/*
* Instrument select instructions similar to how we instrument branches.
*/
void PandaInstrumentVisitor::visitSelectInst(SelectInst &I){
BinaryOperator *BO;
ZExtInst *ZEI;
CallInst *CI;
std::vector<Value*> argValues;
Value *condition;
Function *F = mod->getFunction("log_dynval");
if (!F) {
printf("Instrumentation function not found\n");
assert(1==0);
}
condition = I.getCondition();
BO = static_cast<BinaryOperator*>(IRB.CreateNot(condition));
ZEI = static_cast<ZExtInst*>(IRB.CreateZExt(BO, wordType));
argValues.push_back(ConstantInt::get(ptrType,
(uintptr_t)dynval_buffer));
argValues.push_back(ConstantInt::get(intType, SELECTENTRY));
argValues.push_back(ConstantInt::get(intType, SELECT));
argValues.push_back(static_cast<Value*>(ZEI));
CI = IRB.CreateCall(F, ArrayRef<Value*>(argValues));
CI->insertBefore(static_cast<Instruction*>(&I));
ZEI->insertBefore(static_cast<Instruction*>(CI));
BO->insertBefore(static_cast<Instruction*>(ZEI));
}
/// Insert code in the prolog code when unrolling a loop with a
/// run-time trip-count.
///
/// This method assumes that the loop unroll factor is total number
/// of loop bodes in the loop after unrolling. (Some folks refer
/// to the unroll factor as the number of *extra* copies added).
/// We assume also that the loop unroll factor is a power-of-two. So, after
/// unrolling the loop, the number of loop bodies executed is 2,
/// 4, 8, etc. Note - LLVM converts the if-then-sequence to a switch
/// instruction in SimplifyCFG.cpp. Then, the backend decides how code for
/// the switch instruction is generated.
///
/// extraiters = tripcount % loopfactor
/// if (extraiters == 0) jump Loop:
/// if (extraiters == loopfactor) jump L1
/// if (extraiters == loopfactor-1) jump L2
/// ...
/// L1: LoopBody;
/// L2: LoopBody;
/// ...
/// if tripcount < loopfactor jump End
/// Loop:
/// ...
/// End:
///
bool llvm::UnrollRuntimeLoopProlog(Loop *L, unsigned Count, LoopInfo *LI,
LPPassManager *LPM) {
// for now, only unroll loops that contain a single exit
if (!L->getExitingBlock())
return false;
// Make sure the loop is in canonical form, and there is a single
// exit block only.
if (!L->isLoopSimplifyForm() || !L->getUniqueExitBlock())
return false;
// Use Scalar Evolution to compute the trip count. This allows more
// loops to be unrolled than relying on induction var simplification
if (!LPM)
return false;
ScalarEvolution *SE = LPM->getAnalysisIfAvailable<ScalarEvolution>();
if (!SE)
return false;
// Only unroll loops with a computable trip count and the trip count needs
// to be an int value (allowing a pointer type is a TODO item)
const SCEV *BECount = SE->getBackedgeTakenCount(L);
if (isa<SCEVCouldNotCompute>(BECount) || !BECount->getType()->isIntegerTy())
return false;
// Add 1 since the backedge count doesn't include the first loop iteration
const SCEV *TripCountSC =
SE->getAddExpr(BECount, SE->getConstant(BECount->getType(), 1));
if (isa<SCEVCouldNotCompute>(TripCountSC))
return false;
// We only handle cases when the unroll factor is a power of 2.
