#define DEBUG_TYPE "argpromotion"
#include "llvm/Transforms/IPO.h"
-#include "llvm/Constants.h"
-#include "llvm/DerivedTypes.h"
-#include "llvm/Module.h"
-#include "llvm/CallGraphSCCPass.h"
-#include "llvm/Instructions.h"
-#include "llvm/LLVMContext.h"
+#include "llvm/ADT/DepthFirstIterator.h"
+#include "llvm/ADT/Statistic.h"
+#include "llvm/ADT/StringExtras.h"
#include "llvm/Analysis/AliasAnalysis.h"
#include "llvm/Analysis/CallGraph.h"
-#include "llvm/Support/CallSite.h"
+#include "llvm/CallGraphSCCPass.h"
+#include "llvm/IR/Constants.h"
+#include "llvm/IR/DerivedTypes.h"
+#include "llvm/IR/Instructions.h"
+#include "llvm/IR/LLVMContext.h"
+#include "llvm/IR/Module.h"
#include "llvm/Support/CFG.h"
+#include "llvm/Support/CallSite.h"
#include "llvm/Support/Debug.h"
#include "llvm/Support/raw_ostream.h"
-#include "llvm/ADT/DepthFirstIterator.h"
-#include "llvm/ADT/Statistic.h"
-#include "llvm/ADT/StringExtras.h"
#include <set>
using namespace llvm;
if (PointerArgs.empty()) return 0;
// Second check: make sure that all callers are direct callers. We can't
- // transform functions that have indirect callers.
- if (F->hasAddressTaken())
- return 0;
-
+ // transform functions that have indirect callers. Also see if the function
+ // is self-recursive.
+ bool isSelfRecursive = false;
+ for (Value::use_iterator UI = F->use_begin(), E = F->use_end();
+ UI != E; ++UI) {
+ CallSite CS(*UI);
+ // Must be a direct call.
+ if (CS.getInstruction() == 0 || !CS.isCallee(UI)) return 0;
+
+ if (CS.getInstruction()->getParent()->getParent() == F)
+ isSelfRecursive = true;
+ }
+
// Check to see which arguments are promotable. If an argument is promotable,
// add it to ArgsToPromote.
SmallPtrSet<Argument*, 8> ArgsToPromote;
SmallPtrSet<Argument*, 8> ByValArgsToTransform;
for (unsigned i = 0; i != PointerArgs.size(); ++i) {
- bool isByVal = F->paramHasAttr(PointerArgs[i].second+1, Attribute::ByVal);
+ bool isByVal=F->getAttributes().
+ hasAttribute(PointerArgs[i].second+1, Attribute::ByVal);
+ Argument *PtrArg = PointerArgs[i].first;
+ Type *AgTy = cast<PointerType>(PtrArg->getType())->getElementType();
// If this is a byval argument, and if the aggregate type is small, just
// pass the elements, which is always safe.
- Argument *PtrArg = PointerArgs[i].first;
if (isByVal) {
- const Type *AgTy = cast<PointerType>(PtrArg->getType())->getElementType();
- if (const StructType *STy = dyn_cast<StructType>(AgTy)) {
+ if (StructType *STy = dyn_cast<StructType>(AgTy)) {
if (maxElements > 0 && STy->getNumElements() > maxElements) {
DEBUG(dbgs() << "argpromotion disable promoting argument '"
<< PtrArg->getName() << "' because it would require adding more"
<< " than " << maxElements << " arguments to the function.\n");
- } else {
- // If all the elements are single-value types, we can promote it.
- bool AllSimple = true;
- for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
- if (!STy->getElementType(i)->isSingleValueType()) {
- AllSimple = false;
- break;
- }
-
- // Safe to transform, don't even bother trying to "promote" it.
- // Passing the elements as a scalar will allow scalarrepl to hack on
- // the new alloca we introduce.
- if (AllSimple) {
- ByValArgsToTransform.insert(PtrArg);
- continue;
+ continue;
+ }
+
+ // If all the elements are single-value types, we can promote it.
+ bool AllSimple = true;
+ for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
+ if (!STy->getElementType(i)->isSingleValueType()) {
+ AllSimple = false;
+ break;
}
}
+
+ // Safe to transform, don't even bother trying to "promote" it.
+ // Passing the elements as a scalar will allow scalarrepl to hack on
+ // the new alloca we introduce.
+ if (AllSimple) {
+ ByValArgsToTransform.insert(PtrArg);
+ continue;
+ }
}
}
+ // If the argument is a recursive type and we're in a recursive
+ // function, we could end up infinitely peeling the function argument.
