X-Git-Url: http://plrg.eecs.uci.edu/git/?p=oota-llvm.git;a=blobdiff_plain;f=lib%2FTransforms%2FIPO%2FInlineSimple.cpp;h=45609f891ed87368f65f56c2b5509cefece65195;hp=4bbefa3aacbd4328ea26ba8ffd5db36e27e2ec36;hb=0651a407f6a408caa02b76a1cdaf0d8fa54f29fb;hpb=fd93908ae8b9684fe71c239e3c6cfe13ff6a2663 diff --git a/lib/Transforms/IPO/InlineSimple.cpp b/lib/Transforms/IPO/InlineSimple.cpp index 4bbefa3aacb..45609f891ed 100644 --- a/lib/Transforms/IPO/InlineSimple.cpp +++ b/lib/Transforms/IPO/InlineSimple.cpp @@ -2,8 +2,8 @@ // // The LLVM Compiler Infrastructure // -// This file was developed by the LLVM research group and is distributed under -// the University of Illinois Open Source License. See LICENSE.TXT for details. +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. // //===----------------------------------------------------------------------===// // @@ -11,250 +11,100 @@ // //===----------------------------------------------------------------------===// -#include "Inliner.h" -#include "llvm/Instructions.h" -#include "llvm/IntrinsicInst.h" -#include "llvm/Function.h" -#include "llvm/Type.h" -#include "llvm/Support/CallSite.h" +#include "llvm/Analysis/AssumptionCache.h" +#include "llvm/Analysis/CallGraph.h" +#include "llvm/Analysis/InlineCost.h" +#include "llvm/Analysis/TargetLibraryInfo.h" +#include "llvm/Analysis/TargetTransformInfo.h" +#include "llvm/IR/CallSite.h" +#include "llvm/IR/CallingConv.h" +#include "llvm/IR/DataLayout.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/IntrinsicInst.h" +#include "llvm/IR/Module.h" +#include "llvm/IR/Type.h" #include "llvm/Transforms/IPO.h" +#include "llvm/Transforms/IPO/InlinerPass.h" + using namespace llvm; -namespace { - struct ArgInfo { - unsigned ConstantWeight; - unsigned AllocaWeight; +#define DEBUG_TYPE "inline" - ArgInfo(unsigned CWeight, unsigned AWeight) - : ConstantWeight(CWeight), AllocaWeight(AWeight) {} - }; +namespace { - // FunctionInfo - For each function, calculate the size of it in blocks and - // instructions. - struct FunctionInfo { - // HasAllocas - Keep track of whether or not a function contains an alloca - // instruction that is not in the entry block of the function. Inlining - // this call could cause us to blow out the stack, because the stack memory - // would never be released. - // - // FIXME: LLVM needs a way of dealloca'ing memory, which would make this - // irrelevant! - // - bool HasAllocas; +/// \brief Actual inliner pass implementation. +/// +/// The common implementation of the inlining logic is shared between this +/// inliner pass and the always inliner pass. The two passes use different cost +/// analyses to determine when to inline. +class SimpleInliner : public Inliner { - // NumInsts, NumBlocks - Keep track of how large each function is, which is - // used to estimate the code size cost of inlining it. - unsigned NumInsts, NumBlocks; +public: + SimpleInliner() : Inliner(ID) { + initializeSimpleInlinerPass(*PassRegistry::getPassRegistry()); + } - // ArgumentWeights - Each formal argument of the function is inspected to - // see if it is used in any contexts where making it a constant or alloca - // would reduce the code size. If so, we add some value to the argument - // entry here. - std::vector ArgumentWeights; + SimpleInliner(int Threshold) + : Inliner(ID, Threshold, /*InsertLifetime*/ true) { + initializeSimpleInlinerPass(*PassRegistry::getPassRegistry()); + } - FunctionInfo() : HasAllocas(false), NumInsts(0), NumBlocks(0) {} + static char ID; // Pass identification, replacement for typeid - /// analyzeFunction - Fill in the current structure with information gleaned - /// from the specified function. - void analyzeFunction(Function *F); - }; + InlineCost getInlineCost(CallSite CS) override { + Function *Callee = CS.getCalledFunction(); + TargetTransformInfo &TTI = TTIWP->getTTI(*Callee); + return llvm::getInlineCost(CS, getInlineThreshold(CS), TTI, ACT); + } - class SimpleInliner : public Inliner { - std::map CachedFunctionInfo; - public: - int getInlineCost(CallSite CS); - }; - RegisterOpt X("inline", "Function Integration/Inlining"); + bool runOnSCC(CallGraphSCC &SCC) override; + void