// Count is the loop unroll factor, the number of extra copies added + 1.
if ((Count & (Count-1)) != 0)
return false;
// If this loop is nested, then the loop unroller changes the code in
// parent loop, so the Scalar Evolution pass needs to be run again
if (Loop *ParentLoop = L->getParentLoop())
SE->forgetLoop(ParentLoop);
BasicBlock *PH = L->getLoopPreheader();
BasicBlock *Header = L->getHeader();
BasicBlock *Latch = L->getLoopLatch();
// It helps to splits the original preheader twice, one for the end of the
// prolog code and one for a new loop preheader
BasicBlock *PEnd = SplitEdge(PH, Header, LPM->getAsPass());
BasicBlock *NewPH = SplitBlock(PEnd, PEnd->getTerminator(), LPM->getAsPass());
BranchInst *PreHeaderBR = cast<BranchInst>(PH->getTerminator());
// Compute the number of extra iterations required, which is:
// extra iterations = run-time trip count % (loop unroll factor + 1)
SCEVExpander Expander(*SE, "loop-unroll");
Value *TripCount = Expander.expandCodeFor(TripCountSC, TripCountSC->getType(),
PreHeaderBR);
Type *CountTy = TripCount->getType();
BinaryOperator *ModVal =
BinaryOperator::CreateURem(TripCount,
ConstantInt::get(CountTy, Count),
"xtraiter");
ModVal->insertBefore(PreHeaderBR);
// Check if for no extra iterations, then jump to unrolled loop
Value *BranchVal = new ICmpInst(PreHeaderBR,
ICmpInst::ICMP_NE, ModVal,
ConstantInt::get(CountTy, 0), "lcmp");
// Branch to either the extra iterations or the unrolled loop
// We will fix up the true branch label when adding loop body copies
BranchInst::Create(PEnd, PEnd, BranchVal, PreHeaderBR);
assert(PreHeaderBR->isUnconditional() &&
PreHeaderBR->getSuccessor(0) == PEnd &&
"CFG edges in Preheader are not correct");
PreHeaderBR->eraseFromParent();
ValueToValueMapTy LVMap;
Function *F = Header->getParent();
// These variables are used to update the CFG links in each iteration
BasicBlock *CompareBB = nullptr;
BasicBlock *LastLoopBB = PH;
// Get an ordered list of blocks in the loop to help with the ordering of the
// cloned blocks in the prolog code
LoopBlocksDFS LoopBlocks(L);
LoopBlocks.perform(LI);
//
// For each extra loop iteration, create a copy of the loop's basic blocks
// and generate a condition that branches to the copy depending on the
// number of 'left over' iterations.
//
for (unsigned leftOverIters = Count-1; leftOverIters > 0; --leftOverIters) {
std::vector<BasicBlock*> NewBlocks;
ValueToValueMapTy VMap;
// Clone all the basic blocks in the loop, but we don't clone the loop
// This function adds the appropriate CFG connections.
CloneLoopBlocks(L, (leftOverIters == Count-1), LastLoopBB, PEnd, NewBlocks,
LoopBlocks, VMap, LVMap, LI);
LastLoopBB = cast<BasicBlock>(VMap[Latch]);
// Insert the cloned blocks into function just before the original loop
F->getBasicBlockList().splice(PEnd, F->getBasicBlockList(),
NewBlocks[0], F->end());
// Generate the code for the comparison which determines if the loop
// prolog code needs to be executed.
if (leftOverIters == Count-1) {
// There is no compare block for the fall-thru case when for the last
// left over iteration
CompareBB = NewBlocks[0];
} else {
// Create a new block for the comparison
BasicBlock *NewBB = BasicBlock::Create(CompareBB->getContext(), "unr.cmp",
F, CompareBB);
if (Loop *ParentLoop = L->getParentLoop()) {
// Add the new block to the parent loop, if needed
ParentLoop->addBasicBlockToLoop(NewBB, LI->getBase());
}
// The comparison w/ the extra iteration value and branch
Value *BranchVal = new ICmpInst(*NewBB, ICmpInst::ICMP_EQ, ModVal,
ConstantInt::get(CountTy, leftOverIters),
"un.tmp");
// Branch to either the extra iterations or the unrolled loop
BranchInst::Create(NewBlocks[0], CompareBB,
BranchVal, NewBB);
CompareBB = NewBB;
PH->getTerminator()->setSuccessor(0, NewBB);
VMap[NewPH] = CompareBB;
}
// Rewrite the cloned instruction operands to use the values
// created when the clone is created.
for (unsigned i = 0, e = NewBlocks.size(); i != e; ++i) {
for (BasicBlock::iterator I = NewBlocks[i]->begin(),
E = NewBlocks[i]->end(); I != E; ++I) {
RemapInstruction(I, VMap,
RF_NoModuleLevelChanges|RF_IgnoreMissingEntries);
}
}
}
// Connect the prolog code to the original loop and update the
// PHI functions.
