1 //===---- MachineCombiner.cpp - Instcombining on SSA form machine code ----===// 2 // 3 // The LLVM Compiler Infrastructure 4 // 5 // This file is distributed under the University of Illinois Open Source 6 // License. See LICENSE.TXT for details. 7 // 8 //===----------------------------------------------------------------------===// 9 // 10 // The machine combiner pass uses machine trace metrics to ensure the combined 11 // instructions does not lengthen the critical path or the resource depth. 12 //===----------------------------------------------------------------------===// 13 #define DEBUG_TYPE "machine-combiner" 14 15 #include "llvm/ADT/Statistic.h" 16 #include "llvm/ADT/DenseMap.h" 17 #include "llvm/CodeGen/MachineDominators.h" 18 #include "llvm/CodeGen/MachineFunction.h" 19 #include "llvm/CodeGen/MachineFunctionPass.h" 20 #include "llvm/CodeGen/MachineInstrBuilder.h" 21 #include "llvm/CodeGen/MachineLoopInfo.h" 22 #include "llvm/CodeGen/MachineRegisterInfo.h" 23 #include "llvm/CodeGen/MachineTraceMetrics.h" 24 #include "llvm/CodeGen/Passes.h" 25 #include "llvm/CodeGen/TargetSchedule.h" 26 #include "llvm/Support/CommandLine.h" 27 #include "llvm/Support/Debug.h" 28 #include "llvm/Support/raw_ostream.h" 29 #include "llvm/Target/TargetInstrInfo.h" 30 #include "llvm/Target/TargetRegisterInfo.h" 31 #include "llvm/Target/TargetSubtargetInfo.h" 32 33 using namespace llvm; 34 35 STATISTIC(NumInstCombined, "Number of machineinst combined"); 36 37 namespace { 38 class MachineCombiner : public MachineFunctionPass { 39 const TargetInstrInfo *TII; 40 const TargetRegisterInfo *TRI; 41 const MCSchedModel *SchedModel; 42 MachineRegisterInfo *MRI; 43 MachineTraceMetrics *Traces; 44 MachineTraceMetrics::Ensemble *MinInstr; 45 46 TargetSchedModel TSchedModel; 47 48 /// OptSize - True if optimizing for code size. 49 bool OptSize; 50 51 public: 52 static char ID; 53 MachineCombiner() : MachineFunctionPass(ID) { 54 initializeMachineCombinerPass(*PassRegistry::getPassRegistry()); 55 } 56 void getAnalysisUsage(AnalysisUsage &AU) const override; 57 bool runOnMachineFunction(MachineFunction &MF) override; 58 const char *getPassName() const override { return "Machine InstCombiner"; } 59 60 private: 61 bool doSubstitute(unsigned NewSize, unsigned OldSize); 62 bool combineInstructions(MachineBasicBlock *); 63 MachineInstr *getOperandDef(const MachineOperand &MO); 64 unsigned getDepth(SmallVectorImpl<MachineInstr *> &InsInstrs, 65 DenseMap<unsigned, unsigned> &InstrIdxForVirtReg, 66 MachineTraceMetrics::Trace BlockTrace); 67 unsigned getLatency(MachineInstr *Root, MachineInstr *NewRoot, 68 MachineTraceMetrics::Trace BlockTrace); 69 bool 70 preservesCriticalPathLen(MachineBasicBlock *MBB, MachineInstr *Root, 71 MachineTraceMetrics::Trace BlockTrace, 72 SmallVectorImpl<MachineInstr *> &InsInstrs, 73 DenseMap<unsigned, unsigned> &InstrIdxForVirtReg); 74 bool preservesResourceLen(MachineBasicBlock *MBB, 75 MachineTraceMetrics::Trace BlockTrace, 76 SmallVectorImpl<MachineInstr *> &InsInstrs, 77 SmallVectorImpl<MachineInstr *> &DelInstrs); 78 void instr2instrSC(SmallVectorImpl<MachineInstr *> &Instrs, 79 SmallVectorImpl<const MCSchedClassDesc *> &InstrsSC); 80 }; 81 } 82 83 char MachineCombiner::ID = 0; 84 char &llvm::MachineCombinerID = MachineCombiner::ID; 85 86 INITIALIZE_PASS_BEGIN(MachineCombiner, "machine-combiner", 87 "Machine InstCombiner", false, false) 88 INITIALIZE_PASS_DEPENDENCY(MachineTraceMetrics) 89 INITIALIZE_PASS_END(MachineCombiner, "machine-combiner", "Machine InstCombiner", 90 false, false) 91 92 void MachineCombiner::getAnalysisUsage(AnalysisUsage &AU) const { 93 AU.setPreservesCFG(); 94 AU.addPreserved<MachineDominatorTree>(); 95 AU.addPreserved<MachineLoopInfo>(); 96 AU.addRequired<MachineTraceMetrics>(); 97 AU.addPreserved<MachineTraceMetrics>(); 98 MachineFunctionPass::getAnalysisUsage(AU); 99 } 100 101 MachineInstr *MachineCombiner::getOperandDef(const MachineOperand &MO) { 102 MachineInstr *DefInstr = nullptr; 103 // We need a virtual register definition. 