#include "llvm/Pass.h"
#include "llvm/Analysis/CFG.h"
+#include "llvm/Analysis/InstructionSimplify.h"
+#include "llvm/Analysis/TargetTransformInfo.h"
#include "llvm/ADT/SetOperations.h"
#include "llvm/ADT/Statistic.h"
#include "llvm/ADT/DenseSet.h"
#include "llvm/ADT/SetVector.h"
+#include "llvm/ADT/StringRef.h"
+#include "llvm/ADT/MapVector.h"
#include "llvm/IR/BasicBlock.h"
#include "llvm/IR/CallSite.h"
#include "llvm/IR/Dominators.h"
#include "llvm/IR/Intrinsics.h"
#include "llvm/IR/IntrinsicInst.h"
#include "llvm/IR/Module.h"
+#include "llvm/IR/MDBuilder.h"
#include "llvm/IR/Statepoint.h"
#include "llvm/IR/Value.h"
#include "llvm/IR/Verifier.h"
using namespace llvm;
-// Print tracing output
-static cl::opt<bool> TraceLSP("trace-rewrite-statepoints", cl::Hidden,
- cl::init(false));
-
// Print the liveset found at the insert location
static cl::opt<bool> PrintLiveSet("spp-print-liveset", cl::Hidden,
cl::init(false));
static cl::opt<bool> PrintBasePointers("spp-print-base-pointers", cl::Hidden,
cl::init(false));
+// Cost threshold measuring when it is profitable to rematerialize value instead
+// of relocating it
+static cl::opt<unsigned>
+RematerializationThreshold("spp-rematerialization-threshold", cl::Hidden,
+ cl::init(6));
+
#ifdef XDEBUG
static bool ClobberNonLive = true;
#else
cl::Hidden);
namespace {
-struct RewriteStatepointsForGC : public FunctionPass {
+struct RewriteStatepointsForGC : public ModulePass {
static char ID; // Pass identification, replacement for typeid
- RewriteStatepointsForGC() : FunctionPass(ID) {
+ RewriteStatepointsForGC() : ModulePass(ID) {
initializeRewriteStatepointsForGCPass(*PassRegistry::getPassRegistry());
}
- bool runOnFunction(Function &F) override;
+ bool runOnFunction(Function &F);
+ bool runOnModule(Module &M) override {
+ bool Changed = false;
+ for (Function &F : M)
+ Changed |= runOnFunction(F);
+
+ if (Changed) {
+ // stripDereferenceabilityInfo asserts that shouldRewriteStatepointsIn
+ // returns true for at least one function in the module. Since at least
+ // one function changed, we know that the precondition is satisfied.
+ stripDereferenceabilityInfo(M);
+ }
+
+ return Changed;
+ }
void getAnalysisUsage(AnalysisUsage &AU) const override {
// We add and rewrite a bunch of instructions, but don't really do much
// else. We could in theory preserve a lot more analyses here.
AU.addRequired<DominatorTreeWrapperPass>();
- }
+ AU.addRequired<TargetTransformInfoWrapperPass>();
+ }
+
+ /// The IR fed into RewriteStatepointsForGC may have had attributes implying
+ /// dereferenceability that are no longer valid/correct after
+ /// RewriteStatepointsForGC has run. This is because semantically, after
+ /// RewriteStatepointsForGC runs, all calls to gc.statepoint "free" the entire
+ /// heap. stripDereferenceabilityInfo (conservatively) restores correctness
+ /// by erasing all attributes in the module that externally imply
+ /// dereferenceability.
+ ///
+ void stripDereferenceabilityInfo(Module &M);
+
+ // Helpers for stripDereferenceabilityInfo
+ void stripDereferenceabilityInfoFromBody(Function &F);
+ void stripDereferenceabilityInfoFromPrototype(Function &F);
};
} // namespace
char RewriteStatepointsForGC::ID = 0;
-FunctionPass *llvm::createRewriteStatepointsForGCPass() {
+ModulePass *llvm::createRewriteStatepointsForGCPass() {
return new RewriteStatepointsForGC();
}
// types, then update all the second type to the first type
typedef DenseMap<Value *, Value *> DefiningValueMapTy;
typedef DenseSet<llvm::Value *> StatepointLiveSetTy;
+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
/// Instruction to which exceptional gc relocates are attached
/// Makes it easier to iterate through them during relocationViaAlloca.
