typedef DenseMap<Instruction *, Value *> RematerializedValueMapTy;
struct PartiallyConstructedSafepointRecord {
- /// The set of values known to be live accross this safepoint
+ /// The set of values known to be live across this safepoint
StatepointLiveSetTy liveset;
/// Mapping from live pointers to a base-defining-value
// TODO: Once we can get to the GCStrategy, this becomes
// Optional<bool> isGCManagedPointer(const Value *V) const override {
-static bool isGCPointerType(const Type *T) {
- if (const PointerType *PT = dyn_cast<PointerType>(T))
+static bool isGCPointerType(Type *T) {
+ if (auto *PT = dyn_cast<PointerType>(T))
// For the sake of this example GC, we arbitrarily pick addrspace(1) as our
// GC managed heap. We know that a pointer into this heap needs to be
// updated and that no other pointer does.
if (PrintLiveSet) {
// Note: This output is used by several of the test cases
- // The order of elemtns in a set is not stable, put them in a vec and sort
+ // The order of elements in a set is not stable, put them in a vec and sort
// by name
SmallVector<Value *, 64> temp;
temp.insert(temp.end(), liveset.begin(), liveset.end());
static Value *findBaseDefiningValue(Value *I);
-/// If we can trivially determine that the index specified in the given vector
-/// is a base pointer, return it. In cases where the entire vector is known to
-/// consist of base pointers, the entire vector will be returned. This
-/// indicates that the relevant extractelement is a valid base pointer and
-/// should be used directly.
-static Value *findBaseOfVector(Value *I, Value *Index) {
+/// Return a base defining value for the 'Index' element of the given vector
+/// instruction 'I'. If Index is null, returns a BDV for the entire vector
+/// 'I'. As an optimization, this method will try to determine when the
+/// element is known to already be a base pointer. If this can be established,
+/// the second value in the returned pair will be true. Note that either a
+/// vector or a pointer typed value can be returned. For the former, the
+/// vector returned is a BDV (and possibly a base) of the entire vector 'I'.
+/// If the later, the return pointer is a BDV (or possibly a base) for the
+/// particular element in 'I'.
+static std::pair<Value *, bool>
+findBaseDefiningValueOfVector(Value *I, Value *Index = nullptr) {
assert(I->getType()->isVectorTy() &&
cast<VectorType>(I->getType())->getElementType()->isPointerTy() &&
"Illegal to ask for the base pointer of a non-pointer type");
if (isa<Argument>(I))
// An incoming argument to the function is a base pointer
- return I;
+ return std::make_pair(I, true);
// We shouldn't see the address of a global as a vector value?
assert(!isa<GlobalVariable>(I) &&
if (isa<UndefValue>(I))
// utterly meaningless, but useful for dealing with partially optimized
// code.
- return I;
+ return std::make_pair(I, true);
// Due to inheritance, this must be _after_ the global variable and undef
// checks
assert(!isa<GlobalVariable>(I) && !isa<UndefValue>(I) &&
"order of checks wrong!");
assert(Con->isNullValue() && "null is the only case which makes sense");
- return Con;
+ return std::make_pair(Con, true);
}
-
+
if (isa<LoadInst>(I))
- return I;
-
+ return std::make_pair(I, true);
+
// For an insert element, we might be able to look through it if we know
- // something about the indexes, but if the indices are arbitrary values, we
- // can't without much more extensive scalarization.
+ // something about the indexes.
if (InsertElementInst *IEI = dyn_cast<InsertElementInst>(I)) {
- Value *InsertIndex = IEI->getOperand(2);
- // This index is inserting the value, look for it's base
- if (InsertIndex == Index)
- return findBaseDefiningValue(IEI->getOperand(1));
- // Both constant, and can't be equal per above. This insert is definitely
- // not relevant, look back at the rest of the vector and keep trying.
- if (isa<ConstantInt>(Index) && isa<ConstantInt>(InsertIndex))
- return findBaseOfVector(IEI->getOperand(0), Index);
- }
-
- // Note: This code is currently rather incomplete. We are essentially only
- // handling cases where the vector element is trivially a base pointer. We
- // need to update the entire base pointer construction algorithm to know how
- // to track vector elements and potentially scalarize, but the case which
- // would motivate the work hasn't shown up in real workloads yet.
- llvm_unreachable("no base found for vector element");
+ if (Index) {
+ Value *InsertIndex = IEI->getOperand(2);
+ // This index is inserting the value, look for its BDV
+ if (InsertIndex == Index)
+ return std::make_pair(findBaseDefiningValue(IEI->getOperand(1)), false);
+ // Both constant, and can't be equal per above. This insert is definitely
+ // not relevant, look back at the rest of the vector and keep trying.
+ if (isa<ConstantInt>(Index) && isa<ConstantInt>(InsertIndex))
+ return findBaseDefiningValueOfVector(IEI->getOperand(0), Index);
+ }
+
+ // We don't know whether this vector contains entirely base pointers or
+ // not. To be conservatively correct, we treat it as a BDV and will
+ // duplicate code as needed to construct a parallel vector of bases.
