Rong Xu | 6e34c49 | 2016-04-27 23:20:27 +0000 | [diff] [blame] | 1 | //===-- IndirectCallPromotion.cpp - Promote indirect calls to direct calls ===// |
| 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 | // This file implements the transformation that promotes indirect calls to |
| 11 | // conditional direct calls when the indirect-call value profile metadata is |
| 12 | // available. |
| 13 | // |
| 14 | //===----------------------------------------------------------------------===// |
| 15 | |
| 16 | #include "IndirectCallSiteVisitor.h" |
| 17 | #include "llvm/ADT/STLExtras.h" |
| 18 | #include "llvm/ADT/Statistic.h" |
| 19 | #include "llvm/ADT/Triple.h" |
| 20 | #include "llvm/Analysis/CFG.h" |
| 21 | #include "llvm/IR/CallSite.h" |
| 22 | #include "llvm/IR/DiagnosticInfo.h" |
| 23 | #include "llvm/IR/IRBuilder.h" |
| 24 | #include "llvm/IR/InstIterator.h" |
| 25 | #include "llvm/IR/InstVisitor.h" |
| 26 | #include "llvm/IR/Instructions.h" |
| 27 | #include "llvm/IR/IntrinsicInst.h" |
| 28 | #include "llvm/IR/MDBuilder.h" |
| 29 | #include "llvm/IR/Module.h" |
| 30 | #include "llvm/Pass.h" |
| 31 | #include "llvm/ProfileData/InstrProfReader.h" |
| 32 | #include "llvm/Support/Debug.h" |
| 33 | #include "llvm/Transforms/Instrumentation.h" |
Xinliang David Li | f3c7a35 | 2016-05-16 16:31:07 +0000 | [diff] [blame] | 34 | #include "llvm/Transforms/PGOInstrumentation.h" |
Rong Xu | 6e34c49 | 2016-04-27 23:20:27 +0000 | [diff] [blame] | 35 | #include "llvm/Transforms/Utils/BasicBlockUtils.h" |
| 36 | #include <string> |
| 37 | #include <utility> |
| 38 | #include <vector> |
| 39 | |
| 40 | using namespace llvm; |
| 41 | |
Xinliang David Li | 0b29330 | 2016-06-02 01:52:05 +0000 | [diff] [blame^] | 42 | #define DEBUG_TYPE "pgo-icall-prom" |
Rong Xu | 6e34c49 | 2016-04-27 23:20:27 +0000 | [diff] [blame] | 43 | |
| 44 | STATISTIC(NumOfPGOICallPromotion, "Number of indirect call promotions."); |
| 45 | STATISTIC(NumOfPGOICallsites, "Number of indirect call candidate sites."); |
| 46 | |
| 47 | // Command line option to disable indirect-call promotion with the default as |
| 48 | // false. This is for debug purpose. |
| 49 | static cl::opt<bool> DisableICP("disable-icp", cl::init(false), cl::Hidden, |
| 50 | cl::desc("Disable indirect call promotion")); |
| 51 | |
| 52 | // The minimum call count for the direct-call target to be considered as the |
| 53 | // promotion candidate. |
| 54 | static cl::opt<unsigned> |
| 55 | ICPCountThreshold("icp-count-threshold", cl::Hidden, cl::ZeroOrMore, |
| 56 | cl::init(1000), |
| 57 | cl::desc("The minimum count to the direct call target " |
| 58 | "for the promotion")); |
| 59 | |
| 60 | // The percent threshold for the direct-call target (this call site vs the |
| 61 | // total call count) for it to be considered as the promotion target. |
| 62 | static cl::opt<unsigned> |
| 63 | ICPPercentThreshold("icp-percent-threshold", cl::init(33), cl::Hidden, |
| 64 | cl::ZeroOrMore, |
| 65 | cl::desc("The percentage threshold for the promotion")); |
| 66 | |
| 67 | // Set the maximum number of targets to promote for a single indirect-call |
| 68 | // callsite. |
| 69 | static cl::opt<unsigned> |
| 70 | MaxNumPromotions("icp-max-prom", cl::init(2), cl::Hidden, cl::ZeroOrMore, |
| 71 | cl::desc("Max number of promotions for a single indirect " |
| 72 | "call callsite")); |
| 73 | |
| 74 | // Set the cutoff value for the promotion. If the value is other than 0, we |
| 75 | // stop the transformation once the total number of promotions equals the cutoff |
| 76 | // value. |
| 77 | // For debug use only. |
| 78 | static cl::opt<unsigned> |
| 79 | ICPCutOff("icp-cutoff", cl::init(0), cl::Hidden, cl::ZeroOrMore, |
| 80 | cl::desc("Max number of promotions for this compilaiton")); |
| 81 | |
| 82 | // If ICPCSSkip is non zero, the first ICPCSSkip callsites will be skipped. |
| 83 | // For debug use only. |
| 84 | static cl::opt<unsigned> |
| 85 | ICPCSSkip("icp-csskip", cl::init(0), cl::Hidden, cl::ZeroOrMore, |
| 86 | cl::desc("Skip Callsite up to this number for this compilaiton")); |
| 87 | |