Instruction *UnwindToken;
+
+ /// Record live values we are rematerialized instead of relocating.
+ /// They are not included into 'liveset' field.
+ /// Maps rematerialized copy to it's original value.
+ RematerializedValueMapTy RematerializedValues;
};
}
// 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());
- std::sort(temp.begin(), temp.end(), order_by_name);
+ SmallVector<Value *, 64> Temp;
+ Temp.insert(Temp.end(), liveset.begin(), liveset.end());
+ std::sort(Temp.begin(), Temp.end(), order_by_name);
errs() << "Live Variables:\n";
- for (Value *V : temp) {
- errs() << " " << V->getName(); // no newline
- V->dump();
- }
+ for (Value *V : Temp)
+ dbgs() << " " << V->getName() << " " << *V << "\n";
}
if (PrintLiveSetSize) {
errs() << "Safepoint For: " << CS.getCalledValue()->getName() << "\n";
result.liveset = liveset;
}
-/// If we can trivially determine that this vector contains only base pointers,
-/// return the base instruction.
-static Value *findBaseOfVector(Value *I) {
+static bool isKnownBaseResult(Value *V);
+namespace {
+/// A single base defining value - An immediate base defining value for an
+/// instruction 'Def' is an input to 'Def' whose base is also a base of 'Def'.
+/// For instructions which have multiple pointer [vector] inputs or that
+/// transition between vector and scalar types, there is no immediate base
+/// defining value. The 'base defining value' for 'Def' is the transitive
+/// closure of this relation stopping at the first instruction which has no
+/// immediate base defining value. The b.d.v. might itself be a base pointer,
+/// but it can also be an arbitrary derived pointer.
+struct BaseDefiningValueResult {
+ /// Contains the value which is the base defining value.
+ Value * const BDV;
+ /// True if the base defining value is also known to be an actual base
+ /// pointer.
+ const bool IsKnownBase;
+ BaseDefiningValueResult(Value *BDV, bool IsKnownBase)
+ : BDV(BDV), IsKnownBase(IsKnownBase) {
+#ifndef NDEBUG
+ // Check consistency between new and old means of checking whether a BDV is
+ // a base.
+ bool MustBeBase = isKnownBaseResult(BDV);
+ assert(!MustBeBase || MustBeBase == IsKnownBase);
+#endif
+ }
+};
+}
+
+static BaseDefiningValueResult findBaseDefiningValue(Value *I);
+
+/// 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 BaseDefiningValueResult
+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 BaseDefiningValueResult(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 BaseDefiningValueResult(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 BaseDefiningValueResult(Con, true);
}
-
+
if (isa<LoadInst>(I))
- return I;
-
- // 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");
+ return BaseDefiningValueResult(I, true);
+
+ // For an insert element, we might be able to look through it if we know
+ // something about the indexes.
+ if (InsertElementInst *IEI = dyn_cast<InsertElementInst>(I)) {
+ if (Index) {
+ Value *InsertIndex = IEI->getOperand(2);
+ // This index is inserting the value, look for its BDV
+ 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 findBaseDefiningValueOfVector(IEI->getOperand(0), Index);
+ }
+
+ // If both inputs to the insertelement are known bases, then so is the
+ // insertelement itself. NOTE: This should be handled within the generic
+ // base pointer inference code and after http://reviews.llvm.org/D12583,
+ // will be. However, when strengthening asserts I needed to add this to
+ // keep an existing test passing which was 'working'. FIXME
+ if (findBaseDefiningValue(IEI->getOperand(0)).IsKnownBase &&
+ findBaseDefiningValue(IEI->getOperand(1)).IsKnownBase)
+ return BaseDefiningValueResult(IEI, true);
+
+ // 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 BaseDefiningValueResult(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 BaseDefiningValueResult(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 BaseDefiningValueResult(I, false);
}
/// 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) {
+/// defines the base pointer for the input, b) blocks the simple search
+/// (i.e. a PHI or Select of two derived pointers), or c) involves a change
+/// from pointer to vector type or back.
