#include "llvm/Module.h"
#include "llvm/Instructions.h"
#include "llvm/Intrinsics.h"
+#include "llvm/Analysis/CallGraph.h"
+#include "llvm/ADT/SmallVector.h"
#include "llvm/Support/CallSite.h"
using namespace llvm;
-bool llvm::InlineFunction(CallInst *CI) { return InlineFunction(CallSite(CI)); }
-bool llvm::InlineFunction(InvokeInst *II) {return InlineFunction(CallSite(II));}
+bool llvm::InlineFunction(CallInst *CI, CallGraph *CG, const TargetData *TD) {
+ return InlineFunction(CallSite(CI), CG, TD);
+}
+bool llvm::InlineFunction(InvokeInst *II, CallGraph *CG, const TargetData *TD) {
+ return InlineFunction(CallSite(II), CG, TD);
+}
+
+/// HandleInlinedInvoke - If we inlined an invoke site, we need to convert calls
+/// in the body of the inlined function into invokes and turn unwind
+/// instructions into branches to the invoke unwind dest.
+///
+/// II is the invoke instruction begin inlined. FirstNewBlock is the first
+/// block of the inlined code (the last block is the end of the function),
+/// and InlineCodeInfo is information about the code that got inlined.
+static void HandleInlinedInvoke(InvokeInst *II, BasicBlock *FirstNewBlock,
+ ClonedCodeInfo &InlinedCodeInfo) {
+ BasicBlock *InvokeDest = II->getUnwindDest();
+ std::vector<Value*> InvokeDestPHIValues;
+
+ // If there are PHI nodes in the unwind destination block, we need to
+ // keep track of which values came into them from this invoke, then remove
+ // the entry for this block.
+ BasicBlock *InvokeBlock = II->getParent();
+ for (BasicBlock::iterator I = InvokeDest->begin(); isa<PHINode>(I); ++I) {
+ PHINode *PN = cast<PHINode>(I);
+ // Save the value to use for this edge.
+ InvokeDestPHIValues.push_back(PN->getIncomingValueForBlock(InvokeBlock));
+ }
+
+ Function *Caller = FirstNewBlock->getParent();
+
+ // The inlined code is currently at the end of the function, scan from the
+ // start of the inlined code to its end, checking for stuff we need to
+ // rewrite.
+ if (InlinedCodeInfo.ContainsCalls || InlinedCodeInfo.ContainsUnwinds) {
+ for (Function::iterator BB = FirstNewBlock, E = Caller->end();
+ BB != E; ++BB) {
+ if (InlinedCodeInfo.ContainsCalls) {
+ for (BasicBlock::iterator BBI = BB->begin(), E = BB->end(); BBI != E; ){
+ Instruction *I = BBI++;
+
+ // We only need to check for function calls: inlined invoke
+ // instructions require no special handling.
+ if (!isa<CallInst>(I)) continue;
+ CallInst *CI = cast<CallInst>(I);
+
+ // If this call cannot unwind, don't convert it to an invoke.
+ if (CI->doesNotThrow())
+ continue;
+
+ // Convert this function call into an invoke instruction.
+ // 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.
+ SmallVector<Value*, 8> InvokeArgs(CI->op_begin()+1, CI->op_end());
+ InvokeInst *II =
+ new InvokeInst(CI->getCalledValue(), Split, InvokeDest,
+ InvokeArgs.begin(), InvokeArgs.end(),
+ CI->getName(), BB->getTerminator());
+ II->setCallingConv(CI->getCallingConv());
+ II->setParamAttrs(CI->getParamAttrs());
+
+ // Make sure that anything using the call now uses the invoke!
+ CI->replaceAllUsesWith(II);
+
+ // Delete the unconditional branch inserted by splitBasicBlock
+ BB->getInstList().pop_back();
+ Split->getInstList().pop_front(); // Delete the original call
+
+ // Update any PHI nodes in the exceptional block to indicate that
+ // there is now a new entry in them.
+ unsigned i = 0;
+ for (BasicBlock::iterator I = InvokeDest->begin();
+ isa<PHINode>(I); ++I, ++i) {
+ PHINode *PN = cast<PHINode>(I);
+ PN->addIncoming(InvokeDestPHIValues[i], BB);
+ }
+
+ // This basic block is now complete, start scanning the next one.