+ return std::make_pair(IEI, false);
+ }
+
+ if (isa<ShuffleVectorInst>(I))
+ // We don't know whether this vector contains entirely base pointers or
+ // not. To be conservatively correct, we treat it as a BDV and will
+ // duplicate code as needed to construct a parallel vector of bases.
+ // TODO: There a number of local optimizations which could be applied here
+ // for particular sufflevector patterns.
+ return std::make_pair(I, false);
+
+ // A PHI or Select is a base defining value. The outer findBasePointer
+ // algorithm is responsible for constructing a base value for this BDV.
+ assert((isa<SelectInst>(I) || isa<PHINode>(I)) &&
+ "unknown vector instruction - no base found for vector element");
+ return std::make_pair(I, false);
}
+static bool isKnownBaseResult(Value *V);
+
/// Helper function for findBasePointer - Will return a value which either a)
/// defines the base pointer for the input or b) blocks the simple search
/// (i.e. a PHI or Select of two derived pointers)
static Value *findBaseDefiningValue(Value *I) {
+ if (I->getType()->isVectorTy())
+ return findBaseDefiningValueOfVector(I).first;
+
assert(I->getType()->isPointerTy() &&
"Illegal to ask for the base pointer of a non-pointer type");
if (auto *EEI = dyn_cast<ExtractElementInst>(I)) {
Value *VectorOperand = EEI->getVectorOperand();
Value *Index = EEI->getIndexOperand();
- Value *VectorBase = findBaseOfVector(VectorOperand, Index);
- // If the result returned is a vector, we know the entire vector must
- // contain base pointers. In that case, the extractelement is a valid base
- // for this value.
- if (VectorBase->getType()->isVectorTy())
- return EEI;
- // Otherwise, we needed to look through the vector to find the base for
- // this particular element.
- assert(VectorBase->getType()->isPointerTy());
- return VectorBase;
+ std::pair<Value *, bool> pair =
+ findBaseDefiningValueOfVector(VectorOperand, Index);
+ Value *VectorBase = pair.first;
+ if (VectorBase->getType()->isPointerTy())
+ // We found a BDV for this specific element with the vector. This is an
+ // optimization, but in practice it covers most of the useful cases
+ // created via scalarization.
+ return VectorBase;
+ else {
+ assert(VectorBase->getType()->isVectorTy());
+ if (pair.second)
+ // If the entire vector returned is known to be entirely base pointers,
+ // then the extractelement is valid base for this value.
+ return EEI;
+ else {
+ // Otherwise, we have an instruction which potentially produces a
+ // derived pointer and we need findBasePointers to clone code for us
+ // such that we can create an instruction which produces the
+ // accompanying base pointer.
+ // Note: This code is currently rather incomplete. We don't currently
+ // support the general form of shufflevector of insertelement.
+ // Conceptually, these are just 'base defining values' of the same
+ // variety as phi or select instructions. We need to update the
+ // findBasePointers algorithm to insert new 'base-only' versions of the
+ // original instructions. This is relative straight forward to do, but
+ // the case which would motivate the work hasn't shown up in real
+ // workloads yet.
+ assert((isa<PHINode>(VectorBase) || isa<SelectInst>(VectorBase)) &&
+ "need to extend findBasePointers for generic vector"
+ "instruction cases");
+ return VectorBase;
+ }
+ }
}
if (isa<Argument>(I))
return I;
// I have absolutely no idea how to implement this part yet. It's not
- // neccessarily hard, I just haven't really looked at it yet.
+ // necessarily hard, I just haven't really looked at it yet.
assert(!isa<LandingPadInst>(I) && "Landing Pad is unimplemented");
if (isa<AtomicCmpXchgInst>(I))
"Base pointer for a struct is meaningless");
// The last two cases here don't return a base pointer. Instead, they
- // return a value which dynamically selects from amoung several base
+ // return a value which dynamically selects from among several base
// derived pointers (each with it's own base potentially). It's the job of
// the caller to resolve these.
assert((isa<SelectInst>(I) || isa<PHINode>(I)) &&
Value *&Cached = Cache[I];
if (!Cached) {
Cached = findBaseDefiningValue(I);
+ DEBUG(dbgs() << "fBDV-cached: " << I->getName() << " -> "
+ << Cached->getName() << "\n");
}
assert(Cache[I] != nullptr);
-
- if (TraceLSP) {
- dbgs() << "fBDV-cached: " << I->getName() << " -> " << Cached->getName()
- << "\n";
- }
return Cached;
}
return false;
}
-// TODO: find a better name for this
namespace {
-class PhiState {
+/// Models the state of a single base defining value in the findBasePointer
+/// algorithm for determining where a new instruction is needed to propagate
+/// the base of this BDV.
