//
// The LLVM Compiler Infrastructure
//
-// This file was developed by the LLVM research group and is distributed under
-// the University of Illinois Open Source License. See LICENSE.TXT for details.
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
//
//===----------------------------------------------------------------------===//
//
#include "llvm/Module.h"
#include "llvm/Instructions.h"
#include "llvm/Intrinsics.h"
+#include "llvm/ParameterAttributes.h"
#include "llvm/Analysis/CallGraph.h"
+#include "llvm/Target/TargetData.h"
#include "llvm/ADT/SmallVector.h"
#include "llvm/Support/CallSite.h"
using namespace llvm;
CallInst *CI = cast<CallInst>(I);
// If this call cannot unwind, don't convert it to an invoke.
- if (CI->isNoUnwind())
+ if (CI->doesNotThrow())
continue;
// Convert this function call into an invoke instruction.
bool MustClearTailCallFlags =
isa<CallInst>(TheCall) && !cast<CallInst>(TheCall)->isTailCall();
+ // If the call to the callee cannot throw, set the 'nounwind' flag on any
+ // calls that we inline.
+ bool MarkNoUnwind = CS.doesNotThrow();
+
BasicBlock *OrigBB = TheCall->getParent();
Function *Caller = OrigBB->getParent();
-
+ BasicBlock *UnwindBB = OrigBB->getUnwindDest();
+
+ // GC poses two hazards to inlining, which only occur when the callee has GC:
+ // 1. If the caller has no GC, then the callee's GC must be propagated to the
+ // caller.
+ // 2. If the caller has a differing GC, it is invalid to inline.
+ if (CalledFunc->hasCollector()) {
+ if (!Caller->hasCollector())
+ Caller->setCollector(CalledFunc->getCollector());
+ else if (CalledFunc->getCollector() != Caller->getCollector())
+ return false;
+ }
+
// Get an iterator to the last basic block in the function, which will have
// the new function inlined after it.
//
std::vector<ReturnInst*> Returns;
ClonedCodeInfo InlinedFunctionInfo;
Function::iterator FirstNewBlock;
-
+
{ // Scope to destroy ValueMap after cloning.
DenseMap<const Value*, Value*> ValueMap;
- // Calculate the vector of arguments to pass into the function cloner, which
- // matches up the formal to the actual argument values.
assert(std::distance(CalledFunc->arg_begin(), CalledFunc->arg_end()) ==
std::distance(CS.arg_begin(), CS.arg_end()) &&
"No varargs calls can be inlined!");
+
+ // Calculate the vector of arguments to pass into the function cloner, which
+ // matches up the formal to the actual argument values.
CallSite::arg_iterator AI = CS.arg_begin();
+ unsigned ArgNo = 0;
for (Function::const_arg_iterator I = CalledFunc->arg_begin(),
- E = CalledFunc->arg_end(); I != E; ++I, ++AI)
- ValueMap[I] = *AI;
+ E = CalledFunc->arg_end(); I != E; ++I, ++AI, ++ArgNo) {
+ Value *ActualArg = *AI;
+
+ // When byval arguments actually inlined, we need to make the copy implied
+ // by them explicit. However, we don't do this if the callee is readonly
+ // or readnone, because the copy would be unneeded: the callee doesn't
+ // modify the struct.
+ if (CalledFunc->paramHasAttr(ArgNo+1, ParamAttr::ByVal) &&
+ !CalledFunc->onlyReadsMemory()) {
+ const Type *AggTy = cast<PointerType>(I->getType())->getElementType();
+ const Type *VoidPtrTy = PointerType::getUnqual(Type::Int8Ty);
+
+ // Create the alloca. If we have TargetData, use nice alignment.
+ unsigned Align = 1;
+ if (TD) Align = TD->getPrefTypeAlignment(AggTy);
+ Value *NewAlloca = new AllocaInst(AggTy, 0, Align, I->getName(),
+ Caller->begin()->begin());
+ // Emit a memcpy.
+ Function *MemCpyFn = Intrinsic::getDeclaration(Caller->getParent(),
+ Intrinsic::memcpy_i64);
+ Value *DestCast = new BitCastInst(NewAlloca, VoidPtrTy, "tmp", TheCall);
+ Value *SrcCast = new BitCastInst(*AI, VoidPtrTy, "tmp", TheCall);
+
+ Value *Size;
+ if (TD == 0)
+ Size = ConstantExpr::getSizeOf(AggTy);
+ else
+ Size = ConstantInt::get(Type::Int64Ty, TD->getTypeStoreSize(AggTy));
+
+ // Always generate a memcpy of alignment 1 here because we don't know
+ // the alignment of the src pointer. Other optimizations can infer
+ // better alignment.
