X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTransforms%2FUtils%2FCloneFunction.cpp;h=be8d39e128a53e7d3f118f2124fd43417c028490;hb=defce4cfd607e013169033164def7a3e558bfd54;hp=1b28c35238e9ca264324d3506dd1a7e4a2e6548c;hpb=b7febfbafae97bce00bd239d34e91b9997a5338d;p=oota-llvm.git diff --git a/lib/Transforms/Utils/CloneFunction.cpp b/lib/Transforms/Utils/CloneFunction.cpp index 1b28c35238e..be8d39e128a 100644 --- a/lib/Transforms/Utils/CloneFunction.cpp +++ b/lib/Transforms/Utils/CloneFunction.cpp @@ -14,19 +14,22 @@ //===----------------------------------------------------------------------===// #include "llvm/Transforms/Utils/Cloning.h" -#include "llvm/Constants.h" -#include "llvm/DerivedTypes.h" -#include "llvm/Instructions.h" -#include "llvm/IntrinsicInst.h" -#include "llvm/GlobalVariable.h" -#include "llvm/Function.h" -#include "llvm/LLVMContext.h" -#include "llvm/Metadata.h" +#include "llvm/ADT/SmallVector.h" +#include "llvm/Analysis/ConstantFolding.h" +#include "llvm/Analysis/InstructionSimplify.h" +#include "llvm/DebugInfo.h" +#include "llvm/IR/Constants.h" +#include "llvm/IR/DerivedTypes.h" +#include "llvm/IR/Function.h" +#include "llvm/IR/GlobalVariable.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/IntrinsicInst.h" +#include "llvm/IR/LLVMContext.h" +#include "llvm/IR/Metadata.h" #include "llvm/Support/CFG.h" +#include "llvm/Transforms/Utils/BasicBlockUtils.h" +#include "llvm/Transforms/Utils/Local.h" #include "llvm/Transforms/Utils/ValueMapper.h" -#include "llvm/Analysis/ConstantFolding.h" -#include "llvm/Analysis/DebugInfo.h" -#include "llvm/ADT/SmallVector.h" #include using namespace llvm; @@ -84,24 +87,26 @@ void llvm::CloneFunctionInto(Function *NewFunc, const Function *OldFunc, assert(VMap.count(I) && "No mapping from source argument specified!"); #endif - // Clone any attributes. - if (NewFunc->arg_size() == OldFunc->arg_size()) - NewFunc->copyAttributesFrom(OldFunc); - else { - //Some arguments were deleted with the VMap. Copy arguments one by one - for (Function::const_arg_iterator I = OldFunc->arg_begin(), - E = OldFunc->arg_end(); I != E; ++I) - if (Argument* Anew = dyn_cast(VMap[I])) - Anew->addAttr( OldFunc->getAttributes() - .getParamAttributes(I->getArgNo() + 1)); - NewFunc->setAttributes(NewFunc->getAttributes() - .addAttr(0, OldFunc->getAttributes() - .getRetAttributes())); - NewFunc->setAttributes(NewFunc->getAttributes() - .addAttr(~0, OldFunc->getAttributes() - .getFnAttributes())); + AttributeSet OldAttrs = OldFunc->getAttributes(); + // Clone any argument attributes that are present in the VMap. + for (Function::const_arg_iterator I = OldFunc->arg_begin(), + E = OldFunc->arg_end(); + I != E; ++I) + if (Argument *Anew = dyn_cast(VMap[I])) { + AttributeSet attrs = + OldAttrs.getParamAttributes(I->getArgNo() + 1); + if (attrs.getNumSlots() > 0) + Anew->addAttr(attrs); + } - } + NewFunc->setAttributes(NewFunc->getAttributes() + .addAttributes(NewFunc->getContext(), + AttributeSet::ReturnIndex, + OldAttrs.getRetAttributes())); + NewFunc->setAttributes(NewFunc->getAttributes() + .addAttributes(NewFunc->getContext(), + AttributeSet::FunctionIndex, + OldAttrs.getFnAttributes())); // Loop over all of the basic blocks in the function, cloning them as // appropriate. Note that we save BE this way in order to handle cloning of @@ -197,32 +202,25 @@ namespace { const Function *OldFunc; ValueToValueMapTy &VMap; bool ModuleLevelChanges; - SmallVectorImpl &Returns; const char *NameSuffix; ClonedCodeInfo *CodeInfo; - const TargetData *TD; + const DataLayout *TD; public: PruningFunctionCloner(Function *newFunc, const Function *oldFunc, ValueToValueMapTy &valueMap, bool moduleLevelChanges, - SmallVectorImpl &returns, const char *nameSuffix, ClonedCodeInfo *codeInfo, - const TargetData *td) + const DataLayout *td) : NewFunc(newFunc), OldFunc(oldFunc), VMap(valueMap), ModuleLevelChanges(moduleLevelChanges), - Returns(returns), NameSuffix(nameSuffix), CodeInfo(codeInfo), TD(td) { + NameSuffix(nameSuffix), CodeInfo(codeInfo), TD(td) { } /// CloneBlock - The specified block is found to be reachable, clone it and /// anything that it can reach. void CloneBlock(const BasicBlock *BB, std::vector &ToClone); - - public: - /// ConstantFoldMappedInstruction - Constant fold the specified instruction, - /// mapping