+class BDVState {
public:
enum Status { Unknown, Base, Conflict };
- PhiState(Status s, Value *b = nullptr) : status(s), base(b) {
+ BDVState(Status s, Value *b = nullptr) : status(s), base(b) {
assert(status != Base || b);
}
- PhiState(Value *b) : status(Base), base(b) {}
- PhiState() : status(Unknown), base(nullptr) {}
+ explicit BDVState(Value *b) : status(Base), base(b) {}
+ BDVState() : status(Unknown), base(nullptr) {}
Status getStatus() const { return status; }
Value *getBase() const { return base; }
bool isUnknown() const { return getStatus() == Unknown; }
bool isConflict() const { return getStatus() == Conflict; }
- bool operator==(const PhiState &other) const {
+ bool operator==(const BDVState &other) const {
return base == other.base && status == other.status;
}
- bool operator!=(const PhiState &other) const { return !(*this == other); }
+ bool operator!=(const BDVState &other) const { return !(*this == other); }
- void dump() {
- errs() << status << " (" << base << " - "
- << (base ? base->getName() : "nullptr") << "): ";
+ LLVM_DUMP_METHOD
+ void dump() const { print(dbgs()); dbgs() << '\n'; }
+
+ void print(raw_ostream &OS) const {
+ OS << status << " (" << base << " - "
+ << (base ? base->getName() : "nullptr") << "): ";
}
private:
Value *base; // non null only if status == base
};
-typedef DenseMap<Value *, PhiState> ConflictStateMapTy;
-// Values of type PhiState form a lattice, and this is a helper
+inline raw_ostream &operator<<(raw_ostream &OS, const BDVState &State) {
+ State.print(OS);
+ return OS;
+}
+
+
+typedef DenseMap<Value *, BDVState> ConflictStateMapTy;
+// Values of type BDVState form a lattice, and this is a helper
// class that implementes the meet operation. The meat of the meet
-// operation is implemented in MeetPhiStates::pureMeet
-class MeetPhiStates {
+// operation is implemented in MeetBDVStates::pureMeet
+class MeetBDVStates {
public:
- // phiStates is a mapping from PHINodes and SelectInst's to PhiStates.
- explicit MeetPhiStates(const ConflictStateMapTy &phiStates)
- : phiStates(phiStates) {}
-
- // Destructively meet the current result with the base V. V can
- // either be a merge instruction (SelectInst / PHINode), in which
- // case its status is looked up in the phiStates map; or a regular
- // SSA value, in which case it is assumed to be a base.
- void meetWith(Value *V) {
- PhiState otherState = getStateForBDV(V);
- assert((MeetPhiStates::pureMeet(otherState, currentResult) ==
- MeetPhiStates::pureMeet(currentResult, otherState)) &&
- "math is wrong: meet does not commute!");
- currentResult = MeetPhiStates::pureMeet(otherState, currentResult);
+ /// Initializes the currentResult to the TOP state so that if can be met with
+ /// any other state to produce that state.
+ MeetBDVStates() {}
+
+ // Destructively meet the current result with the given BDVState
+ void meetWith(BDVState otherState) {
+ currentResult = meet(otherState, currentResult);
}
- PhiState getResult() const { return currentResult; }
+ BDVState getResult() const { return currentResult; }
private:
- const ConflictStateMapTy &phiStates;
- PhiState currentResult;
-
- /// Return a phi state for a base defining value. We'll generate a new
- /// base state for known bases and expect to find a cached state otherwise
- PhiState getStateForBDV(Value *baseValue) {
- if (isKnownBaseResult(baseValue)) {
- return PhiState(baseValue);
- } else {
- return lookupFromMap(baseValue);
- }
- }
+ BDVState currentResult;
- PhiState lookupFromMap(Value *V) {
- auto I = phiStates.find(V);
- assert(I != phiStates.end() && "lookup failed!");
- return I->second;
+ /// Perform a meet operation on two elements of the BDVState lattice.
+ static BDVState meet(BDVState LHS, BDVState RHS) {
+ assert((pureMeet(LHS, RHS) == pureMeet(RHS, LHS)) &&
+ "math is wrong: meet does not commute!");
+ BDVState Result = pureMeet(LHS, RHS);
+ DEBUG(dbgs() << "meet of " << LHS << " with " << RHS
+ << " produced " << Result << "\n");
+ return Result;
}
- static PhiState pureMeet(const PhiState &stateA, const PhiState &stateB) {
+ static BDVState pureMeet(const BDVState &stateA, const BDVState &stateB) {
switch (stateA.getStatus()) {
- case PhiState::Unknown:
+ case BDVState::Unknown:
return stateB;
- case PhiState::Base:
+ case BDVState::Base:
assert(stateA.getBase() && "can't be null");
if (stateB.isUnknown())
return stateA;
assert(stateA == stateB && "equality broken!");
return stateA;
}
- return PhiState(PhiState::Conflict);
+ return BDVState(BDVState::Conflict);
}
assert(stateB.isConflict() && "only three states!");
- return PhiState(PhiState::Conflict);
+ return BDVState(BDVState::Conflict);
- case PhiState::Conflict:
+ case BDVState::Conflict:
return stateA;
}
llvm_unreachable("only three states!");
//
// Note: A simpler form of this would be to add the conflict form of all
// PHIs without running the optimistic algorithm. This would be
- // analougous to pessimistic data flow and would likely lead to an
+ // analogous to pessimistic data flow and would likely lead to an
// overall worse solution.