getAnalysisUsage(AnalysisUsage &AU) const override; + +private: + TargetTransformInfoWrapperPass *TTIWP; +}; + +static int computeThresholdFromOptLevels(unsigned OptLevel, + unsigned SizeOptLevel) { + if (OptLevel > 2) + return 275; + if (SizeOptLevel == 1) // -Os + return 75; + if (SizeOptLevel == 2) // -Oz + return 25; + return 225; } -ModulePass *llvm::createFunctionInliningPass() { return new SimpleInliner(); } - -// CountCodeReductionForConstant - Figure out an approximation for how many -// instructions will be constant folded if the specified value is constant. -// -static unsigned CountCodeReductionForConstant(Value *V) { - unsigned Reduction = 0; - for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E; ++UI) - if (isa(*UI)) - Reduction += 40; // Eliminating a conditional branch is a big win - else if (SwitchInst *SI = dyn_cast(*UI)) - // Eliminating a switch is a big win, proportional to the number of edges - // deleted. - Reduction += (SI->getNumSuccessors()-1) * 40; - else if (CallInst *CI = dyn_cast(*UI)) { - // Turning an indirect call into a direct call is a BIG win - Reduction += CI->getCalledValue() == V ? 500 : 0; - } else if (InvokeInst *II = dyn_cast(*UI)) { - // Turning an indirect call into a direct call is a BIG win - Reduction += II->getCalledValue() == V ? 500 : 0; - } else { - // Figure out if this instruction will be removed due to simple constant - // propagation. - Instruction &Inst = cast(**UI); - bool AllOperandsConstant = true; - for (unsigned i = 0, e = Inst.getNumOperands(); i != e; ++i) - if (!isa(Inst.getOperand(i)) && Inst.getOperand(i) != V) { - AllOperandsConstant = false; - break; - } +} // end anonymous namespace - if (AllOperandsConstant) { - // We will get to remove this instruction... - Reduction += 7; +char SimpleInliner::ID = 0; +INITIALIZE_PASS_BEGIN(SimpleInliner, "inline", + "Function Integration/Inlining", false, false) +INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker) +INITIALIZE_PASS_DEPENDENCY(CallGraphWrapperPass) +INITIALIZE_PASS_DEPENDENCY(TargetTransformInfoWrapperPass) +INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfoWrapperPass) +INITIALIZE_PASS_END(SimpleInliner, "inline", + "Function Integration/Inlining", false, false) - // And any other instructions that use it which become constants - // themselves. - Reduction += CountCodeReductionForConstant(&Inst); - } - } +Pass *llvm::createFunctionInliningPass() { return new SimpleInliner(); } - return Reduction; +Pass *llvm::createFunctionInliningPass(int Threshold) { + return new SimpleInliner(Threshold); } -// CountCodeReductionForAlloca - Figure out an approximation of how much smaller -// the function will be if it is inlined into a context where an argument -// becomes an alloca. -// -static unsigned CountCodeReductionForAlloca(Value *V) { - if (!isa(V->getType())) return 0; // Not a pointer - unsigned Reduction = 0; - for (Value::use_iterator UI = V->use_begin(), E = V->use_end(); UI != E;++UI){ - Instruction *I = cast(*UI); - if (isa(I) || isa(I)) - Reduction += 10; - else if (GetElementPtrInst *GEP = dyn_cast(I)) { - // If the GEP has variable indices, we won't be able to do much with it. - for (Instruction::op_iterator I = GEP->op_begin()+1, E = GEP->op_end(); - I != E; ++I) - if (!isa(*I)) return 0; - Reduction += CountCodeReductionForAlloca(GEP)+15; - } else { - // If there is some other strange instruction, we're not going to be able - // to do much if we inline this. - return 0; - } - } - - return Reduction; +Pass *llvm::createFunctionInliningPass(unsigned OptLevel, + unsigned SizeOptLevel) { + return new SimpleInliner( + computeThresholdFromOptLevels(OptLevel, SizeOptLevel)); } -/// analyzeFunction - Fill in the current structure with information gleaned -/// from the specified function. -void FunctionInfo::analyzeFunction(Function *F) { - unsigned NumInsts = 0, NumBlocks = 0; - - // Look at the size of the callee. Each basic block counts as 20 units, and - // each instruction counts as 10. - for (Function::const_iterator BB = F->begin(), E = F->end(); BB != E; ++BB) { - for (BasicBlock::const_iterator II = BB->begin(), E = BB->end(); - II != E; ++II) { - if (!isa(II)) ++NumInsts; - - // If there is an alloca in the body of the function, we cannot currently - // inline the function without the risk of exploding the stack. - if (isa(II) && BB != F->begin()) { - HasAllocas = true; - this->NumBlocks = this->NumInsts = 1; - return; - } - } - - ++NumBlocks; - } - - this->NumBlocks = NumBlocks; - this->NumInsts = NumInsts; - - // Check out all of the arguments to the function, figuring out how much - // code can be eliminated if one of the arguments is a constant. - for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(); I != E; ++I) - ArgumentWeights.push_back(ArgInfo(CountCodeReductionForConstant(I), - CountCodeReductionForAlloca(I))); +bool SimpleInliner::runOnSCC(CallGraphSCC &SCC) { + TTIWP = &getAnalysis(); + return Inliner::runOnSCC(SCC); } - -// getInlineCost - The heuristic used to determine if we should inline the -// function call or not. -// -int SimpleInliner::getInlineCost(CallSite CS) { - Instruction *TheCall = CS.getInstruction(); - Function *Callee = CS.getCalledFunction(); - const Function *Caller = TheCall->getParent()->getParent(); - - // Don't inline a directly recursive call. - if (Caller == Callee) return 2000000000; - - // InlineCost - This value measures how good of an inline candidate this call - // site is to inline. A lower inline cost make is more likely for the call to - // be inlined. This value may go negative. - // - int InlineCost = 0; - - // If there is only one call of the function, and it has internal linkage, - // make it almost guaranteed to be inlined. - // - if (Callee->hasInternalLinkage() && Callee->hasOneUse()) - InlineCost -= 30000; - - // Get information about the callee... - FunctionInfo &CalleeFI = CachedFunctionInfo[Callee]; - - // If we haven't calculated this information yet, do so now. - if (CalleeFI.NumBlocks == 0) - CalleeFI.analyzeFunction(Callee); - - // Don't inline calls to functions with allocas that are not in the entry - // block of the function. - if (CalleeFI.HasAllocas) - return 2000000000; - - // Add to the inline quality for properties that make the call valuable to - // inline. This includes factors that indicate that the result of inlining - // the function will be optimizable. Currently this just looks at arguments - // passed into the function. - // - unsigned ArgNo = 0; - for (CallSite::arg_iterator I = CS.arg_begin(), E = CS.arg_end(); - I != E; ++I, ++ArgNo) { - // Each argument passed in has a cost at both the caller and the callee - // sides. This favors functions that take many arguments over functions - // that take few arguments. - InlineCost -= 20; - - // If this is a function being passed in, it is very likely that we will be - // able to turn an indirect function call into a direct function call. - if (isa(I)) - InlineCost -= 100; - - // If an alloca is passed in, inlining this function is likely to allow - // significant future optimization possibilities (like scalar promotion, and - // scalarization), so encourage the inlining of the function. - // - else if (AllocaInst *AI = dyn_cast(I)) { - if (ArgNo < CalleeFI.ArgumentWeights.size()) - InlineCost -= CalleeFI.ArgumentWeights[ArgNo].AllocaWeight; - - // If this is a constant being passed into the function, use the argument - // weights calculated for the callee to determine how much will be folded - // away with this information. - } else if (isa(I)) { - if (ArgNo < CalleeFI.ArgumentWeights.size()) - InlineCost -= CalleeFI.ArgumentWeights[ArgNo].ConstantWeight; - } - } - - // Now that we have considered all of the factors that make the call site more - // likely to be inlined, look at factors that make us not want to inline it. - - // Don't inline into something too big, which would make it bigger. Here, we - // count each basic block as a single unit. - // - InlineCost += Caller->size()/20; - - - // Look at the size of the callee. Each basic block counts as 20 units, and - // each instruction counts as 5. - InlineCost += CalleeFI.NumInsts*5 + CalleeFI.NumBlocks*20; - return InlineCost; +void SimpleInliner::getAnalysisUsage(AnalysisUsage &AU) const { + AU.addRequired(); + Inliner::getAnalysisUsage(AU); } -