+static BaseDefiningValueResult findBaseDefiningValue(Value *I) {
+ if (I->getType()->isVectorTy())
+ return findBaseDefiningValueOfVector(I);
+
assert(I->getType()->isPointerTy() &&
"Illegal to ask for the base pointer of a non-pointer type");
- // This case is a bit of a hack - it only handles extracts from vectors which
- // trivially contain only base pointers. See note inside the function for
- // how to improve this.
- if (auto *EEI = dyn_cast<ExtractElementInst>(I)) {
- Value *VectorOperand = EEI->getVectorOperand();
- Value *VectorBase = findBaseOfVector(VectorOperand);
- (void)VectorBase;
- assert(VectorBase && "extract element not known to be a trivial base");
- return EEI;
- }
-
if (isa<Argument>(I))
// An incoming argument to the function is a base pointer
// We should have never reached here if this argument isn't an gc value
- return I;
+ return BaseDefiningValueResult(I, true);
if (isa<GlobalVariable>(I))
// base case
- return I;
+ return BaseDefiningValueResult(I, true);
// inlining could possibly introduce phi node that contains
// undef if callee has multiple returns
if (isa<UndefValue>(I))
// utterly meaningless, but useful for dealing with
// partially optimized code.
- return I;
+ return BaseDefiningValueResult(I, true);
// Due to inheritance, this must be _after_ the global variable and undef
// checks
- if (Constant *Con = dyn_cast<Constant>(I)) {
+ if (isa<Constant>(I)) {
assert(!isa<GlobalVariable>(I) && !isa<UndefValue>(I) &&
"order of checks wrong!");
// Note: Finding a constant base for something marked for relocation
// off a potentially null value and have proven it null. We also use
// null pointers in dead paths of relocation phis (which we might later
// want to find a base pointer for).
- assert(isa<ConstantPointerNull>(Con) &&
+ assert(isa<ConstantPointerNull>(I) &&
"null is the only case which makes sense");
- return Con;
+ return BaseDefiningValueResult(I, true);
}
if (CastInst *CI = dyn_cast<CastInst>(I)) {
}
if (isa<LoadInst>(I))
- return I; // The value loaded is an gc base itself
+ // The value loaded is an gc base itself
+ return BaseDefiningValueResult(I, true);
+
if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(I))
// The base of this GEP is the base
// pointers. This should probably be generalized via attributes to support
// both source language and internal functions.
if (isa<CallInst>(I) || isa<InvokeInst>(I))
- return I;
+ return BaseDefiningValueResult(I, true);
// 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))
// A CAS is effectively a atomic store and load combined under a
// predicate. From the perspective of base pointers, we just treat it
// like a load.
- return I;
+ return BaseDefiningValueResult(I, true);
assert(!isa<AtomicRMWInst>(I) && "Xchg handled above, all others are "
"binary ops which don't apply to pointers");
// stack, but in either case, this is simply a field load. As a result,
// this is a defining definition of the base just like a load is.
if (isa<ExtractValueInst>(I))
- return I;
+ return BaseDefiningValueResult(I, true);
// We should never see an insert vector since that would require we be
// tracing back a struct value not a pointer value.
assert(!isa<InsertValueInst>(I) &&
"Base pointer for a struct is meaningless");
+ // An extractelement produces a base result exactly when it's input does.
+ // We may need to insert a parallel instruction to extract the appropriate
+ // element out of the base vector corresponding to the input. Given this,
+ // it's analogous to the phi and select case even though it's not a merge.
+ if (auto *EEI = dyn_cast<ExtractElementInst>(I)) {
+ Value *VectorOperand = EEI->getVectorOperand();
+ Value *Index = EEI->getIndexOperand();
+ auto VecResult = findBaseDefiningValueOfVector(VectorOperand, Index);
+ Value *VectorBase = VecResult.BDV;
+ 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. Note: The peephole optimization here is
+ // currently needed for correctness since the general algorithm doesn't
+ // yet handle insertelements. That will change shortly.
+ return BaseDefiningValueResult(VectorBase, VecResult.IsKnownBase);
+ else {
+ assert(VectorBase->getType()->isVectorTy());
+ // 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.
