LoopUnswitch.cpp

//===-- LoopUnswitch.cpp - Hoist loop-invariant conditionals in loop ------===//
//
//                     The LLVM Compiler Infrastructure
//
// This file is distributed under the University of Illinois Open Source
// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
// This pass transforms loops that contain branches on loop-invariant conditions
// to have multiple loops.  For example, it turns the left into the right code:
//
//  for (...)                  if (lic)
//    A                          for (...)
//    if (lic)                     A; B; C
//      B                      else
//    C                          for (...)
//                                 A; C
//
// This can increase the size of the code exponentially (doubling it every time
// a loop is unswitched) so we only unswitch if the resultant code will be
// smaller than a threshold.
//
// This pass expects LICM to be run before it to hoist invariant conditions out
// of the loop, to make the unswitching opportunity obvious.
//
//===----------------------------------------------------------------------===//

#define DEBUG_TYPE "loop-unswitch"
#include "llvm/Transforms/Scalar.h"
#include "llvm/Constants.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Function.h"
#include "llvm/Instructions.h"
#include "llvm/Analysis/ConstantFolding.h"
#include "llvm/Analysis/LoopInfo.h"
#include "llvm/Analysis/LoopPass.h"
#include "llvm/Analysis/Dominators.h"
#include "llvm/Transforms/Utils/Cloning.h"
#include "llvm/Transforms/Utils/Local.h"
#include "llvm/Transforms/Utils/BasicBlockUtils.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/SmallPtrSet.h"
#include "llvm/ADT/STLExtras.h"
#include "llvm/Support/CommandLine.h"
#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include <algorithm>
#include <set>
using namespace llvm;

STATISTIC(NumBranches, "Number of branches unswitched");
STATISTIC(NumSwitches, "Number of switches unswitched");
STATISTIC(NumSelects , "Number of selects unswitched");
STATISTIC(NumTrivial , "Number of unswitches that are trivial");
STATISTIC(NumSimplify, "Number of simplifications of unswitched code");

static cl::opt<unsigned>
Threshold("loop-unswitch-threshold", cl::desc("Max loop size to unswitch"),
          cl::init(10), cl::Hidden);
  
namespace {
  class VISIBILITY_HIDDEN LoopUnswitch : public LoopPass {
    LoopInfo *LI;  // Loop information
    LPPassManager *LPM;

    // LoopProcessWorklist - Used to check if second loop needs processing
    // after RewriteLoopBodyWithConditionConstant rewrites first loop.
    std::vector<Loop*> LoopProcessWorklist;
    SmallPtrSet<Value *,8> UnswitchedVals;
    
    bool OptimizeForSize;
    bool redoLoop;

    Loop *currentLoop;
    DominanceFrontier *DF;
    DominatorTree *DT;
    BasicBlock *loopHeader;
    BasicBlock *loopPreheader;
    
    // LoopBlocks contains all of the basic blocks of the loop, including the
    // preheader of the loop, the body of the loop, and the exit blocks of the 
    // loop, in that order.
    std::vector<BasicBlock*> LoopBlocks;
    // NewBlocks contained cloned copy of basic blocks from LoopBlocks.
    std::vector<BasicBlock*> NewBlocks;

  public:
    static char ID; // Pass ID, replacement for typeid
    explicit LoopUnswitch(bool Os = false) : 
      LoopPass(&ID), OptimizeForSize(Os), redoLoop(false), 
      currentLoop(NULL), DF(NULL), DT(NULL), loopHeader(NULL),
      loopPreheader(NULL) {}

    bool runOnLoop(Loop *L, LPPassManager &LPM);
    bool processCurrentLoop();

    /// This transformation requires natural loop information & requires that
    /// loop preheaders be inserted into the CFG...
    ///
    virtual void getAnalysisUsage(AnalysisUsage &AU) const {
      AU.addRequiredID(LoopSimplifyID);
      AU.addPreservedID(LoopSimplifyID);
      AU.addRequired<LoopInfo>();
      AU.addPreserved<LoopInfo>();
      AU.addRequiredID(LCSSAID);
      AU.addPreservedID(LCSSAID);
      AU.addPreserved<DominatorTree>();
      AU.addPreserved<DominanceFrontier>();
    }

  private:

