM_DEBUG(dbgs()
163 << "Encountered an implicit kill, cannot proceed: ");
164 LLVM_DEBUG(AfterBBI->dump());
165 break;
166 }
167
168 if (KillIdx != -1) {
169 assert(!DeadOrKillToUnset && "Shouldn't kill same register twice");
170 DeadOrKillToUnset = &AfterBBI->getOperand(KillIdx);
171 LLVM_DEBUG(dbgs()
172 << " Kill flag of " << *DeadOrKillToUnset << " from "
173 << *AfterBBI << " is a unsetting candidate\n");
174 }
175
176 if (!AfterBBI->modifiesRegister(Reg, TRI))
177 continue;
178 // Finish scanning because Reg is overwritten by a non-load
179 // instruction.
180 if (AfterBBI->getOpcode() != Opc)
181 break;
182 assert(AfterBBI->getOperand(0).isReg() &&
183 "Expected a register for the first operand");
184 // Finish scanning because Reg is overwritten by a relocation or a
185 // different value.
186 if (!AfterBBI->getOperand(1).isImm() ||
187 AfterBBI->getOperand(1).getImm() != Imm)
188 break;
189
190 // It loads same immediate value to the same Reg, which is redundant.
191 // We would unset kill flag in previous Reg usage to extend live range
192 // of Reg first, then remove the redundancy.
193 if (DeadOrKillToUnset) {
194 LLVM_DEBUG(dbgs()
195 << " Unset dead/kill flag of " << *DeadOrKillToUnset
196 << " from " << *DeadOrKillToUnset->getParent());
197 if (DeadOrKillToUnset->isDef())
198 DeadOrKillToUnset->setIsDead(false);
199 else
200 DeadOrKillToUnset->setIsKill(false);
201 }
202 DeadOrKillToUnset =
203 AfterBBI->findRegisterDefOperand(Reg, TRI, true, true);
204 if (DeadOrKillToUnset)
205 LLVM_DEBUG(dbgs()
206 << " Dead flag of " << *DeadOrKillToUnset << " from "
207 << *AfterBBI << " is a unsetting candidate\n");
208 InstrsToErase.insert(&*AfterBBI);
209 LLVM_DEBUG(dbgs() << " Remove redundant load immediate: ";
210 AfterBBI->dump());
211 }
212 }
213
214 for (MachineInstr *MI : InstrsToErase) {
215 MI->eraseFromParent();
216 }
217 NumRemovedInPreEmit += InstrsToErase.size();
218 return !InstrsToErase.empty();
219 }
220
221 // Check if this instruction is a PLDpc that is part of a GOT indirect
222 // access.
isGOTPLDpc(MachineInstr & Instr)223 bool isGOTPLDpc(MachineInstr &Instr) {
224 if (Instr.getOpcode() != PPC::PLDpc)
225 return false;
226
227 // The result must be a register.
228 const MachineOperand &LoadedAddressReg = Instr.getOperand(0);
229 if (!LoadedAddressReg.isReg())
230 return false;
231
232 // Make sure that this is a global symbol.
233 const MachineOperand &SymbolOp = Instr.getOperand(1);
234 if (!SymbolOp.isGlobal())
235 return false;
236
237 // Finally return true only if the GOT flag is present.
238 return PPCInstrInfo::hasGOTFlag(SymbolOp.getTargetFlags());
239 }
240
addLinkerOpt(MachineBasicBlock & MBB,const TargetRegisterInfo * TRI)241 bool addLinkerOpt(MachineBasicBlock &MBB, const TargetRegisterInfo *TRI) {
242 MachineFunction *MF = MBB.getParent();
243 // If the linker opt is disabled then just return.
244 if (!EnablePCRelLinkerOpt)
245 return false;
246
247 // Add this linker opt only if we are using PC Relative memops.
248 if (!MF->getSubtarget<PPCSubtarget>().isUsingPCRelativeCalls())
249 return false;
250
251 // Struct to keep track of one def/use pair for a GOT indirect access.
252 struct GOTDefUsePair {
253 MachineBasicBlock::iterator DefInst;
254 MachineBasicBlock::iterator UseInst;
255 Register DefReg;
256 Register UseReg;
257 bool StillValid;
258 };
259 // Vector of def/ues pairs in this basic block.