| 88 | // Set if the pass is called in LTO optimization. The difference for LTO mode |
| 89 | // is the pass won't prefix the source module name to the internal linkage |
| 90 | // symbols. |
| 91 | static cl::opt<bool> ICPLTOMode("icp-lto", cl::init(false), cl::Hidden, |
| 92 | cl::desc("Run indirect-call promotion in LTO " |
| 93 | "mode")); |
| 94 | // If the option is set to true, only call instructions will be considered for |
| 95 | // transformation -- invoke instructions will be ignored. |
| 96 | static cl::opt<bool> |
| 97 | ICPCallOnly("icp-call-only", cl::init(false), cl::Hidden, |
| 98 | cl::desc("Run indirect-call promotion for call instructions " |
| 99 | "only")); |
| 100 | |
| 101 | // If the option is set to true, only invoke instructions will be considered for |
| 102 | // transformation -- call instructions will be ignored. |
| 103 | static cl::opt<bool> ICPInvokeOnly("icp-invoke-only", cl::init(false), |
| 104 | cl::Hidden, |
| 105 | cl::desc("Run indirect-call promotion for " |
| 106 | "invoke instruction only")); |
| 107 | |
| 108 | // Dump the function level IR if the transformation happened in this |
| 109 | // function. For debug use only. |
| 110 | static cl::opt<bool> |
| 111 | ICPDUMPAFTER("icp-dumpafter", cl::init(false), cl::Hidden, |
| 112 | cl::desc("Dump IR after transformation happens")); |
| 113 | |
| 114 | namespace { |
Xinliang David Li | 7261618 | 2016-05-15 01:04:24 +0000 | [diff] [blame] | 115 | class PGOIndirectCallPromotionLegacyPass : public ModulePass { |
Rong Xu | 6e34c49 | 2016-04-27 23:20:27 +0000 | [diff] [blame] | 116 | public: |
| 117 | static char ID; |
| 118 | |
Xinliang David Li | 7261618 | 2016-05-15 01:04:24 +0000 | [diff] [blame] | 119 | PGOIndirectCallPromotionLegacyPass(bool InLTO = false) |
| 120 | : ModulePass(ID), InLTO(InLTO) { |
| 121 | initializePGOIndirectCallPromotionLegacyPassPass( |
| 122 | *PassRegistry::getPassRegistry()); |
Rong Xu | 6e34c49 | 2016-04-27 23:20:27 +0000 | [diff] [blame] | 123 | } |
| 124 | |
| 125 | const char *getPassName() const override { |
| 126 | return "PGOIndirectCallPromotion"; |
| 127 | } |
| 128 | |
| 129 | private: |
| 130 | bool runOnModule(Module &M) override; |
| 131 | |
| 132 | // If this pass is called in LTO. We need to special handling the PGOFuncName |
| 133 | // for the static variables due to LTO's internalization. |
| 134 | bool InLTO; |
| 135 | }; |
| 136 | } // end anonymous namespace |
| 137 | |
Xinliang David Li | 7261618 | 2016-05-15 01:04:24 +0000 | [diff] [blame] | 138 | char PGOIndirectCallPromotionLegacyPass::ID = 0; |
| 139 | INITIALIZE_PASS(PGOIndirectCallPromotionLegacyPass, "pgo-icall-prom", |
Rong Xu | 6e34c49 | 2016-04-27 23:20:27 +0000 | [diff] [blame] | 140 | "Use PGO instrumentation profile to promote indirect calls to " |
| 141 | "direct calls.", |
| 142 | false, false) |
| 143 | |
Xinliang David Li | 7261618 | 2016-05-15 01:04:24 +0000 | [diff] [blame] | 144 | ModulePass *llvm::createPGOIndirectCallPromotionLegacyPass(bool InLTO) { |
| 145 | return new PGOIndirectCallPromotionLegacyPass(InLTO); |
Rong Xu | 6e34c49 | 2016-04-27 23:20:27 +0000 | [diff] [blame] | 146 | } |
| 147 | |
Benjamin Kramer | a65b610 | 2016-05-15 15:18:11 +0000 | [diff] [blame] | 148 | namespace { |
Rong Xu | 6e34c49 | 2016-04-27 23:20:27 +0000 | [diff] [blame] | 149 | // The class for main data structure to promote indirect calls to conditional |
| 150 | // direct calls. |
| 151 | class ICallPromotionFunc { |
| 152 | private: |
| 153 | Function &F; |
| 154 | Module *M; |
| 155 | |
| 156 | // Symtab that maps indirect call profile values to function names and |
| 157 | // defines. |
| 158 | InstrProfSymtab *Symtab; |
| 159 | |
| 160 | // Allocate space to read the profile annotation. |
| 161 | std::unique_ptr<InstrProfValueData[]> ValueDataArray; |
| 162 | |
| 163 | // Count is the call count for the direct-call target and |
| 164 | // TotalCount is the call count for the indirect-call callsite. |
| 165 | // Return true we should promote this indirect-call target. |
| 166 | bool isPromotionProfitable(uint64_t Count, uint64_t TotalCount); |
| 167 | |