104 if (MO.isReg() && TargetRegisterInfo::isVirtualRegister(MO.getReg())) 105 DefInstr = MRI->getUniqueVRegDef(MO.getReg()); 106 // PHI's have no depth etc. 107 if (DefInstr && DefInstr->isPHI()) 108 DefInstr = nullptr; 109 return DefInstr; 110 } 111 112 /// getDepth - Computes depth of instructions in vector \InsInstr. 113 /// 114 /// \param InsInstrs is a vector of machine instructions 115 /// \param InstrIdxForVirtReg is a dense map of virtual register to index 116 /// of defining machine instruction in \p InsInstrs 117 /// \param BlockTrace is a trace of machine instructions 118 /// 119 /// \returns Depth of last instruction in \InsInstrs ("NewRoot") 120 unsigned 121 MachineCombiner::getDepth(SmallVectorImpl<MachineInstr *> &InsInstrs, 122 DenseMap<unsigned, unsigned> &InstrIdxForVirtReg, 123 MachineTraceMetrics::Trace BlockTrace) { 124 125 SmallVector<unsigned, 16> InstrDepth; 126 assert(TSchedModel.hasInstrSchedModel() && "Missing machine model\n"); 127 128 // Foreach instruction in in the new sequence compute the depth based on the 129 // operands. Use the trace information when possible. For new operands which 130 // are tracked in the InstrIdxForVirtReg map depth is looked up in InstrDepth 131 for (auto *InstrPtr : InsInstrs) { // for each Use 132 unsigned IDepth = 0; 133 DEBUG(dbgs() << "NEW INSTR "; InstrPtr->dump(); dbgs() << "\n";); 134 for (unsigned i = 0, e = InstrPtr->getNumOperands(); i != e; ++i) { 135 const MachineOperand &MO = InstrPtr->getOperand(i); 136 // Check for virtual register operand. 137 if (!(MO.isReg() && TargetRegisterInfo::isVirtualRegister(MO.getReg()))) 138 continue; 139 if (!MO.isUse()) 140 continue; 141 unsigned DepthOp = 0; 142 unsigned LatencyOp = 0; 143 DenseMap<unsigned, unsigned>::iterator II = 144 InstrIdxForVirtReg.find(MO.getReg()); 145 if (II != InstrIdxForVirtReg.end()) { 146 // Operand is new virtual register not in trace 147 assert(II->second < InstrDepth.size() && "Bad Index"); 148 MachineInstr *DefInstr = InsInstrs[II->second]; 149 assert(DefInstr && 150 "There must be a definition for a new virtual register"); 151 DepthOp = InstrDepth[II->second]; 152 LatencyOp = TSchedModel.computeOperandLatency( 153 DefInstr, DefInstr->findRegisterDefOperandIdx(MO.getReg()), 154 InstrPtr, InstrPtr->findRegisterUseOperandIdx(MO.getReg())); 155 } else { 156 MachineInstr *DefInstr = getOperandDef(MO); 157 if (DefInstr) { 158 DepthOp = BlockTrace.getInstrCycles(DefInstr).Depth; 159 LatencyOp = TSchedModel.computeOperandLatency( 160 DefInstr, DefInstr->findRegisterDefOperandIdx(MO.getReg()), 161 InstrPtr, InstrPtr->findRegisterUseOperandIdx(MO.getReg())); 162 } 163 } 164 IDepth = std::max(IDepth, DepthOp + LatencyOp); 165 } 166 InstrDepth.push_back(IDepth); 167 } 168 unsigned NewRootIdx = InsInstrs.size() - 1; 169 return InstrDepth[NewRootIdx]; 170 } 171 172 /// getLatency - Computes instruction latency as max of latency of defined 173 /// operands 174 /// 175 /// \param Root is a machine instruction that could be replaced by NewRoot. 