+ return BaseDefiningValueResult(I, VecResult.IsKnownBase);
+ }
+ }
+
// 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)) &&
"missing instruction case in findBaseDefiningValing");
- return I;
+ return BaseDefiningValueResult(I, false);
}
/// Returns the base defining value for this value.
static Value *findBaseDefiningValueCached(Value *I, DefiningValueMapTy &Cache) {
Value *&Cached = Cache[I];
if (!Cached) {
- Cached = findBaseDefiningValue(I);
+ Cached = findBaseDefiningValue(I).BDV;
+ 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;
}
/// Given the result of a call to findBaseDefiningValue, or findBaseOrBDV,
/// is it known to be a base pointer? Or do we need to continue searching.
static bool isKnownBaseResult(Value *V) {
- if (!isa<PHINode>(V) && !isa<SelectInst>(V)) {
+ if (!isa<PHINode>(V) && !isa<SelectInst>(V) && !isa<ExtractElementInst>(V)) {
// no recursion possible
return true;
}
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 {
+ switch (status) {
+ case Unknown:
+ OS << "U";
+ break;
+ case Base:
+ OS << "B";
+ break;
+ case Conflict:
+ OS << "C";
+ break;
+ };
+ OS << " (" << base << " - "
+ << (base ? base->getName() : "nullptr") << "): ";
}
private:
Status status;
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
+#ifndef NDEBUG
+static raw_ostream &operator<<(raw_ostream &OS, const BDVState &State) {
+ State.print(OS);
+ return OS;
+}
+#endif
+
+namespace {
+// 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!");
}
};
}
+
+
/// For a given value or instruction, figure out what base ptr it's derived
/// from. For gc objects, this is simply itself. On success, returns a value
/// which is the base pointer. (This is reliable and can be used for
//
// 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.
- 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;
- }
+#ifndef NDEBUG
+ auto isExpectedBDVType = [](Value *BDV) {
+ return isa<PHINode>(BDV) || isa<SelectInst>(BDV) || isa<ExtractElementInst>(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.
+ // We use the order of insertion (DFS over the def/use graph) to provide a
+ // stable deterministic ordering for visiting DenseMaps (which are unordered)
+ // below. This is important for deterministic compilation.
+ MapVector<Value *, BDVState> states;
+
+ // Recursively fill in all base defining values reachable from the initial
+ // one for which we don't already know a definite base value for
+ /* scope */ {
+ SmallVector<Value*, 16> Worklist;
+ Worklist.push_back(def);
+ states.insert(std::make_pair(def, BDVState()));
+ 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 (states.insert(std::make_pair(Base, BDVState())).second)
+ Worklist.push_back(Base);
+ };
+ if (PHINode *Phi = dyn_cast<PHINode>(Current)) {
+ for (Value *InVal : Phi->incoming_values())
+ visitIncomingValue(InVal);
+ } else if (SelectInst *Sel = dyn_cast<SelectInst>(Current)) {
+ visitIncomingValue(Sel->getTrueValue());
+ visitIncomingValue(Sel->getFalseValue());
+ } else if (auto *EE = dyn_cast<ExtractElementInst>(Current)) {
+ visitIncomingValue(EE->getVectorOperand());
+ } else {
+ // There are two classes of instructions we know we don't handle.
+ assert(isa<ShuffleVectorInst>(Current) ||
+ isa<InsertElementInst>(Current));
+ llvm_unreachable("unimplemented instruction case");
}
}
}
- 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();
- }
+#ifndef NDEBUG
+ DEBUG(dbgs() << "States after initialization:\n");
+ for (auto Pair : states) {
+ DEBUG(dbgs() << " " << Pair.second << " for " << *Pair.first << "\n");
}
+#endif
- // 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) {
size_t oldSize = states.size();
#endif
progress = false;
- // We're only changing keys in this loop, thus safe to keep iterators
+ // We're only changing values in this loop, thus safe to keep iterators.
+ // Since this is computing a fixed point, the order of visit does not
+ // effect the result. TODO: We could use a worklist here and make this run
+ // much faster.
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));
- } else
- for (Value *Val : cast<PHINode>(v)->incoming_values())
- calculateMeet.meetWith(findBaseOrBDV(Val, cache));
-
- PhiState oldState = states[v];
- PhiState newState = calculateMeet.getResult();
+ calculateMeet.meetWith(getStateForInput(select->getTrueValue()));
+ calculateMeet.meetWith(getStateForInput(select->getFalseValue()));
+ } else if (PHINode *Phi = dyn_cast<PHINode>(v)) {
+ for (Value *Val : Phi->incoming_values())
+ calculateMeet.meetWith(getStateForInput(Val));
+ } else {
+ // The 'meet' for an extractelement is slightly trivial, but it's still
+ // useful in that it drives us to conflict if our input is.