| 168 | enum TargetStatus { |
| 169 | OK, // Should be able to promote. |
| 170 | NotAvailableInModule, // Cannot find the target in current module. |
| 171 | ReturnTypeMismatch, // Return type mismatch b/w target and indirect-call. |
| 172 | NumArgsMismatch, // Number of arguments does not match. |
| 173 | ArgTypeMismatch // Type mismatch in the arguments (cannot bitcast). |
| 174 | }; |
| 175 | |
| 176 | // Test if we can legally promote this direct-call of Target. |
| 177 | TargetStatus isPromotionLegal(Instruction *Inst, uint64_t Target, |
| 178 | Function *&F); |
| 179 | |
| 180 | // A struct that records the direct target and it's call count. |
| 181 | struct PromotionCandidate { |
| 182 | Function *TargetFunction; |
| 183 | uint64_t Count; |
| 184 | PromotionCandidate(Function *F, uint64_t C) : TargetFunction(F), Count(C) {} |
| 185 | }; |
| 186 | |
| 187 | // Check if the indirect-call call site should be promoted. Return the number |
| 188 | // of promotions. |
| 189 | std::vector<PromotionCandidate> getPromotionCandidatesForCallSite( |
| 190 | Instruction *Inst, const ArrayRef<InstrProfValueData> &ValueDataRef, |
| 191 | uint64_t TotalCount); |
| 192 | |
| 193 | // Main function that transforms Inst (either a indirect-call instruction, or |
| 194 | // an invoke instruction , to a conditional call to F. This is like: |
| 195 | // if (Inst.CalledValue == F) |
| 196 | // F(...); |
| 197 | // else |
| 198 | // Inst(...); |
| 199 | // end |
| 200 | // TotalCount is the profile count value that the instruction executes. |
| 201 | // Count is the profile count value that F is the target function. |
| 202 | // These two values are being used to update the branch weight. |
| 203 | void promote(Instruction *Inst, Function *F, uint64_t Count, |
| 204 | uint64_t TotalCount); |
| 205 | |
| 206 | // Promote a list of targets for one indirect-call callsite. Return |
| 207 | // the number of promotions. |
| 208 | uint32_t tryToPromote(Instruction *Inst, |
| 209 | const std::vector<PromotionCandidate> &Candidates, |
| 210 | uint64_t &TotalCount); |
| 211 | |
| 212 | static const char *StatusToString(const TargetStatus S) { |
| 213 | switch (S) { |
| 214 | case OK: |
| 215 | return "OK to promote"; |
| 216 | case NotAvailableInModule: |
| 217 | return "Cannot find the target"; |
| 218 | case ReturnTypeMismatch: |
| 219 | return "Return type mismatch"; |
| 220 | case NumArgsMismatch: |
| 221 | return "The number of arguments mismatch"; |
| 222 | case ArgTypeMismatch: |
| 223 | return "Argument Type mismatch"; |
| 224 | } |
| 225 | llvm_unreachable("Should not reach here"); |
| 226 | } |
| 227 | |
| 228 | // Noncopyable |
| 229 | ICallPromotionFunc(const ICallPromotionFunc &other) = delete; |
| 230 | ICallPromotionFunc &operator=(const ICallPromotionFunc &other) = delete; |
| 231 | |
| 232 | public: |
| 233 | ICallPromotionFunc(Function &Func, Module *Modu, InstrProfSymtab *Symtab) |
| 234 | : F(Func), M(Modu), Symtab(Symtab) { |
| 235 | ValueDataArray = llvm::make_unique<InstrProfValueData[]>(MaxNumPromotions); |
| 236 | } |
| 237 | bool processFunction(); |
| 238 | }; |
Benjamin Kramer | a65b610 | 2016-05-15 15:18:11 +0000 | [diff] [blame] | 239 | } // end anonymous namespace |
Rong Xu | 6e34c49 | 2016-04-27 23:20:27 +0000 | [diff] [blame] | 240 | |
| 241 | bool ICallPromotionFunc::isPromotionProfitable(uint64_t Count, |
| 242 | uint64_t TotalCount) { |
| 243 | if (Count < ICPCountThreshold) |
| 244 | return false; |
| 245 | |
| 246 | unsigned Percentage = (Count * 100) / TotalCount; |
| 247 | return (Percentage >= ICPPercentThreshold); |
| 248 | } |
| 249 | |
| 250 | ICallPromotionFunc::TargetStatus |
| 251 | ICallPromotionFunc::isPromotionLegal(Instruction *Inst, uint64_t Target, |
| 252 | Function *&TargetFunction) { |
| 253 | Function *DirectCallee = Symtab->getFunction(Target); |
| 254 | if (DirectCallee == nullptr) |
| 255 | return NotAvailableInModule; |
| 256 | // Check the return type. |
| 257 | Type *CallRetType = Inst->getType(); |
| 258 | if (!CallRetType->isVoidTy()) { |
| 259 | Type *FuncRetType = DirectCallee->getReturnType(); |