260 SmallVector<GOTDefUsePair, 4> CandPairs;
261 SmallVector<GOTDefUsePair, 4> ValidPairs;
262 bool MadeChange = false;
263
264 // Run through all of the instructions in the basic block and try to
265 // collect potential pairs of GOT indirect access instructions.
266 for (auto BBI = MBB.instr_begin(); BBI != MBB.instr_end(); ++BBI) {
267 // Look for the initial GOT indirect load.
268 if (isGOTPLDpc(*BBI)) {
269 GOTDefUsePair CurrentPair{BBI, MachineBasicBlock::iterator(),
270 BBI->getOperand(0).getReg(),
271 PPC::NoRegister, true};
272 CandPairs.push_back(CurrentPair);
273 continue;
274 }
275
276 // We haven't encountered any new PLD instructions, nothing to check.
277 if (CandPairs.empty())
278 continue;
279
280 // Run through the candidate pairs and see if any of the registers
281 // defined in the PLD instructions are used by this instruction.
282 // Note: the size of CandPairs can change in the loop.
283 for (unsigned Idx = 0; Idx < CandPairs.size(); Idx++) {
284 GOTDefUsePair &Pair = CandPairs[Idx];
285 // The instruction does not use or modify this PLD's def reg,
286 // ignore it.
287 if (!BBI->readsRegister(Pair.DefReg, TRI) &&
288 !BBI->modifiesRegister(Pair.DefReg, TRI))
289 continue;
290
291 // The use needs to be used in the address computation and not
292 // as the register being stored for a store.
293 const MachineOperand *UseOp =
294 hasPCRelativeForm(*BBI) ? &BBI->getOperand(2) : nullptr;
295
296 // Check for a valid use.
297 if (UseOp && UseOp->isReg() && UseOp->getReg() == Pair.DefReg &&
298 UseOp->isUse() && UseOp->isKill()) {
299 Pair.UseInst = BBI;
300 Pair.UseReg = BBI->getOperand(0).getReg();
301 ValidPairs.push_back(Pair);
302 }
303 CandPairs.erase(CandPairs.begin() + Idx);
304 }
305 }
306
307 // Go through all of the pairs and check for any more valid uses.
308 for (auto Pair = ValidPairs.begin(); Pair != ValidPairs.end(); Pair++) {
309 // We shouldn't be here if we don't have a valid pair.
310 assert(Pair->UseInst.isValid() && Pair->StillValid &&
311 "Kept an invalid def/use pair for GOT PCRel opt");
312 // We have found a potential pair. Search through the instructions
313 // between the def and the use to see if it is valid to mark this as a
314 // linker opt.
315 MachineBasicBlock::iterator BBI = Pair->DefInst;
316 ++BBI;
317 for (; BBI != Pair->UseInst; ++BBI) {
318 if (BBI->readsRegister(Pair->UseReg, TRI) ||
319 BBI->modifiesRegister(Pair->UseReg, TRI)) {
320 Pair->StillValid = false;
321 break;
322 }
323 }
324
325 if (!Pair->StillValid)
326 continue;
327
328 // The load/store instruction that uses the address from the PLD will
329 // either use a register (for a store) or define a register (for the
330 // load). That register will be added as an implicit def to the PLD
331 // and as an implicit use on the second memory op. This is a precaution
332 // to prevent future passes from using that register between the two
333 // instructions.
334 MachineOperand ImplDef =
335 MachineOperand::CreateReg(Pair->UseReg, true, true);
336 MachineOperand ImplUse =
337 MachineOperand::CreateReg(Pair->UseReg, false, true);
338 Pair->DefInst->addOperand(ImplDef);
339 Pair->UseInst->addOperand(ImplUse);
340
341 // Create the symbol.
342 MCContext &Context = MF->getContext();
343 MCSymbol *Symbol = Context.createNamedTempSymbol("pcrel");
344 MachineOperand PCRelLabel =
345 MachineOperand::CreateMCSymbol(Symbol, PPCII::MO_PCREL_OPT_FLAG);
346 Pair->DefInst->addOperand(*MF, PCRelLabel);
347 Pair->UseInst->addOperand(*MF, PCRelLabel);
348 MadeChange |= true;
349 }
350 return MadeChange;
351 }
352
353 // This function removes redundant pairs of accumulator prime/unprime
354 // instructions. In some situations, it's possible the compiler inserts an
355 // accumulator prime instruction followed by an unprime instruction (e.g.
