llvm/lib/Transforms/Utils/InlineFunction.cpp
2003-08-24 06:59:16 +00:00

265 lines
11 KiB
C++

//===- InlineFunction.cpp - Code to perform function inlining -------------===//
//
// This file implements inlining of a function into a call site, resolving
// parameters and the return value as appropriate.
//
// FIXME: This pass should transform alloca instructions in the called function
// into malloc/free pairs! Or perhaps it should refuse to inline them!
//
//===----------------------------------------------------------------------===//
#include "llvm/Transforms/Utils/Cloning.h"
#include "llvm/Constant.h"
#include "llvm/DerivedTypes.h"
#include "llvm/Module.h"
#include "llvm/Instructions.h"
#include "llvm/Intrinsics.h"
#include "llvm/Support/CallSite.h"
#include "llvm/Transforms/Utils/Local.h"
bool InlineFunction(CallInst *CI) { return InlineFunction(CallSite(CI)); }
bool InlineFunction(InvokeInst *II) { return InlineFunction(CallSite(II)); }
// InlineFunction - This function inlines the called function into the basic
// block of the caller. This returns false if it is not possible to inline this
// call. The program is still in a well defined state if this occurs though.
//
// Note that this only does one level of inlining. For example, if the
// instruction 'call B' is inlined, and 'B' calls 'C', then the call to 'C' now
// exists in the instruction stream. Similiarly this will inline a recursive
// function by one level.
//
bool InlineFunction(CallSite CS) {
Instruction *TheCall = CS.getInstruction();
assert(TheCall->getParent() && TheCall->getParent()->getParent() &&
"Instruction not in function!");
const Function *CalledFunc = CS.getCalledFunction();
if (CalledFunc == 0 || // Can't inline external function or indirect
CalledFunc->isExternal() || // call, or call to a vararg function!
CalledFunc->getFunctionType()->isVarArg()) return false;
BasicBlock *OrigBB = TheCall->getParent();
Function *Caller = OrigBB->getParent();
// We want to clone the entire callee function into the whole between the
// "starter" and "ender" blocks. How we accomplish this depends on whether
// this is an invoke instruction or a call instruction.
BasicBlock *InvokeDest = 0; // Exception handling destination
BasicBlock *AfterCallBB;
if (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
AfterCallBB = II->getNormalDest();
InvokeDest = II->getExceptionalDest();
// Add an unconditional branch to make this look like the CallInst case...
new BranchInst(AfterCallBB, TheCall);
// Remove (unlink) the InvokeInst from the function...
OrigBB->getInstList().remove(TheCall);
} else { // It's a call
// If this is a call instruction, we need to split the basic block that the
// call lives in.
//
AfterCallBB = OrigBB->splitBasicBlock(TheCall,
CalledFunc->getName()+".entry");
// Remove (unlink) the CallInst from the function...
AfterCallBB->getInstList().remove(TheCall);
}
// If we have a return value generated by this call, convert it into a PHI
// node that gets values from each of the old RET instructions in the original
// function.
//
PHINode *PHI = 0;
if (!TheCall->use_empty()) {
// The PHI node should go at the front of the new basic block to merge all
// possible incoming values.
//
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);
}
// Get an iterator to the last basic block in the function, which will have
// the new function inlined after it.
//
Function::iterator LastBlock = &Caller->back();
// Calculate the vector of arguments to pass into the function cloner...
std::map<const Value*, Value*> ValueMap;
assert(std::distance(CalledFunc->abegin(), CalledFunc->aend()) ==
std::distance(CS.arg_begin(), CS.arg_end()) &&
"No varargs calls can be inlined!");
CallSite::arg_iterator AI = CS.arg_begin();
for (Function::const_aiterator I = CalledFunc->abegin(), E=CalledFunc->aend();
I != E; ++I, ++AI)
ValueMap[I] = *AI;
// Since we are now done with the Call/Invoke, we can delete it.
delete TheCall;
// Make a vector to capture the return instructions in the cloned function...
std::vector<ReturnInst*> Returns;
// Populate the value map with all of the globals in the program.
// FIXME: This should be the default for CloneFunctionInto!
Module &M = *Caller->getParent();
for (Module::iterator I = M.begin(), E = M.end(); I != E; ++I)
ValueMap[I] = I;
for (Module::giterator I = M.gbegin(), E = M.gend(); I != E; ++I)
ValueMap[I] = I;
// Do all of the hard part of cloning the callee into the caller...