| 260 | if (FuncRetType != CallRetType && |
| 261 | !CastInst::isBitCastable(FuncRetType, CallRetType)) |
| 262 | return ReturnTypeMismatch; |
| 263 | } |
| 264 | |
| 265 | // Check if the arguments are compatible with the parameters |
| 266 | FunctionType *DirectCalleeType = DirectCallee->getFunctionType(); |
| 267 | unsigned ParamNum = DirectCalleeType->getFunctionNumParams(); |
| 268 | CallSite CS(Inst); |
| 269 | unsigned ArgNum = CS.arg_size(); |
| 270 | |
| 271 | if (ParamNum != ArgNum && !DirectCalleeType->isVarArg()) |
| 272 | return NumArgsMismatch; |
| 273 | |
| 274 | for (unsigned I = 0; I < ParamNum; ++I) { |
| 275 | Type *PTy = DirectCalleeType->getFunctionParamType(I); |
| 276 | Type *ATy = CS.getArgument(I)->getType(); |
| 277 | if (PTy == ATy) |
| 278 | continue; |
| 279 | if (!CastInst::castIsValid(Instruction::BitCast, CS.getArgument(I), PTy)) |
| 280 | return ArgTypeMismatch; |
| 281 | } |
| 282 | |
| 283 | DEBUG(dbgs() << " #" << NumOfPGOICallPromotion << " Promote the icall to " |
| 284 | << Symtab->getFuncName(Target) << "\n"); |
| 285 | TargetFunction = DirectCallee; |
| 286 | return OK; |
| 287 | } |
| 288 | |
| 289 | // Indirect-call promotion heuristic. The direct targets are sorted based on |
| 290 | // the count. Stop at the first target that is not promoted. |
| 291 | std::vector<ICallPromotionFunc::PromotionCandidate> |
| 292 | ICallPromotionFunc::getPromotionCandidatesForCallSite( |
| 293 | Instruction *Inst, const ArrayRef<InstrProfValueData> &ValueDataRef, |
| 294 | uint64_t TotalCount) { |
| 295 | uint32_t NumVals = ValueDataRef.size(); |
| 296 | std::vector<PromotionCandidate> Ret; |
| 297 | |
| 298 | DEBUG(dbgs() << " \nWork on callsite #" << NumOfPGOICallsites << *Inst |
| 299 | << " Num_targets: " << NumVals << "\n"); |
| 300 | NumOfPGOICallsites++; |
| 301 | if (ICPCSSkip != 0 && NumOfPGOICallsites <= ICPCSSkip) { |
| 302 | DEBUG(dbgs() << " Skip: User options.\n"); |
| 303 | return Ret; |
| 304 | } |
| 305 | |
| 306 | for (uint32_t I = 0; I < MaxNumPromotions && I < NumVals; I++) { |
| 307 | uint64_t Count = ValueDataRef[I].Count; |
| 308 | assert(Count <= TotalCount); |
| 309 | uint64_t Target = ValueDataRef[I].Value; |
| 310 | DEBUG(dbgs() << " Candidate " << I << " Count=" << Count |
| 311 | << " Target_func: " << Target << "\n"); |
| 312 | |
| 313 | if (ICPInvokeOnly && dyn_cast<CallInst>(Inst)) { |
| 314 | DEBUG(dbgs() << " Not promote: User options.\n"); |
| 315 | break; |
| 316 | } |
| 317 | if (ICPCallOnly && dyn_cast<InvokeInst>(Inst)) { |
| 318 | DEBUG(dbgs() << " Not promote: User option.\n"); |
| 319 | break; |
| 320 | } |
| 321 | if (ICPCutOff != 0 && NumOfPGOICallPromotion >= ICPCutOff) { |
| 322 | DEBUG(dbgs() << " Not promote: Cutoff reached.\n"); |
| 323 | break; |
| 324 | } |
| 325 | if (!isPromotionProfitable(Count, TotalCount)) { |
| 326 | DEBUG(dbgs() << " Not promote: Cold target.\n"); |
| 327 | break; |
| 328 | } |
| 329 | Function *TargetFunction = nullptr; |
| 330 | TargetStatus Status = isPromotionLegal(Inst, Target, TargetFunction); |
| 331 | if (Status != OK) { |
| 332 | StringRef TargetFuncName = Symtab->getFuncName(Target); |
| 333 | const char *Reason = StatusToString(Status); |
| 334 | DEBUG(dbgs() << " Not promote: " << Reason << "\n"); |
Rong Xu | 62d5e47 | 2016-04-28 17:49:56 +0000 | [diff] [blame] | 335 | emitOptimizationRemarkMissed( |
Xinliang David Li | 0b29330 | 2016-06-02 01:52:05 +0000 | [diff] [blame^] | 336 | F.getContext(), "pgo-icall-prom", F, Inst->getDebugLoc(), |
Rong Xu | 6e34c49 | 2016-04-27 23:20:27 +0000 | [diff] [blame] | 337 | Twine("Cannot promote indirect call to ") + |
Rong Xu | 62d5e47 | 2016-04-28 17:49:56 +0000 | [diff] [blame] | 338 | (TargetFuncName.empty() ? Twine(Target) : Twine(TargetFuncName)) + |
| 339 | Twine(" with count of ") + Twine(Count) + ": " + Reason); |
Rong Xu | 6e34c49 | 2016-04-27 23:20:27 +0000 | [diff] [blame] | 340 | break; |
| 341 | } |
| 342 | Ret.push_back(PromotionCandidate(TargetFunction, Count)); |
| 343 | TotalCount -= Count; |
| 344 | } |
| 345 | return Ret; |
| 346 | } |
| 347 | |