+ Value *CallArgs[] = {
+ DestCast, SrcCast, Size, ConstantInt::get(Type::Int32Ty, 1)
+ };
+ CallInst *TheMemCpy =
+ new CallInst(MemCpyFn, CallArgs, CallArgs+4, "", TheCall);
+
+ // If we have a call graph, update it.
+ if (CG) {
+ CallGraphNode *MemCpyCGN = CG->getOrInsertFunction(MemCpyFn);
+ CallGraphNode *CallerNode = (*CG)[Caller];
+ CallerNode->addCalledFunction(TheMemCpy, MemCpyCGN);
+ }
+
+ // Uses of the argument in the function should use our new alloca
+ // instead.
+ ActualArg = NewAlloca;
+ }
+
+ ValueMap[I] = ActualArg;
+ }
// We want the inliner to prune the code as it copies. We would LOVE to
// have no dead or constant instructions leftover after inlining occurs
// code with llvm.stacksave/llvm.stackrestore intrinsics.
if (InlinedFunctionInfo.ContainsDynamicAllocas) {
Module *M = Caller->getParent();
- const Type *BytePtr = PointerType::get(Type::Int8Ty);
+ const Type *BytePtr = PointerType::getUnqual(Type::Int8Ty);
// Get the two intrinsics we care about.
Constant *StackSave, *StackRestore;
StackSave = M->getOrInsertFunction("llvm.stacksave", BytePtr, NULL);
// If we are inlining tail call instruction through a call site that isn't
// marked 'tail', we must remove the tail marker for any calls in the inlined
- // code.
- if (MustClearTailCallFlags && InlinedFunctionInfo.ContainsCalls) {
+ // code. Also, calls inlined through a 'nounwind' call site should be marked
+ // 'nounwind'.
+ if (InlinedFunctionInfo.ContainsCalls &&
+ (MustClearTailCallFlags || MarkNoUnwind)) {
for (Function::iterator BB = FirstNewBlock, E = Caller->end();
BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
- if (CallInst *CI = dyn_cast<CallInst>(I))
- CI->setTailCall(false);
+ if (CallInst *CI = dyn_cast<CallInst>(I)) {
+ if (MustClearTailCallFlags)
+ CI->setTailCall(false);
+ if (MarkNoUnwind)
+ CI->setDoesNotThrow();
+ }
}
+ // If we are inlining through a 'nounwind' call site then any inlined 'unwind'
+ // instructions are unreachable.
+ if (InlinedFunctionInfo.ContainsUnwinds && MarkNoUnwind)
+ for (Function::iterator BB = FirstNewBlock, E = Caller->end();
+ BB != E; ++BB) {
+ TerminatorInst *Term = BB->getTerminator();
+ if (isa<UnwindInst>(Term)) {
+ new UnreachableInst(Term);
+ BB->getInstList().erase(Term);
+ }
+ }
+
+ // If we are inlining a function that unwinds into a BB with an unwind dest,
+ // turn the inlined unwinds into branches to the unwind dest.
+ if (InlinedFunctionInfo.ContainsUnwinds && UnwindBB && isa<CallInst>(TheCall))
+ for (Function::iterator BB = FirstNewBlock, E = Caller->end();
+ BB != E; ++BB) {
+ TerminatorInst *Term = BB->getTerminator();
+ if (isa<UnwindInst>(Term)) {
+ new BranchInst(UnwindBB, Term);
+ BB->getInstList().erase(Term);
+ }
+ }
+
// If we are inlining for an invoke instruction, we must make sure to rewrite
// any inlined 'unwind' instructions into branches to the invoke exception
// destination, and call instructions into invoke instructions.
// If the return instruction returned a value, replace uses of the call with
// uses of the returned value.
- if (!TheCall->use_empty())
- TheCall->replaceAllUsesWith(Returns[0]->getReturnValue());
-
+ if (!TheCall->use_empty()) {
+ ReturnInst *R = Returns[0];
+ if (R->getNumOperands() > 1) {
+ // Multiple return values.
+ while (!TheCall->use_empty()) {
+ GetResultInst *GR = cast<GetResultInst>(TheCall->use_back());
+ Value *RV = R->getOperand(GR->getIndex());
+ GR->replaceAllUsesWith(RV);
+ GR->eraseFromParent();
+ }
+ } else
+ TheCall->replaceAllUsesWith(R->getReturnValue());
+ }
// Since we are now done with the Call/Invoke, we can delete it.
TheCall->getParent()->getInstList().erase(TheCall);
// Now that the function is correct, make it a little bit nicer. In
// particular, move the basic blocks inserted from the end of the function
// into the space made by splitting the source basic block.
- //
Caller->getBasicBlockList().splice(AfterCallBB, Caller->getBasicBlockList(),
FirstNewBlock, Caller->end());
// Handle all of the return instructions that we just cloned in, and eliminate
// any users of the original call/invoke instruction.