CloneFunctionInto(Caller, CalledFunc, ValueMap, Returns, ".i");
// Loop over all of the return instructions, turning them into unconditional
// branches to the merge point now...
for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
ReturnInst *RI = Returns[i];
BasicBlock *BB = RI->getParent();
// Add a branch to the merge point where the PHI node lives if it exists.
new BranchInst(AfterCallBB, RI);
if (PHI) { // The PHI node should include this value!
assert(RI->getReturnValue() && "Ret should have value!");
assert(RI->getReturnValue()->getType() == PHI->getType() &&
"Ret value not consistent in function!");
PHI->addIncoming(RI->getReturnValue(), BB);
}
// Delete the return instruction now
BB->getInstList().erase(RI);
}
// Check to see if the PHI node only has one argument. This is a common
// case resulting from there only being a single return instruction in the
// function call. Because this is so common, eliminate the PHI node.
//
if (PHI && PHI->getNumIncomingValues() == 1) {
PHI->replaceAllUsesWith(PHI->getIncomingValue(0));
PHI->getParent()->getInstList().erase(PHI);
}
// Change the branch that used to go to AfterCallBB to branch to the first
// basic block of the inlined function.
//
TerminatorInst *Br = OrigBB->getTerminator();
assert(Br && Br->getOpcode() == Instruction::Br &&
"splitBasicBlock broken!");
Br->setOperand(0, ++LastBlock);
// If there are any alloca instructions in the block that used to be the entry
// block for the callee, move them to the entry block of the caller. First
// calculate which instruction they should be inserted before. We insert the
// instructions at the end of the current alloca list.
//
if (isa<AllocaInst>(LastBlock->begin())) {
BasicBlock::iterator InsertPoint = Caller->begin()->begin();
while (isa<AllocaInst>(InsertPoint)) ++InsertPoint;
for (BasicBlock::iterator I = LastBlock->begin(), E = LastBlock->end();
I != E; )
if (AllocaInst *AI = dyn_cast<AllocaInst>(I)) {
++I; // Move to the next instruction
LastBlock->getInstList().remove(AI);
Caller->front().getInstList().insert(InsertPoint, AI);
} else {
++I;
}
}
// If we just inlined a call due to an invoke instruction, scan the inlined
// function checking for function calls that should now be made into invoke
// instructions, and for llvm.exc.rethrow()'s which should be turned into
// branches.
if (InvokeDest)
for (Function::iterator BB = LastBlock, E = Caller->end(); BB != E; ++BB)
for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ) {
// We only need to check for function calls: inlined invoke instructions
// require no special handling...
if (CallInst *CI = dyn_cast<CallInst>(I)) {
// FIXME: this should use annotations of the LLVM functions themselves
// to determine whether or not the function can throw.
bool ShouldInvokify = true;
if (Function *F = CI->getCalledFunction())
if (unsigned ID = F->getIntrinsicID())
if (ID == LLVMIntrinsic::exc_rethrow) {
// llvm.exc.rethrow requires special handling when it gets
// inlined into an invoke site. Once this happens, we know that
// the rethrow would cause a control transfer to the invoke
// exception destination, so we can transform it into a direct
// branch to the exception destination.
BranchInst *BI = new BranchInst(InvokeDest, CI);
// Note that since any instructions after the rethrow/branch are
// dead, we must delete them now (otherwise the terminator we
// just inserted wouldn't be at the end of the basic block!)
BasicBlock *CurBB = BB;
while (&CurBB->back() != BI) {
Instruction *I = &CurBB->back();
if (!I->use_empty())
I->replaceAllUsesWith(Constant::getNullValue(I->getType()));
CurBB->getInstList().pop_back();
}
break; // Done with this basic block!
} else if (ID == LLVMIntrinsic::exc_throw ||
ID == LLVMIntrinsic::exc_getcurrent) {
ShouldInvokify = false; // Not correct to invokify exc.throw!
}
// If we should convert this function into an invoke instruction, do
// so now.
if (ShouldInvokify) {
// First, split the basic block...
BasicBlock *Split = BB->splitBasicBlock(CI, CI->getName()+".noexc");
// Next, create the new invoke instruction, inserting it at the end
// of the old basic block.
new InvokeInst(CI->getCalledValue(), Split, InvokeDest,
std::vector<Value*>(CI->op_begin()+1, CI->op_end()),
CI->getName(), BB->getTerminator());
// Delete the unconditional branch inserted by splitBasicBlock
BB->getInstList().pop_back();
Split->getInstList().pop_front(); // Delete the original call
// This basic block is now complete, start scanning the next one.
break;
} else {
++I;
}
} else {
++I;
}
}
// 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(),
LastBlock, Caller->end());
// We should always be able to fold the entry block of the function into the
// single predecessor of the block...
assert(cast<BranchInst>(Br)->isUnconditional() && "splitBasicBlock broken!");
BasicBlock *CalleeEntry = cast<BranchInst>(Br)->getSuccessor(0);
SimplifyCFG(CalleeEntry);
// Okay, continue the CFG cleanup. It's often the case that there is only a
// single return instruction in the callee function. If this is the case,
// then we have an unconditional branch from the return block to the
// 'AfterCallBB'. Check for this case, and eliminate the branch is possible.
SimplifyCFG(AfterCallBB);
return true;
}