| 348 | // Create a diamond structure for If_Then_Else. Also update the profile |
| 349 | // count. Do the fix-up for the invoke instruction. |
| 350 | static void createIfThenElse(Instruction *Inst, Function *DirectCallee, |
| 351 | uint64_t Count, uint64_t TotalCount, |
| 352 | BasicBlock **DirectCallBB, |
| 353 | BasicBlock **IndirectCallBB, |
| 354 | BasicBlock **MergeBB) { |
| 355 | CallSite CS(Inst); |
| 356 | Value *OrigCallee = CS.getCalledValue(); |
| 357 | |
| 358 | IRBuilder<> BBBuilder(Inst); |
| 359 | LLVMContext &Ctx = Inst->getContext(); |
| 360 | Value *BCI1 = |
| 361 | BBBuilder.CreateBitCast(OrigCallee, Type::getInt8PtrTy(Ctx), ""); |
| 362 | Value *BCI2 = |
| 363 | BBBuilder.CreateBitCast(DirectCallee, Type::getInt8PtrTy(Ctx), ""); |
| 364 | Value *PtrCmp = BBBuilder.CreateICmpEQ(BCI1, BCI2, ""); |
| 365 | |
| 366 | uint64_t ElseCount = TotalCount - Count; |
| 367 | uint64_t MaxCount = (Count >= ElseCount ? Count : ElseCount); |
| 368 | uint64_t Scale = calculateCountScale(MaxCount); |
| 369 | MDBuilder MDB(Inst->getContext()); |
| 370 | MDNode *BranchWeights = MDB.createBranchWeights( |
| 371 | scaleBranchCount(Count, Scale), scaleBranchCount(ElseCount, Scale)); |
| 372 | TerminatorInst *ThenTerm, *ElseTerm; |
| 373 | SplitBlockAndInsertIfThenElse(PtrCmp, Inst, &ThenTerm, &ElseTerm, |
| 374 | BranchWeights); |
| 375 | *DirectCallBB = ThenTerm->getParent(); |
| 376 | (*DirectCallBB)->setName("if.true.direct_targ"); |
| 377 | *IndirectCallBB = ElseTerm->getParent(); |
| 378 | (*IndirectCallBB)->setName("if.false.orig_indirect"); |
| 379 | *MergeBB = Inst->getParent(); |
| 380 | (*MergeBB)->setName("if.end.icp"); |
| 381 | |
| 382 | // Special handing of Invoke instructions. |
| 383 | InvokeInst *II = dyn_cast<InvokeInst>(Inst); |
| 384 | if (!II) |
| 385 | return; |
| 386 | |
| 387 | // We don't need branch instructions for invoke. |
| 388 | ThenTerm->eraseFromParent(); |
| 389 | ElseTerm->eraseFromParent(); |
| 390 | |
| 391 | // Add jump from Merge BB to the NormalDest. This is needed for the newly |
| 392 | // created direct invoke stmt -- as its NormalDst will be fixed up to MergeBB. |
| 393 | BranchInst::Create(II->getNormalDest(), *MergeBB); |
| 394 | } |
| 395 | |
| 396 | // Find the PHI in BB that have the CallResult as the operand. |
| 397 | static bool getCallRetPHINode(BasicBlock *BB, Instruction *Inst) { |
| 398 | BasicBlock *From = Inst->getParent(); |
| 399 | for (auto &I : *BB) { |
| 400 | PHINode *PHI = dyn_cast<PHINode>(&I); |
| 401 | if (!PHI) |
| 402 | continue; |
| 403 | int IX = PHI->getBasicBlockIndex(From); |
| 404 | if (IX == -1) |
| 405 | continue; |
| 406 | Value *V = PHI->getIncomingValue(IX); |
| 407 | if (dyn_cast<Instruction>(V) == Inst) |
| 408 | return true; |
| 409 | } |
| 410 | return false; |
| 411 | } |
| 412 | |
| 413 | // This method fixes up PHI nodes in BB where BB is the UnwindDest of an |
| 414 | // invoke instruction. In BB, there may be PHIs with incoming block being |
| 415 | // OrigBB (the MergeBB after if-then-else splitting). After moving the invoke |
| 416 | // instructions to its own BB, OrigBB is no longer the predecessor block of BB. |
| 417 | // Instead two new predecessors are added: IndirectCallBB and DirectCallBB, |
| 418 | // so the PHI node's incoming BBs need to be fixed up accordingly. |
| 419 | static void fixupPHINodeForUnwind(Instruction *Inst, BasicBlock *BB, |
| 420 | BasicBlock *OrigBB, |
| 421 | BasicBlock *IndirectCallBB, |
| 422 | BasicBlock *DirectCallBB) { |
| 423 | for (auto &I : *BB) { |
| 424 | PHINode *PHI = dyn_cast<PHINode>(&I); |
| 425 | if (!PHI) |
| 426 | continue; |
| 427 | int IX = PHI->getBasicBlockIndex(OrigBB); |
| 428 | if (IX == -1) |
| 429 | continue; |
| 430 | Value *V = PHI->getIncomingValue(IX); |
| 431 | PHI->addIncoming(V, IndirectCallBB); |
| 432 | PHI->setIncomingBlock(IX, DirectCallBB); |
| 433 | } |
| 434 | } |
| 435 | |
| 436 | // This method fixes up PHI nodes in BB where BB is the NormalDest of an |
| 437 | // invoke instruction. In BB, there may be PHIs with incoming block being |
| 438 | // OrigBB (the MergeBB after if-then-else splitting). After moving the invoke |