+ auto *EE = cast<ExtractElementInst>(v);
+ calculateMeet.meetWith(getStateForInput(EE->getVectorOperand()));
+ }
+
+ BDVState oldState = states[v];
+ BDVState newState = calculateMeet.getResult();
if (oldState != newState) {
progress = true;
states[v] = newState;
assert(oldSize == states.size() || progress);
}
- 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();
- }
- }
-
- // Insert Phis for all conflicts
- // We want to keep naming deterministic in the loop that follows, so
- // sort the keys before iteration. This is useful in allowing us to
- // write stable tests. Note that there is no invalidation issue here.
- SmallVector<Value *, 16> Keys;
- Keys.reserve(states.size());
+#ifndef NDEBUG
+ DEBUG(dbgs() << "States after meet iteration:\n");
for (auto Pair : states) {
- Value *V = Pair.first;
- Keys.push_back(V);
+ DEBUG(dbgs() << " " << Pair.second << " for " << *Pair.first << "\n");
}
- std::sort(Keys.begin(), Keys.end(), order_by_name);
+#endif
+
+ // Insert Phis for all conflicts
// 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())
+ for (auto Pair : states) {
+ Instruction *I = cast<Instruction>(Pair.first);
+ BDVState State = Pair.second;
+ assert(!isKnownBaseResult(I) && "why did it get added?");
+ assert(!State.isUnknown() && "Optimistic algorithm didn't complete!");
+
+ // extractelement instructions are a bit special in that we may need to
+ // insert an extract even when we know an exact base for the instruction.
+ // The problem is that we need to convert from a vector base to a scalar
+ // base for the particular indice we're interested in.
+ if (State.isBase() && isa<ExtractElementInst>(I) &&
+ isa<VectorType>(State.getBase()->getType())) {
+ auto *EE = cast<ExtractElementInst>(I);
+ // TODO: In many cases, the new instruction is just EE itself. We should
+ // exploit this, but can't do it here since it would break the invariant
+ // about the BDV not being known to be a base.
+ auto *BaseInst = ExtractElementInst::Create(State.getBase(),
+ EE->getIndexOperand(),
+ "base_ee", EE);
+ BaseInst->setMetadata("is_base_value", MDNode::get(I->getContext(), {}));
+ states[I] = BDVState(BDVState::Base, BaseInst);
+ }
+
+ 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);
+ /// 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);
+ } else if (SelectInst *Sel = dyn_cast<SelectInst>(I)) {
+ // The undef will be replaced later
+ 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);
+ } else {
+ auto *EE = cast<ExtractElementInst>(I);
+ UndefValue *Undef = UndefValue::get(EE->getVectorOperand()->getType());
+ std::string Name = I->hasName() ?
+ (I->getName() + ".base").str() : "base_ee";
+ return ExtractElementInst::Create(Undef, EE->getIndexOperand(), Name,
+ EE);
+ }
+ };
+ 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);
+ }
+
+ // Returns a instruction which produces the base pointer for a given
+ // instruction. The instruction is assumed to be an input to one of the BDVs
+ // seen in the inference algorithm above. As such, we must either already
+ // know it's base defining value is a base, or have inserted a new
+ // instruction to propagate the base of it's BDV and have entered that newly
+ // introduced instruction into the state table. In either case, we are
+ // assured to be able to determine an instruction which produces it's base
+ // pointer.
+ auto getBaseForInput = [&](Value *Input, Instruction *InsertPt) {
+ Value *BDV = findBaseOrBDV(Input, cache);
+ Value *Base = nullptr;
+ if (isKnownBaseResult(BDV)) {
+ Base = BDV;
} else {
- SelectInst *sel = cast<SelectInst>(v);
- // 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);
+ // Either conflict or base.