+ break;
+ }
+ }
+
+ if (UnwindInst *UI = dyn_cast<UnwindInst>(BB->getTerminator())) {
+ // An UnwindInst requires special handling when it gets inlined into an
+ // invoke site. Once this happens, we know that the unwind would cause
+ // a control transfer to the invoke exception destination, so we can
+ // transform it into a direct branch to the exception destination.
+ new BranchInst(InvokeDest, UI);
+
+ // Delete the unwind instruction!
+ UI->getParent()->getInstList().pop_back();
+
+ // Update any PHI nodes in the exceptional block to indicate that
+ // there is now a new entry in them.
+ unsigned i = 0;
+ for (BasicBlock::iterator I = InvokeDest->begin();
+ isa<PHINode>(I); ++I, ++i) {
+ PHINode *PN = cast<PHINode>(I);
+ PN->addIncoming(InvokeDestPHIValues[i], BB);
+ }
+ }
+ }
+ }
+
+ // Now that everything is happy, we have one final detail. The PHI nodes in
+ // the exception destination block still have entries due to the original
+ // invoke instruction. Eliminate these entries (which might even delete the
+ // PHI node) now.
+ InvokeDest->removePredecessor(II->getParent());
+}
+
+/// UpdateCallGraphAfterInlining - Once we have cloned code over from a callee
+/// into the caller, update the specified callgraph to reflect the changes we
+/// made. Note that it's possible that not all code was copied over, so only
+/// some edges of the callgraph will be remain.
+static void UpdateCallGraphAfterInlining(const Function *Caller,
+ const Function *Callee,
+ Function::iterator FirstNewBlock,
+ DenseMap<const Value*, Value*> &ValueMap,
+ CallGraph &CG) {
+ // Update the call graph by deleting the edge from Callee to Caller
+ CallGraphNode *CalleeNode = CG[Callee];
+ CallGraphNode *CallerNode = CG[Caller];
+ CallerNode->removeCallEdgeTo(CalleeNode);
+
+ // Since we inlined some uninlined call sites in the callee into the caller,
+ // add edges from the caller to all of the callees of the callee.
+ for (CallGraphNode::iterator I = CalleeNode->begin(),
+ E = CalleeNode->end(); I != E; ++I) {
+ const Instruction *OrigCall = I->first.getInstruction();
+
+ DenseMap<const Value*, Value*>::iterator VMI = ValueMap.find(OrigCall);
+ // Only copy the edge if the call was inlined!
+ if (VMI != ValueMap.end() && VMI->second) {
+ // If the call was inlined, but then constant folded, there is no edge to
+ // add. Check for this case.
+ if (Instruction *NewCall = dyn_cast<Instruction>(VMI->second))
+ CallerNode->addCalledFunction(CallSite::get(NewCall), I->second);
+ }
+ }
+}
+
// 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
// exists in the instruction stream. Similiarly this will inline a recursive
// function by one level.
//
-bool llvm::InlineFunction(CallSite CS) {
+bool llvm::InlineFunction(CallSite CS, CallGraph *CG, const TargetData *TD) {
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->isDeclaration() || // call, or call to a vararg function!
CalledFunc->getFunctionType()->isVarArg()) return false;
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();
+
+ // 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.
//
// Make sure to capture all of the return instructions from the cloned
// function.
std::vector<ReturnInst*> Returns;
+ ClonedCodeInfo InlinedFunctionInfo;
+ Function::iterator FirstNewBlock;
+
{ // Scope to destroy ValueMap after cloning.
- // Calculate the vector of arguments to pass into the function cloner...
- std::map<const Value*, Value*> ValueMap;
+ 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!");
-
CallSite::arg_iterator AI = CS.arg_begin();
for (Function::const_arg_iterator I = CalledFunc->arg_begin(),
E = CalledFunc->arg_end(); I != E; ++I, ++AI)
ValueMap[I] = *AI;
- // Clone the entire body of the callee into the caller.
- CloneFunctionInto(Caller, CalledFunc, ValueMap, Returns, ".i");
+ // 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
+ // (which can happen, e.g., because an argument was constant), but we'll be
+ // happy with whatever the cloner can do.