| 439 | // instructions to its own BB, a new incoming edge will be added to the original |
| 440 | // NormalDstBB from the IndirectCallBB. |
| 441 | static void fixupPHINodeForNormalDest(Instruction *Inst, BasicBlock *BB, |
| 442 | BasicBlock *OrigBB, |
| 443 | BasicBlock *IndirectCallBB, |
| 444 | Instruction *NewInst) { |
| 445 | for (auto &I : *BB) { |
| 446 | PHINode *PHI = dyn_cast<PHINode>(&I); |
| 447 | if (!PHI) |
| 448 | continue; |
| 449 | int IX = PHI->getBasicBlockIndex(OrigBB); |
| 450 | if (IX == -1) |
| 451 | continue; |
| 452 | Value *V = PHI->getIncomingValue(IX); |
| 453 | if (dyn_cast<Instruction>(V) == Inst) { |
| 454 | PHI->setIncomingBlock(IX, IndirectCallBB); |
| 455 | PHI->addIncoming(NewInst, OrigBB); |
| 456 | continue; |
| 457 | } |
| 458 | PHI->addIncoming(V, IndirectCallBB); |
| 459 | } |
| 460 | } |
| 461 | |
| 462 | // Add a bitcast instruction to the direct-call return value if needed. |
Rong Xu | 6e34c49 | 2016-04-27 23:20:27 +0000 | [diff] [blame] | 463 | static Instruction *insertCallRetCast(const Instruction *Inst, |
| 464 | Instruction *DirectCallInst, |
| 465 | Function *DirectCallee) { |
| 466 | if (Inst->getType()->isVoidTy()) |
| 467 | return DirectCallInst; |
| 468 | |
| 469 | Type *CallRetType = Inst->getType(); |
| 470 | Type *FuncRetType = DirectCallee->getReturnType(); |
| 471 | if (FuncRetType == CallRetType) |
| 472 | return DirectCallInst; |
| 473 | |
| 474 | BasicBlock *InsertionBB; |
| 475 | if (CallInst *CI = dyn_cast<CallInst>(DirectCallInst)) |
| 476 | InsertionBB = CI->getParent(); |
| 477 | else |
| 478 | InsertionBB = (dyn_cast<InvokeInst>(DirectCallInst))->getNormalDest(); |
| 479 | |
| 480 | return (new BitCastInst(DirectCallInst, CallRetType, "", |
| 481 | InsertionBB->getTerminator())); |
| 482 | } |
| 483 | |
| 484 | // Create a DirectCall instruction in the DirectCallBB. |
| 485 | // Parameter Inst is the indirect-call (invoke) instruction. |
| 486 | // DirectCallee is the decl of the direct-call (invoke) target. |
| 487 | // DirecallBB is the BB that the direct-call (invoke) instruction is inserted. |
| 488 | // MergeBB is the bottom BB of the if-then-else-diamond after the |
| 489 | // transformation. For invoke instruction, the edges from DirectCallBB and |
| 490 | // IndirectCallBB to MergeBB are removed before this call (during |
| 491 | // createIfThenElse). |
| 492 | static Instruction *createDirectCallInst(const Instruction *Inst, |
| 493 | Function *DirectCallee, |
| 494 | BasicBlock *DirectCallBB, |
| 495 | BasicBlock *MergeBB) { |
| 496 | Instruction *NewInst = Inst->clone(); |
| 497 | if (CallInst *CI = dyn_cast<CallInst>(NewInst)) { |
| 498 | CI->setCalledFunction(DirectCallee); |
| 499 | CI->mutateFunctionType(DirectCallee->getFunctionType()); |
| 500 | } else { |
| 501 | // Must be an invoke instruction. Direct invoke's normal destination is |
| 502 | // fixed up to MergeBB. MergeBB is the place where return cast is inserted. |
| 503 | // Also since IndirectCallBB does not have an edge to MergeBB, there is no |
| 504 | // need to insert new PHIs into MergeBB. |
| 505 | InvokeInst *II = dyn_cast<InvokeInst>(NewInst); |
| 506 | assert(II); |
| 507 | II->setCalledFunction(DirectCallee); |
| 508 | II->mutateFunctionType(DirectCallee->getFunctionType()); |
| 509 | II->setNormalDest(MergeBB); |
| 510 | } |
| 511 | |
| 512 | DirectCallBB->getInstList().insert(DirectCallBB->getFirstInsertionPt(), |
| 513 | NewInst); |
| 514 | |
| 515 | // Clear the value profile data. |
| 516 | NewInst->setMetadata(LLVMContext::MD_prof, 0); |
| 517 | CallSite NewCS(NewInst); |
| 518 | FunctionType *DirectCalleeType = DirectCallee->getFunctionType(); |
| 519 | unsigned ParamNum = DirectCalleeType->getFunctionNumParams(); |
| 520 | for (unsigned I = 0; I < ParamNum; ++I) { |
| 521 | Type *ATy = NewCS.getArgument(I)->getType(); |
| 522 | Type *PTy = DirectCalleeType->getParamType(I); |
| 523 | if (ATy != PTy) { |
| 524 | BitCastInst *BI = new BitCastInst(NewCS.getArgument(I), PTy, "", NewInst); |
| 525 | NewCS.setArgument(I, BI); |
| 526 | } |
| 527 | } |