+ assert(states.count(BDV));
+ Base = states[BDV].getBase();
}
- }
+ assert(Base && "can't be null");
+ // The cast is needed since base traversal may strip away bitcasts
+ if (Base->getType() != Input->getType() &&
+ InsertPt) {
+ Base = new BitCastInst(Base, Input->getType(), "cast",
+ InsertPt);
+ }
+ return Base;
+ };
- // Fixup all the inputs of the new PHIs
+ // Fixup all the inputs of the new PHIs. Visit order needs to be
+ // deterministic and predictable because we're naming newly created
+ // instructions.
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!");
if (blockIndex != -1) {
Value *oldBase = basephi->getIncomingValue(blockIndex);
basephi->addIncoming(oldBase, InBB);
+
#ifndef NDEBUG
- Value *base = findBaseOrBDV(InVal, cache);
- if (!isKnownBaseResult(base)) {
- // Either conflict or base.
- assert(states.count(base));
- base = states[base].getBase();
- assert(base != nullptr && "unknown PhiState!");
- }
-
- // In essense this assert states: the only way two
+ Value *Base = getBaseForInput(InVal, nullptr);
+ // 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
// findBaseOrBDV to return an llvm::Value of the correct
// type (and still remain pure). This will remove the
// need to add bitcasts.
- assert(base->stripPointerCasts() == oldBase->stripPointerCasts() &&
+ assert(Base->stripPointerCasts() == oldBase->stripPointerCasts() &&
"sanity -- findBaseOrBDV should be pure!");
#endif
continue;
}
- // Find either the defining value for the PHI or the normal base for
- // a non-phi node
- Value *base = findBaseOrBDV(InVal, cache);
- if (!isKnownBaseResult(base)) {
- // Either conflict or base.
- assert(states.count(base));
- base = states[base].getBase();
- assert(base != nullptr && "unknown PhiState!");
- }
- assert(base && "can't be null");
- // Must use original input BB since base may not be Instruction
- // The cast is needed since base traversal may strip away bitcasts
- if (base->getType() != basephi->getType()) {
- base = new BitCastInst(base, basephi->getType(), "cast",
- InBB->getTerminator());
- }
- basephi->addIncoming(base, InBB);
+ // Find the instruction which produces the base for each input. We may
+ // need to insert a bitcast in the incoming block.
+ // TODO: Need to split critical edges if insertion is needed
+ Value *Base = getBaseForInput(InVal, InBB->getTerminator());
+ basephi->addIncoming(Base, InBB);
}
assert(basephi->getNumIncomingValues() == NumPHIValues);
- } else {
- SelectInst *basesel = cast<SelectInst>(state.getBase());
- SelectInst *sel = cast<SelectInst>(v);
+ } else if (SelectInst *BaseSel = dyn_cast<SelectInst>(state.getBase())) {
+ SelectInst *Sel = cast<SelectInst>(v);
// Operand 1 & 2 are true, false path respectively. TODO: refactor to
// something more safe and less hacky.
for (int i = 1; i <= 2; i++) {
- Value *InVal = sel->getOperand(i);
- // Find either the defining value for the PHI or the normal base for
- // a non-phi node
- Value *base = findBaseOrBDV(InVal, cache);
- if (!isKnownBaseResult(base)) {
- // Either conflict or base.
- assert(states.count(base));
- base = states[base].getBase();
- assert(base != nullptr && "unknown PhiState!");
- }
- assert(base && "can't be null");
- // Must use original input BB since base may not be Instruction
- // The cast is needed since base traversal may strip away bitcasts
- if (base->getType() != basesel->getType()) {
- base = new BitCastInst(base, basesel->getType(), "cast", basesel);
- }
- basesel->setOperand(i, base);
+ Value *InVal = Sel->getOperand(i);
+ // Find the instruction which produces the base for each input. We may
+ // need to insert a bitcast.
+ Value *Base = getBaseForInput(InVal, BaseSel);
+ BaseSel->setOperand(i, Base);
}
+ } else {
+ auto *BaseEE = cast<ExtractElementInst>(state.getBase());
+ Value *InVal = cast<ExtractElementInst>(v)->getVectorOperand();
+ // Find the instruction which produces the base for each input. We may
+ // need to insert a bitcast.