176 /// It is used to compute a more accurate latency information for NewRoot in 177 /// case there is a dependent instruction in the same trace (\p BlockTrace) 178 /// \param NewRoot is the instruction for which the latency is computed 179 /// \param BlockTrace is a trace of machine instructions 180 /// 181 /// \returns Latency of \p NewRoot 182 unsigned MachineCombiner::getLatency(MachineInstr *Root, MachineInstr *NewRoot, 183 MachineTraceMetrics::Trace BlockTrace) { 184 185 assert(TSchedModel.hasInstrSchedModel() && "Missing machine model\n"); 186 187 // Check each definition in NewRoot and compute the latency 188 unsigned NewRootLatency = 0; 189 190 for (unsigned i = 0, e = NewRoot->getNumOperands(); i != e; ++i) { 191 const MachineOperand &MO = NewRoot->getOperand(i); 192 // Check for virtual register operand. 193 if (!(MO.isReg() && TargetRegisterInfo::isVirtualRegister(MO.getReg()))) 194 continue; 195 if (!MO.isDef()) 196 continue; 197 // Get the first instruction that uses MO 198 MachineRegisterInfo::reg_iterator RI = MRI->reg_begin(MO.getReg()); 199 RI++; 200 MachineInstr *UseMO = RI->getParent(); 201 unsigned LatencyOp = 0; 202 if (UseMO && BlockTrace.isDepInTrace(Root, UseMO)) { 203 LatencyOp = TSchedModel.computeOperandLatency( 204 NewRoot, NewRoot->findRegisterDefOperandIdx(MO.getReg()), UseMO, 205 UseMO->findRegisterUseOperandIdx(MO.getReg())); 206 } else { 207 LatencyOp = TSchedModel.computeInstrLatency(NewRoot->getOpcode()); 208 } 209 NewRootLatency = std::max(NewRootLatency, LatencyOp); 210 } 211 return NewRootLatency; 212 } 213 214 /// preservesCriticalPathlen - True when the new instruction sequence does not 215 /// lengthen the critical path. The DAGCombine code sequence ends in MI 216 /// (Machine Instruction) Root. The new code sequence ends in MI NewRoot. A 217 /// necessary condition for the new sequence to replace the old sequence is that 218 /// is cannot lengthen the critical path. This is decided by the formula 219 /// (NewRootDepth + NewRootLatency) <= (RootDepth + RootLatency + RootSlack)). 220 /// The slack is the number of cycles Root can be delayed before the critical 221 /// patch becomes longer. 222 bool MachineCombiner::preservesCriticalPathLen( 223 MachineBasicBlock *MBB, MachineInstr *Root, 224 MachineTraceMetrics::Trace BlockTrace, 225 SmallVectorImpl<MachineInstr *> &InsInstrs, 226 DenseMap<unsigned, unsigned> &InstrIdxForVirtReg) { 227 228 assert(TSchedModel.hasInstrSchedModel() && "Missing machine model\n"); 229 // NewRoot is the last instruction in the \p InsInstrs vector 230 // Get depth and latency of NewRoot 231 unsigned NewRootIdx = InsInstrs.size() - 1; 232 MachineInstr *NewRoot = InsInstrs[NewRootIdx]; 233 unsigned NewRootDepth = getDepth(InsInstrs, InstrIdxForVirtReg, BlockTrace); 234 unsigned NewRootLatency = getLatency(Root, NewRoot, BlockTrace); 235 236 // Get depth, latency and slack of Root 237 unsigned RootDepth = BlockTrace.getInstrCycles(Root).Depth; 238 unsigned RootLatency = TSchedModel.computeInstrLatency(Root); 239 unsigned RootSlack = BlockTrace.getInstrSlack(Root); 240 241 DEBUG(dbgs() << "DEPENDENCE DATA FOR " << Root << "\n"; 242 dbgs() << " NewRootDepth: " << NewRootDepth 243 << " NewRootLatency: " << NewRootLatency << "\n"; 244 dbgs() << " RootDepth: " << RootDepth << " RootLatency: " << RootLatency 245 << " RootSlack: " << RootSlack << "\n"; 246 dbgs() << " NewRootDepth + NewRootLatency " 247 << NewRootDepth + NewRootLatency << "\n"; 248 dbgs() << " RootDepth + RootLatency + RootSlack " 249 << RootDepth + RootLatency + RootSlack << "\n";); 250 251 /// True when the new sequence does not lenghten the critical path. 252 return ((NewRootDepth + NewRootLatency) <= 253 (RootDepth + RootLatency + RootSlack)); 254 } 255 256 /// helper routine to convert instructions into SC 257 void MachineCombiner::instr2instrSC( 258 SmallVectorImpl<MachineInstr *> &Instrs, 259 SmallVectorImpl<const MCSchedClassDesc *> &InstrsSC) { 260 for (auto *InstrPtr : Instrs) { 261 unsigned Opc = InstrPtr->getOpcode(); 262 unsigned Idx = TII->get(Opc).getSchedClass(); 263 const MCSchedClassDesc *SC = SchedModel->getSchedClassDesc(Idx); 264 InstrsSC.push_back(SC); 265 } 266 } 267 /// preservesResourceLen - True when the new instructions do not increase 268 /// resource length 269 bool MachineCombiner::preservesResourceLen( 270 MachineBasicBlock *MBB, MachineTraceMetrics::Trace BlockTrace, 271 SmallVectorImpl<MachineInstr *> &InsInstrs, 272 SmallVectorImpl<MachineInstr *> &DelInstrs) { 273 274 // Compute current resource length 275 276 ArrayRef<const MachineBasicBlock *> MBBarr(MBB); 277 unsigned ResLenBeforeCombine = BlockTrace.getResourceLength(MBBarr); 278 279 // Deal with SC rather than Instructions. 280 SmallVector<const MCSchedClassDesc *, 16> InsInstrsSC; 281 SmallVector<const MCSchedClassDesc *, 16> DelInstrsSC; 282 283 instr2instrSC(InsInstrs, InsInstrsSC); 284 instr2instrSC(DelInstrs, DelInstrsSC); 285 286 ArrayRef<const MCSchedClassDesc *> MSCInsArr = makeArrayRef(InsInstrsSC); 287 ArrayRef<const MCSchedClassDesc *> MSCDelArr = makeArrayRef(DelInstrsSC); 288 289 // Compute new resource length 290 unsigned ResLenAfterCombine = 291 BlockTrace.getResourceLength(MBBarr, MSCInsArr, MSCDelArr); 292 293 DEBUG(dbgs() << "RESOURCE DATA: \n"; 294 dbgs() << " resource len before: " << ResLenBeforeCombine 295 << " after: " << ResLenAfterCombine << "\n";); 296 297 return ResLenAfterCombine <= ResLenBeforeCombine; 298 } 299 300 /// \returns true when new instruction sequence should be generated 301 /// independent if it lenghtens critical path or not 302 bool MachineCombiner::doSubstitute(unsigned NewSize, unsigned OldSize) { 303 if (OptSize && (NewSize < OldSize)) 304 return true; 305 if (!TSchedModel.hasInstrSchedModel()) 306 return true; 307 return false; 308 } 309 310 /// combineInstructions - substitute a slow code sequence with a faster one by 311 /// evaluating instruction combining pattern. 312 /// The prototype of such a pattern is MUl + ADD -> MADD. Performs instruction 313 /// combining based on machine trace metrics. Only combine a sequence of 314 /// instructions when this neither lengthens the critical path nor increases 315 /// resource pressure. When optimizing for codesize always combine when the new 316 /// sequence is shorter. 317 bool MachineCombiner::combineInstructions(MachineBasicBlock *MBB) { 318 bool Changed = false; 319 DEBUG(dbgs() << "Combining MBB " << MBB->getName() << "\n"); 320 321 auto BlockIter = MBB->begin(); 322 323 while (BlockIter != MBB->end()) { 324 auto &MI = *BlockIter++; 325 326 DEBUG(dbgs() << "INSTR "; MI.dump(); dbgs() << "\n";); 327 SmallVector<MachineCombinerPattern::MC_PATTERN, 16> Pattern; 328 // The motivating example is: 329 // 330 // MUL Other MUL_op1 MUL_op2 Other 331 // \ / \ | / 332 // ADD/SUB => MADD/MSUB 333 // (=Root) (=NewRoot) 334 335 // The DAGCombine code always replaced MUL + ADD/SUB by MADD. While this is 336 // usually beneficial for code size it unfortunately can hurt performance 337 // when the ADD is on the critical path, but the MUL is not. With the 338 // substitution the MUL becomes part of the critical path (in form of the 339 // MADD) and can lengthen it on architectures where the MADD latency is 340 // longer than the ADD latency. 