| 528 | |
| 529 | return insertCallRetCast(Inst, NewInst, DirectCallee); |
| 530 | } |
| 531 | |
| 532 | // Create a PHI to unify the return values of calls. |
| 533 | static void insertCallRetPHI(Instruction *Inst, Instruction *CallResult, |
| 534 | Function *DirectCallee) { |
| 535 | if (Inst->getType()->isVoidTy()) |
| 536 | return; |
| 537 | |
| 538 | BasicBlock *RetValBB = CallResult->getParent(); |
| 539 | |
| 540 | BasicBlock *PHIBB; |
| 541 | if (InvokeInst *II = dyn_cast<InvokeInst>(CallResult)) |
| 542 | RetValBB = II->getNormalDest(); |
| 543 | |
| 544 | PHIBB = RetValBB->getSingleSuccessor(); |
| 545 | if (getCallRetPHINode(PHIBB, Inst)) |
| 546 | return; |
| 547 | |
| 548 | PHINode *CallRetPHI = PHINode::Create(Inst->getType(), 0); |
| 549 | PHIBB->getInstList().push_front(CallRetPHI); |
| 550 | Inst->replaceAllUsesWith(CallRetPHI); |
| 551 | CallRetPHI->addIncoming(Inst, Inst->getParent()); |
| 552 | CallRetPHI->addIncoming(CallResult, RetValBB); |
| 553 | } |
| 554 | |
| 555 | // This function does the actual indirect-call promotion transformation: |
| 556 | // For an indirect-call like: |
| 557 | // Ret = (*Foo)(Args); |
| 558 | // It transforms to: |
| 559 | // if (Foo == DirectCallee) |
| 560 | // Ret1 = DirectCallee(Args); |
| 561 | // else |
| 562 | // Ret2 = (*Foo)(Args); |
| 563 | // Ret = phi(Ret1, Ret2); |
| 564 | // It adds type casts for the args do not match the parameters and the return |
| 565 | // value. Branch weights metadata also updated. |
| 566 | void ICallPromotionFunc::promote(Instruction *Inst, Function *DirectCallee, |
| 567 | uint64_t Count, uint64_t TotalCount) { |
| 568 | assert(DirectCallee != nullptr); |
| 569 | BasicBlock *BB = Inst->getParent(); |
| 570 | // Just to suppress the non-debug build warning. |
| 571 | (void)BB; |
| 572 | DEBUG(dbgs() << "\n\n== Basic Block Before ==\n"); |
| 573 | DEBUG(dbgs() << *BB << "\n"); |
| 574 | |
| 575 | BasicBlock *DirectCallBB, *IndirectCallBB, *MergeBB; |
| 576 | createIfThenElse(Inst, DirectCallee, Count, TotalCount, &DirectCallBB, |
| 577 | &IndirectCallBB, &MergeBB); |
| 578 | |
| 579 | Instruction *NewInst = |
| 580 | createDirectCallInst(Inst, DirectCallee, DirectCallBB, MergeBB); |
| 581 | |
| 582 | // Move Inst from MergeBB to IndirectCallBB. |
| 583 | Inst->removeFromParent(); |
| 584 | IndirectCallBB->getInstList().insert(IndirectCallBB->getFirstInsertionPt(), |
| 585 | Inst); |
| 586 | |
| 587 | if (InvokeInst *II = dyn_cast<InvokeInst>(Inst)) { |
| 588 | // At this point, the original indirect invoke instruction has the original |
| 589 | // UnwindDest and NormalDest. For the direct invoke instruction, the |
| 590 | // NormalDest points to MergeBB, and MergeBB jumps to the original |
| 591 | // NormalDest. MergeBB might have a new bitcast instruction for the return |
| 592 | // value. The PHIs are with the original NormalDest. Since we now have two |
| 593 | // incoming edges to NormalDest and UnwindDest, we have to do some fixups. |
| 594 | // |
| 595 | // UnwindDest will not use the return value. So pass nullptr here. |
| 596 | fixupPHINodeForUnwind(Inst, II->getUnwindDest(), MergeBB, IndirectCallBB, |
| 597 | DirectCallBB); |
| 598 | // We don't need to update the operand from NormalDest for DirectCallBB. |
| 599 | // Pass nullptr here. |
| 600 | fixupPHINodeForNormalDest(Inst, II->getNormalDest(), MergeBB, |
| 601 | IndirectCallBB, NewInst); |
| 602 | } |
| 603 | |
| 604 | insertCallRetPHI(Inst, NewInst, DirectCallee); |
| 605 | |
| 606 | DEBUG(dbgs() << "\n== Basic Blocks After ==\n"); |
| 607 | DEBUG(dbgs() << *BB << *DirectCallBB << *IndirectCallBB << *MergeBB << "\n"); |
| 608 | |
Rong Xu | 62d5e47 | 2016-04-28 17:49:56 +0000 | [diff] [blame] | 609 | emitOptimizationRemark( |
Xinliang David Li | 0b29330 | 2016-06-02 01:52:05 +0000 | [diff] [blame^] | 610 | F.getContext(), "pgo-icall-prom", F, Inst->getDebugLoc(), |
Rong Xu | 62d5e47 | 2016-04-28 17:49:56 +0000 | [diff] [blame] | 611 | Twine("Promote indirect call to ") + DirectCallee->getName() + |
| 612 | " with count " + Twine(Count) + " out of " + Twine(TotalCount)); |