+ if (isSelfRecursive) {
+ if (StructType *STy = dyn_cast<StructType>(AgTy)) {
+ bool RecursiveType = false;
+ for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
+ if (STy->getElementType(i) == PtrArg->getType()) {
+ RecursiveType = true;
+ break;
+ }
+ }
+ if (RecursiveType)
+ continue;
+ }
+ }
+
// Otherwise, see if we can promote the pointer to its value.
if (isSafeToPromoteArgument(PtrArg, isByVal))
ArgsToPromote.insert(PtrArg);
return DoPromotion(F, ArgsToPromote, ByValArgsToTransform);
}
-/// AllCalleesPassInValidPointerForArgument - Return true if we can prove that
+/// AllCallersPassInValidPointerForArgument - Return true if we can prove that
/// all callees pass in a valid pointer for the specified function argument.
-static bool AllCalleesPassInValidPointerForArgument(Argument *Arg) {
+static bool AllCallersPassInValidPointerForArgument(Argument *Arg) {
Function *Callee = Arg->getParent();
unsigned ArgNo = std::distance(Callee->arg_begin(),
const ArgPromotion::IndicesVector &Longer) {
if (Prefix.size() > Longer.size())
return false;
- for (unsigned i = 0, e = Prefix.size(); i != e; ++i)
- if (Prefix[i] != Longer[i])
- return false;
- return true;
+ return std::equal(Prefix.begin(), Prefix.end(), Longer.begin());
}
GEPIndicesSet ToPromote;
// If the pointer is always valid, any load with first index 0 is valid.
- if (isByVal || AllCalleesPassInValidPointerForArgument(Arg))
+ if (isByVal || AllCallersPassInValidPointerForArgument(Arg))
SafeToUnconditionallyLoad.insert(IndicesVector(1, 0));
// First, iterate the entry block and mark loads of (geps of) arguments as
User *U = *UI;
Operands.clear();
if (LoadInst *LI = dyn_cast<LoadInst>(U)) {
- if (LI->isVolatile()) return false; // Don't hack volatile loads
+ // Don't hack volatile/atomic loads
+ if (!LI->isSimple()) return false;
Loads.push_back(LI);
// Direct loads are equivalent to a GEP with a zero index and then a load.
Operands.push_back(0);
for (Value::use_iterator UI = GEP->use_begin(), E = GEP->use_end();
UI != E; ++UI)
if (LoadInst *LI = dyn_cast<LoadInst>(*UI)) {
- if (LI->isVolatile()) return false; // Don't hack volatile loads
+ // Don't hack volatile/atomic loads
+ if (!LI->isSimple()) return false;
Loads.push_back(LI);
} else {
// Other uses than load?
LoadInst *Load = Loads[i];
BasicBlock *BB = Load->getParent();
- AliasAnalysis::Location Loc(Load->getPointerOperand(),
- AA.getTypeStoreSize(Load->getType()),
- Load->getMetadata(LLVMContext::MD_tbaa));
-
+ AliasAnalysis::Location Loc = AA.getLocation(Load);
if (AA.canInstructionRangeModify(BB->front(), *Load, Loc))
return false; // Pointer is invalidated!
// Start by computing a new prototype for the function, which is the same as
// the old function, but has modified arguments.
- const FunctionType *FTy = F->getFunctionType();
- std::vector<const Type*> Params;
+ FunctionType *FTy = F->getFunctionType();
+ std::vector<Type*> Params;
typedef std::set<IndicesVector> ScalarizeTable;
// what the new GEP/Load instructions we are inserting look like.
std::map<IndicesVector, LoadInst*> OriginalLoads;
- // Attributes - Keep track of the parameter attributes for the arguments
+ // Attribute - Keep track of the parameter attributes for the arguments
// that we are *not* promoting. For the ones that we do promote, the parameter
// attributes are lost
SmallVector<AttributeWithIndex, 8> AttributesVec;
- const AttrListPtr &PAL = F->getAttributes();
+ const AttributeSet &PAL = F->getAttributes();
// Add any return attributes.
- if (Attributes attrs = PAL.getRetAttributes())
- AttributesVec.push_back(AttributeWithIndex::get(0, attrs));
+ Attribute attrs = PAL.getRetAttributes();
+ if (attrs.hasAttributes())
+ AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::ReturnIndex,
+ attrs));
// First, determine the new argument list
unsigned ArgIndex = 1;
++I, ++ArgIndex) {
if (ByValArgsToTransform.count(I)) {
// Simple byval argument? Just add all the struct element types.
