llvm/lib/Transforms/Scalar/PartiallyInlineLibCalls.cpp

   1 //===--- PartiallyInlineLibCalls.cpp - Partially inline libcalls ----------===//
   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 // This pass tries to partially inline the fast path of well-known library
  10 // functions, such as using square-root instructions for cases where sqrt()
  11 // does not need to set errno.
  12 //
  13 //===----------------------------------------------------------------------===//
  14
  15 #include "llvm/Transforms/Scalar/PartiallyInlineLibCalls.h"
  16 #include "llvm/Analysis/DomTreeUpdater.h"
  17 #include "llvm/Analysis/TargetLibraryInfo.h"
  18 #include "llvm/Analysis/TargetTransformInfo.h"
  19 #include "llvm/IR/Dominators.h"
  20 #include "llvm/IR/IRBuilder.h"
  21 #include "llvm/InitializePasses.h"
  22 #include "llvm/Support/DebugCounter.h"
  23 #include "llvm/Transforms/Scalar.h"
  24 #include "llvm/Transforms/Utils/BasicBlockUtils.h"
  25
  26 using namespace llvm;
  27
  28 #define DEBUG_TYPE "partially-inline-libcalls"
  29
  30 DEBUG_COUNTER(PILCounter, "partially-inline-libcalls-transform",
  31               "Controls transformations in partially-inline-libcalls");
  32
  33 static bool optimizeSQRT(CallInst *Call, Function *CalledFunc,
  34                          BasicBlock &CurrBB, Function::iterator &BB,
  35                          const TargetTransformInfo *TTI, DomTreeUpdater *DTU) {
  36   // There is no need to change the IR, since backend will emit sqrt
  37   // instruction if the call has already been marked read-only.
  38   if (Call->onlyReadsMemory())
  39     return false;
  40
  41   if (!DebugCounter::shouldExecute(PILCounter))
  42     return false;
  43
  44   // Do the following transformation:
  45   //
  46   // (before)
  47   // dst = sqrt(src)
  48   //
  49   // (after)
  50   // v0 = sqrt_noreadmem(src) # native sqrt instruction.
  51   // [if (v0 is a NaN) || if (src < 0)]
  52   //   v1 = sqrt(src)         # library call.
  53   // dst = phi(v0, v1)
  54   //
  55
  56   Type *Ty = Call->getType();
  57   IRBuilder<> Builder(Call->getNextNode());
  58
  59   // Split CurrBB right after the call, create a 'then' block (that branches
  60   // back to split-off tail of CurrBB) into which we'll insert a libcall.
  61   Instruction *LibCallTerm = SplitBlockAndInsertIfThen(
  62       Builder.getTrue(), Call->getNextNode(), /*Unreachable=*/false,
  63       /*BranchWeights*/ nullptr, DTU);
  64
  65   auto *CurrBBTerm = cast<BranchInst>(CurrBB.getTerminator());
  66   // We want an 'else' block though, not a 'then' block.
  67   cast<BranchInst>(CurrBBTerm)->swapSuccessors();
  68
  69   // Create phi that will merge results of either sqrt and replace all uses.
  70   BasicBlock *JoinBB = LibCallTerm->getSuccessor(0);
  71   JoinBB->setName(CurrBB.getName() + ".split");
  72   Builder.SetInsertPoint(JoinBB, JoinBB->begin());
  73   PHINode *Phi = Builder.CreatePHI(Ty, 2);
  74   Call->replaceAllUsesWith(Phi);
  75
  76   // Finally, insert the libcall into 'else' block.
  77   BasicBlock *LibCallBB = LibCallTerm->getParent();
  78   LibCallBB->setName("call.sqrt");
  79   Builder.SetInsertPoint(LibCallTerm);
  80   Instruction *LibCall = Call->clone();
  81   Builder.Insert(LibCall);
  82
  83   // Add attribute "readnone" so that backend can use a native sqrt instruction
  84   // for this call.
  85   Call->addFnAttr(Attribute::ReadNone);
  86
  87   // Insert a FP compare instruction and use it as the CurrBB branch condition.
  88   Builder.SetInsertPoint(CurrBBTerm);
  89   Value *FCmp = TTI->isFCmpOrdCheaperThanFCmpZero(Ty)
  90                     ? Builder.CreateFCmpORD(Call, Call)
  91                     : Builder.CreateFCmpOGE(Call->getOperand(0),
  92                                             ConstantFP::get(Ty, 0.0));
  93   CurrBBTerm->setCondition(FCmp);
  94
  95   // Add phi operands.
