#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/Pass.h"
-#include "llvm/Target/TargetAsmInfo.h"
-#include "llvm/Target/TargetData.h"
-#include "llvm/Target/TargetLowering.h"
-#include "llvm/Target/TargetMachine.h"
-#include "llvm/Transforms/Utils/BasicBlockUtils.h"
-#include "llvm/Transforms/Utils/Local.h"
#include "llvm/ADT/DenseMap.h"
#include "llvm/ADT/SmallSet.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/ADT/ValueMap.h"
+#include "llvm/Analysis/DominatorInternals.h"
+#include "llvm/Analysis/Dominators.h"
+#include "llvm/Analysis/InstructionSimplify.h"
+#include "llvm/Analysis/ProfileInfo.h"
+#include "llvm/Assembly/Writer.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DataLayout.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Function.h"
+#include "llvm/IR/IRBuilder.h"
+#include "llvm/IR/InlineAsm.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/IntrinsicInst.h"
+#include "llvm/Pass.h"
#include "llvm/Support/CallSite.h"
#include "llvm/Support/CommandLine.h"
-#include "llvm/Support/Compiler.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/GetElementPtrTypeIterator.h"
#include "llvm/Support/PatternMatch.h"
+#include "llvm/Support/ValueHandle.h"
+#include "llvm/Support/raw_ostream.h"
+#include "llvm/Target/TargetLibraryInfo.h"
+#include "llvm/Target/TargetLowering.h"
+#include "llvm/Transforms/Utils/BasicBlockUtils.h"
+#include "llvm/Transforms/Utils/BuildLibCalls.h"
+#include "llvm/Transforms/Utils/BypassSlowDivision.h"
+#include "llvm/Transforms/Utils/Local.h"
using namespace llvm;
using namespace llvm::PatternMatch;
-static cl::opt<bool> FactorCommonPreds("split-critical-paths-tweak",
- cl::init(false), cl::Hidden);
+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");
+STATISTIC(NumRetsDup, "Number of return instructions duplicated");
+STATISTIC(NumDbgValueMoved, "Number of debug value instructions moved");
+STATISTIC(NumSelectsExpanded, "Number of selects turned into branches");
+
+static cl::opt<bool> DisableBranchOpts(
+ "disable-cgp-branch-opts", cl::Hidden, cl::init(false),
+ cl::desc("Disable branch optimizations in CodeGenPrepare"));
+
+static cl::opt<bool> DisableSelectToBranch(
+ "disable-cgp-select2branch", cl::Hidden, cl::init(false),
+ cl::desc("Disable select to branch conversion."));
namespace {
- class VISIBILITY_HIDDEN CodeGenPrepare : public FunctionPass {
+ class CodeGenPrepare : public FunctionPass {
/// TLI - Keep a pointer of a TargetLowering to consult for determining
/// transformation profitability.
const TargetLowering *TLI;
+ const TargetLibraryInfo *TLInfo;
+ 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;
+
+ /// 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.
+ ValueMap<Value*, Value*> SunkAddrs;
+
+ /// ModifiedDT - If CFG is modified in anyway, dominator tree may need to
+ /// be updated.
+ bool ModifiedDT;
+
+ /// OptSize - True if optimizing for size.
+ bool OptSize;
- /// BackEdges - Keep a set of all the loop back edges.
- ///
- SmallSet<std::pair<BasicBlock*,BasicBlock*>, 8> BackEdges;
public:
static char ID; // Pass identification, replacement for typeid
explicit CodeGenPrepare(const TargetLowering *tli = 0)
- : FunctionPass(&ID), TLI(tli) {}
+ : FunctionPass(ID), TLI(tli) {
+ initializeCodeGenPreparePass(*PassRegistry::getPassRegistry());
+ }
bool runOnFunction(Function &F);
+ const char *getPassName() const { return "CodeGen Prepare"; }
+
+ virtual void getAnalysisUsage(AnalysisUsage &AU) const {
+ AU.addPreserved<DominatorTree>();
+ AU.addPreserved<ProfileInfo>();
+ AU.addRequired<TargetLibraryInfo>();
+ }
+
private:
+ bool EliminateFallThrough(Function &F);
bool EliminateMostlyEmptyBlocks(Function &F);
bool CanMergeBlocks(const BasicBlock *BB, const BasicBlock *DestBB) const;
void EliminateMostlyEmptyBlock(BasicBlock *BB);
bool OptimizeBlock(BasicBlock &BB);
- bool OptimizeMemoryInst(Instruction *I, Value *Addr, const Type *AccessTy,
- DenseMap<Value*,Value*> &SunkAddrs);
- bool OptimizeInlineAsmInst(Instruction *I, CallSite CS,
- DenseMap<Value*,Value*> &SunkAddrs);
+ bool OptimizeInst(Instruction *I);
+ bool OptimizeMemoryInst(Instruction *I, Value *Addr, Type *AccessTy);
+ bool OptimizeInlineAsmInst(CallInst *CS);
+ bool OptimizeCallInst(CallInst *CI);
+ bool MoveExtToFormExtLoad(Instruction *I);
bool OptimizeExtUses(Instruction *I);
- void findLoopBackEdges(Function &F);
+ bool OptimizeSelectInst(SelectInst *SI);
+ bool DupRetToEnableTailCallOpts(BasicBlock *BB);
+ bool PlaceDbgValues(Function &F);
};
}
char CodeGenPrepare::ID = 0;
-static RegisterPass<CodeGenPrepare> X("codegenprepare",
- "Optimize for code generation");
+INITIALIZE_PASS_BEGIN(CodeGenPrepare, "codegenprepare",
+ "Optimize for code generation", false, false)
+INITIALIZE_PASS_DEPENDENCY(TargetLibraryInfo)
+INITIALIZE_PASS_END(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(Function &F) {
- SmallPtrSet<BasicBlock*, 8> Visited;
- SmallVector<std::pair<BasicBlock*, succ_iterator>, 8> VisitStack;
- SmallPtrSet<BasicBlock*, 8> InStack;
-
- BasicBlock *BB = &F.getEntryBlock();
- if (succ_begin(BB) == succ_end(BB))
- return;
- Visited.insert(BB);
- VisitStack.push_back(std::make_pair(BB, succ_begin(BB)));
- InStack.insert(BB);
- do {
- std::pair<BasicBlock*, succ_iterator> &Top = VisitStack.back();
- BasicBlock *ParentBB = Top.first;
- succ_iterator &I = Top.second;
-
- bool FoundNew = false;
- while (I != succ_end(ParentBB)) {
- BB = *I++;
- if (Visited.insert(BB)) {
- FoundNew = true;
- break;
- }
- // Successor is in VisitStack, it's a back edge.
- if (InStack.count(BB))
- BackEdges.insert(std::make_pair(ParentBB, BB));
- }
-
- if (FoundNew) {
- // Go down one level if there is a unvisited successor.
- InStack.insert(BB);
- VisitStack.push_back(std::make_pair(BB, succ_begin(BB)));
- } else {
- // Go up one level.
- std::pair<BasicBlock*, succ_iterator> &Pop = VisitStack.back();
- InStack.erase(Pop.first);
- VisitStack.pop_back();
- }
- } while (!VisitStack.empty());
-}
-
-
bool CodeGenPrepare::runOnFunction(Function &F) {
bool EverMadeChange = false;
- findLoopBackEdges(F);
+ ModifiedDT = false;
+ TLInfo = &getAnalysis<TargetLibraryInfo>();
+ DT = getAnalysisIfAvailable<DominatorTree>();
+ PFI = getAnalysisIfAvailable<ProfileInfo>();
+ OptSize = F.getAttributes().hasAttribute(AttributeSet::FunctionIndex,
+ Attribute::OptimizeForSize);
+
+ /// This optimization identifies DIV instructions that can be
+ /// profitably bypassed and carried out with a shorter, faster divide.
+ if (!OptSize && TLI && TLI->isSlowDivBypassed()) {
+ const DenseMap<unsigned int, unsigned int> &BypassWidths =
+ TLI->getBypassSlowDivWidths();
+ for (Function::iterator I = F.begin(); I != F.end(); I++)
+ EverMadeChange |= bypassSlowDivision(F, I, BypassWidths);
+ }
- // First pass, eliminate blocks that contain only PHI nodes and an
+ // Eliminate blocks that contain only PHI nodes and an
// unconditional branch.
EverMadeChange |= EliminateMostlyEmptyBlocks(F);
+ // llvm.dbg.value is far away from the value then iSel may not be able
+ // handle it properly. iSel will drop llvm.dbg.value if it can not
+ // find a node corresponding to the value.
+ EverMadeChange |= PlaceDbgValues(F);
+
bool MadeChange = true;
while (MadeChange) {
MadeChange = false;
- for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB)
+ for (Function::iterator I = F.begin(); I != F.end(); ) {
+ BasicBlock *BB = I++;
MadeChange |= OptimizeBlock(*BB);
+ }
+ EverMadeChange |= MadeChange;
+ }
+
+ SunkAddrs.clear();
+
+ if (!DisableBranchOpts) {
+ MadeChange = false;
+ SmallPtrSet<BasicBlock*, 8> WorkList;
+ for (Function::iterator BB = F.begin(), E = F.end(); BB != E; ++BB) {
+ SmallVector<BasicBlock*, 2> Successors(succ_begin(BB), succ_end(BB));
+ MadeChange |= ConstantFoldTerminator(BB, true);
+ if (!MadeChange) continue;
+
+ for (SmallVectorImpl<BasicBlock*>::iterator
+ II = Successors.begin(), IE = Successors.end(); II != IE; ++II)
+ if (pred_begin(*II) == pred_end(*II))
+ WorkList.insert(*II);
+ }
+
+ // Delete the dead blocks and any of their dead successors.