356 // when we store an accumulator after restoring it from a spill). If the
357 // accumulator is not used between the two, they can be removed. This
358 // function removes these redundant pairs from basic blocks.
359 // The algorithm is quite straightforward - every time we encounter a prime
360 // instruction, the primed register is added to a candidate set. Any use
361 // other than a prime removes the candidate from the set and any de-prime
362 // of a current candidate marks both the prime and de-prime for removal.
363 // This way we ensure we only remove prime/de-prime *pairs* with no
364 // intervening uses.
removeAccPrimeUnprime(MachineBasicBlock & MBB)365 bool removeAccPrimeUnprime(MachineBasicBlock &MBB) {
366 DenseSet<MachineInstr *> InstrsToErase;
367 // Initially, none of the acc registers are candidates.
368 SmallVector<MachineInstr *, 8> Candidates(
369 PPC::UACCRCRegClass.getNumRegs(), nullptr);
370
371 for (MachineInstr &BBI : MBB.instrs()) {
372 unsigned Opc = BBI.getOpcode();
373 // If we are visiting a xxmtacc instruction, we add it and its operand
374 // register to the candidate set.
375 if (Opc == PPC::XXMTACC) {
376 Register Acc = BBI.getOperand(0).getReg();
377 assert(PPC::ACCRCRegClass.contains(Acc) &&
378 "Unexpected register for XXMTACC");
379 Candidates[Acc - PPC::ACC0] = &BBI;
380 }
381 // If we are visiting a xxmfacc instruction and its operand register is
382 // in the candidate set, we mark the two instructions for removal.
383 else if (Opc == PPC::XXMFACC) {
384 Register Acc = BBI.getOperand(0).getReg();
385 assert(PPC::ACCRCRegClass.contains(Acc) &&
386 "Unexpected register for XXMFACC");
387 if (!Candidates[Acc - PPC::ACC0])
388 continue;
389 InstrsToErase.insert(&BBI);
390 InstrsToErase.insert(Candidates[Acc - PPC::ACC0]);
391 }
392 // If we are visiting an instruction using an accumulator register
393 // as operand, we remove it from the candidate set.
394 else {
395 for (MachineOperand &Operand : BBI.operands()) {
396 if (!Operand.isReg())
397 continue;
398 Register Reg = Operand.getReg();
399 if (PPC::ACCRCRegClass.contains(Reg))
400 Candidates[Reg - PPC::ACC0] = nullptr;
401 }
402 }
403 }
404
405 for (MachineInstr *MI : InstrsToErase)
406 MI->eraseFromParent();
407 NumRemovedInPreEmit += InstrsToErase.size();
408 return !InstrsToErase.empty();
409 }
410
runOnMachineFunction(MachineFunction & MF)411 bool runOnMachineFunction(MachineFunction &MF) override {
412 // If the user wants to set the DSCR using command-line options,
413 // load in the specified value at the start of main.
414 if (DSCRValue.getNumOccurrences() > 0 && MF.getName() == "main" &&
415 MF.getFunction().hasExternalLinkage()) {
416 DSCRValue = (uint32_t)(DSCRValue & 0x01FFFFFF); // 25-bit DSCR mask
417 RegScavenger RS;
418 MachineBasicBlock &MBB = MF.front();
419 // Find an unused GPR according to register liveness
420 RS.enterBasicBlock(MBB);
421 unsigned InDSCR = RS.FindUnusedReg(&PPC::GPRCRegClass);
422 if (InDSCR) {
423 const PPCInstrInfo *TII =
424 MF.getSubtarget<PPCSubtarget>().getInstrInfo();
425 DebugLoc dl;
426 MachineBasicBlock::iterator IP = MBB.begin(); // Insert Point
427 // Copy the 32-bit DSCRValue integer into the GPR InDSCR using LIS and
428 // ORI, then move to DSCR. If the requested DSCR value is contained
429 // in a 16-bit signed number, we can emit a single `LI`, but the
430 // impact of saving one instruction in one function does not warrant
431 // any additional complexity in the logic here.