+ CloneAndPruneFunctionInto(Caller, CalledFunc, ValueMap, Returns, ".i",
+ &InlinedFunctionInfo, TD);
+
+ // Remember the first block that is newly cloned over.
+ FirstNewBlock = LastBlock; ++FirstNewBlock;
+
+ // Update the callgraph if requested.
+ if (CG)
+ UpdateCallGraphAfterInlining(Caller, CalledFunc, FirstNewBlock, ValueMap,
+ *CG);
}
-
- // Remember the first block that is newly cloned over.
- Function::iterator FirstNewBlock = LastBlock; ++FirstNewBlock;
-
+
// 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
BasicBlock::iterator InsertPoint = Caller->begin()->begin();
for (BasicBlock::iterator I = FirstNewBlock->begin(),
E = FirstNewBlock->end(); I != E; )
- if (AllocaInst *AI = dyn_cast<AllocaInst>(I++))
+ if (AllocaInst *AI = dyn_cast<AllocaInst>(I++)) {
+ // If the alloca is now dead, remove it. This often occurs due to code
+ // specialization.
+ if (AI->use_empty()) {
+ AI->eraseFromParent();
+ continue;
+ }
+
if (isa<Constant>(AI->getArraySize())) {
// Scan for the block of allocas that we can move over, and move them
// all at once.
++I;
// Transfer all of the allocas over in a block. Using splice means
- // that they instructions aren't removed from the symbol table, then
+ // that the instructions aren't removed from the symbol table, then
// reinserted.
- Caller->front().getInstList().splice(InsertPoint,
- FirstNewBlock->getInstList(),
- AI, I);
+ Caller->getEntryBlock().getInstList().splice(
+ InsertPoint,
+ FirstNewBlock->getInstList(),
+ AI, I);
}
+ }
}
- // If we are inlining tail call instruction through an invoke or
- if (MustClearTailCallFlags) {
- 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 the inlined code contained dynamic alloca instructions, wrap the inlined
+ // code with llvm.stacksave/llvm.stackrestore intrinsics.
+ if (InlinedFunctionInfo.ContainsDynamicAllocas) {
+ Module *M = Caller->getParent();
+ const Type *BytePtr = PointerType::getUnqual(Type::Int8Ty);
+ // Get the two intrinsics we care about.
+ Constant *StackSave, *StackRestore;
+ StackSave = M->getOrInsertFunction("llvm.stacksave", BytePtr, NULL);
+ StackRestore = M->getOrInsertFunction("llvm.stackrestore", Type::VoidTy,
+ BytePtr, NULL);
+
+ // If we are preserving the callgraph, add edges to the stacksave/restore
+ // functions for the calls we insert.
+ CallGraphNode *StackSaveCGN = 0, *StackRestoreCGN = 0, *CallerNode = 0;
+ if (CG) {
+ // We know that StackSave/StackRestore are Function*'s, because they are
+ // intrinsics which must have the right types.
+ StackSaveCGN = CG->getOrInsertFunction(cast<Function>(StackSave));
+ StackRestoreCGN = CG->getOrInsertFunction(cast<Function>(StackRestore));
+ CallerNode = (*CG)[Caller];
+ }
+
+ // Insert the llvm.stacksave.
+ CallInst *SavedPtr = new CallInst(StackSave, "savedstack",
+ FirstNewBlock->begin());
+ if (CG) CallerNode->addCalledFunction(SavedPtr, StackSaveCGN);
+
+ // Insert a call to llvm.stackrestore before any return instructions in the
+ // inlined function.
+ for (unsigned i = 0, e = Returns.size(); i != e; ++i) {
+ CallInst *CI = new CallInst(StackRestore, SavedPtr, "", Returns[i]);
+ if (CG) CallerNode->addCalledFunction(CI, StackRestoreCGN);
+ }
- // 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 (InvokeInst *II = dyn_cast<InvokeInst>(TheCall)) {
- BasicBlock *InvokeDest = II->getUnwindDest();
- std::vector<Value*> InvokeDestPHIValues;
-
- // If there are PHI nodes in the exceptional destination block, we need to
- // keep track of which values came into them from this invoke, then remove
- // the entry for this block.
- for (BasicBlock::iterator I = InvokeDest->begin(); isa<PHINode>(I); ++I) {
- PHINode *PN = cast<PHINode>(I);
- // Save the value to use for this edge...