+ MadeChange |= !WorkList.empty();
+ while (!WorkList.empty()) {
+ BasicBlock *BB = *WorkList.begin();
+ WorkList.erase(BB);
+ SmallVector<BasicBlock*, 2> Successors(succ_begin(BB), succ_end(BB));
+
+ DeleteDeadBlock(BB);
+
+ for (SmallVectorImpl<BasicBlock*>::iterator
+ II = Successors.begin(), IE = Successors.end(); II != IE; ++II)
+ if (pred_begin(*II) == pred_end(*II))
+ WorkList.insert(*II);
+ }
+
+ // Merge pairs of basic blocks with unconditional branches, connected by
+ // a single edge.
+ if (EverMadeChange || MadeChange)
+ MadeChange |= EliminateFallThrough(F);
+
+ if (MadeChange)
+ ModifiedDT = true;
EverMadeChange |= MadeChange;
}
+
+ if (ModifiedDT && DT)
+ DT->DT->recalculate(F);
+
return EverMadeChange;
}
-/// EliminateMostlyEmptyBlocks - eliminate blocks that contain only PHI nodes
-/// 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.
+/// EliminateFallThrough - Merge basic blocks which are connected
+/// by a single edge, where one of the basic blocks has a single successor
+/// pointing to the other basic block, which has a single predecessor.
+bool CodeGenPrepare::EliminateFallThrough(Function &F) {
+ bool Changed = false;
+ // Scan all of the blocks in the function, except for the entry block.
+ for (Function::iterator I = ++F.begin(), E = F.end(); I != E; ) {
+ BasicBlock *BB = I++;
+ // If the destination block has a single pred, then this is a trivial
+ // edge, just collapse it.
+ BasicBlock *SinglePred = BB->getSinglePredecessor();
+
+ // Don't merge if BB's address is taken.
+ if (!SinglePred || SinglePred == BB || BB->hasAddressTaken()) continue;
+
+ BranchInst *Term = dyn_cast<BranchInst>(SinglePred->getTerminator());
+ if (Term && !Term->isConditional()) {
+ Changed = true;
+ DEBUG(dbgs() << "To merge:\n"<< *SinglePred << "\n\n\n");
+ // Remember if SinglePred was the entry block of the function.
+ // If so, we will need to move BB back to the entry position.
+ bool isEntry = SinglePred == &SinglePred->getParent()->getEntryBlock();
+ MergeBasicBlockIntoOnlyPred(BB, this);
+
+ if (isEntry && BB != &BB->getParent()->getEntryBlock())
+ BB->moveBefore(&BB->getParent()->getEntryBlock());
+
+ // We have erased a block. Update the iterator.
+ I = BB;
+ }
+ }
+ return Changed;
+}
+
+/// 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.
if (!BI || !BI->isUnconditional())
continue;
- // If the instruction before the branch isn't a phi node, then other stuff
- // is happening here.
+ // 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;
- if (!isa<PHINode>(BBI)) continue;
+ while (isa<DbgInfoIntrinsic>(BBI)) {
+ if (BBI == BB->begin())
+ break;
+ --BBI;
+ }
+ if (!isa<DbgInfoIntrinsic>(BBI) && !isa<PHINode>(BBI))
+ continue;
}
// Do not break infinite loops.
// don't mess around with them.
BasicBlock::const_iterator BBI = BB->begin();
while (const PHINode *PN = dyn_cast<PHINode>(BBI++)) {
- for (Value::use_const_iterator UI = PN->use_begin(), E = PN->use_end();
+ for (Value::const_use_iterator UI = PN->use_begin(), E = PN->use_end();
UI != E; ++UI) {
const Instruction *User = cast<Instruction>(*UI);
if (User->getParent() != DestBB || !isa<PHINode>(User))
BranchInst *BI = cast<BranchInst>(BB->getTerminator());
BasicBlock *DestBB = BI->getSuccessor(0);
- DOUT << "MERGING MOSTLY EMPTY BLOCKS - BEFORE:\n" << *BB << *DestBB;
+ DEBUG(dbgs() << "MERGING MOSTLY EMPTY BLOCKS - BEFORE:\n" << *BB << *DestBB);
// If the destination block has a single pred, then this is a trivial edge,
// just collapse it.
// Remember if SinglePred was the entry block of the function. If so, we
// will need to move BB back to the entry position.
bool isEntry = SinglePred == &SinglePred->getParent()->getEntryBlock();
- MergeBasicBlockIntoOnlyPred(DestBB);
+ MergeBasicBlockIntoOnlyPred(DestBB, this);
if (isEntry && BB != &BB->getParent()->getEntryBlock())
BB->moveBefore(&BB->getParent()->getEntryBlock());
-
- DOUT << "AFTER:\n" << *DestBB << "\n\n\n";
+
+ DEBUG(dbgs() << "AFTER:\n" << *DestBB << "\n\n\n");
return;
}
}
// The PHIs are now updated, change everything that refers to BB to use
// DestBB and remove BB.
BB->replaceAllUsesWith(DestBB);
- BB->eraseFromParent();
-
- DOUT << "AFTER:\n" << *DestBB << "\n\n\n";
-}
-
-
-/// SplitEdgeNicely - Split the critical edge from TI to its specified
-/// successor if it will improve codegen. We only do this if the successor has
-/// phi nodes (otherwise critical edges are ok). If there is already another
-/// predecessor of the succ that is empty (and thus has no phi nodes), use it
-/// instead of introducing a new block.
-static void SplitEdgeNicely(TerminatorInst *TI, unsigned SuccNum,
- SmallSet<std::pair<BasicBlock*,BasicBlock*>, 8> &BackEdges,
- Pass *P) {
- BasicBlock *TIBB = TI->getParent();
- BasicBlock *Dest = TI->getSuccessor(SuccNum);
- assert(isa<PHINode>(Dest->begin()) &&
- "This should only be called if Dest has a PHI!");
-
- // As a hack, never split backedges of loops. Even though the copy for any
- // PHIs inserted on the backedge would be dead for exits from the loop, we
- // assume that the cost of *splitting* the backedge would be too high.
- if (BackEdges.count(std::make_pair(TIBB, Dest)))
- return;
-
- if (!FactorCommonPreds) {
- /// TIPHIValues - This array is lazily computed to determine the values of
- /// PHIs in Dest that TI would provide.
- SmallVector<Value*, 32> TIPHIValues;
-
- // Check to see if Dest has any blocks that can be used as a split edge for
- // this terminator.
- for (pred_iterator PI = pred_begin(Dest), E = pred_end(Dest); PI != E; ++PI) {
- BasicBlock *Pred = *PI;
- // To be usable, the pred has to end with an uncond branch to the dest.
- BranchInst *PredBr = dyn_cast<BranchInst>(Pred->getTerminator());
- if (!PredBr || !PredBr->isUnconditional() ||
- // Must be empty other than the branch.
- &Pred->front() != PredBr ||
- // Cannot be the entry block; its label does not get emitted.
- Pred == &(Dest->getParent()->getEntryBlock()))
- continue;
-
- // Finally, since we know that Dest has phi nodes in it, we have to make
- // sure that jumping to Pred will have the same affect as going to Dest in
- // terms of PHI values.
- PHINode *PN;
- unsigned PHINo = 0;
- bool FoundMatch = true;
- for (BasicBlock::iterator I = Dest->begin();
- (PN = dyn_cast<PHINode>(I)); ++I, ++PHINo) {
- if (PHINo == TIPHIValues.size())
- TIPHIValues.push_back(PN->getIncomingValueForBlock(TIBB));
-
- // If the PHI entry doesn't work, we can't use this pred.
- if (TIPHIValues[PHINo] != PN->getIncomingValueForBlock(Pred)) {
- FoundMatch = false;
- break;
- }
- }
-
- // If we found a workable predecessor, change TI to branch to Succ.
- if (FoundMatch) {
- Dest->removePredecessor(TIBB);
- TI->setSuccessor(SuccNum, Pred);
- return;
- }
- }
-
- SplitCriticalEdge(TI, SuccNum, P, true);
- return;
+ if (DT && !ModifiedDT) {
+ BasicBlock *BBIDom = DT->getNode(BB)->getIDom()->getBlock();
+ BasicBlock *DestBBIDom = DT->getNode(DestBB)->getIDom()->getBlock();
+ BasicBlock *NewIDom = DT->findNearestCommonDominator(BBIDom, DestBBIDom);
+ DT->changeImmediateDominator(DestBB, NewIDom);
+ DT->eraseNode(BB);
}
-
- PHINode *PN;
- SmallVector<Value*, 8> TIPHIValues;
- for (BasicBlock::iterator I = Dest->begin();
- (PN = dyn_cast<PHINode>(I)); ++I)
- TIPHIValues.push_back(PN->getIncomingValueForBlock(TIBB));
-
- SmallVector<BasicBlock*, 8> IdenticalPreds;
- for (pred_iterator PI = pred_begin(Dest), E = pred_end(Dest); PI != E; ++PI) {
- BasicBlock *Pred = *PI;
- if (BackEdges.count(std::make_pair(Pred, Dest)))
- continue;
- if (PI == TIBB)
- IdenticalPreds.push_back(Pred);
- else {
- bool Identical = true;
- unsigned PHINo = 0;
- for (BasicBlock::iterator I = Dest->begin();
- (PN = dyn_cast<PHINode>(I)); ++I, ++PHINo)
- if (TIPHIValues[PHINo] != PN->getIncomingValueForBlock(Pred)) {
- Identical = false;
- break;
- }
- if (Identical)
- IdenticalPreds.push_back(Pred);
- }
+ if (PFI) {
+ PFI->replaceAllUses(BB, DestBB);
+ PFI->removeEdge(ProfileInfo::getEdge(BB, DestBB));
}
+ BB->eraseFromParent();
+ ++NumBlocksElim;
- assert(!IdenticalPreds.empty());
- SplitBlockPredecessors(Dest, &IdenticalPreds[0], IdenticalPreds.size(),
- ".critedge", P);
+ DEBUG(dbgs() << "AFTER:\n" << *DestBB << "\n\n\n");
}
-
/// OptimizeNoopCopyExpression - If the specified cast instruction is a noop
-/// copy (e.g. it's casting from one pointer type to another, int->uint, or
-/// int->sbyte on PPC), sink it into user blocks to reduce the number of virtual
+/// copy (e.g. it's casting from one pointer type to another, i32->i8 on PPC),
+/// sink it into user blocks to reduce the number of virtual
/// registers that must be created and coalesced.