341 // 342 // For each instruction we check if it can be the root of a combiner 343 // pattern. Then for each pattern the new code sequence in form of MI is 344 // generated and evaluated. When the efficiency criteria (don't lengthen 345 // critical path, don't use more resources) is met the new sequence gets 346 // hooked up into the basic block before the old sequence is removed. 347 // 348 // The algorithm does not try to evaluate all patterns and pick the best. 349 // This is only an artificial restriction though. In practice there is 350 // mostly one pattern and hasPattern() can order patterns based on an 351 // internal cost heuristic. 352 353 if (TII->hasPattern(MI, Pattern)) { 354 for (auto P : Pattern) { 355 SmallVector<MachineInstr *, 16> InsInstrs; 356 SmallVector<MachineInstr *, 16> DelInstrs; 357 DenseMap<unsigned, unsigned> InstrIdxForVirtReg; 358 if (!MinInstr) 359 MinInstr = Traces->getEnsemble(MachineTraceMetrics::TS_MinInstrCount); 360 MachineTraceMetrics::Trace BlockTrace = MinInstr->getTrace(MBB); 361 Traces->verifyAnalysis(); 362 TII->genAlternativeCodeSequence(MI, P, InsInstrs, DelInstrs, 363 InstrIdxForVirtReg); 364 // Found pattern, but did not generate alternative sequence. 365 // This can happen e.g. when an immediate could not be materialized 366 // in a single instruction. 367 if (!InsInstrs.size()) 368 continue; 369 // Substitute when we optimize for codesize and the new sequence has 370 // fewer instructions OR 371 // the new sequence neither lenghten the critical path nor increases 372 // resource pressure. 373 if (doSubstitute(InsInstrs.size(), DelInstrs.size()) || 374 (preservesCriticalPathLen(MBB, &MI, BlockTrace, InsInstrs, 375 InstrIdxForVirtReg) && 376 preservesResourceLen(MBB, BlockTrace, InsInstrs, DelInstrs))) { 377 for (auto *InstrPtr : InsInstrs) 378 MBB->insert((MachineBasicBlock::iterator) & MI, 379 (MachineInstr *)InstrPtr); 380 for (auto *InstrPtr : DelInstrs) 381 InstrPtr->eraseFromParent(); 382 383 Changed = true; 384 ++NumInstCombined; 385 386 Traces->invalidate(MBB); 387 Traces->verifyAnalysis(); 388 // Eagerly stop after the first pattern fired 389 break; 390 } else { 391 // Cleanup instructions of the alternative code sequence. There is no 392 // use for them. 393 for (auto *InstrPtr : InsInstrs) { 394 MachineFunction *MF = MBB->getParent(); 395 MF->DeleteMachineInstr((MachineInstr *)InstrPtr); 396 } 397 } 398 InstrIdxForVirtReg.clear(); 399 } 400 } 401 } 402 403 return Changed; 404 } 405 406 bool MachineCombiner::runOnMachineFunction(MachineFunction &MF) { 407 TII = MF.getTarget().getInstrInfo(); 408 TRI = MF.getTarget().getRegisterInfo(); 409 const TargetSubtargetInfo &STI = 410 MF.getTarget().getSubtarget<TargetSubtargetInfo>(); 411 SchedModel = STI.getSchedModel(); 412 TSchedModel.init(*SchedModel, &STI, TII); 413 MRI = &MF.getRegInfo(); 414 Traces = &getAnalysis<MachineTraceMetrics>(); 415 MinInstr = 0; 416 417 OptSize = MF.getFunction()->getAttributes().hasAttribute( 418 AttributeSet::FunctionIndex, Attribute::OptimizeForSize); 419 420 DEBUG(dbgs() << getPassName() << ": " << MF.getName() << '\n'); 421 if (!TII->useMachineCombiner()) { 422 DEBUG(dbgs() << " Skipping pass: Target does not support machine combiner\n"); 423 return false; 424 } 425 426 bool Changed = false; 427 428 // Try to combine instructions. 429 for (auto &MBB : MF) 430 Changed |= combineInstructions(&MBB); 431 432 return Changed; 433 } 434