-
const Type *AgTy = cast<PointerType>(I->getType())->getElementType();
- const StructType *STy = cast<StructType>(AgTy);
+ Type *AgTy = cast<PointerType>(I->getType())->getElementType();
+ StructType *STy = cast<StructType>(AgTy);
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i)
Params.push_back(STy->getElementType(i));
++NumByValArgsPromoted;
} else if (!ArgsToPromote.count(I)) {
// Unchanged argument
Params.push_back(I->getType());
- if (Attributes attrs = PAL.getParamAttributes(ArgIndex))
+ Attribute attrs = PAL.getParamAttributes(ArgIndex);
+ if (attrs.hasAttributes())
AttributesVec.push_back(AttributeWithIndex::get(Params.size(), attrs));
} else if (I->use_empty()) {
// Dead argument (which are always marked as promotable)
for (ScalarizeTable::iterator SI = ArgIndices.begin(),
E = ArgIndices.end(); SI != E; ++SI) {
// not allowed to dereference ->begin() if size() is 0
- Params.push_back(GetElementPtrInst::getIndexedType(I->getType(),
- SI->begin(),
- SI->end()));
+ Params.push_back(GetElementPtrInst::getIndexedType(I->getType(), *SI));
assert(Params.back());
}
}
// Add any function attributes.
- if (Attributes attrs = PAL.getFnAttributes())
- AttributesVec.push_back(AttributeWithIndex::get(~0, attrs));
+ attrs = PAL.getFnAttributes();
+ if (attrs.hasAttributes())
+ AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::FunctionIndex,
+ attrs));
- const Type *RetTy = FTy->getReturnType();
-
- // Work around LLVM bug PR56: the CWriter cannot emit varargs functions which
- // have zero fixed arguments.
- bool ExtraArgHack = false;
- if (Params.empty() && FTy->isVarArg()) {
- ExtraArgHack = true;
- Params.push_back(Type::getInt32Ty(F->getContext()));
- }
+ Type *RetTy = FTy->getReturnType();
// Construct the new function type using the new arguments.
FunctionType *NFTy = FunctionType::get(RetTy, Params, FTy->isVarArg());
// Recompute the parameter attributes list based on the new arguments for
// the function.
- NF->setAttributes(AttrListPtr::get(AttributesVec.begin(),
- AttributesVec.end()));
+ NF->setAttributes(AttributeSet::get(F->getContext(), AttributesVec));
AttributesVec.clear();
F->getParent()->getFunctionList().insert(F, NF);
CallSite CS(F->use_back());
assert(CS.getCalledFunction() == F);
Instruction *Call = CS.getInstruction();
- const AttrListPtr &CallPAL = CS.getAttributes();
+ const AttributeSet &CallPAL = CS.getAttributes();
// Add any return attributes.
- if (Attributes attrs = CallPAL.getRetAttributes())
- AttributesVec.push_back(AttributeWithIndex::get(0, attrs));
+ Attribute attrs = CallPAL.getRetAttributes();
+ if (attrs.hasAttributes())
+ AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::ReturnIndex,
+ attrs));
// Loop over the operands, inserting GEP and loads in the caller as
// appropriate.
if (!ArgsToPromote.count(I) && !ByValArgsToTransform.count(I)) {
Args.push_back(*AI); // Unmodified argument
- if (Attributes Attrs = CallPAL.getParamAttributes(ArgIndex))
+ Attribute Attrs = CallPAL.getParamAttributes(ArgIndex);
+ if (Attrs.hasAttributes())
AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs));
} else if (ByValArgsToTransform.count(I)) {
// Emit a GEP and load for each element of the struct.
-
const Type *AgTy = cast<PointerType>(I->getType())->getElementType();
- const StructType *STy = cast<StructType>(AgTy);
+ Type *AgTy = cast<PointerType>(I->getType())->getElementType();
+ StructType *STy = cast<StructType>(AgTy);
Value *Idxs[2] = {
ConstantInt::get(Type::getInt32Ty(F->getContext()), 0), 0 };
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
Idxs[1] = ConstantInt::get(Type::getInt32Ty(F->getContext()), i);
- Value *Idx = GetElementPtrInst::Create(*AI, Idxs, Idxs+2,
+ Value *Idx = GetElementPtrInst::Create(*AI, Idxs,
(*AI)->getName()+"."+utostr(i),
Call);
// TODO: Tell AA about the new values?
LoadInst *OrigLoad = OriginalLoads[*SI];
if (!SI->empty()) {
Ops.reserve(SI->size());
- const Type *ElTy = V->getType();
+ Type *ElTy = V->getType();
for (IndicesVector::const_iterator II = SI->begin(),
IE = SI->end(); II != IE; ++II) {
// Use i32 to index structs, and i64 for others (pointers/arrays).
// This satisfies GEP constraints.
- const Type *IdxTy = (ElTy->isStructTy() ?
+ Type *IdxTy = (ElTy->isStructTy() ?
Type::getInt32Ty(F->getContext()) :
Type::getInt64Ty(F->getContext()));
Ops.push_back(ConstantInt::get(IdxTy, *II));
ElTy = cast<CompositeType>(ElTy)->getTypeAtIndex(*II);
}
// And create a GEP to extract those indices.