///
/// Return true if any changes are made.
///
static bool OptimizeNoopCopyExpression(CastInst *CI, const TargetLowering &TLI){
// If this is a noop copy,
- MVT SrcVT = TLI.getValueType(CI->getOperand(0)->getType());
- MVT DstVT = TLI.getValueType(CI->getType());
+ EVT SrcVT = TLI.getValueType(CI->getOperand(0)->getType());
+ EVT DstVT = TLI.getValueType(CI->getType());
// This is an fp<->int conversion?
if (SrcVT.isInteger() != DstVT.isInteger())
// If these values will be promoted, find out what they will be promoted
// to. This helps us consider truncates on PPC as noop copies when they
// are.
- if (TLI.getTypeAction(SrcVT) == TargetLowering::Promote)
- SrcVT = TLI.getTypeToTransformTo(SrcVT);
- if (TLI.getTypeAction(DstVT) == TargetLowering::Promote)
- DstVT = TLI.getTypeToTransformTo(DstVT);
+ if (TLI.getTypeAction(CI->getContext(), SrcVT) ==
+ TargetLowering::TypePromoteInteger)
+ SrcVT = TLI.getTypeToTransformTo(CI->getContext(), SrcVT);
+ if (TLI.getTypeAction(CI->getContext(), DstVT) ==
+ TargetLowering::TypePromoteInteger)
+ DstVT = TLI.getTypeToTransformTo(CI->getContext(), DstVT);
// If, after promotion, these are the same types, this is a noop copy.
if (SrcVT != DstVT)
// appropriate predecessor block.
BasicBlock *UserBB = User->getParent();
if (PHINode *PN = dyn_cast<PHINode>(User)) {
- unsigned OpVal = UI.getOperandNo()/2;
- UserBB = PN->getIncomingBlock(OpVal);
+ UserBB = PN->getIncomingBlock(UI);
}
// Preincrement use iterator so we don't invalidate it.
CastInst *&InsertedCast = InsertedCasts[UserBB];
if (!InsertedCast) {
- BasicBlock::iterator InsertPt = UserBB->getFirstNonPHI();
-
+ BasicBlock::iterator InsertPt = UserBB->getFirstInsertionPt();
InsertedCast =
CastInst::Create(CI->getOpcode(), CI->getOperand(0), CI->getType(), "",
InsertPt);
// Replace a use of the cast with a use of the new cast.
TheUse = InsertedCast;
+ ++NumCastUses;
}
// If we removed all uses, nuke the cast.
CmpInst *&InsertedCmp = InsertedCmps[UserBB];
if (!InsertedCmp) {
- BasicBlock::iterator InsertPt = UserBB->getFirstNonPHI();
-
+ BasicBlock::iterator InsertPt = UserBB->getFirstInsertionPt();
InsertedCmp =
- CmpInst::Create(CI->getOpcode(), CI->getPredicate(), CI->getOperand(0),
+ CmpInst::Create(CI->getOpcode(),
+ CI->getPredicate(), CI->getOperand(0),
CI->getOperand(1), "", InsertPt);
MadeChange = true;
}
// Replace a use of the cmp with a use of the new cmp.
TheUse = InsertedCmp;
+ ++NumCmpUses;
}
// If we removed all uses, nuke the cmp.
return MadeChange;
}
+namespace {
+class CodeGenPrepareFortifiedLibCalls : public SimplifyFortifiedLibCalls {
+protected:
+ void replaceCall(Value *With) {
+ CI->replaceAllUsesWith(With);
+ CI->eraseFromParent();
+ }
+ bool isFoldable(unsigned SizeCIOp, unsigned, bool) const {
+ if (ConstantInt *SizeCI =
+ dyn_cast<ConstantInt>(CI->getArgOperand(SizeCIOp)))
+ return SizeCI->isAllOnesValue();
+ return false;
+ }
+};
+} // end anonymous namespace
+
+bool CodeGenPrepare::OptimizeCallInst(CallInst *CI) {
+ BasicBlock *BB = CI->getParent();
+
+ // Lower inline assembly if we can.
+ // If we found an inline asm expession, and if the target knows how to
+ // lower it to normal LLVM code, do so now.
+ if (TLI && isa<InlineAsm>(CI->getCalledValue())) {
+ if (TLI->ExpandInlineAsm(CI)) {
+ // Avoid invalidating the iterator.
+ CurInstIterator = BB->begin();
+ // Avoid processing instructions out of order, which could cause
+ // reuse before a value is defined.
+ SunkAddrs.clear();
+ return true;
+ }
+ // Sink address computing for memory operands into the block.
+ if (OptimizeInlineAsmInst(CI))
+ return true;
+ }
+
+ // Lower all uses of llvm.objectsize.*
+ IntrinsicInst *II = dyn_cast<IntrinsicInst>(CI);
+ if (II && II->getIntrinsicID() == Intrinsic::objectsize) {
+ bool Min = (cast<ConstantInt>(II->getArgOperand(1))->getZExtValue() == 1);
+ Type *ReturnTy = CI->getType();
+ Constant *RetVal = ConstantInt::get(ReturnTy, Min ? 0 : -1ULL);
+
+ // Substituting this can cause recursive simplifications, which can
+ // invalidate our iterator. Use a WeakVH to hold onto it in case this
+ // happens.
+ WeakVH IterHandle(CurInstIterator);
+
+ replaceAndRecursivelySimplify(CI, RetVal, TLI ? TLI->getDataLayout() : 0,
+ TLInfo, ModifiedDT ? 0 : DT);
+
+ // If the iterator instruction was recursively deleted, start over at the
+ // start of the block.
+ if (IterHandle != CurInstIterator) {
+ CurInstIterator = BB->begin();
+ SunkAddrs.clear();
+ }
+ return true;
+ }
+
+ if (II && TLI) {
+ SmallVector<Value*, 2> PtrOps;
+ Type *AccessTy;
+ if (TLI->GetAddrModeArguments(II, PtrOps, AccessTy))
+ while (!PtrOps.empty())
+ if (OptimizeMemoryInst(II, PtrOps.pop_back_val(), AccessTy))
+ return true;
+ }
+
+ // From here on out we're working with named functions.
+ if (CI->getCalledFunction() == 0) return false;
+
+ // We'll need DataLayout from here on out.
+ const DataLayout *TD = TLI ? TLI->getDataLayout() : 0;
+ if (!TD) return false;
+
+ // Lower all default uses of _chk calls. This is very similar
+ // to what InstCombineCalls does, but here we are only lowering calls
+ // that have the default "don't know" as the objectsize. Anything else
+ // should be left alone.
+ CodeGenPrepareFortifiedLibCalls Simplifier;
+ return Simplifier.fold(CI, TD, TLInfo);
+}
+
+/// DupRetToEnableTailCallOpts - Look for opportunities to duplicate return
+/// instructions to the predecessor to enable tail call optimizations. The
+/// case it is currently looking for is:
+/// @code
+/// bb0:
+/// %tmp0 = tail call i32 @f0()
+/// br label %return
+/// bb1:
+/// %tmp1 = tail call i32 @f1()
+/// br label %return
+/// bb2:
+/// %tmp2 = tail call i32 @f2()
+/// br label %return
+/// return:
+/// %retval = phi i32 [ %tmp0, %bb0 ], [ %tmp1, %bb1 ], [ %tmp2, %bb2 ]
+/// ret i32 %retval
+/// @endcode
+///
+/// =>
+///
+/// @code
+/// bb0:
+/// %tmp0 = tail call i32 @f0()
+/// ret i32 %tmp0
+/// bb1:
+/// %tmp1 = tail call i32 @f1()
+/// ret i32 %tmp1
+/// bb2:
+/// %tmp2 = tail call i32 @f2()
+/// ret i32 %tmp2
+/// @endcode
+bool CodeGenPrepare::DupRetToEnableTailCallOpts(BasicBlock *BB) {
+ if (!TLI)
+ return false;
+
+ ReturnInst *RI = dyn_cast<ReturnInst>(BB->getTerminator());
+ if (!RI)
+ return false;
+
+ PHINode *PN = 0;
+ BitCastInst *BCI = 0;
+ Value *V = RI->getReturnValue();
+ if (V) {
+ BCI = dyn_cast<BitCastInst>(V);
+ if (BCI)
+ V = BCI->getOperand(0);
+
+ PN = dyn_cast<PHINode>(V);
+ if (!PN)
+ return false;
+ }
+
+ if (PN && PN->getParent() != BB)
+ return false;
+
+ // It's not safe to eliminate the sign / zero extension of the return value.
+ // See llvm::isInTailCallPosition().
+ const Function *F = BB->getParent();
+ AttributeSet CallerAttrs = F->getAttributes();
+ if (CallerAttrs.hasAttribute(AttributeSet::ReturnIndex, Attribute::ZExt) ||
+ CallerAttrs.hasAttribute(AttributeSet::ReturnIndex, Attribute::SExt))
+ return false;
+
+ // Make sure there are no instructions between the PHI and return, or that the
+ // return is the first instruction in the block.
+ if (PN) {
+ BasicBlock::iterator BI = BB->begin();
+ do { ++BI; } while (isa<DbgInfoIntrinsic>(BI));
+ if (&*BI == BCI)
+ // Also skip over the bitcast.
+ ++BI;
+ if (&*BI != RI)
+ return false;
+ } else {
+ BasicBlock::iterator BI = BB->begin();
+ while (isa<DbgInfoIntrinsic>(BI)) ++BI;
+ if (&*BI != RI)
+ return false;
+ }
+
+ /// Only dup the ReturnInst if the CallInst is likely to be emitted as a tail
+ /// call.
+ SmallVector<CallInst*, 4> TailCalls;
+ if (PN) {
+ for (unsigned I = 0, E = PN->getNumIncomingValues(); I != E; ++I) {
+ CallInst *CI = dyn_cast<CallInst>(PN->getIncomingValue(I));
+ // Make sure the phi value is indeed produced by the tail call.
+ if (CI && CI->hasOneUse() && CI->getParent() == PN->getIncomingBlock(I) &&
+ TLI->mayBeEmittedAsTailCall(CI))
+ TailCalls.push_back(CI);
+ }
+ } else {
+ SmallPtrSet<BasicBlock*, 4> VisitedBBs;
+ for (pred_iterator PI = pred_begin(BB), PE = pred_end(BB); PI != PE; ++PI) {
+ if (!VisitedBBs.insert(*PI))
+ continue;
+
+ BasicBlock::InstListType &InstList = (*PI)->getInstList();
+ BasicBlock::InstListType::reverse_iterator RI = InstList.rbegin();
+ BasicBlock::InstListType::reverse_iterator RE = InstList.rend();
+ do { ++RI; } while (RI != RE && isa<DbgInfoIntrinsic>(&*RI));
+ if (RI == RE)
+ continue;
+
+ CallInst *CI = dyn_cast<CallInst>(&*RI);
+ if (CI && CI->use_empty() && TLI->mayBeEmittedAsTailCall(CI))
+ TailCalls.push_back(CI);
+ }
+ }
+
+ bool Changed = false;
+ for (unsigned i = 0, e = TailCalls.size(); i != e; ++i) {
+ CallInst *CI = TailCalls[i];
+ CallSite CS(CI);
+
+ // Conservatively require the attributes of the call to match those of the
+ // return. Ignore noalias because it doesn't affect the call sequence.
+ AttributeSet CalleeAttrs = CS.getAttributes();
+ if (AttrBuilder(CalleeAttrs, AttributeSet::ReturnIndex).
+ removeAttribute(Attribute::NoAlias) !=
+ AttrBuilder(CalleeAttrs, AttributeSet::ReturnIndex).
+ removeAttribute(Attribute::NoAlias))
+ continue;
+
+ // Make sure the call instruction is followed by an unconditional branch to
+ // the return block.
+ BasicBlock *CallBB = CI->getParent();
+ BranchInst *BI = dyn_cast<BranchInst>(CallBB->getTerminator());
+ if (!BI || !BI->isUnconditional() || BI->getSuccessor(0) != BB)
+ continue;
+
+ // Duplicate the return into CallBB.
+ (void)FoldReturnIntoUncondBranch(RI, BB, CallBB);
+ ModifiedDT = Changed = true;
+ ++NumRetsDup;
+ }
+
+ // If we eliminated all predecessors of the block, delete the block now.
+ if (Changed && !BB->hasAddressTaken() && pred_begin(BB) == pred_end(BB))
+ BB->eraseFromParent();
+
+ return Changed;
+}
+
//===----------------------------------------------------------------------===//
-// Addressing Mode Analysis and Optimization
+// Memory Optimization
//===----------------------------------------------------------------------===//
namespace {
- /// ExtAddrMode - This is an extended version of TargetLowering::AddrMode
- /// which holds actual Value*'s for register values.
- struct ExtAddrMode : public TargetLowering::AddrMode {
- Value *BaseReg;
- Value *ScaledReg;
- ExtAddrMode() : BaseReg(0), ScaledReg(0) {}
- void print(OStream &OS) const;
- void dump() const {
- print(cerr);
- cerr << '\n';
- }
- };
-} // end anonymous namespace
-static inline OStream &operator<<(OStream &OS, const ExtAddrMode &AM) {
+/// ExtAddrMode - This is an extended version of TargetLowering::AddrMode
+/// which holds actual Value*'s for register values.
+struct ExtAddrMode : public TargetLowering::AddrMode {
+ Value *BaseReg;
+ Value *ScaledReg;
+ ExtAddrMode() : BaseReg(0), ScaledReg(0) {}
+ void print(raw_ostream &OS) const;
+ void dump() const;
+
+ bool operator==(const ExtAddrMode& O) const {
+ return (BaseReg == O.BaseReg) && (ScaledReg == O.ScaledReg) &&
+ (BaseGV == O.BaseGV) && (BaseOffs == O.BaseOffs) &&
+ (HasBaseReg == O.HasBaseReg) && (Scale == O.Scale);
+ }
+};
+
+static inline raw_ostream &operator<<(raw_ostream &OS, const ExtAddrMode &AM) {
AM.print(OS);
return OS;
}
-void ExtAddrMode::print(OStream &OS) const {
+void ExtAddrMode::print(raw_ostream &OS) const {
bool NeedPlus = false;
OS << "[";
- if (BaseGV)
+ if (BaseGV) {
OS << (NeedPlus ? " + " : "")
- << "GV:%" << BaseGV->getName(), NeedPlus = true;
+ << "GV:";
+ WriteAsOperand(OS, BaseGV, /*PrintType=*/false);
+ NeedPlus = true;
+ }
if (BaseOffs)
OS << (NeedPlus ? " + " : "") << BaseOffs, NeedPlus = true;
- if (BaseReg)
+ if (BaseReg) {
OS << (NeedPlus ? " + " : "")
- << "Base:%" << BaseReg->getName(), NeedPlus = true;
- if (Scale)
+ << "Base:";
+ WriteAsOperand(OS, BaseReg, /*PrintType=*/false);
+ NeedPlus = true;
+ }
+ if (Scale) {
OS << (NeedPlus ? " + " : "")
- << Scale << "*%" << ScaledReg->getName(), NeedPlus = true;
+ << Scale << "*";
+ WriteAsOperand(OS, ScaledReg, /*PrintType=*/false);
+ NeedPlus = true;
+ }
OS << ']';
}
-namespace {
-/// AddressingModeMatcher - This class exposes a single public method, which is
-/// used to construct a "maximal munch" of the addressing mode for the target
-/// specified by TLI for an access to "V" with an access type of AccessTy. This
-/// returns the addressing mode that is actually matched by value, but also
-/// returns the list of instructions involved in that addressing computation in
-/// AddrModeInsts.
+#if !defined(NDEBUG) || defined(LLVM_ENABLE_DUMP)
+void ExtAddrMode::dump() const {
+ print(dbgs());
+ dbgs() << '\n';
+}
+#endif
+
+
+/// \brief A helper class for matching addressing modes.
+///
+/// This encapsulates the logic for matching the target-legal addressing modes.
class AddressingModeMatcher {
SmallVectorImpl<Instruction*> &AddrModeInsts;
const TargetLowering &TLI;
/// AccessTy/MemoryInst - This is the type for the access (e.g. double) and
/// the memory instruction that we're computing this address for.
- const Type *AccessTy;
+ Type *AccessTy;
Instruction *MemoryInst;
/// AddrMode - This is the addressing mode that we're building up. This is
bool IgnoreProfitability;
AddressingModeMatcher(SmallVectorImpl<Instruction*> &AMI,
- const TargetLowering &T, const Type *AT,
+ const TargetLowering &T, Type *AT,
Instruction *MI, ExtAddrMode &AM)
: AddrModeInsts(AMI), TLI(T), AccessTy(AT), MemoryInst(MI), AddrMode(AM) {
IgnoreProfitability = false;
/// Match - Find the maximal addressing mode that a load/store of V can fold,
/// give an access type of AccessTy. This returns a list of involved
/// instructions in AddrModeInsts.
- static ExtAddrMode Match(Value *V, const Type *AccessTy,
+ static ExtAddrMode Match(Value *V, Type *AccessTy,
Instruction *MemoryInst,
SmallVectorImpl<Instruction*> &AddrModeInsts,
const TargetLowering &TLI) {
bool Success =
AddressingModeMatcher(AddrModeInsts, TLI, AccessTy,
MemoryInst, Result).MatchAddr(V, 0);
- Success = Success; assert(Success && "Couldn't select *anything*?");
+ (void)Success; assert(Success && "Couldn't select *anything*?");
return Result;
}
private:
ExtAddrMode &AMAfter);
bool ValueAlreadyLiveAtInst(Value *Val, Value *KnownLive1, Value *KnownLive2);
};
-} // end anonymous namespace
/// MatchScaledValue - Try adding ScaleReg*Scale to the current addressing mode.
/// Return true and update AddrMode if this addr mode is legal for the target,
// Okay, we decided that we can add ScaleReg+Scale to AddrMode. Check now
// to see if ScaleReg is actually X+C. If so, we can turn this into adding
// X*Scale + C*Scale to addr mode.
- ConstantInt *CI; Value *AddLHS;
- if (match(ScaleReg, m_Add(m_Value(AddLHS), m_ConstantInt(CI)))) {
+ ConstantInt *CI = 0; Value *AddLHS = 0;
+ if (isa<Instruction>(ScaleReg) && // not a constant expr.
