LoopUnswitch.cpp

//===-- LoopUnswitch.cpp - Hoist loop-invariant conditionals in loop ------===//
//
//                     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 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/PostOrderIterator.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");

namespace {
  cl::opt<unsigned>
  Threshold("loop-unswitch-threshold", cl::desc("Max loop size to unswitch"),
            cl::init(10), cl::Hidden);
  
  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;
  public:
    static char ID; // Pass ID, replacement for typeid
    LoopUnswitch(bool Os = false) : 
      LoopPass((intptr_t)&ID), OptimizeForSize(Os), redoLoop(false) {}

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

    /// 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);
    }
      
    bool UnswitchIfProfitable(Value *LoopCond, Constant *Val,Loop *L);
    unsigned getLoopUnswitchCost(Loop *L, 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);
  };
  char LoopUnswitch::ID = 0;
  RegisterPass<LoopUnswitch> X("loop-unswitch", "Unswitch loops");
}

LoopPass *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 false;

  // 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;
  bool Changed = false;

  do {
    redoLoop = false;
    Changed |= processLoop(L);
  } while(redoLoop);

  return Changed;
}

/// processLoop - Do actual work and unswitch loop if possible and profitable.
bool LoopUnswitch::processLoop(Loop *L) {
  assert(L->isLCSSAForm());
  bool Changed = false;

  // Loop over all of the basic blocks in the loop.  If we find an interior
  // block that is branching on a loop-invariant condition, we can unswitch this
  // loop.
  for (Loop::block_iterator I = L->block_begin(), E = L->block_end();
       I != E; ++I) {
    TerminatorInst *TI = (*I)->getTerminator();
    if (BranchInst *BI = dyn_cast<BranchInst>(TI)) {
      // If this isn't branching on an invariant condition, we can't unswitch
      // it.
      if (BI->isConditional()) {
        // See if this, or some part of it, is loop invariant.  If so, we can
        // unswitch on it if we desire.
        Value *LoopCond = FindLIVLoopCondition(BI->getCondition(), L, Changed);
        if (LoopCond && UnswitchIfProfitable(LoopCond, ConstantInt::getTrue(),
                                             L)) {
          ++NumBranches;
          return true;
        }
      }      
    } else if (SwitchInst *SI = dyn_cast<SwitchInst>(TI)) {
      Value *LoopCond = FindLIVLoopCondition(SI->getCondition(), L, Changed);
      if (LoopCond && SI->getNumCases() > 1) {
        // Find a value to unswitch on: