CodeGenPrepare.cpp

//===- CodeGenPrepare.cpp - Prepare a function for code generation --------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This pass munges the code in the input function to better prepare it for
// SelectionDAG-based code generation. This works around limitations in it's
// basic-block-at-a-time approach. It should eventually be removed.
//
//===----------------------------------------------------------------------===//

#define DEBUG_TYPE "codegenprepare"
#include "llvm/Transforms/Scalar.h"
#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Function.h"
#include "llvm/InlineAsm.h"
#include "llvm/Instructions.h"
#include "llvm/IntrinsicInst.h"
#include "llvm/Pass.h"
#include "llvm/Analysis/Dominators.h"
#include "llvm/Analysis/InstructionSimplify.h"
#include "llvm/Analysis/ProfileInfo.h"
#include "llvm/Target/TargetData.h"
#include "llvm/Target/TargetLowering.h"
#include "llvm/Transforms/Utils/AddrModeMatcher.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Transforms/Utils/BuildLibCalls.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/SmallSet.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/Assembly/Writer.h"
#include "llvm/Support/CallSite.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/PatternMatch.h"
#include "llvm/Support/raw_ostream.h"
#include "llvm/Support/IRBuilder.h"
#include "llvm/Support/ValueHandle.h"
using namespace llvm;
using namespace llvm::PatternMatch;

STATISTIC(NumBlocksElim, "Number of blocks eliminated");
STATISTIC(NumPHIsElim, "Number of trivial PHIs eliminated");
STATISTIC(NumGEPsElim, "Number of GEPs converted to casts");
STATISTIC(NumCmpUses, "Number of uses of Cmp expressions replaced with uses of "
                      "sunken Cmps");
STATISTIC(NumCastUses, "Number of uses of Cast expressions replaced with uses "
                       "of sunken Casts");
STATISTIC(NumMemoryInsts, "Number of memory instructions whose address "
                          "computations were sunk");
STATISTIC(NumExtsMoved, "Number of [s|z]ext instructions combined with loads");
STATISTIC(NumExtUses, "Number of uses of [s|z]ext instructions optimized");

static cl::opt<bool>
CriticalEdgeSplit("cgp-critical-edge-splitting",
                  cl::desc("Split critical edges during codegen prepare"),
                  cl::init(false), cl::Hidden);

namespace {
  class CodeGenPrepare : public FunctionPass {
    /// TLI - Keep a pointer of a TargetLowering to consult for determining
    /// transformation profitability.
    const TargetLowering *TLI;
    DominatorTree *DT;
    ProfileInfo *PFI;
    
    /// CurInstIterator - As we scan instructions optimizing them, this is the
    /// next instruction to optimize.  Xforms that can invalidate this should
    /// update it.
    BasicBlock::iterator CurInstIterator;

    /// BackEdges - Keep a set of all the loop back edges.
    ///
    SmallSet<std::pair<const BasicBlock*, const BasicBlock*>, 8> BackEdges;

    // Keeps track of non-local addresses that have been sunk into a block. This
    // allows us to avoid inserting duplicate code for blocks with multiple
    // load/stores of the same address.
    DenseMap<Value*, Value*> SunkAddrs;

  public:
    static char ID; // Pass identification, replacement for typeid
    explicit CodeGenPrepare(const TargetLowering *tli = 0)
      : FunctionPass(ID), TLI(tli) {
        initializeCodeGenPreparePass(*PassRegistry::getPassRegistry());
      }
    bool runOnFunction(Function &F);

    virtual void getAnalysisUsage(AnalysisUsage &AU) const {
      AU.addPreserved<DominatorTree>();
      AU.addPreserved<ProfileInfo>();
    }

    virtual void releaseMemory() {
      BackEdges.clear();
    }

  private:
    bool EliminateMostlyEmptyBlocks(Function &F);
    bool CanMergeBlocks(const BasicBlock *BB, const BasicBlock *DestBB) const;
    void EliminateMostlyEmptyBlock(BasicBlock *BB);
    bool OptimizeBlock(BasicBlock &BB);
    bool OptimizeInst(Instruction *I);
    bool OptimizeMemoryInst(Instruction *I, Value *Addr, const Type *AccessTy,
                            DenseMap<Value*,Value*> &SunkAddrs);
    bool OptimizeInlineAsmInst(CallInst *CS);
    bool OptimizeCallInst(CallInst *CI);
    bool MoveExtToFormExtLoad(Instruction *I);
    bool OptimizeExtUses(Instruction *I);
    void findLoopBackEdges(const Function &F);
  };
}

char CodeGenPrepare::ID = 0;
INITIALIZE_PASS(CodeGenPrepare, "codegenprepare",
                "Optimize for code generation", false, false)

FunctionPass *llvm::createCodeGenPreparePass(const TargetLowering *TLI) {
  return new CodeGenPrepare(TLI);
}

/// findLoopBackEdges - Do a DFS walk to find loop back edges.
///
void CodeGenPrepare::findLoopBackEdges(const Function &F) {
  SmallVector<std::pair<const BasicBlock*,const BasicBlock*>, 32> Edges;
  FindFunctionBackedges(F, Edges);
  
  BackEdges.insert(Edges.begin(), Edges.end());
}


bool CodeGenPrepare::runOnFunction(Function &F) {
  bool EverMadeChange = false;

  DT = getAnalysisIfAvailable<DominatorTree>();
  PFI = getAnalysisIfAvailable<ProfileInfo>();
  // First pass, eliminate blocks that contain only PHI nodes and an
  // unconditional branch.
  EverMadeChange |= EliminateMostlyEmptyBlocks(F);

  // Now find loop back edges, but only if they are being used to decide which
  // critical edges to split.
  if (CriticalEdgeSplit)
    findLoopBackEdges(F);

  bool MadeChange = true;
  while (MadeChange) {
    MadeChange = false;
    for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
      MadeChange |= OptimizeBlock(*BB);
    EverMadeChange |= MadeChange;
  }

  SunkAddrs.clear();

  return EverMadeChange;
}

/// EliminateMostlyEmptyBlocks - eliminate blocks that contain only PHI nodes,
/// debug info directives, and an unconditional branch.  Passes before isel
/// (e.g. LSR/loopsimplify) often split edges in ways that are non-optimal for
/// isel.  Start by eliminating these blocks so we can split them the way we
/// want them.
bool CodeGenPrepare::EliminateMostlyEmptyBlocks(Function &F) {
  bool MadeChange = false;
  // Note that this intentionally skips the entry block.
  for (Function::iterator I = ++F.begin(), E = F.end(); I != E; ) {
    BasicBlock *BB = I++;

    // If this block doesn't end with an uncond branch, ignore it.
    BranchInst *BI = dyn_cast<BranchInst>(BB->getTerminator());
    if (!BI || !BI->isUnconditional())
      continue;

    // If the instruction before the branch (skipping debug info) isn't a phi
    // node, then other stuff is happening here.
    BasicBlock::iterator BBI = BI;
    if (BBI != BB->begin()) {
      --BBI;
      while (isa<DbgInfoIntrinsic>(BBI)) {
        if (BBI == BB->begin())
          break;
        --BBI;
      }
      if (!isa<DbgInfoIntrinsic>(BBI) && !isa<PHINode>(BBI))
        continue;
    }

    // Do not break infinite loops.
    BasicBlock *DestBB = BI->getSuccessor(0);
    if (DestBB == BB)
      continue;