+ match(ScaleReg, m_Add(m_Value(AddLHS), m_ConstantInt(CI)))) {
TestAddrMode.ScaledReg = AddLHS;
TestAddrMode.BaseOffs += CI->getSExtValue()*TestAddrMode.Scale;
// Don't touch identity bitcasts.
if (I->getType() == I->getOperand(0)->getType())
return false;
- return isa<PointerType>(I->getType()) || isa<IntegerType>(I->getType());
+ return I->getType()->isPointerTy() || I->getType()->isIntegerTy();
case Instruction::PtrToInt:
// PtrToInt is always a noop, as we know that the int type is pointer sized.
return true;
}
}
-
/// MatchOperationAddr - Given an instruction or constant expr, see if we can
/// fold the operation into the addressing mode. If so, update the addressing
/// mode and return true, otherwise return false without modifying AddrMode.
case Instruction::BitCast:
// BitCast is always a noop, and we can handle it as long as it is
// int->int or pointer->pointer (we don't want int<->fp or something).
- if ((isa<PointerType>(AddrInst->getOperand(0)->getType()) ||
- isa<IntegerType>(AddrInst->getOperand(0)->getType())) &&
+ if ((AddrInst->getOperand(0)->getType()->isPointerTy() ||
+ AddrInst->getOperand(0)->getType()->isIntegerTy()) &&
// Don't touch identity bitcasts. These were probably put here by LSR,
// and we don't want to mess around with them. Assume it knows what it
// is doing.
if (!RHS) return false;
int64_t Scale = RHS->getSExtValue();
if (Opcode == Instruction::Shl)
- Scale = 1 << Scale;
+ Scale = 1LL << Scale;
return MatchScaledValue(AddrInst->getOperand(0), Scale, Depth);
}
unsigned VariableScale = 0;
int64_t ConstantOffset = 0;
- const TargetData *TD = TLI.getTargetData();
+ const DataLayout *TD = TLI.getDataLayout();
gep_type_iterator GTI = gep_type_begin(AddrInst);
for (unsigned i = 1, e = AddrInst->getNumOperands(); i != e; ++i, ++GTI) {
- if (const StructType *STy = dyn_cast<StructType>(*GTI)) {
+ if (StructType *STy = dyn_cast<StructType>(*GTI)) {
const StructLayout *SL = TD->getStructLayout(STy);
unsigned Idx =
cast<ConstantInt>(AddrInst->getOperand(i))->getZExtValue();
ConstantOffset += SL->getElementOffset(Idx);
} else {
- uint64_t TypeSize = TD->getABITypeSize(GTI.getIndexedType());
+ uint64_t TypeSize = TD->getTypeAllocSize(GTI.getIndexedType());
if (ConstantInt *CI = dyn_cast<ConstantInt>(AddrInst->getOperand(i))) {
ConstantOffset += CI->getSExtValue()*TypeSize;
} else if (TypeSize) { // Scales of zero don't do anything.
// Save the valid addressing mode in case we can't match.
ExtAddrMode BackupAddrMode = AddrMode;
-
- // Check that this has no base reg yet. If so, we won't have a place to
- // put the base of the GEP (assuming it is not a null ptr).
- bool SetBaseReg = true;
- if (isa<ConstantPointerNull>(AddrInst->getOperand(0)))
- SetBaseReg = false; // null pointer base doesn't need representation.
- else if (AddrMode.HasBaseReg)
- return false; // Base register already specified, can't match GEP.
- else {
- // Otherwise, we'll use the GEP base as the BaseReg.
+ unsigned OldSize = AddrModeInsts.size();
+
+ // See if the scale and offset amount is valid for this target.
+ AddrMode.BaseOffs += ConstantOffset;
+
+ // Match the base operand of the GEP.
+ if (!MatchAddr(AddrInst->getOperand(0), Depth+1)) {
+ // If it couldn't be matched, just stuff the value in a register.
+ if (AddrMode.HasBaseReg) {
+ AddrMode = BackupAddrMode;
+ AddrModeInsts.resize(OldSize);
+ return false;
+ }
AddrMode.HasBaseReg = true;
AddrMode.BaseReg = AddrInst->getOperand(0);
}
-
- // See if the scale and offset amount is valid for this target.
- AddrMode.BaseOffs += ConstantOffset;
-
+
+ // Match the remaining variable portion of the GEP.
if (!MatchScaledValue(AddrInst->getOperand(VariableOperand), VariableScale,
Depth)) {
+ // If it couldn't be matched, try stuffing the base into a register
+ // instead of matching it, and retrying the match of the scale.
AddrMode = BackupAddrMode;
- return false;
+ AddrModeInsts.resize(OldSize);
+ if (AddrMode.HasBaseReg)
+ return false;
+ AddrMode.HasBaseReg = true;
+ AddrMode.BaseReg = AddrInst->getOperand(0);
+ AddrMode.BaseOffs += ConstantOffset;
+ if (!MatchScaledValue(AddrInst->getOperand(VariableOperand),
+ VariableScale, Depth)) {
+ // If even that didn't work, bail.
+ AddrMode = BackupAddrMode;
+ AddrModeInsts.resize(OldSize);
+ return false;
+ }
}
-
- // If we have a null as the base of the GEP, folding in the constant offset
- // plus variable scale is all we can do.
- if (!SetBaseReg) return true;
-
- // If this match succeeded, we know that we can form an address with the
- // GepBase as the basereg. Match the base pointer of the GEP more
- // aggressively by zeroing out BaseReg and rematching. If the base is
- // (for example) another GEP, this allows merging in that other GEP into
- // the addressing mode we're forming.
- AddrMode.HasBaseReg = false;
- AddrMode.BaseReg = 0;
- bool Success = MatchAddr(AddrInst->getOperand(0), Depth+1);
- assert(Success && "MatchAddr should be able to fill in BaseReg!");
- Success=Success;
+
return true;
}
}
return false;
}
-
/// IsOperandAMemoryOperand - Check to see if all uses of OpVal by the specified
/// inline asm call are due to memory operands. If so, return true, otherwise
/// return false.
static bool IsOperandAMemoryOperand(CallInst *CI, InlineAsm *IA, Value *OpVal,
const TargetLowering &TLI) {
- std::vector<InlineAsm::ConstraintInfo>
- Constraints = IA->ParseConstraints();
-
- unsigned ArgNo = 1; // ArgNo - The operand of the CallInst.
- for (unsigned i = 0, e = Constraints.size(); i != e; ++i) {
- TargetLowering::AsmOperandInfo OpInfo(Constraints[i]);
-
- // Compute the value type for each operand.
- switch (OpInfo.Type) {
- case InlineAsm::isOutput:
- if (OpInfo.isIndirect)
- OpInfo.CallOperandVal = CI->getOperand(ArgNo++);
- break;
- case InlineAsm::isInput:
- OpInfo.CallOperandVal = CI->getOperand(ArgNo++);
- break;
- case InlineAsm::isClobber:
- // Nothing to do.
- break;
- }
+ TargetLowering::AsmOperandInfoVector TargetConstraints = TLI.ParseConstraints(ImmutableCallSite(CI));
+ for (unsigned i = 0, e = TargetConstraints.size(); i != e; ++i) {
+ TargetLowering::AsmOperandInfo &OpInfo = TargetConstraints[i];
// Compute the constraint code and ConstraintType to use.
- TLI.ComputeConstraintToUse(OpInfo, SDValue(),
- OpInfo.ConstraintType == TargetLowering::C_Memory);
-
+ TLI.ComputeConstraintToUse(OpInfo, SDValue());
+
// If this asm operand is our Value*, and if it isn't an indirect memory
// operand, we can't fold it!
if (OpInfo.CallOperandVal == OpVal &&
!OpInfo.isIndirect))
return false;
}
-
+
return true;
}
-
/// FindAllMemoryUses - Recursively walk all the uses of I until we find a
/// memory use. If we find an obviously non-foldable instruction, return true.
/// Add the ultimately found memory instructions to MemoryUses.
// Loop over all the uses, recursively processing them.
for (Value::use_iterator UI = I->use_begin(), E = I->use_end();
UI != E; ++UI) {
- if (LoadInst *LI = dyn_cast<LoadInst>(*UI)) {
+ User *U = *UI;
+
+ if (LoadInst *LI = dyn_cast<LoadInst>(U)) {
MemoryUses.push_back(std::make_pair(LI, UI.getOperandNo()));
continue;
}
- if (StoreInst *SI = dyn_cast<StoreInst>(*UI)) {
- if (UI.getOperandNo() == 0) return true; // Storing addr, not into addr.
- MemoryUses.push_back(std::make_pair(SI, UI.getOperandNo()));
+ if (StoreInst *SI = dyn_cast<StoreInst>(U)) {
+ unsigned opNo = UI.getOperandNo();
+ if (opNo == 0) return true; // Storing addr, not into addr.
+ MemoryUses.push_back(std::make_pair(SI, opNo));
continue;
}
- if (CallInst *CI = dyn_cast<CallInst>(*UI)) {
+ if (CallInst *CI = dyn_cast<CallInst>(U)) {
InlineAsm *IA = dyn_cast<InlineAsm>(CI->getCalledValue());
- if (IA == 0) return true;
+ if (!IA) return true;
// If this is a memory operand, we're cool, otherwise bail out.
if (!IsOperandAMemoryOperand(CI, IA, I, TLI))
continue;
}
- if (FindAllMemoryUses(cast<Instruction>(*UI), MemoryUses, ConsideredInsts,
+ if (FindAllMemoryUses(cast<Instruction>(U), MemoryUses, ConsideredInsts,
TLI))
return true;
}
return false;
}
-
/// ValueAlreadyLiveAtInst - Retrn true if Val is already known to be live at
/// the use site that we're folding it into. If so, there is no cost to
/// include it in the addressing mode. KnownLive1 and KnownLive2 are two values
// Check to see if this value is already used in the memory instruction's
// block. If so, it's already live into the block at the very least, so we
// can reasonably fold it.
- BasicBlock *MemBB = MemoryInst->getParent();
- for (Value::use_iterator UI = Val->use_begin(), E = Val->use_end();
- UI != E; ++UI)
- // We know that uses of arguments and instructions have to be instructions.