    /// RemoveLoopFromWorklist - If the specified loop is on the loop worklist,
    /// remove it.
    void RemoveLoopFromWorklist(Loop *L) {
      std::vector<Loop*>::iterator I = std::find(LoopProcessWorklist.begin(),
                                                 LoopProcessWorklist.end(), L);
      if (I != LoopProcessWorklist.end())
        LoopProcessWorklist.erase(I);
    }

    void initLoopData() {
      loopHeader = currentLoop->getHeader();
      loopPreheader = currentLoop->getLoopPreheader();
    }

    /// Split all of the edges from inside the loop to their exit blocks.
    /// Update the appropriate Phi nodes as we do so.
    void SplitExitEdges(Loop *L, const SmallVector<BasicBlock *, 8> &ExitBlocks);

    bool UnswitchIfProfitable(Value *LoopCond, Constant *Val);
    unsigned getLoopUnswitchCost(Value *LIC);
    void UnswitchTrivialCondition(Loop *L, Value *Cond, Constant *Val,
                                  BasicBlock *ExitBlock);
    void UnswitchNontrivialCondition(Value *LIC, Constant *OnVal, Loop *L);

    void RewriteLoopBodyWithConditionConstant(Loop *L, Value *LIC,
                                              Constant *Val, bool isEqual);

    void EmitPreheaderBranchOnCondition(Value *LIC, Constant *Val,
                                        BasicBlock *TrueDest, 
                                        BasicBlock *FalseDest,
                                        Instruction *InsertPt);

    void SimplifyCode(std::vector<Instruction*> &Worklist, Loop *L);
    void RemoveBlockIfDead(BasicBlock *BB,
                           std::vector<Instruction*> &Worklist, Loop *l);
    void RemoveLoopFromHierarchy(Loop *L);
    bool IsTrivialUnswitchCondition(Value *Cond, Constant **Val = 0,
                                    BasicBlock **LoopExit = 0);

  };
}
char LoopUnswitch::ID = 0;
static RegisterPass<LoopUnswitch> X("loop-unswitch", "Unswitch loops");

Pass *llvm::createLoopUnswitchPass(bool Os) { 
  return new LoopUnswitch(Os); 
}

/// FindLIVLoopCondition - Cond is a condition that occurs in L.  If it is
/// invariant in the loop, or has an invariant piece, return the invariant.
/// Otherwise, return null.
static Value *FindLIVLoopCondition(Value *Cond, Loop *L, bool &Changed) {
  // Constants should be folded, not unswitched on!
  if (isa<Constant>(Cond)) return 0;

  // TODO: Handle: br (VARIANT|INVARIANT).
  // TODO: Hoist simple expressions out of loops.
  if (L->isLoopInvariant(Cond)) return Cond;

  if (BinaryOperator *BO = dyn_cast<BinaryOperator>(Cond))
    if (BO->getOpcode() == Instruction::And ||
        BO->getOpcode() == Instruction::Or) {
      // If either the left or right side is invariant, we can unswitch on this,
      // which will cause the branch to go away in one loop and the condition to
      // simplify in the other one.
      if (Value *LHS = FindLIVLoopCondition(BO->getOperand(0), L, Changed))
        return LHS;
      if (Value *RHS = FindLIVLoopCondition(BO->getOperand(1), L, Changed))
        return RHS;
    }
  
  return 0;
}

bool LoopUnswitch::runOnLoop(Loop *L, LPPassManager &LPM_Ref) {
  LI = &getAnalysis<LoopInfo>();
  LPM = &LPM_Ref;
  DF = getAnalysisToUpdate<DominanceFrontier>();
  DT = getAnalysisToUpdate<DominatorTree>();
  currentLoop = L;
  Function *F = currentLoop->getHeader()->getParent();
  bool Changed = false;
  do {
    assert(currentLoop->isLCSSAForm());
    redoLoop = false;
    Changed |= processCurrentLoop();
  } while(redoLoop);