- V = GetElementPtrInst::Create(V, Ops.begin(), Ops.end(),
- V->getName()+".idx", Call);
+ V = GetElementPtrInst::Create(V, Ops, V->getName()+".idx", Call);
Ops.clear();
AA.copyValue(OrigLoad->getOperand(0), V);
}
}
}
- if (ExtraArgHack)
- Args.push_back(Constant::getNullValue(Type::getInt32Ty(F->getContext())));
-
// Push any varargs arguments on the list.
for (; AI != CS.arg_end(); ++AI, ++ArgIndex) {
Args.push_back(*AI);
- if (Attributes Attrs = CallPAL.getParamAttributes(ArgIndex))
+ Attribute Attrs = CallPAL.getParamAttributes(ArgIndex);
+ if (Attrs.hasAttributes())
AttributesVec.push_back(AttributeWithIndex::get(Args.size(), Attrs));
}
// Add any function attributes.
- if (Attributes attrs = CallPAL.getFnAttributes())
- AttributesVec.push_back(AttributeWithIndex::get(~0, attrs));
+ attrs = CallPAL.getFnAttributes();
+ if (attrs.hasAttributes())
+ AttributesVec.push_back(AttributeWithIndex::get(AttributeSet::FunctionIndex,
+ attrs));
Instruction *New;
if (InvokeInst *II = dyn_cast<InvokeInst>(Call)) {
New = InvokeInst::Create(NF, II->getNormalDest(), II->getUnwindDest(),
- Args.begin(), Args.end(), "", Call);
+ Args, "", Call);
cast<InvokeInst>(New)->setCallingConv(CS.getCallingConv());
- cast<InvokeInst>(New)->setAttributes(AttrListPtr::get(AttributesVec.begin(),
- AttributesVec.end()));
+ cast<InvokeInst>(New)->setAttributes(AttributeSet::get(II->getContext(),
+ AttributesVec));
} else {
- New = CallInst::Create(NF, Args.begin(), Args.end(), "", Call);
+ New = CallInst::Create(NF, Args, "", Call);
cast<CallInst>(New)->setCallingConv(CS.getCallingConv());
- cast<CallInst>(New)->setAttributes(AttrListPtr::get(AttributesVec.begin(),
- AttributesVec.end()));
+ cast<CallInst>(New)->setAttributes(AttributeSet::get(New->getContext(),
+ AttributesVec));
if (cast<CallInst>(Call)->isTailCall())
cast<CallInst>(New)->setTailCall();
}
// function empty.
NF->getBasicBlockList().splice(NF->begin(), F->getBasicBlockList());
- // Loop over the argument list, transfering uses of the old arguments over to
- // the new arguments, also transfering over the names as well.
+ // Loop over the argument list, transferring uses of the old arguments over to
+ // the new arguments, also transferring over the names as well.
//
for (Function::arg_iterator I = F->arg_begin(), E = F->arg_end(),
I2 = NF->arg_begin(); I != E; ++I) {
Instruction *InsertPt = NF->begin()->begin();
// Just add all the struct element types.
-
const Type *AgTy = cast<PointerType>(I->getType())->getElementType();
+ Type *AgTy = cast<PointerType>(I->getType())->getElementType();
Value *TheAlloca = new AllocaInst(AgTy, 0, "", InsertPt);
- const StructType *STy = cast<StructType>(AgTy);
+ StructType *STy = cast<StructType>(AgTy);
Value *Idxs[2] = {
ConstantInt::get(Type::getInt32Ty(F->getContext()), 0), 0 };
for (unsigned i = 0, e = STy->getNumElements(); i != e; ++i) {
Idxs[1] = ConstantInt::get(Type::getInt32Ty(F->getContext()), i);
Value *Idx =
- GetElementPtrInst::Create(TheAlloca, Idxs, Idxs+2,
+ GetElementPtrInst::Create(TheAlloca, Idxs,
TheAlloca->getName()+"."+Twine(i),
InsertPt);
I2->setName(I->getName()+"."+Twine(i));
}
// Increment I2 past all of the arguments added for this promoted pointer.
- for (unsigned i = 0, e = ArgIndices.size(); i != e; ++i)
- ++I2;
+ std::advance(I2, ArgIndices.size());
}
- // Notify the alias analysis implementation that we inserted a new argument.
- if (ExtraArgHack)
- AA.copyValue(Constant::getNullValue(Type::getInt32Ty(F->getContext())),
- NF->arg_begin());
-
-
// Tell the alias analysis that the old function is about to disappear.
AA.replaceWithNewValue(F, NF);
projects / oota-llvm.git / blobdiff