- if (cast<Instruction>(*UI)->getParent() == MemBB)
- return true;
-
- return false;
+ return Val->isUsedInBasicBlock(MemoryInst->getParent());
}
-
-
/// IsProfitableToFoldIntoAddressingMode - It is possible for the addressing
/// mode of the machine to fold the specified instruction into a load or store
/// that ultimately uses it. However, the specified instruction has multiple
// Get the access type of this use. If the use isn't a pointer, we don't
// know what it accesses.
Value *Address = User->getOperand(OpNo);
- if (!isa<PointerType>(Address->getType()))
+ if (!Address->getType()->isPointerTy())
return false;
- const Type *AddressAccessTy =
+ Type *AddressAccessTy =
cast<PointerType>(Address->getType())->getElementType();
// Do a match against the root of this address, ignoring profitability. This
MemoryInst, Result);
Matcher.IgnoreProfitability = true;
bool Success = Matcher.MatchAddr(Address, 0);
- Success = Success; assert(Success && "Couldn't select *anything*?");
+ (void)Success; assert(Success && "Couldn't select *anything*?");
// If the match didn't cover I, then it won't be shared by it.
if (std::find(MatchedAddrModeInsts.begin(), MatchedAddrModeInsts.end(),
return true;
}
-
-//===----------------------------------------------------------------------===//
-// Memory Optimization
-//===----------------------------------------------------------------------===//
+} // end anonymous namespace
/// IsNonLocalValue - Return true if the specified values are defined in a
/// different basic block than BB.
return false;
}
-/// OptimizeMemoryInst - Load and Store Instructions have often have
+/// OptimizeMemoryInst - Load and Store Instructions often have
/// addressing modes that can do significant amounts of computation. As such,
/// instruction selection will try to get the load or store to do as much
/// computation as possible for the program. The problem is that isel can only
/// This method is used to optimize both load/store and inline asms with memory
/// operands.
bool CodeGenPrepare::OptimizeMemoryInst(Instruction *MemoryInst, Value *Addr,
- const Type *AccessTy,
- DenseMap<Value*,Value*> &SunkAddrs) {
- // Figure out what addressing mode will be built up for this operation.
+ Type *AccessTy) {
+ Value *Repl = Addr;
+
+ // Try to collapse single-value PHI nodes. This is necessary to undo
+ // unprofitable PRE transformations.
+ SmallVector<Value*, 8> worklist;
+ SmallPtrSet<Value*, 16> Visited;
+ worklist.push_back(Addr);
+
+ // Use a worklist to iteratively look through PHI nodes, and ensure that
+ // the addressing mode obtained from the non-PHI roots of the graph
+ // are equivalent.
+ Value *Consensus = 0;
+ unsigned NumUsesConsensus = 0;
+ bool IsNumUsesConsensusValid = false;
SmallVector<Instruction*, 16> AddrModeInsts;
- ExtAddrMode AddrMode = AddressingModeMatcher::Match(Addr, AccessTy,MemoryInst,
- AddrModeInsts, *TLI);
+ ExtAddrMode AddrMode;
+ while (!worklist.empty()) {
+ Value *V = worklist.back();
+ worklist.pop_back();
+
+ // Break use-def graph loops.
+ if (!Visited.insert(V)) {
+ Consensus = 0;
+ break;
+ }
+
+ // For a PHI node, push all of its incoming values.
+ if (PHINode *P = dyn_cast<PHINode>(V)) {
+ for (unsigned i = 0, e = P->getNumIncomingValues(); i != e; ++i)
+ worklist.push_back(P->getIncomingValue(i));
+ continue;
+ }
+
+ // For non-PHIs, determine the addressing mode being computed.
+ SmallVector<Instruction*, 16> NewAddrModeInsts;
+ ExtAddrMode NewAddrMode =
+ AddressingModeMatcher::Match(V, AccessTy, MemoryInst,
+ NewAddrModeInsts, *TLI);
+
+ // This check is broken into two cases with very similar code to avoid using
+ // getNumUses() as much as possible. Some values have a lot of uses, so
+ // calling getNumUses() unconditionally caused a significant compile-time
+ // regression.
+ if (!Consensus) {
+ Consensus = V;
+ AddrMode = NewAddrMode;
+ AddrModeInsts = NewAddrModeInsts;
+ continue;
+ } else if (NewAddrMode == AddrMode) {
+ if (!IsNumUsesConsensusValid) {
+ NumUsesConsensus = Consensus->getNumUses();
+ IsNumUsesConsensusValid = true;
+ }
+
+ // Ensure that the obtained addressing mode is equivalent to that obtained
+ // for all other roots of the PHI traversal. Also, when choosing one
+ // such root as representative, select the one with the most uses in order
+ // to keep the cost modeling heuristics in AddressingModeMatcher
+ // applicable.
+ unsigned NumUses = V->getNumUses();
+ if (NumUses > NumUsesConsensus) {
+ Consensus = V;
+ NumUsesConsensus = NumUses;
+ AddrModeInsts = NewAddrModeInsts;
+ }
+ continue;
+ }
+
+ Consensus = 0;
+ break;
+ }
+
+ // If the addressing mode couldn't be determined, or if multiple different
+ // ones were determined, bail out now.
+ if (!Consensus) return false;
// Check to see if any of the instructions supersumed by this addr mode are
// non-local to I's BB.
// If all the instructions matched are already in this BB, don't do anything.
if (!AnyNonLocal) {
- DEBUG(cerr << "CGP: Found local addrmode: " << AddrMode << "\n");
+ DEBUG(dbgs() << "CGP: Found local addrmode: " << AddrMode << "\n");
return false;
}
// Insert this computation right after this user. Since our caller is
// scanning from the top of the BB to the bottom, reuse of the expr are
// guaranteed to happen later.
- BasicBlock::iterator InsertPt = MemoryInst;
+ IRBuilder<> Builder(MemoryInst);
// Now that we determined the addressing expression we want to use and know
// that we have to sink it into this block. Check to see if we have already
// computation.
Value *&SunkAddr = SunkAddrs[Addr];
if (SunkAddr) {
- DEBUG(cerr << "CGP: Reusing nonlocal addrmode: " << AddrMode << "\n");
+ DEBUG(dbgs() << "CGP: Reusing nonlocal addrmode: " << AddrMode << " for "
+ << *MemoryInst);
if (SunkAddr->getType() != Addr->getType())
- SunkAddr = new BitCastInst(SunkAddr, Addr->getType(), "tmp", InsertPt);
+ SunkAddr = Builder.CreateBitCast(SunkAddr, Addr->getType());
} else {
- DEBUG(cerr << "CGP: SINKING nonlocal addrmode: " << AddrMode << "\n");
- const Type *IntPtrTy = TLI->getTargetData()->getIntPtrType();
+ DEBUG(dbgs() << "CGP: SINKING nonlocal addrmode: " << AddrMode << " for "
+ << *MemoryInst);
+ Type *IntPtrTy =
+ TLI->getDataLayout()->getIntPtrType(AccessTy->getContext());
Value *Result = 0;
- // Start with the scale value.
+
+ // Start with the base register. Do this first so that subsequent address
+ // matching finds it last, which will prevent it from trying to match it
+ // as the scaled value in case it happens to be a mul. That would be
+ // problematic if we've sunk a different mul for the scale, because then
+ // we'd end up sinking both muls.
+ if (AddrMode.BaseReg) {
+ Value *V = AddrMode.BaseReg;
+ if (V->getType()->isPointerTy())
+ V = Builder.CreatePtrToInt(V, IntPtrTy, "sunkaddr");
+ if (V->getType() != IntPtrTy)
+ V = Builder.CreateIntCast(V, IntPtrTy, /*isSigned=*/true, "sunkaddr");
+ Result = V;
+ }
+
+ // Add the scale value.
if (AddrMode.Scale) {
Value *V = AddrMode.ScaledReg;
if (V->getType() == IntPtrTy) {
// done.
- } else if (isa<PointerType>(V->getType())) {
- V = new PtrToIntInst(V, IntPtrTy, "sunkaddr", InsertPt);
+ } else if (V->getType()->isPointerTy()) {
+ V = Builder.CreatePtrToInt(V, IntPtrTy, "sunkaddr");
} else if (cast<IntegerType>(IntPtrTy)->getBitWidth() <
cast<IntegerType>(V->getType())->getBitWidth()) {
- V = new TruncInst(V, IntPtrTy, "sunkaddr", InsertPt);
+ V = Builder.CreateTrunc(V, IntPtrTy, "sunkaddr");
} else {
- V = new SExtInst(V, IntPtrTy, "sunkaddr", InsertPt);
+ V = Builder.CreateSExt(V, IntPtrTy, "sunkaddr");
}
if (AddrMode.Scale != 1)
- V = BinaryOperator::CreateMul(V, ConstantInt::get(IntPtrTy,
- AddrMode.Scale),
- "sunkaddr", InsertPt);
- Result = V;
- }
-
- // Add in the base register.