+#ifndef NDEBUG
+ auto isExpectedBDVType = [](Value *BDV) {
+ return isa<PHINode>(BDV) || isa<SelectInst>(BDV);
+ };
+#endif
+
+ // Once populated, will contain a mapping from each potentially non-base BDV
+ // to a lattice value (described above) which corresponds to that BDV.
ConflictStateMapTy states;
- states[def] = PhiState();
// Recursively fill in all phis & selects reachable from the initial one
// for which we don't already know a definite base value for
- // TODO: This should be rewritten with a worklist
- bool done = false;
- while (!done) {
- done = true;
- // Since we're adding elements to 'states' as we run, we can't keep
- // iterators into the set.
- SmallVector<Value *, 16> Keys;
- Keys.reserve(states.size());
- for (auto Pair : states) {
- Value *V = Pair.first;
- Keys.push_back(V);
- }
- for (Value *v : Keys) {
- assert(!isKnownBaseResult(v) && "why did it get added?");
- if (PHINode *phi = dyn_cast<PHINode>(v)) {
- assert(phi->getNumIncomingValues() > 0 &&
- "zero input phis are illegal");
- for (Value *InVal : phi->incoming_values()) {
- Value *local = findBaseOrBDV(InVal, cache);
- if (!isKnownBaseResult(local) && states.find(local) == states.end()) {
- states[local] = PhiState();
- done = false;
- }
- }
- } else if (SelectInst *sel = dyn_cast<SelectInst>(v)) {
- Value *local = findBaseOrBDV(sel->getTrueValue(), cache);
- if (!isKnownBaseResult(local) && states.find(local) == states.end()) {
- states[local] = PhiState();
- done = false;
- }
- local = findBaseOrBDV(sel->getFalseValue(), cache);
- if (!isKnownBaseResult(local) && states.find(local) == states.end()) {
- states[local] = PhiState();
- done = false;
- }
+ /* scope */ {
+ DenseSet<Value *> Visited;
+ SmallVector<Value*, 16> Worklist;
+ Worklist.push_back(def);
+ Visited.insert(def);
+ while (!Worklist.empty()) {
+ Value *Current = Worklist.pop_back_val();
+ assert(!isKnownBaseResult(Current) && "why did it get added?");
+
+ auto visitIncomingValue = [&](Value *InVal) {
+ Value *Base = findBaseOrBDV(InVal, cache);
+ if (isKnownBaseResult(Base))
+ // Known bases won't need new instructions introduced and can be
+ // ignored safely
+ return;
+ assert(isExpectedBDVType(Base) && "the only non-base values "
+ "we see should be base defining values");
+ if (Visited.insert(Base).second)
+ Worklist.push_back(Base);
+ };
+ if (PHINode *Phi = dyn_cast<PHINode>(Current)) {
+ for (Value *InVal : Phi->incoming_values())
+ visitIncomingValue(InVal);
+ } else {
+ SelectInst *Sel = cast<SelectInst>(Current);
+ visitIncomingValue(Sel->getTrueValue());
+ visitIncomingValue(Sel->getFalseValue());
}
}
+ // The frontier of visited instructions are the ones we might need to
+ // duplicate, so fill in the starting state for the optimistic algorithm
+ // that follows.
+ for (Value *BDV : Visited) {
+ states[BDV] = BDVState();
+ }
}
if (TraceLSP) {
errs() << "States after initialization:\n";
- for (auto Pair : states) {
- Instruction *v = cast<Instruction>(Pair.first);
- PhiState state = Pair.second;
- state.dump();
- v->dump();
- }
+ for (auto Pair : states)
+ dbgs() << " " << Pair.second << " for " << Pair.first << "\n";
}
// TODO: come back and revisit the state transitions around inputs which
// have reached conflict state. The current version seems too conservative.
+ // Return a phi state for a base defining value. We'll generate a new
+ // base state for known bases and expect to find a cached state otherwise.
+ auto getStateForBDV = [&](Value *baseValue) {
+ if (isKnownBaseResult(baseValue))
+ return BDVState(baseValue);
+ auto I = states.find(baseValue);
+ assert(I != states.end() && "lookup failed!");
+ return I->second;
+ };
+
bool progress = true;
while (progress) {
#ifndef NDEBUG
progress = false;
// We're only changing keys in this loop, thus safe to keep iterators
for (auto Pair : states) {
- MeetPhiStates calculateMeet(states);
Value *v = Pair.first;
assert(!isKnownBaseResult(v) && "why did it get added?");
+
+ // Given an input value for the current instruction, return a BDVState
+ // instance which represents the BDV of that value.
