X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTransforms%2FIPO%2FInlineSimple.cpp;h=9b01d81b3c7c1ac8596e0d9e4a6c955e46bb66fa;hb=0e2c9c8c4ad6d658fd8f61896e5a9ff5d445a2b0;hp=64a210d30f3dfebc30f808b397c32f30ed9b43a4;hpb=e54453387486c1d5e61401e1d4febd3f6ebe86cf;p=oota-llvm.git diff --git a/lib/Transforms/IPO/InlineSimple.cpp b/lib/Transforms/IPO/InlineSimple.cpp index 64a210d30f3..9b01d81b3c7 100644 --- a/lib/Transforms/IPO/InlineSimple.cpp +++ b/lib/Transforms/IPO/InlineSimple.cpp @@ -1,189 +1,105 @@ -//===- FunctionInlining.cpp - Code to perform function inlining -----------===// +//===- InlineSimple.cpp - Code to perform simple function inlining --------===// +// +// The LLVM Compiler Infrastructure +// +// This file is distributed under the University of Illinois Open Source +// License. See LICENSE.TXT for details. +// +//===----------------------------------------------------------------------===// // // This file implements bottom-up inlining of functions into callees. // //===----------------------------------------------------------------------===// -#define DEBUG_TYPE "inline" #include "llvm/Transforms/IPO.h" -#include "llvm/Module.h" -#include "llvm/Pass.h" -#include "llvm/iOther.h" -#include "llvm/iMemory.h" -#include "llvm/iTerminators.h" -#include "llvm/Support/CallSite.h" -#include "llvm/Transforms/Utils/Cloning.h" -#include "Support/CommandLine.h" -#include "Support/Debug.h" -#include "Support/Statistic.h" -#include +#include "llvm/Analysis/AliasAnalysis.h" +#include "llvm/Analysis/AssumptionCache.h" +#include "llvm/Analysis/CallGraph.h" +#include "llvm/Analysis/InlineCost.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/InlinerPass.h" + +using namespace llvm; -namespace { - Statistic<> NumInlined("inline", "Number of functions inlined"); - Statistic<> NumDeleted("inline", "Number of functions deleted because all callers found"); - cl::opt // FIXME: 200 is VERY conservative - InlineLimit("inline-threshold", cl::Hidden, cl::init(200), - cl::desc("Control the amount of inlining to perform (default = 200)")); - - struct FunctionInlining : public Pass { - virtual bool run(Module &M) { - bool Changed = false; - for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I) - Changed |= doInlining(I); - ProcessedFunctions.clear(); - return Changed; - } - - private: - std::set ProcessedFunctions; // Prevent infinite recursion - bool doInlining(Function *F); - }; - RegisterOpt X("inline", "Function Integration/Inlining"); -} +#define DEBUG_TYPE "inline" -Pass *createFunctionInliningPass() { return new FunctionInlining(); } +namespace { +/// \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 { + InlineCostAnalysis *ICA; + +public: + SimpleInliner() : Inliner(ID), ICA(nullptr) { + initializeSimpleInlinerPass(*PassRegistry::getPassRegistry()); + } -// ShouldInlineFunction - The heuristic used to determine if we should inline -// the function call or not. -// -static inline bool ShouldInlineFunction(CallSite CS) { - Instruction *TheCall = CS.getInstruction(); - assert(TheCall->getParent() && TheCall->getParent()->getParent() && - "Call not embedded into a function!"); - - const Function *Callee = CS.getCalledFunction(); - if (Callee == 0 || Callee->isExternal()) - return false; // Cannot inline an indirect call... or external function. - - // Don't inline a recursive call. - const Function *Caller = TheCall->getParent()->getParent(); - if (Caller == Callee) return false; - - // InlineQuality - This value measures how good of an inline candidate this - // call site is to inline. The initial value determines how aggressive the - // inliner is. If this value is negative after the final computation, - // inlining is not performed. - // - int InlineQuality = InlineLimit; - - // If there is only one call of the function, and it has internal linkage, - // make it almost guaranteed to be inlined. - // - if (Callee->use_size() == 1 && Callee->hasInternalLinkage()) - InlineQuality += 30000; - - // Add to the inline quality for properties that make the call valueable 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. - // - for (CallSite::arg_iterator I = CS.arg_begin(), E = CS.arg_end(); - I != E; ++I) { - // 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. - InlineQuality += 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)) - InlineQuality += 100; - - // If a constant, global variable or alloca is passed in, inlining this - // function is likely