- if (AddrMode.BaseReg) {
- Value *V = AddrMode.BaseReg;
- if (V->getType() != IntPtrTy)
- V = new PtrToIntInst(V, IntPtrTy, "sunkaddr", InsertPt);
+ V = Builder.CreateMul(V, ConstantInt::get(IntPtrTy, AddrMode.Scale),
+ "sunkaddr");
if (Result)
- Result = BinaryOperator::CreateAdd(Result, V, "sunkaddr", InsertPt);
+ Result = Builder.CreateAdd(Result, V, "sunkaddr");
else
Result = V;
}
// Add in the BaseGV if present.
if (AddrMode.BaseGV) {
- Value *V = new PtrToIntInst(AddrMode.BaseGV, IntPtrTy, "sunkaddr",
- InsertPt);
+ Value *V = Builder.CreatePtrToInt(AddrMode.BaseGV, IntPtrTy, "sunkaddr");
if (Result)
- Result = BinaryOperator::CreateAdd(Result, V, "sunkaddr", InsertPt);
+ Result = Builder.CreateAdd(Result, V, "sunkaddr");
else
Result = V;
}
if (AddrMode.BaseOffs) {
Value *V = ConstantInt::get(IntPtrTy, AddrMode.BaseOffs);
if (Result)
- Result = BinaryOperator::CreateAdd(Result, V, "sunkaddr", InsertPt);
+ Result = Builder.CreateAdd(Result, V, "sunkaddr");
else
Result = V;
}
if (Result == 0)
SunkAddr = Constant::getNullValue(Addr->getType());
else
- SunkAddr = new IntToPtrInst(Result, Addr->getType(), "sunkaddr",InsertPt);
+ SunkAddr = Builder.CreateIntToPtr(Result, Addr->getType(), "sunkaddr");
}
- MemoryInst->replaceUsesOfWith(Addr, SunkAddr);
+ MemoryInst->replaceUsesOfWith(Repl, SunkAddr);
- if (Addr->use_empty())
- RecursivelyDeleteTriviallyDeadInstructions(Addr);
+ // If we have no uses, recursively delete the value and all dead instructions
+ // using it.
+ if (Repl->use_empty()) {
+ // This can cause recursive deletion, which can invalidate our iterator.
+ // Use a WeakVH to hold onto it in case this happens.
+ WeakVH IterHandle(CurInstIterator);
+ BasicBlock *BB = CurInstIterator->getParent();
+
+ RecursivelyDeleteTriviallyDeadInstructions(Repl, TLInfo);
+
+ if (IterHandle != CurInstIterator) {
+ // If the iterator instruction was recursively deleted, start over at the
+ // start of the block.
+ CurInstIterator = BB->begin();
+ SunkAddrs.clear();
+ }
+ }
+ ++NumMemoryInsts;
return true;
}
/// OptimizeInlineAsmInst - If there are any memory operands, use
/// OptimizeMemoryInst to sink their address computing into the block when
/// possible / profitable.
-bool CodeGenPrepare::OptimizeInlineAsmInst(Instruction *I, CallSite CS,
- DenseMap<Value*,Value*> &SunkAddrs) {
+bool CodeGenPrepare::OptimizeInlineAsmInst(CallInst *CS) {
bool MadeChange = false;
- InlineAsm *IA = cast<InlineAsm>(CS.getCalledValue());
-
- // Do a prepass over the constraints, canonicalizing them, and building up the
- // ConstraintOperands list.
- std::vector<InlineAsm::ConstraintInfo>
- ConstraintInfos = IA->ParseConstraints();
-
- /// ConstraintOperands - Information about all of the constraints.
- std::vector<TargetLowering::AsmOperandInfo> ConstraintOperands;
- unsigned ArgNo = 0; // ArgNo - The argument of the CallInst.
- for (unsigned i = 0, e = ConstraintInfos.size(); i != e; ++i) {
- ConstraintOperands.
- push_back(TargetLowering::AsmOperandInfo(ConstraintInfos[i]));
- TargetLowering::AsmOperandInfo &OpInfo = ConstraintOperands.back();
-
- // Compute the value type for each operand.
- switch (OpInfo.Type) {
- case InlineAsm::isOutput:
- if (OpInfo.isIndirect)
- OpInfo.CallOperandVal = CS.getArgument(ArgNo++);
- break;
- case InlineAsm::isInput:
- OpInfo.CallOperandVal = CS.getArgument(ArgNo++);
- break;
- case InlineAsm::isClobber:
- // Nothing to do.
- break;
- }
+
+ TargetLowering::AsmOperandInfoVector
+ TargetConstraints = TLI->ParseConstraints(CS);
+ unsigned ArgNo = 0;
+ for (unsigned i = 0, e = TargetConstraints.size(); i != e; ++i) {
+ TargetLowering::AsmOperandInfo &OpInfo = TargetConstraints[i];
// Compute the constraint code and ConstraintType to use.
- TLI->ComputeConstraintToUse(OpInfo, SDValue(),
- OpInfo.ConstraintType == TargetLowering::C_Memory);
+ TLI->ComputeConstraintToUse(OpInfo, SDValue());
if (OpInfo.ConstraintType == TargetLowering::C_Memory &&
OpInfo.isIndirect) {
- Value *OpVal = OpInfo.CallOperandVal;
- MadeChange |= OptimizeMemoryInst(I, OpVal, OpVal->getType(), SunkAddrs);
- }
+ Value *OpVal = CS->getArgOperand(ArgNo++);
+ MadeChange |= OptimizeMemoryInst(CS, OpVal, OpVal->getType());
+ } else if (OpInfo.Type == InlineAsm::isInput)
+ ArgNo++;
}
return MadeChange;
}
+/// MoveExtToFormExtLoad - Move a zext or sext fed by a load into the same
+/// basic block as the load, unless conditions are unfavorable. This allows
+/// SelectionDAG to fold the extend into the load.
+///
+bool CodeGenPrepare::MoveExtToFormExtLoad(Instruction *I) {
+ // Look for a load being extended.
+ LoadInst *LI = dyn_cast<LoadInst>(I->getOperand(0));
+ if (!LI) return false;
+
+ // If they're already in the same block, there's nothing to do.
+ if (LI->getParent() == I->getParent())
+ return false;
+
+ // If the load has other users and the truncate is not free, this probably
+ // isn't worthwhile.
+ if (!LI->hasOneUse() &&
+ TLI && (TLI->isTypeLegal(TLI->getValueType(LI->getType())) ||
+ !TLI->isTypeLegal(TLI->getValueType(I->getType()))) &&
+ !TLI->isTruncateFree(I->getType(), LI->getType()))
+ return false;
+
+ // Check whether the target supports casts folded into loads.
+ unsigned LType;
+ if (isa<ZExtInst>(I))
+ LType = ISD::ZEXTLOAD;
+ else {
+ assert(isa<SExtInst>(I) && "Unexpected ext type!");
+ LType = ISD::SEXTLOAD;
+ }
+ if (TLI && !TLI->isLoadExtLegal(LType, TLI->getValueType(LI->getType())))
+ return false;
+
+ // Move the extend into the same block as the load, so that SelectionDAG
+ // can fold it.
+ I->removeFromParent();
+ I->insertAfter(LI);
+ ++NumExtsMoved;
+ return true;
+}
+
bool CodeGenPrepare::OptimizeExtUses(Instruction *I) {
BasicBlock *DefBB = I->getParent();
- // If both result of the {s|z}xt and its source are live out, rewrite all
+ // If the result of a {s|z}ext and its source are both live out, rewrite all
// other uses of the source with result of extension.
Value *Src = I->getOperand(0);
if (Src->hasOneUse())
if (!DefIsLiveOut)
return false;
- // Make sure non of the uses are PHI nodes.
+ // Make sure none of the uses are PHI nodes.
for (Value::use_iterator UI = Src->use_begin(), E = Src->use_end();
UI != E; ++UI) {
Instruction *User = cast<Instruction>(*UI);
Instruction *&InsertedTrunc = InsertedTruncs[UserBB];
if (!InsertedTrunc) {
- BasicBlock::iterator InsertPt = UserBB->getFirstNonPHI();
-
+ BasicBlock::iterator InsertPt = UserBB->getFirstInsertionPt();
InsertedTrunc = new TruncInst(I, Src->getType(), "", InsertPt);
}
// Replace a use of the {s|z}ext source with a use of the result.
TheUse = InsertedTrunc;
-
+ ++NumExtUses;
MadeChange = true;
}
return MadeChange;
}
+/// isFormingBranchFromSelectProfitable - Returns true if a SelectInst should be
+/// turned into an explicit branch.
+static bool isFormingBranchFromSelectProfitable(SelectInst *SI) {
+ // FIXME: This should use the same heuristics as IfConversion to determine
+ // whether a select is better represented as a branch. This requires that
+ // branch probability metadata is preserved for the select, which is not the
+ // case currently.
+
+ CmpInst *Cmp = dyn_cast<CmpInst>(SI->getCondition());
+
+ // If the branch is predicted right, an out of order CPU can avoid blocking on
+ // the compare. Emit cmovs on compares with a memory operand as branches to
+ // avoid stalls on the load from memory. If the compare has more than one use
+ // there's probably another cmov or setcc around so it's not worth emitting a
+ // branch.
+ if (!Cmp)
+ return false;
+
+ Value *CmpOp0 = Cmp->getOperand(0);
+ Value *CmpOp1 = Cmp->getOperand(1);
+
+ // We check that the memory operand has one use to avoid uses of the loaded
+ // value directly after the compare, making branches unprofitable.
+ return Cmp->hasOneUse() &&
+ ((isa<LoadInst>(CmpOp0) && CmpOp0->hasOneUse()) ||
+ (isa<LoadInst>(CmpOp1) && CmpOp1->hasOneUse()));
+}
+
+
+/// If we have a SelectInst that will likely profit from branch prediction,
+/// turn it into a branch.
+bool CodeGenPrepare::OptimizeSelectInst(SelectInst *SI) {
+ bool VectorCond = !SI->getCondition()->getType()->isIntegerTy(1);
+
+ // Can we convert the 'select' to CF ?
+ if (DisableSelectToBranch || OptSize || !TLI || VectorCond)
+ return false;
+
+ TargetLowering::SelectSupportKind SelectKind;
+ if (VectorCond)
+ SelectKind = TargetLowering::VectorMaskSelect;
+ else if (SI->getType()->isVectorTy())
+ SelectKind = TargetLowering::ScalarCondVectorVal;
+ else
+ SelectKind = TargetLowering::ScalarValSelect;
+
+ // Do we have efficient codegen support for this kind of 'selects' ?