+ auto getStateForInput = [&](Value *V) mutable {
+ Value *BDV = findBaseOrBDV(V, cache);
+ return getStateForBDV(BDV);
+ };
+
+ MeetBDVStates calculateMeet;
if (SelectInst *select = dyn_cast<SelectInst>(v)) {
- calculateMeet.meetWith(findBaseOrBDV(select->getTrueValue(), cache));
- calculateMeet.meetWith(findBaseOrBDV(select->getFalseValue(), cache));
+ calculateMeet.meetWith(getStateForInput(select->getTrueValue()));
+ calculateMeet.meetWith(getStateForInput(select->getFalseValue()));
} else
for (Value *Val : cast<PHINode>(v)->incoming_values())
- calculateMeet.meetWith(findBaseOrBDV(Val, cache));
+ calculateMeet.meetWith(getStateForInput(Val));
- PhiState oldState = states[v];
- PhiState newState = calculateMeet.getResult();
+ BDVState oldState = states[v];
+ BDVState newState = calculateMeet.getResult();
if (oldState != newState) {
progress = true;
states[v] = newState;
if (TraceLSP) {
errs() << "States after meet iteration:\n";
- for (auto Pair : states) {
- Instruction *v = cast<Instruction>(Pair.first);
- PhiState state = Pair.second;
- state.dump();
- v->dump();
- }
+ for (auto Pair : states)
+ dbgs() << " " << Pair.second << " for " << Pair.first << "\n";
}
// Insert Phis for all conflicts
std::sort(Keys.begin(), Keys.end(), order_by_name);
// TODO: adjust naming patterns to avoid this order of iteration dependency
for (Value *V : Keys) {
- Instruction *v = cast<Instruction>(V);
- PhiState state = states[V];
- assert(!isKnownBaseResult(v) && "why did it get added?");
- assert(!state.isUnknown() && "Optimistic algorithm didn't complete!");
- if (!state.isConflict())
+ Instruction *I = cast<Instruction>(V);
+ BDVState State = states[I];
+ assert(!isKnownBaseResult(I) && "why did it get added?");
+ assert(!State.isUnknown() && "Optimistic algorithm didn't complete!");
+ if (!State.isConflict())
continue;
- if (isa<PHINode>(v)) {
- int num_preds =
- std::distance(pred_begin(v->getParent()), pred_end(v->getParent()));
- assert(num_preds > 0 && "how did we reach here");
- PHINode *phi = PHINode::Create(v->getType(), num_preds, "base_phi", v);
- // Add metadata marking this as a base value
- auto *const_1 = ConstantInt::get(
- Type::getInt32Ty(
- v->getParent()->getParent()->getParent()->getContext()),
- 1);
- auto MDConst = ConstantAsMetadata::get(const_1);
- MDNode *md = MDNode::get(
- v->getParent()->getParent()->getParent()->getContext(), MDConst);
- phi->setMetadata("is_base_value", md);
- states[v] = PhiState(PhiState::Conflict, phi);
- } else {
- SelectInst *sel = cast<SelectInst>(v);
+ /// Create and insert a new instruction which will represent the base of
+ /// the given instruction 'I'.
+ auto MakeBaseInstPlaceholder = [](Instruction *I) -> Instruction* {
+ if (isa<PHINode>(I)) {
+ BasicBlock *BB = I->getParent();
+ int NumPreds = std::distance(pred_begin(BB), pred_end(BB));
+ assert(NumPreds > 0 && "how did we reach here");
+ std::string Name = I->hasName() ?
+ (I->getName() + ".base").str() : "base_phi";
+ return PHINode::Create(I->getType(), NumPreds, Name, I);
+ }
+ SelectInst *Sel = cast<SelectInst>(I);
// The undef will be replaced later
- UndefValue *undef = UndefValue::get(sel->getType());
- SelectInst *basesel = SelectInst::Create(sel->getCondition(), undef,
- undef, "base_select", sel);
- // Add metadata marking this as a base value
- auto *const_1 = ConstantInt::get(
- Type::getInt32Ty(
- v->getParent()->getParent()->getParent()->getContext()),
- 1);
- auto MDConst = ConstantAsMetadata::get(const_1);
- MDNode *md = MDNode::get(
- v->getParent()->getParent()->getParent()->getContext(), MDConst);
- basesel->setMetadata("is_base_value", md);
- states[v] = PhiState(PhiState::Conflict, basesel);
- }
+ UndefValue *Undef = UndefValue::get(Sel->getType());
+ std::string Name = I->hasName() ?
+ (I->getName() + ".base").str() : "base_select";
+ return SelectInst::Create(Sel->getCondition(), Undef,
+ Undef, Name, Sel);
+ };
+ Instruction *BaseInst = MakeBaseInstPlaceholder(I);
+ // Add metadata marking this as a base value
+ BaseInst->setMetadata("is_base_value", MDNode::get(I->getContext(), {}));
+ states[I] = BDVState(BDVState::Conflict, BaseInst);
}
// Fixup all the inputs of the new PHIs
for (auto Pair : states) {
Instruction *v = cast<Instruction>(Pair.first);
- PhiState state = Pair.second;
+ BDVState state = Pair.second;
assert(!isKnownBaseResult(v) && "why did it get added?");
assert(!state.isUnknown() && "Optimistic algorithm didn't complete!");
// Either conflict or base.