Rong Xu | 6e34c49 | 2016-04-27 23:20:27 +0000 | [diff] [blame] | 613 | } |
| 614 | |
| 615 | // Promote indirect-call to conditional direct-call for one callsite. |
| 616 | uint32_t ICallPromotionFunc::tryToPromote( |
| 617 | Instruction *Inst, const std::vector<PromotionCandidate> &Candidates, |
| 618 | uint64_t &TotalCount) { |
| 619 | uint32_t NumPromoted = 0; |
| 620 | |
| 621 | for (auto &C : Candidates) { |
| 622 | uint64_t Count = C.Count; |
| 623 | promote(Inst, C.TargetFunction, Count, TotalCount); |
| 624 | assert(TotalCount >= Count); |
| 625 | TotalCount -= Count; |
| 626 | NumOfPGOICallPromotion++; |
| 627 | NumPromoted++; |
| 628 | } |
| 629 | return NumPromoted; |
| 630 | } |
| 631 | |
| 632 | // Traverse all the indirect-call callsite and get the value profile |
| 633 | // annotation to perform indirect-call promotion. |
| 634 | bool ICallPromotionFunc::processFunction() { |
| 635 | bool Changed = false; |
| 636 | for (auto &I : findIndirectCallSites(F)) { |
| 637 | uint32_t NumVals; |
| 638 | uint64_t TotalCount; |
| 639 | bool Res = |
| 640 | getValueProfDataFromInst(*I, IPVK_IndirectCallTarget, MaxNumPromotions, |
| 641 | ValueDataArray.get(), NumVals, TotalCount); |
| 642 | if (!Res) |
| 643 | continue; |
| 644 | ArrayRef<InstrProfValueData> ValueDataArrayRef(ValueDataArray.get(), |
| 645 | NumVals); |
| 646 | auto PromotionCandidates = |
| 647 | getPromotionCandidatesForCallSite(I, ValueDataArrayRef, TotalCount); |
| 648 | uint32_t NumPromoted = tryToPromote(I, PromotionCandidates, TotalCount); |
| 649 | if (NumPromoted == 0) |
| 650 | continue; |
| 651 | |
| 652 | Changed = true; |
| 653 | // Adjust the MD.prof metadata. First delete the old one. |
| 654 | I->setMetadata(LLVMContext::MD_prof, 0); |
| 655 | // If all promoted, we don't need the MD.prof metadata. |
| 656 | if (TotalCount == 0 || NumPromoted == NumVals) |
| 657 | continue; |
| 658 | // Otherwise we need update with the un-promoted records back. |
| 659 | annotateValueSite(*M, *I, ValueDataArrayRef.slice(NumPromoted), TotalCount, |
| 660 | IPVK_IndirectCallTarget, MaxNumPromotions); |
| 661 | } |
| 662 | return Changed; |
| 663 | } |
| 664 | |
| 665 | // A wrapper function that does the actual work. |
| 666 | static bool promoteIndirectCalls(Module &M, bool InLTO) { |
| 667 | if (DisableICP) |
| 668 | return false; |
| 669 | InstrProfSymtab Symtab; |
| 670 | Symtab.create(M, InLTO); |
| 671 | bool Changed = false; |
| 672 | for (auto &F : M) { |
| 673 | if (F.isDeclaration()) |
| 674 | continue; |
| 675 | if (F.hasFnAttribute(Attribute::OptimizeNone)) |
| 676 | continue; |
| 677 | ICallPromotionFunc ICallPromotion(F, &M, &Symtab); |
| 678 | bool FuncChanged = ICallPromotion.processFunction(); |
| 679 | if (ICPDUMPAFTER && FuncChanged) { |
| 680 | DEBUG(dbgs() << "\n== IR Dump After =="; F.print(dbgs())); |
| 681 | DEBUG(dbgs() << "\n"); |
| 682 | } |
| 683 | Changed |= FuncChanged; |
| 684 | if (ICPCutOff != 0 && NumOfPGOICallPromotion >= ICPCutOff) { |
| 685 | DEBUG(dbgs() << " Stop: Cutoff reached.\n"); |
| 686 | break; |
| 687 | } |
| 688 | } |
| 689 | return Changed; |
| 690 | } |
| 691 | |
Xinliang David Li | 7261618 | 2016-05-15 01:04:24 +0000 | [diff] [blame] | 692 | bool PGOIndirectCallPromotionLegacyPass::runOnModule(Module &M) { |
Rong Xu | 6e34c49 | 2016-04-27 23:20:27 +0000 | [diff] [blame] | 693 | // Command-line option has the priority for InLTO. |
Xinliang David Li | 7d0fed7 | 2016-05-17 21:06:16 +0000 | [diff] [blame] | 694 | return promoteIndirectCalls(M, InLTO | ICPLTOMode); |
Rong Xu | 6e34c49 | 2016-04-27 23:20:27 +0000 | [diff] [blame] | 695 | } |
Xinliang David Li | f3c7a35 | 2016-05-16 16:31:07 +0000 | [diff] [blame] | 696 | |
| 697 | PreservedAnalyses PGOIndirectCallPromotion::run(Module &M, AnalysisManager<Module> &AM) { |
Xinliang David Li | 7d0fed7 | 2016-05-17 21:06:16 +0000 | [diff] [blame] | 698 | if (!promoteIndirectCalls(M, InLTO | ICPLTOMode)) |
Xinliang David Li | f3c7a35 | 2016-05-16 16:31:07 +0000 | [diff] [blame] | 699 | return PreservedAnalyses::all(); |
| 700 | |
| 701 | return PreservedAnalyses::none(); |
| 702 | } |