+ if (TLI->isSelectSupported(SelectKind)) {
+ // We have efficient codegen support for the select instruction.
+ // Check if it is profitable to keep this 'select'.
+ if (!TLI->isPredictableSelectExpensive() ||
+ !isFormingBranchFromSelectProfitable(SI))
+ return false;
+ }
+
+ ModifiedDT = true;
+
+ // First, we split the block containing the select into 2 blocks.
+ BasicBlock *StartBlock = SI->getParent();
+ BasicBlock::iterator SplitPt = ++(BasicBlock::iterator(SI));
+ BasicBlock *NextBlock = StartBlock->splitBasicBlock(SplitPt, "select.end");
+
+ // Create a new block serving as the landing pad for the branch.
+ BasicBlock *SmallBlock = BasicBlock::Create(SI->getContext(), "select.mid",
+ NextBlock->getParent(), NextBlock);
+
+ // Move the unconditional branch from the block with the select in it into our
+ // landing pad block.
+ StartBlock->getTerminator()->eraseFromParent();
+ BranchInst::Create(NextBlock, SmallBlock);
+
+ // Insert the real conditional branch based on the original condition.
+ BranchInst::Create(NextBlock, SmallBlock, SI->getCondition(), SI);
+
+ // The select itself is replaced with a PHI Node.
+ PHINode *PN = PHINode::Create(SI->getType(), 2, "", NextBlock->begin());
+ PN->takeName(SI);
+ PN->addIncoming(SI->getTrueValue(), StartBlock);
+ PN->addIncoming(SI->getFalseValue(), SmallBlock);
+ SI->replaceAllUsesWith(PN);
+ SI->eraseFromParent();
+
+ // Instruct OptimizeBlock to skip to the next block.
+ CurInstIterator = StartBlock->end();
+ ++NumSelectsExpanded;
+ return true;
+}
+
+bool CodeGenPrepare::OptimizeInst(Instruction *I) {
+ if (PHINode *P = dyn_cast<PHINode>(I)) {
+ // It is possible for very late stage optimizations (such as SimplifyCFG)
+ // to introduce PHI nodes too late to be cleaned up. If we detect such a
+ // trivial PHI, go ahead and zap it here.
+ if (Value *V = SimplifyInstruction(P)) {
+ P->replaceAllUsesWith(V);
+ P->eraseFromParent();
+ ++NumPHIsElim;
+ return true;
+ }
+ return false;
+ }
+
+ if (CastInst *CI = dyn_cast<CastInst>(I)) {
+ // If the source of the cast is a constant, then this should have
+ // already been constant folded. The only reason NOT to constant fold
+ // it is if something (e.g. LSR) was careful to place the constant
+ // evaluation in a block other than then one that uses it (e.g. to hoist
+ // the address of globals out of a loop). If this is the case, we don't
+ // want to forward-subst the cast.
+ if (isa<Constant>(CI->getOperand(0)))
+ return false;
+
+ if (TLI && OptimizeNoopCopyExpression(CI, *TLI))
+ return true;
+
+ if (isa<ZExtInst>(I) || isa<SExtInst>(I)) {
+ bool MadeChange = MoveExtToFormExtLoad(I);
+ return MadeChange | OptimizeExtUses(I);
+ }
+ return false;
+ }
+
+ if (CmpInst *CI = dyn_cast<CmpInst>(I))
+ return OptimizeCmpExpression(CI);
+
+ if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
+ if (TLI)
+ return OptimizeMemoryInst(I, I->getOperand(0), LI->getType());
+ return false;
+ }
+
+ if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
+ if (TLI)
+ return OptimizeMemoryInst(I, SI->getOperand(1),
+ SI->getOperand(0)->getType());
+ return false;
+ }
+
+ if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(I)) {
+ if (GEPI->hasAllZeroIndices()) {
+ /// The GEP operand must be a pointer, so must its result -> BitCast
+ Instruction *NC = new BitCastInst(GEPI->getOperand(0), GEPI->getType(),
+ GEPI->getName(), GEPI);
+ GEPI->replaceAllUsesWith(NC);
+ GEPI->eraseFromParent();
+ ++NumGEPsElim;
+ OptimizeInst(NC);
+ return true;
+ }
+ return false;
+ }
+
+ if (CallInst *CI = dyn_cast<CallInst>(I))
+ return OptimizeCallInst(CI);
+
+ if (SelectInst *SI = dyn_cast<SelectInst>(I))
+ return OptimizeSelectInst(SI);
+
+ return false;
+}
+
// In this pass we look for GEP and cast instructions that are used
// across basic blocks and rewrite them to improve basic-block-at-a-time
// selection.
bool CodeGenPrepare::OptimizeBlock(BasicBlock &BB) {
+ SunkAddrs.clear();
bool MadeChange = false;
- // Split all critical edges where the dest block has a PHI.
- TerminatorInst *BBTI = BB.getTerminator();
- if (BBTI->getNumSuccessors() > 1) {
- for (unsigned i = 0, e = BBTI->getNumSuccessors(); i != e; ++i) {
- BasicBlock *SuccBB = BBTI->getSuccessor(i);
- if (isa<PHINode>(SuccBB->begin()) && isCriticalEdge(BBTI, i, true))
- SplitEdgeNicely(BBTI, i, BackEdges, this);
- }
- }
+ CurInstIterator = BB.begin();
+ while (CurInstIterator != BB.end())
+ MadeChange |= OptimizeInst(CurInstIterator++);
- // Keep track of non-local addresses that have been sunk into this block.
- // This allows us to avoid inserting duplicate code for blocks with multiple
- // load/stores of the same address.
- DenseMap<Value*, Value*> SunkAddrs;
-
- for (BasicBlock::iterator BBI = BB.begin(), E = BB.end(); BBI != E; ) {
- Instruction *I = BBI++;
-
- if (CastInst *CI = dyn_cast<CastInst>(I)) {
- // If the source of the cast is a constant, then this should have
- // already been constant folded. The only reason NOT to constant fold
- // it is if something (e.g. LSR) was careful to place the constant
- // evaluation in a block other than then one that uses it (e.g. to hoist
- // the address of globals out of a loop). If this is the case, we don't
- // want to forward-subst the cast.
- if (isa<Constant>(CI->getOperand(0)))
- continue;
+ MadeChange |= DupRetToEnableTailCallOpts(&BB);
+
+ return MadeChange;
+}
- bool Change = false;
- if (TLI) {
- Change = OptimizeNoopCopyExpression(CI, *TLI);
- MadeChange |= Change;
+// llvm.dbg.value is far away from the value then iSel may not be able
+// handle it properly. iSel will drop llvm.dbg.value if it can not
+// find a node corresponding to the value.
+bool CodeGenPrepare::PlaceDbgValues(Function &F) {
+ bool MadeChange = false;
+ for (Function::iterator I = F.begin(), E = F.end(); I != E; ++I) {
+ Instruction *PrevNonDbgInst = NULL;
+ for (BasicBlock::iterator BI = I->begin(), BE = I->end(); BI != BE;) {
+ Instruction *Insn = BI; ++BI;
+ DbgValueInst *DVI = dyn_cast<DbgValueInst>(Insn);
+ if (!DVI) {
+ PrevNonDbgInst = Insn;
+ continue;
}
- if (!Change && (isa<ZExtInst>(I) || isa<SExtInst>(I)))
- MadeChange |= OptimizeExtUses(I);
- } else if (CmpInst *CI = dyn_cast<CmpInst>(I)) {
- MadeChange |= OptimizeCmpExpression(CI);
- } else if (LoadInst *LI = dyn_cast<LoadInst>(I)) {
- if (TLI)
- MadeChange |= OptimizeMemoryInst(I, I->getOperand(0), LI->getType(),
- SunkAddrs);
- } else if (StoreInst *SI = dyn_cast<StoreInst>(I)) {
- if (TLI)
- MadeChange |= OptimizeMemoryInst(I, SI->getOperand(1),
- SI->getOperand(0)->getType(),
- SunkAddrs);
- } else if (GetElementPtrInst *GEPI = dyn_cast<GetElementPtrInst>(I)) {
- if (GEPI->hasAllZeroIndices()) {
- /// The GEP operand must be a pointer, so must its result -> BitCast
- Instruction *NC = new BitCastInst(GEPI->getOperand(0), GEPI->getType(),
- GEPI->getName(), GEPI);
- GEPI->replaceAllUsesWith(NC);
- GEPI->eraseFromParent();
+ Instruction *VI = dyn_cast_or_null<Instruction>(DVI->getValue());
+ if (VI && VI != PrevNonDbgInst && !VI->isTerminator()) {
+ DEBUG(dbgs() << "Moving Debug Value before :\n" << *DVI << ' ' << *VI);
+ DVI->removeFromParent();
+ if (isa<PHINode>(VI))
+ DVI->insertBefore(VI->getParent()->getFirstInsertionPt());
+ else
+ DVI->insertAfter(VI);
MadeChange = true;
- BBI = NC;
+ ++NumDbgValueMoved;
}
- } else if (CallInst *CI = dyn_cast<CallInst>(I)) {
- // If we found an inline asm expession, and if the target knows how to
- // lower it to normal LLVM code, do so now.
- if (TLI && isa<InlineAsm>(CI->getCalledValue()))
- if (const TargetAsmInfo *TAI =
- TLI->getTargetMachine().getTargetAsmInfo()) {
- if (TAI->ExpandInlineAsm(CI))
- BBI = BB.begin();
- else
- // Sink address computing for memory operands into the block.
- MadeChange |= OptimizeInlineAsmInst(I, &(*CI), SunkAddrs);
- }
}
}
-
return MadeChange;
}