its operands through VMap if they are available. - Constant *ConstantFoldMappedInstruction(const Instruction *I); }; } @@ -230,7 +228,7 @@ namespace { /// anything that it can reach. void PruningFunctionCloner::CloneBlock(const BasicBlock *BB, std::vector &ToClone){ - TrackingVH &BBEntry = VMap[BB]; + WeakVH &BBEntry = VMap[BB]; // Have we already cloned this block? if (BBEntry) return; @@ -262,19 +260,33 @@ void PruningFunctionCloner::CloneBlock(const BasicBlock *BB, // loop doesn't include the terminator. for (BasicBlock::const_iterator II = BB->begin(), IE = --BB->end(); II != IE; ++II) { - // If this instruction constant folds, don't bother cloning the instruction, - // instead, just add the constant to the value map. - if (Constant *C = ConstantFoldMappedInstruction(II)) { - VMap[II] = C; - continue; + Instruction *NewInst = II->clone(); + + // Eagerly remap operands to the newly cloned instruction, except for PHI + // nodes for which we defer processing until we update the CFG. + if (!isa(NewInst)) { + RemapInstruction(NewInst, VMap, + ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges); + + // If we can simplify this instruction to some other value, simply add + // a mapping to that value rather than inserting a new instruction into + // the basic block. + if (Value *V = SimplifyInstruction(NewInst, TD)) { + // On the off-chance that this simplifies to an instruction in the old + // function, map it back into the new function. + if (Value *MappedV = VMap.lookup(V)) + V = MappedV; + + VMap[II] = V; + delete NewInst; + continue; + } } - Instruction *NewInst = II->clone(); if (II->hasName()) NewInst->setName(II->getName()+NameSuffix); - NewBB->getInstList().push_back(NewInst); VMap[II] = NewInst; // Add instruction map to value. - + NewBB->getInstList().push_back(NewInst); hasCalls |= (isa(II) && !isa(II)); if (const AllocaInst *AI = dyn_cast(II)) { if (isa(AI->getArraySize())) @@ -340,33 +352,6 @@ void PruningFunctionCloner::CloneBlock(const BasicBlock *BB, CodeInfo->ContainsDynamicAllocas |= hasStaticAllocas && BB != &BB->getParent()->front(); } - - if (ReturnInst *RI = dyn_cast(NewBB->getTerminator())) - Returns.push_back(RI); -} - -/// ConstantFoldMappedInstruction - Constant fold the specified instruction, -/// mapping its operands through VMap if they are available. -Constant *PruningFunctionCloner:: -ConstantFoldMappedInstruction(const Instruction *I) { - SmallVector Ops; - for (unsigned i = 0, e = I->getNumOperands(); i != e; ++i) - if (Constant *Op = dyn_cast_or_null(MapValue(I->getOperand(i), - VMap, - ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges))) - Ops.push_back(Op); - else - return 0; // All operands not constant! - - if (const CmpInst *CI = dyn_cast(I)) - return ConstantFoldCompareInstOperands(CI->getPredicate(), Ops[0], Ops[1], - TD); - - if (const LoadInst *LI = dyn_cast(I)) - if (!LI->isVolatile()) - return ConstantFoldLoadFromConstPtr(Ops[0], TD); - - return ConstantFoldInstOperands(I->getOpcode(), I->getType(), Ops, TD); } /// CloneAndPruneFunctionInto - This works exactly like CloneFunctionInto, @@ -382,7 +367,7 @@ void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc, SmallVectorImpl &Returns, const char *NameSuffix, ClonedCodeInfo *CodeInfo, - const TargetData *TD, + const DataLayout *TD, Instruction *TheCall) { assert(NameSuffix && "NameSuffix cannot be null!"); @@ -393,7 +378,7 @@ void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc, #endif PruningFunctionCloner PFC(NewFunc, OldFunc, VMap, ModuleLevelChanges, - Returns, NameSuffix, CodeInfo, TD); + NameSuffix, CodeInfo, TD); // Clone the entry block, and anything recursively reachable from it. std::vector CloneWorklist; @@ -418,25 +403,19 @@ void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc, // Add the new block to the new function. NewFunc->getBasicBlockList().push_back(NewBB); - - // Loop over all of the instructions in the block, fixing up operand - // references as we go. This uses VMap to do all the hard work. - // - BasicBlock::iterator I = NewBB->begin(); // Handle PHI nodes specially, as we have to remove references to dead // blocks. - if (PHINode *PN = dyn_cast(I)) { - // Skip over all PHI nodes, remembering them for later. - BasicBlock::const_iterator OldI = BI->begin(); - for (; (PN = dyn_cast(I)); ++I, ++OldI) - PHIToResolve.push_back(cast(OldI)); - } - - // Otherwise, remap the rest of the instructions normally. - for (; I != NewBB->end(); ++I) - RemapInstruction(I, VMap, - ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges); + for (BasicBlock::const_iterator I = BI->begin(), E = BI->end(); I != E; ++I) + if (const PHINode *PN = dyn_cast(I)) + PHIToResolve.push_back(PN); + else + break; + + // Finally, remap the terminator instructions, as those can't be remapped + // until all BBs are mapped. + RemapInstruction(NewBB->getTerminator(), VMap, + ModuleLevelChanges ? RF_None : RF_NoModuleLevelChanges); } // Defer PHI resolution until rest of function is resolved, PHI resolution @@ -518,31 +497,55 @@ void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc, ++OldI; } } - // NOTE: We cannot eliminate single entry phi nodes here, because of - // VMap. Single entry phi nodes can have multiple VMap entries - // pointing at them. Thus, deleting one would require scanning the VMap - // to update any entries in it that would require that. This would be - // really slow. } - + + // Make a second pass over the PHINodes now that all of them have been + // remapped into the new function, simplifying the PHINode and performing any + // recursive simplifications exposed. This will transparently update the + // WeakVH in the VMap. Notably, we rely on that so that if we coalesce + // two PHINodes, the iteration over the old PHIs remains valid, and the + // mapping will just map us to the new node (which may not even be a PHI + // node). + for (unsigned Idx = 0, Size = PHIToResolve.size(); Idx != Size; ++Idx) + if (PHINode *PN = dyn_cast(VMap[PHIToResolve[Idx]])) + recursivelySimplifyInstruction(PN, TD); + // Now that the inlined function body has been fully constructed, go through // and zap unconditional fall-through branches. This happen all the time when // specializing code: code specialization turns conditional branches into // uncond branches, and this code folds them. - Function::iterator I = cast(VMap[&OldFunc->getEntryBlock()]); + Function::iterator Begin = cast(VMap[&OldFunc->getEntryBlock()]); + Function::iterator I = Begin; while (I != NewFunc->end()) { + // Check if this block has become dead during inlining or other + // simplifications. Note that the first block will appear dead, as it has + // not yet been wired up properly. + if (I != Begin && (pred_begin(I) == pred_end(I) || + I->getSinglePredecessor() == I)) { + BasicBlock *DeadBB = I++; + DeleteDeadBlock(DeadBB); + continue; + } + + // We need to simplify conditional branches and switches with a constant + // operand. We try to prune these out when cloning, but if the + // simplification required looking through PHI nodes, those are only + // available after forming the full basic block. That may leave some here, + // and we still want to prune the dead code as early as possible. + ConstantFoldTerminator(I); + BranchInst *BI = dyn_cast(I->getTerminator()); if (!BI || BI->isConditional()) { ++I; continue; } - // Note that we can't eliminate uncond branches if the destination has - // single-entry PHI nodes. Eliminating the single-entry phi nodes would - // require scanning the VMap to update any entries that point to the phi - // node. BasicBlock *Dest = BI->getSuccessor(0); - if (!Dest->getSinglePredecessor() || isa(Dest->begin())) { + if (!Dest->getSinglePredecessor()) { ++I; continue; } - + + // We shouldn't be able to get single-entry PHI nodes here, as instsimplify + // above should have zapped all of them.. + assert(!isa(Dest->begin())); + // We know all single-entry PHI nodes in the inlined function have been // removed, so we just need to splice the blocks. BI->eraseFromParent(); @@ -558,4 +561,13 @@ void llvm::CloneAndPruneFunctionInto(Function *NewFunc, const Function *OldFunc, // Do not increment I, iteratively merge all things this block branches to. } + + // Make a final pass over the basic blocks from theh old function to gather + // any return instructions which survived folding. We have to do this here + // because we can iteratively remove and merge returns above. + for (Function::iterator I = cast(VMap[&OldFunc->getEntryBlock()]), + E = NewFunc->end(); + I != E; ++I) + if (ReturnInst *RI = dyn_cast(I->getTerminator())) + Returns.push_back(RI); }