432 BuildMI(MBB, IP, dl, TII->get(PPC::LIS), InDSCR)
433 .addImm(DSCRValue >> 16);
434 BuildMI(MBB, IP, dl, TII->get(PPC::ORI), InDSCR)
435 .addReg(InDSCR)
436 .addImm(DSCRValue & 0xFFFF);
437 BuildMI(MBB, IP, dl, TII->get(PPC::MTUDSCR))
438 .addReg(InDSCR, RegState::Kill);
439 } else
440 errs() << "Warning: Ran out of registers - Unable to set DSCR as "
441 "requested";
442 }
443
444 if (skipFunction(MF.getFunction()) || !RunPreEmitPeephole) {
445 // Remove UNENCODED_NOP even when this pass is disabled.
446 // This needs to be done unconditionally so we don't emit zeros
447 // in the instruction stream.
448 SmallVector<MachineInstr *, 4> InstrsToErase;
449 for (MachineBasicBlock &MBB : MF)
450 for (MachineInstr &MI : MBB)
451 if (MI.getOpcode() == PPC::UNENCODED_NOP)
452 InstrsToErase.push_back(&MI);
453 for (MachineInstr *MI : InstrsToErase)
454 MI->eraseFromParent();
455 return false;
456 }
457 bool Changed = false;
458 const PPCInstrInfo *TII = MF.getSubtarget<PPCSubtarget>().getInstrInfo();
459 const TargetRegisterInfo *TRI = MF.getSubtarget().getRegisterInfo();
460 SmallVector<MachineInstr *, 4> InstrsToErase;
461 for (MachineBasicBlock &MBB : MF) {
462 Changed |= removeRedundantLIs(MBB, TRI);
463 Changed |= addLinkerOpt(MBB, TRI);
464 Changed |= removeAccPrimeUnprime(MBB);
465 for (MachineInstr &MI : MBB) {
466 unsigned Opc = MI.getOpcode();
467 if (Opc == PPC::UNENCODED_NOP) {
468 InstrsToErase.push_back(&MI);
469 continue;
470 }
471 // Detect self copies - these can result from running AADB.
472 if (PPCInstrInfo::isSameClassPhysRegCopy(Opc)) {
473 const MCInstrDesc &MCID = TII->get(Opc);
474 if (MCID.getNumOperands() == 3 &&
475 MI.getOperand(0).getReg() == MI.getOperand(1).getReg() &&
476 MI.getOperand(0).getReg() == MI.getOperand(2).getReg()) {
477 NumberOfSelfCopies++;
478 LLVM_DEBUG(dbgs() << "Deleting self-copy instruction: ");
479 LLVM_DEBUG(MI.dump());
480 InstrsToErase.push_back(&MI);
481 continue;
482 }
483 else if (MCID.getNumOperands() == 2 &&
484 MI.getOperand(0).getReg() == MI.getOperand(1).getReg()) {
485 NumberOfSelfCopies++;
486 LLVM_DEBUG(dbgs() << "Deleting self-copy instruction: ");
487 LLVM_DEBUG(MI.dump());
488 InstrsToErase.push_back(&MI);
489 continue;
490 }
491 }
492 MachineInstr *DefMIToErase = nullptr;
493 SmallSet<Register, 4> UpdatedRegs;
494 if (TII->convertToImmediateForm(MI, UpdatedRegs, &DefMIToErase)) {
495 Changed = true;
496 NumRRConvertedInPreEmit++;
497 LLVM_DEBUG(dbgs() << "Converted instruction to imm form: ");
498 LLVM_DEBUG(MI.dump());
499 if (DefMIToErase) {
500 InstrsToErase.push_back(DefMIToErase);
501 }
502 }
503 if (TII->foldFrameOffset(MI)) {
504 Changed = true;
505 NumFrameOffFoldInPreEmit++;
506 LLVM_DEBUG(dbgs() << "Frame offset folding by using index form: ");
507 LLVM_DEBUG(MI.dump());
508 }
509 if (TII->optimizeCmpPostRA(MI)) {
510 Changed = true;
511 NumCmpsInPreEmit++;
512 LLVM_DEBUG(dbgs() << "Optimize compare by using record form: ");
513 LLVM_DEBUG(MI.dump());
514 InstrsToErase.push_back(&MI);
515 }
516 }
517
518 // Eliminate conditional branch based on a constant CR bit by
519 // CRSET or CRUNSET. We eliminate the conditional branch or
520 // convert it into an unconditional branch. Also, if the CR bit
521 // is not used by other instructions, we eliminate CRSET as well.