ConnectProlog(L, TripCount, Count, LastLoopBB, PEnd, PH, NewPH, LVMap,
LPM->getAsPass());
NumRuntimeUnrolled++;
return true;
}
/*
* Call the logging function, logging the branch target. Target[0] is the true
* branch, and target[1] is the false branch. So when logging, we NOT the
* condition to actually log the target taken. We are also logging and
* processing unconditional branches for the time being.
*/
void PandaInstrumentVisitor::visitBranchInst(BranchInst &I){
BinaryOperator *BO;
ZExtInst *ZEI;
CallInst *CI;
std::vector<Value*> argValues;
Value *condition;
Function *F = mod->getFunction("log_dynval");
if (!F) {
printf("Instrumentation function not found\n");
assert(1==0);
}
if (I.isConditional()){
condition = I.getCondition();
if(isa<UndefValue>(condition)){
BO = static_cast<BinaryOperator*>(IRB.CreateNot(condition));
ZEI = static_cast<ZExtInst*>(IRB.CreateZExt(BO, wordType));
argValues.push_back(ConstantInt::get(ptrType,
(uintptr_t)dynval_buffer));
argValues.push_back(ConstantInt::get(intType, BRANCHENTRY));
argValues.push_back(ConstantInt::get(intType, BRANCHOP));
argValues.push_back(static_cast<Value*>(ZEI));
CI = IRB.CreateCall(F, ArrayRef<Value*>(argValues));
CI->insertBefore(static_cast<Instruction*>(&I));
}
else if (isa<Constant>(condition)){
CallInst *CI;
std::vector<Value*> argValues;
uint64_t constcond = static_cast<ConstantInt*>(
I.getCondition())->getZExtValue();
argValues.push_back(ConstantInt::get(ptrType,
(uintptr_t)dynval_buffer));
argValues.push_back(ConstantInt::get(intType, BRANCHENTRY));
argValues.push_back(ConstantInt::get(intType, BRANCHOP));
argValues.push_back(ConstantInt::get(wordType, !constcond));
CI = IRB.CreateCall(F, ArrayRef<Value*>(argValues));
CI->insertBefore(static_cast<Instruction*>(&I));
}
else {
BO = static_cast<BinaryOperator*>(IRB.CreateNot(condition));
ZEI = static_cast<ZExtInst*>(IRB.CreateZExt(BO, wordType));
argValues.push_back(ConstantInt::get(ptrType,
(uintptr_t)dynval_buffer));
argValues.push_back(ConstantInt::get(intType, BRANCHENTRY));
argValues.push_back(ConstantInt::get(intType, BRANCHOP));
argValues.push_back(static_cast<Value*>(ZEI));
CI = IRB.CreateCall(F, ArrayRef<Value*>(argValues));
CI->insertBefore(static_cast<Instruction*>(&I));
ZEI->insertBefore(static_cast<Instruction*>(CI));
BO->insertBefore(static_cast<Instruction*>(ZEI));
}
}
else {
argValues.push_back(ConstantInt::get(ptrType,
(uintptr_t)dynval_buffer));
argValues.push_back(ConstantInt::get(intType, BRANCHENTRY));
argValues.push_back(ConstantInt::get(intType, BRANCHOP));
argValues.push_back(ConstantInt::get(wordType, 0));
CI = IRB.CreateCall(F, ArrayRef<Value*>(argValues));
CI->insertBefore(static_cast<Instruction*>(&I));
}
}