- InvokeDestPHIValues.push_back(PN->getIncomingValueForBlock(OrigBB));
+ // Count the number of StackRestore calls we insert.
+ unsigned NumStackRestores = Returns.size();
+
+ // If we are inlining an invoke instruction, insert restores before each
+ // unwind. These unwinds will be rewritten into branches later.
+ if (InlinedFunctionInfo.ContainsUnwinds && isa<InvokeInst>(TheCall)) {
+ for (Function::iterator BB = FirstNewBlock, E = Caller->end();
+ BB != E; ++BB)
+ if (UnwindInst *UI = dyn_cast<UnwindInst>(BB->getTerminator())) {
+ new CallInst(StackRestore, SavedPtr, "", UI);
+ ++NumStackRestores;
+ }
}
+ }
+ // 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. 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; ) {
- // We only need to check for function calls: inlined invoke instructions
- // require no special handling...
+ BB != E; ++BB)
+ for (BasicBlock::iterator I = BB->begin(), E = BB->end(); I != E; ++I)
if (CallInst *CI = dyn_cast<CallInst>(I)) {
- // Convert this function call into an invoke instruction... if it's
- // not an intrinsic function call (which are known to not unwind).
- if (CI->getCalledFunction() &&
- CI->getCalledFunction()->getIntrinsicID()) {
- ++I;
- } else {
- // 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.
- InvokeInst *II =
- new InvokeInst(CI->getCalledValue(), Split, InvokeDest,
- std::vector<Value*>(CI->op_begin()+1, CI->op_end()),
- CI->getName(), BB->getTerminator());
- II->setCallingConv(CI->getCallingConv());
-
- // Make sure that anything using the call now uses the invoke!
- CI->replaceAllUsesWith(II);
-
- // Delete the unconditional branch inserted by splitBasicBlock
- BB->getInstList().pop_back();
- Split->getInstList().pop_front(); // Delete the original call
-
- // Update any PHI nodes in the exceptional block to indicate that
- // there is now a new entry in them.
- unsigned i = 0;
- for (BasicBlock::iterator I = InvokeDest->begin();
- isa<PHINode>(I); ++I, ++i) {
- PHINode *PN = cast<PHINode>(I);
- PN->addIncoming(InvokeDestPHIValues[i], BB);
- }
-
- // This basic block is now complete, start scanning the next one.
- break;
- }
- } else {
- ++I;
+ if (MustClearTailCallFlags)
+ CI->setTailCall(false);
+ if (MarkNoUnwind)
+ CI->setDoesNotThrow();
}
- }
-
- if (UnwindInst *UI = dyn_cast<UnwindInst>(BB->getTerminator())) {
- // An UnwindInst requires special handling when it gets inlined into an
- // invoke site. Once this happens, we know that the unwind would cause
- // a control transfer to the invoke exception destination, so we can
- // transform it into a direct branch to the exception destination.
- new BranchInst(InvokeDest, UI);
-
- // Delete the unwind instruction!
- UI->getParent()->getInstList().pop_back();
+ }
- // Update any PHI nodes in the exceptional block to indicate that
- // there is now a new entry in them.
- unsigned i = 0;
- for (BasicBlock::iterator I = InvokeDest->begin();
- isa<PHINode>(I); ++I, ++i) {
- PHINode *PN = cast<PHINode>(I);
- PN->addIncoming(InvokeDestPHIValues[i], BB);
- }
+ // 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);
}
}
- // Now that everything is happy, we have one final detail. The PHI nodes in
- // the exception destination block still have entries due to the original
- // invoke instruction. Eliminate these entries (which might even delete the
- // PHI node) now.
- InvokeDest->removePredecessor(II->getParent());
- }
+ // 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 (InvokeInst *II = dyn_cast<InvokeInst>(TheCall))
+ HandleInlinedInvoke(II, FirstNewBlock, InlinedFunctionInfo);
// If we cloned in _exactly one_ basic block, and if that block ends in a
// return instruction, we splice the body of the inlined callee directly into
// Now we can remove the CalleeEntry block, which is now empty.
Caller->getBasicBlockList().erase(CalleeEntry);
+
return true;
}
projects / oota-llvm.git / blobdiff