  96   Phi->addIncoming(Call, &CurrBB);
  97   Phi->addIncoming(LibCall, LibCallBB);
  98
  99   BB = JoinBB->getIterator();
 100   return true;
 101 }
 102
 103 static bool runPartiallyInlineLibCalls(Function &F, TargetLibraryInfo *TLI,
 104                                        const TargetTransformInfo *TTI,
 105                                        DominatorTree *DT) {
 106   Optional<DomTreeUpdater> DTU;
 107   if (DT)
 108     DTU.emplace(DT, DomTreeUpdater::UpdateStrategy::Lazy);
 109
 110   bool Changed = false;
 111
 112   Function::iterator CurrBB;
 113   for (Function::iterator BB = F.begin(), BE = F.end(); BB != BE;) {
 114     CurrBB = BB++;
 115
 116     for (BasicBlock::iterator II = CurrBB->begin(), IE = CurrBB->end();
 117          II != IE; ++II) {
 118       CallInst *Call = dyn_cast<CallInst>(&*II);
 119       Function *CalledFunc;
 120
 121       if (!Call || !(CalledFunc = Call->getCalledFunction()))
 122         continue;
 123
 124       if (Call->isNoBuiltin() || Call->isStrictFP())
 125         continue;
 126
 127       // Skip if function either has local linkage or is not a known library
 128       // function.
 129       LibFunc LF;
 130       if (CalledFunc->hasLocalLinkage() ||
 131           !TLI->getLibFunc(*CalledFunc, LF) || !TLI->has(LF))
 132         continue;
 133
 134       switch (LF) {
 135       case LibFunc_sqrtf:
 136       case LibFunc_sqrt:
 137         if (TTI->haveFastSqrt(Call->getType()) &&
 138             optimizeSQRT(Call, CalledFunc, *CurrBB, BB, TTI,
 139                          DTU.hasValue() ? DTU.getPointer() : nullptr))
 140           break;
 141         continue;
 142       default:
 143         continue;
 144       }
 145
 146       Changed = true;
 147       break;
 148     }
 149   }
 150
 151   return Changed;
 152 }
 153
 154 PreservedAnalyses
 155 PartiallyInlineLibCallsPass::run(Function &F, FunctionAnalysisManager &AM) {
 156   auto &TLI = AM.getResult<TargetLibraryAnalysis>(F);
 157   auto &TTI = AM.getResult<TargetIRAnalysis>(F);
 158   auto *DT = AM.getCachedResult<DominatorTreeAnalysis>(F);
 159   if (!runPartiallyInlineLibCalls(F, &TLI, &TTI, DT))
 160     return PreservedAnalyses::all();
 161   PreservedAnalyses PA;
 162   PA.preserve<DominatorTreeAnalysis>();
 163   return PA;
 164 }
 165
 166 namespace {
 167 class PartiallyInlineLibCallsLegacyPass : public FunctionPass {
 168 public:
 169   static char ID;
 170
 171   PartiallyInlineLibCallsLegacyPass() : FunctionPass(ID) {
 172     initializePartiallyInlineLibCallsLegacyPassPass(
 173         *PassRegistry::getPassRegistry());
 174   }
 175
 176   void getAnalysisUsage(AnalysisUsage &AU) const override {
 177     AU.addRequired<TargetLibraryInfoWrapperPass>();
 178     AU.addRequired<TargetTransformInfoWrapperPass>();
 179     AU.addPreserved<DominatorTreeWrapperPass>();
 180     FunctionPass::getAnalysisUsage(AU);
 181   }
 182
 183   bool runOnFunction(Function &F) override {
 184     if (skipFunction(F))
 185       return false;
 186
 187     TargetLibraryInfo *TLI =
 188         &getAnalysis<TargetLibraryInfoWrapperPass>().getTLI(F);
 189     const TargetTransformInfo *TTI =
 190         &getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
 191     DominatorTree *DT = nullptr;
 192     if (auto *DTWP = getAnalysisIfAvailable<DominatorTreeWrapperPass>())
 193       DT = &DTWP->getDomTree();
 194     return runPartiallyInlineLibCalls(F, TLI, TTI, DT);
 195   }
 196 };
 197 }
 198
 199 char PartiallyInlineLibCallsLegacyPass::ID = 0;
 200 INITIALIZE_PASS_BEGIN(PartiallyInlineLibCallsLegacyPass,
 201                       "partially-inline-libcalls",
 202                       "Partially inline calls to library functions", false,
 203                       false)
 204 INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass)
 205 INITIALIZE_PASS_DEPENDENCY(DominatorTreeWrapperPass)
 206 INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass)
 207 INITIALIZE_PASS_END(PartiallyInlineLibCallsLegacyPass,
 208                     "partially-inline-libcalls",
 209                     "Partially inline calls to library functions", false, false)
 210
 211 FunctionPass *llvm::createPartiallyInlineLibCallsPass() {
 212   return new PartiallyInlineLibCallsLegacyPass();
 213 }