assert(states.count(base));
base = states[base].getBase();
- assert(base != nullptr && "unknown PhiState!");
+ assert(base != nullptr && "unknown BDVState!");
}
- // In essense this assert states: the only way two
+ // In essence this assert states: the only way two
// values incoming from the same basic block may be
// different is by being different bitcasts of the same
// value. A cleanup that remains TODO is changing
// Either conflict or base.
assert(states.count(base));
base = states[base].getBase();
- assert(base != nullptr && "unknown PhiState!");
+ assert(base != nullptr && "unknown BDVState!");
}
assert(base && "can't be null");
// Must use original input BB since base may not be Instruction
// Either conflict or base.
assert(states.count(base));
base = states[base].getBase();
- assert(base != nullptr && "unknown PhiState!");
+ assert(base != nullptr && "unknown BDVState!");
}
assert(base && "can't be null");
// Must use original input BB since base may not be Instruction
// If you see this trip and like to live really dangerously, the code should
// be correct, just with idioms the verifier can't handle. You can try
- // disabling the verifier at your own substaintial risk.
+ // disabling the verifier at your own substantial risk.
assert(!isa<ConstantPointerNull>(base) &&
"the relocation code needs adjustment to handle the relocation of "
"a null pointer constant without causing false positives in the "
Function &F, DominatorTree &DT, Pass *P, ArrayRef<CallSite> toUpdate,
MutableArrayRef<struct PartiallyConstructedSafepointRecord> records) {
// TODO-PERF: reuse the original liveness, then simply run the dataflow
- // again. The old values are still live and will help it stablize quickly.
+ // again. The old values are still live and will help it stabilize quickly.
GCPtrLivenessData RevisedLivenessData;
computeLiveInValues(DT, F, RevisedLivenessData);
for (size_t i = 0; i < records.size(); i++) {
DominatorTree &DT) {
BasicBlock *Ret = BB;
if (!BB->getUniquePredecessor()) {
- Ret = SplitBlockPredecessors(BB, InvokeParent, "", nullptr, &DT);
+ Ret = SplitBlockPredecessors(BB, InvokeParent, "", &DT);
}
// Now that 'ret' has unique predecessor we can safely remove all phi nodes
return index;
}
-// Create new attribute set containing only attributes which can be transfered
+// Create new attribute set containing only attributes which can be transferred
// from original call to the safepoint.
static AttributeSet legalizeCallAttributes(AttributeSet AS) {
AttributeSet ret;
ArrayRef<llvm::Value *> BasePtrs,
Instruction *StatepointToken,
IRBuilder<> Builder) {
- SmallVector<Instruction *, 64> NewDefs;
- NewDefs.reserve(LiveVariables.size());
-
- Module *M = StatepointToken->getParent()->getParent()->getParent();
+ if (LiveVariables.empty())
+ return;
+
+ // All gc_relocate are set to i8 addrspace(1)* type. We originally generated
+ // unique declarations for each pointer type, but this proved problematic
+ // because the intrinsic mangling code is incomplete and fragile. Since
+ // we're moving towards a single unified pointer type anyways, we can just
+ // cast everything to an i8* of the right address space. A bitcast is added
+ // later to convert gc_relocate to the actual value's type.
+ Module *M = StatepointToken->getModule();
+ auto AS = cast<PointerType>(LiveVariables[0]->getType())->getAddressSpace();
+ Type *Types[] = {Type::getInt8PtrTy(M->getContext(), AS)};
+ Value *GCRelocateDecl =
+ Intrinsic::getDeclaration(M, Intrinsic::experimental_gc_relocate, Types);
for (unsigned i = 0; i < LiveVariables.size(); i++) {
- // We generate a (potentially) unique declaration for every pointer type
- // combination. This results is some blow up the function declarations in
- // the IR, but removes the need for argument bitcasts which shrinks the IR
- // greatly and makes it much more readable.
- SmallVector<Type *, 1> Types; // one per 'any' type
- // All gc_relocate are set to i8 addrspace(1)* type. This could help avoid
- // cases where the actual value's type mangling is not supported by llvm. A
- // bitcast is added later to convert gc_relocate to the actual value's type.
- Types.push_back(Type::getInt8PtrTy(M->getContext(), 1));
- Value *GCRelocateDecl = Intrinsic::getDeclaration(
- M, Intrinsic::experimental_gc_relocate, Types);
-
// Generate the gc.relocate call and save the result
Value *BaseIdx =
- ConstantInt::get(Type::getInt32Ty(M->getContext()),
- LiveStart + find_index(LiveVariables, BasePtrs[i]));
- Value *LiveIdx = ConstantInt::get(
- Type::getInt32Ty(M->getContext()),
- LiveStart + find_index(LiveVariables, LiveVariables[i]));
+ Builder.getInt32(LiveStart + find_index(LiveVariables, BasePtrs[i]));
+ Value *LiveIdx =
+ Builder.getInt32(LiveStart + find_index(LiveVariables, LiveVariables[i]));
// only specify a debug name if we can give a useful one
- Value *Reloc = Builder.CreateCall(
+ CallInst *Reloc = Builder.CreateCall(
GCRelocateDecl, {StatepointToken, BaseIdx, LiveIdx},
LiveVariables[i]->hasName() ? LiveVariables[i]->getName() + ".relocated"
: "");
// Trick CodeGen into thinking there are lots of free registers at this
// fake call.