- if (Returns.size() > 1) {
+ if (!Returns.empty()) {
// The PHI node should go at the front of the new basic block to merge all
// possible incoming values.
- //
- PHINode *PHI = 0;
+ SmallVector<PHINode *, 4> PHIs;
if (!TheCall->use_empty()) {
- PHI = new PHINode(CalledFunc->getReturnType(),
- TheCall->getName(), AfterCallBB->begin());
-
- // Anything that used the result of the function call should now use the
- // PHI node as their operand.
- //
- TheCall->replaceAllUsesWith(PHI);
+ const Type *RTy = CalledFunc->getReturnType();
+ if (const StructType *STy = dyn_cast<StructType>(RTy)) {
+ unsigned NumRetVals = STy->getNumElements();
+ // Create new phi nodes such that phi node number in the PHIs vector
+ // match corresponding return value operand number.
+ for (unsigned i = 0; i < NumRetVals; ++i) {
+ PHINode *PHI = new PHINode(STy->getElementType(i),
+ TheCall->getName(), AfterCallBB->begin());
+ PHIs.push_back(PHI);
+ }
+ // TheCall results are used by GetResult instructions.
+ while (!TheCall->use_empty()) {
+ GetResultInst *GR = cast<GetResultInst>(TheCall->use_back());
+ GR->replaceAllUsesWith(PHIs[GR->getIndex()]);
+ GR->eraseFromParent();
+ }
+ } else {
+ PHINode *PHI = new PHINode(RTy, TheCall->getName(), AfterCallBB->begin());
+ PHIs.push_back(PHI);
+ // Anything that used the result of the function call should now use the
+ // PHI node as their operand.
+ TheCall->replaceAllUsesWith(PHI);
+ }
}
- // Loop over all of the return instructions, turning them into unconditional
- // branches to the merge point now, and adding entries to the PHI node as
+ // Loop over all of the return instructions adding entries to the PHI node as
// appropriate.
- for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
- ReturnInst *RI = Returns[i];
-
- if (PHI) {
- assert(RI->getReturnValue() && "Ret should have value!");
- assert(RI->getReturnValue()->getType() == PHI->getType() &&
- "Ret value not consistent in function!");
- PHI->addIncoming(RI->getReturnValue(), RI->getParent());
+ if (!PHIs.empty()) {
+ const Type *RTy = CalledFunc->getReturnType();
+ if (const StructType *STy = dyn_cast<StructType>(RTy)) {
+ unsigned NumRetVals = STy->getNumElements();
+ for (unsigned j = 0; j < NumRetVals; ++j) {
+ PHINode *PHI = PHIs[j];
+ // Each PHI node will receive one value from each return instruction.
+ for(unsigned i = 0, e = Returns.size(); i != e; ++i) {
+ ReturnInst *RI = Returns[i];
+ PHI->addIncoming(RI->getReturnValue(j /*PHI number matches operand number*/),
+ RI->getParent());
+ }
+ }
+ } else {
+ for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
+ ReturnInst *RI = Returns[i];
+ assert(PHIs.size() == 1 && "Invalid number of PHI nodes");
+ assert(RI->getReturnValue() && "Ret should have value!");
+ assert(RI->getReturnValue()->getType() == PHIs[0]->getType() &&
+ "Ret value not consistent in function!");
+ PHIs[0]->addIncoming(RI->getReturnValue(), RI->getParent());
+ }
}
+ }
- // Add a branch to the merge point where the PHI node lives if it exists.
+ // Add a branch to the merge points and remove retrun instructions.
+ for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
+ ReturnInst *RI = Returns[i];
new BranchInst(AfterCallBB, RI);
-
- // Delete the return instruction now
RI->getParent()->getInstList().erase(RI);
}
-
- } else if (!Returns.empty()) {
- // Otherwise, if there is exactly one return value, just replace anything
- // using the return value of the call with the computed value.
- if (!TheCall->use_empty())
- TheCall->replaceAllUsesWith(Returns[0]->getReturnValue());
-
- // Splice the code from the return block into the block that it will return
- // to, which contains the code that was after the call.
- BasicBlock *ReturnBB = Returns[0]->getParent();
- AfterCallBB->getInstList().splice(AfterCallBB->begin(),
- ReturnBB->getInstList());
-
- // Update PHI nodes that use the ReturnBB to use the AfterCallBB.
- ReturnBB->replaceAllUsesWith(AfterCallBB);
-
- // Delete the return instruction now and empty ReturnBB now.
- Returns[0]->eraseFromParent();
- ReturnBB->eraseFromParent();
} else if (!TheCall->use_empty()) {
// No returns, but something is using the return value of the call. Just
// nuke the result.