522 auto I = MBB.getFirstInstrTerminator();
523 if (I == MBB.instr_end())
524 continue;
525 MachineInstr *Br = &*I;
526 if (Br->getOpcode() != PPC::BC && Br->getOpcode() != PPC::BCn)
527 continue;
528 MachineInstr *CRSetMI = nullptr;
529 Register CRBit = Br->getOperand(0).getReg();
530 unsigned CRReg = getCRFromCRBit(CRBit);
531 bool SeenUse = false;
532 MachineBasicBlock::reverse_iterator It = Br, Er = MBB.rend();
533 for (It++; It != Er; It++) {
534 if (It->modifiesRegister(CRBit, TRI)) {
535 if ((It->getOpcode() == PPC::CRUNSET ||
536 It->getOpcode() == PPC::CRSET) &&
537 It->getOperand(0).getReg() == CRBit)
538 CRSetMI = &*It;
539 break;
540 }
541 if (It->readsRegister(CRBit, TRI))
542 SeenUse = true;
543 }
544 if (!CRSetMI) continue;
545
546 unsigned CRSetOp = CRSetMI->getOpcode();
547 if ((Br->getOpcode() == PPC::BCn && CRSetOp == PPC::CRSET) ||
548 (Br->getOpcode() == PPC::BC && CRSetOp == PPC::CRUNSET)) {
549 // Remove this branch since it cannot be taken.
550 InstrsToErase.push_back(Br);
551 MBB.removeSuccessor(Br->getOperand(1).getMBB());
552 }
553 else {
554 // This conditional branch is always taken. So, remove all branches
555 // and insert an unconditional branch to the destination of this.
556 MachineBasicBlock::iterator It = Br, Er = MBB.end();
557 for (; It != Er; It++) {
558 if (It->isDebugInstr()) continue;
559 assert(It->isTerminator() && "Non-terminator after a terminator");
560 InstrsToErase.push_back(&*It);
561 }
562 if (!MBB.isLayoutSuccessor(Br->getOperand(1).getMBB())) {
563 ArrayRef<MachineOperand> NoCond;
564 TII->insertBranch(MBB, Br->getOperand(1).getMBB(), nullptr,
565 NoCond, Br->getDebugLoc());
566 }
567 for (auto &Succ : MBB.successors())
568 if (Succ != Br->getOperand(1).getMBB()) {
569 MBB.removeSuccessor(Succ);
570 break;
571 }
572 }
573
574 // If the CRBit is not used by another instruction, we can eliminate
575 // CRSET/CRUNSET instruction.
576 if (!SeenUse) {
577 // We need to check use of the CRBit in successors.
578 for (auto &SuccMBB : MBB.successors())
579 if (SuccMBB->isLiveIn(CRBit) || SuccMBB->isLiveIn(CRReg)) {
580 SeenUse = true;
581 break;
582 }
583 if (!SeenUse)
584 InstrsToErase.push_back(CRSetMI);
585 }
586 }
587 for (MachineInstr *MI : InstrsToErase) {
588 LLVM_DEBUG(dbgs() << "PPC pre-emit peephole: erasing instruction: ");
589 LLVM_DEBUG(MI->dump());
590 MI->eraseFromParent();
591 NumRemovedInPreEmit++;
592 }
593 return Changed;
594 }
595 };
596 }
597
598 INITIALIZE_PASS(PPCPreEmitPeephole, DEBUG_TYPE, "PowerPC Pre-Emit Peephole",
599 false, false)
600 char PPCPreEmitPeephole::ID = 0;
601
createPPCPreEmitPeepholePass()602 FunctionPass *llvm::createPPCPreEmitPeepholePass() {
603 return new PPCPreEmitPeephole();
604 }
605