- cast<CallInst>(Reloc)->setCallingConv(CallingConv::Cold);
-
- NewDefs.push_back(cast<Instruction>(Reloc));
+ Reloc->setCallingConv(CallingConv::Cold);
}
- assert(NewDefs.size() == LiveVariables.size() &&
- "missing or extra redefinition at safepoint");
}
static void
// Currently we will fail on parameter attributes and on certain
// function attributes.
AttributeSet new_attrs = legalizeCallAttributes(toReplace->getAttributes());
- // In case if we can handle this set of sttributes - set up function attrs
+ // In case if we can handle this set of attributes - set up function attrs
// directly on statepoint and return attrs later for gc_result intrinsic.
call->setAttributes(new_attrs.getFnAttributes());
return_attributes = new_attrs.getRetAttributes();
// original block.
InvokeInst *invoke = InvokeInst::Create(
gc_statepoint_decl, toReplace->getNormalDest(),
- toReplace->getUnwindDest(), args, "", toReplace->getParent());
+ toReplace->getUnwindDest(), args, "statepoint_token", toReplace->getParent());
invoke->setCallingConv(toReplace->getCallingConv());
// Currently we will fail on parameter attributes and on certain
// function attributes.
AttributeSet new_attrs = legalizeCallAttributes(toReplace->getAttributes());
- // In case if we can handle this set of sttributes - set up function attrs
+ // In case if we can handle this set of attributes - set up function attrs
// directly on statepoint and return attrs later for gc_result intrinsic.
invoke->setAttributes(new_attrs.getFnAttributes());
return_attributes = new_attrs.getRetAttributes();
unwindBlock->getLandingPadInst(), idx, "relocate_token"));
result.UnwindToken = exceptional_token;
- // Just throw away return value. We will use the one we got for normal
- // block.
- (void)CreateGCRelocates(liveVariables, live_start, basePtrs,
- exceptional_token, Builder);
+ CreateGCRelocates(liveVariables, live_start, basePtrs,
+ exceptional_token, Builder);
// Generate gc relocates and returns for normal block
BasicBlock *normalDest = toReplace->getNormalDest();
basevec.reserve(liveset.size());
for (Value *L : liveset) {
livevec.push_back(L);
-
- assert(PointerToBase.find(L) != PointerToBase.end());
+ assert(PointerToBase.count(L));
Value *base = PointerToBase[L];
basevec.push_back(base);
}
VisitedLiveValues);
if (ClobberNonLive) {
- // As a debuging aid, pretend that an unrelocated pointer becomes null at
+ // As a debugging aid, pretend that an unrelocated pointer becomes null at
// the gc.statepoint. This will turn some subtle GC problems into
// slightly easier to debug SEGVs. Note that on large IR files with
// lots of gc.statepoints this is extremely costly both memory and time
/// Remove any vector of pointers from the liveset by scalarizing them over the
/// statepoint instruction. Adds the scalarized pieces to the liveset. It
-/// would be preferrable to include the vector in the statepoint itself, but
+/// would be preferable to include the vector in the statepoint itself, but
/// the lowering code currently does not handle that. Extending it would be
/// slightly non-trivial since it requires a format change. Given how rare
-/// such cases are (for the moment?) scalarizing is an acceptable comprimise.
+/// such cases are (for the moment?) scalarizing is an acceptable compromise.
static void splitVectorValues(Instruction *StatepointInst,
- StatepointLiveSetTy &LiveSet, DominatorTree &DT) {
+ StatepointLiveSetTy &LiveSet,
+ DenseMap<Value *, Value *>& PointerToBase,
+ DominatorTree &DT) {
SmallVector<Value *, 16> ToSplit;
for (Value *V : LiveSet)
if (isa<VectorType>(V->getType()))
if (ToSplit.empty())
return;
+ DenseMap<Value *, SmallVector<Value *, 16>> ElementMapping;
+
Function &F = *(StatepointInst->getParent()->getParent());
DenseMap<Value *, AllocaInst *> AllocaMap;
// First is normal return, second is exceptional return (invoke only)
DenseMap<Value *, std::pair<Value *, Value *>> Replacements;
for (Value *V : ToSplit) {
- LiveSet.erase(V);
-
AllocaInst *Alloca =
new AllocaInst(V->getType(), "", F.getEntryBlock().getFirstNonPHI());
AllocaMap[V] = Alloca;
SmallVector<Value *, 16> Elements;
for (unsigned i = 0; i < VT->getNumElements(); i++)
Elements.push_back(Builder.CreateExtractElement(V, Builder.getInt32(i)));
- LiveSet.insert(Elements.begin(), Elements.end());
+ ElementMapping[V] = Elements;
auto InsertVectorReform = [&](Instruction *IP) {
Builder.SetInsertPoint(IP);
Replacements[V].second = InsertVectorReform(IP);
}
}
+
for (Value *V : ToSplit) {
AllocaInst *Alloca = AllocaMap[V];
for (Value *V : ToSplit)
Allocas.push_back(AllocaMap[V]);
PromoteMemToReg(Allocas, DT);
+
+ // Update our tracking of live pointers and base mappings to account for the
+ // changes we just made.