to allow significant future optimization possibilities - // (constant propagation, scalar promotion, and scalarization), so encourage - // the inlining of the function. - // - else if (isa(I) || isa(I) || isa(I)) - InlineQuality += 60; + SimpleInliner(int Threshold) + : Inliner(ID, Threshold, /*InsertLifetime*/ true), ICA(nullptr) { + initializeSimpleInlinerPass(*PassRegistry::getPassRegistry()); } - // 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. - // As soon as the inline quality gets negative, bail out. + static char ID; // Pass identification, replacement for typeid - // 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 = Callee->begin(), E = Callee->end(); - BB != E; ++BB) { - InlineQuality -= BB->size()*10 + 20; - if (InlineQuality < 0) return false; + InlineCost getInlineCost(CallSite CS) override { + return ICA->getInlineCost(CS, getInlineThreshold(CS)); } - // Don't inline into something too big, which would make it bigger. Here, we - // count each basic block as a single unit. - for (Function::const_iterator BB = Caller->begin(), E = Caller->end(); - BB != E; ++BB) { - --InlineQuality; - if (InlineQuality < 0) return false; - } + bool runOnSCC(CallGraphSCC &SCC) override; + void getAnalysisUsage(AnalysisUsage &AU) const override; +}; + +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; +} + +} // end anonymous namespace + +char SimpleInliner::ID = 0; +INITIALIZE_PASS_BEGIN(SimpleInliner, "inline", + "Function Integration/Inlining", false, false) +INITIALIZE_AG_DEPENDENCY(AliasAnalysis) +INITIALIZE_PASS_DEPENDENCY(AssumptionCacheTracker) +INITIALIZE_PASS_DEPENDENCY(CallGraphWrapperPass) +INITIALIZE_PASS_DEPENDENCY(InlineCostAnalysis) +INITIALIZE_PASS_END(SimpleInliner, "inline", + "Function Integration/Inlining", false, false) - // If we get here, this call site is high enough "quality" to inline. - DEBUG(std::cerr << "Inlining in '" << Caller->getName() - << "', quality = " << InlineQuality << ": " << *TheCall); - return true; +Pass *llvm::createFunctionInliningPass() { return new SimpleInliner(); } + +Pass *llvm::createFunctionInliningPass(int Threshold) { + return new SimpleInliner(Threshold); } +Pass *llvm::createFunctionInliningPass(unsigned OptLevel, + unsigned SizeOptLevel) { + return new SimpleInliner( + computeThresholdFromOptLevels(OptLevel, SizeOptLevel)); +} -// doInlining - Use a heuristic based approach to inline functions that seem to -// look good. -// -bool FunctionInlining::doInlining(Function *F) { - // If we have already processed this function (ie, it is recursive) don't - // revisit. - std::set::iterator PFI = ProcessedFunctions.lower_bound(F); - if (PFI != ProcessedFunctions.end() && *PFI == F) return false; - - // Insert the function in the set so it doesn't get revisited. - ProcessedFunctions.insert(PFI, F); - - bool Changed = false; - for (Function::iterator BB = F->begin(); BB != F->end(); ++BB) - for (BasicBlock::iterator I = BB->begin(); I != BB->end(); ) { - bool ShouldInc = true; - // Found a call or invoke instruction? - if (isa(I) || isa(I)) { - CallSite CS = CallSite::get(I); - if (Function *Callee = CS.getCalledFunction()) { - doInlining(Callee); // Inline in callees before callers! - - // Decide whether we should inline this function... - if (ShouldInlineFunction(CS)) { - // Save an iterator to the instruction before the call if it exists, - // otherwise get an iterator at the end of the block... because the - // call will be destroyed. - // - BasicBlock::iterator SI; - if (I != BB->begin()) { - SI = I; --SI; // Instruction before the call... - } else { - SI = BB->end(); - } - - // Attempt to inline the function... - if (InlineFunction(CS)) { - ++NumInlined; - Changed = true; - // Move to instruction before the call... - I = (SI == BB->end()) ? BB->begin() : SI; - ShouldInc = false; // Don't increment iterator until next time - - // If we inlined the last possible call site to the function, - // delete the function body now. - if (Callee->use_empty() && - (Callee->hasInternalLinkage()||Callee->hasLinkOnceLinkage())){ - F->getParent()->getFunctionList().erase(Callee); - ++NumDeleted; - if (Callee == F) return true; - } - } - } - } - } - if (ShouldInc) ++I; - } - - return Changed; +bool SimpleInliner::runOnSCC(CallGraphSCC &SCC) { + ICA = &getAnalysis(); + return Inliner::runOnSCC(SCC); } +void SimpleInliner::getAnalysisUsage(AnalysisUsage &AU) const { + AU.addRequired(); + Inliner::getAnalysisUsage(AU); +}