+ for (Value *V : ToSplit) {
+ auto &Elements = ElementMapping[V];
+
+ LiveSet.erase(V);
+ LiveSet.insert(Elements.begin(), Elements.end());
+ // We need to update the base mapping as well.
+ assert(PointerToBase.count(V));
+ Value *OldBase = PointerToBase[V];
+ auto &BaseElements = ElementMapping[OldBase];
+ PointerToBase.erase(V);
+ assert(Elements.size() == BaseElements.size());
+ for (unsigned i = 0; i < Elements.size(); i++) {
+ Value *Elem = Elements[i];
+ PointerToBase[Elem] = BaseElements[i];
+ }
+ }
}
// Helper function for the "rematerializeLiveValues". It walks use chain
// starting from the "CurrentValue" until it meets "BaseValue". Only "simple"
// values are visited (currently it is GEP's and casts). Returns true if it
-// sucessfully reached "BaseValue" and false otherwise.
+// successfully reached "BaseValue" and false otherwise.
// Fills "ChainToBase" array with all visited values. "BaseValue" is not
// recorded.
static bool findRematerializableChainToBasePointer(
for (Value *LiveValue: Info.liveset) {
// For each live pointer find it's defining chain
SmallVector<Instruction *, 3> ChainToBase;
- assert(Info.PointerToBase.find(LiveValue) != Info.PointerToBase.end());
+ assert(Info.PointerToBase.count(LiveValue));
bool FoundChain =
findRematerializableChainToBasePointer(ChainToBase,
LiveValue,
}
assert(records.size() == toUpdate.size());
- // A) Identify all gc pointers which are staticly live at the given call
+ // A) Identify all gc pointers which are statically live at the given call
// site.
findLiveReferences(F, DT, P, toUpdate, records);
- // Do a limited scalarization of any live at safepoint vector values which
- // contain pointers. This enables this pass to run after vectorization at
- // the cost of some possible performance loss. TODO: it would be nice to
- // natively support vectors all the way through the backend so we don't need
- // to scalarize here.
- for (size_t i = 0; i < records.size(); i++) {
- struct PartiallyConstructedSafepointRecord &info = records[i];
- Instruction *statepoint = toUpdate[i].getInstruction();
- splitVectorValues(cast<Instruction>(statepoint), info.liveset, DT);
- }
-
// B) Find the base pointers for each live pointer
/* scope for caching */ {
// Cache the 'defining value' relation used in the computation and
}
holders.clear();
+ // Do a limited scalarization of any live at safepoint vector values which
+ // contain pointers. This enables this pass to run after vectorization at
+ // the cost of some possible performance loss. TODO: it would be nice to
+ // natively support vectors all the way through the backend so we don't need
+ // to scalarize here.
+ for (size_t i = 0; i < records.size(); i++) {
+ struct PartiallyConstructedSafepointRecord &info = records[i];
+ Instruction *statepoint = toUpdate[i].getInstruction();
+ splitVectorValues(cast<Instruction>(statepoint), info.liveset,
+ info.PointerToBase, DT);
+ }
+
// In order to reduce live set of statepoint we might choose to rematerialize
- // some values instead of relocating them. This is purelly an optimization and
+ // some values instead of relocating them. This is purely an optimization and
// does not influence correctness.
TargetTransformInfo &TTI =
P->getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
LLVMContext &Ctx = F.getContext();
MDBuilder Builder(Ctx);
- for (Instruction &I : inst_range(F)) {
+ for (Instruction &I : instructions(F)) {
if (const MDNode *MD = I.getMetadata(LLVMContext::MD_tbaa)) {
assert(MD->getNumOperands() < 5 && "unrecognized metadata shape!");
bool IsImmutableTBAA =
// when rewriting. We'll delete the unreachable ones in a moment.
SmallVector<CallSite, 64> ParsePointNeeded;
bool HasUnreachableStatepoint = false;
- for (Instruction &I : inst_range(F)) {
+ for (Instruction &I : instructions(F)) {
// TODO: only the ones with the flag set!
if (isStatepoint(I)) {
if (DT.isReachableFromEntry(I.getParent()))
"support for FCA unimplemented");
if (isHandledGCPointerType(V->getType()) && !isa<Constant>(V)) {
// The choice to exclude all things constant here is slightly subtle.
- // There are two idependent reasons:
+ // There are two independent reasons:
// - We assume that things which are constant (from LLVM's definition)
// do not move at runtime. For example, the address of a global
// variable is fixed, even though it's contents may not be.
} // while( !worklist.empty() )
#ifndef NDEBUG
- // Sanity check our ouput against SSA properties. This helps catch any
+ // Sanity check our output against SSA properties. This helps catch any
// missing kills during the above iteration.
for (BasicBlock &BB : F) {
checkBasicSSA(DT, Data, BB);