+ Value *Base = getBaseForInput(InVal, BaseEE);
+ BaseEE->setOperand(0, Base);
}
}
- // Cache all of our results so we can cheaply reuse them
- // NOTE: This is actually two caches: one of the base defining value
- // relation and one of the base pointer relation! FIXME
- for (auto item : states) {
- Value *v = item.first;
- Value *base = item.second.getBase();
- assert(v && base);
- assert(!isKnownBaseResult(v) && "why did it get added?");
+ // Now that we're done with the algorithm, see if we can optimize the
+ // results slightly by reducing the number of new instructions needed.
+ // Arguably, this should be integrated into the algorithm above, but
+ // doing as a post process step is easier to reason about for the moment.
+ DenseMap<Value *, Value *> ReverseMap;
+ SmallPtrSet<Instruction *, 16> NewInsts;
+ SmallSetVector<AssertingVH<Instruction>, 16> Worklist;
+ // Note: We need to visit the states in a deterministic order. We uses the
+ // Keys we sorted above for this purpose. Note that we are papering over a
+ // bigger problem with the algorithm above - it's visit order is not
+ // deterministic. A larger change is needed to fix this.
+ for (auto Pair : states) {
+ auto *BDV = Pair.first;
+ auto State = Pair.second;
+ Value *Base = State.getBase();
+ assert(BDV && Base);
+ assert(!isKnownBaseResult(BDV) && "why did it get added?");
+ assert(isKnownBaseResult(Base) &&
+ "must be something we 'know' is a base pointer");
+ if (!State.isConflict())
+ continue;
- if (TraceLSP) {
- std::string fromstr =
- cache.count(v) ? (cache[v]->hasName() ? cache[v]->getName() : "")
- : "none";
- errs() << "Updating base value cache"
- << " for: " << (v->hasName() ? v->getName() : "")
- << " from: " << fromstr
- << " to: " << (base->hasName() ? base->getName() : "") << "\n";
+ ReverseMap[Base] = BDV;
+ if (auto *BaseI = dyn_cast<Instruction>(Base)) {
+ NewInsts.insert(BaseI);
+ Worklist.insert(BaseI);
}
+ }
+ auto ReplaceBaseInstWith = [&](Value *BDV, Instruction *BaseI,
+ Value *Replacement) {
+ // Add users which are new instructions (excluding self references)
+ for (User *U : BaseI->users())
+ if (auto *UI = dyn_cast<Instruction>(U))
+ if (NewInsts.count(UI) && UI != BaseI)
+ Worklist.insert(UI);
+ // Then do the actual replacement
+ NewInsts.erase(BaseI);
+ ReverseMap.erase(BaseI);
+ BaseI->replaceAllUsesWith(Replacement);
+ BaseI->eraseFromParent();
+ assert(states.count(BDV));
+ assert(states[BDV].isConflict() && states[BDV].getBase() == BaseI);
+ states[BDV] = BDVState(BDVState::Conflict, Replacement);
+ };
+ const DataLayout &DL = cast<Instruction>(def)->getModule()->getDataLayout();
+ while (!Worklist.empty()) {
+ Instruction *BaseI = Worklist.pop_back_val();
+ assert(NewInsts.count(BaseI));
+ Value *Bdv = ReverseMap[BaseI];
+ if (auto *BdvI = dyn_cast<Instruction>(Bdv))
+ if (BaseI->isIdenticalTo(BdvI)) {
+ DEBUG(dbgs() << "Identical Base: " << *BaseI << "\n");
+ ReplaceBaseInstWith(Bdv, BaseI, Bdv);
+ continue;
+ }
+ if (Value *V = SimplifyInstruction(BaseI, DL)) {
+ DEBUG(dbgs() << "Base " << *BaseI << " simplified to " << *V << "\n");
+ ReplaceBaseInstWith(Bdv, BaseI, V);
+ continue;
+ }
+ }
- assert(isKnownBaseResult(base) &&
- "must be something we 'know' is a base pointer");
- if (cache.count(v)) {
+ // Cache all of our results so we can cheaply reuse them
+ // NOTE: This is actually two caches: one of the base defining value
+ // relation and one of the base pointer relation! FIXME
+ for (auto Pair : states) {
+ auto *BDV = Pair.first;
+ Value *base = Pair.second.getBase();
+ assert(BDV && base);
+
+ std::string fromstr =
+ cache.count(BDV) ? (cache[BDV]->hasName() ? cache[BDV]->getName() : "")
+ : "none";
+ DEBUG(dbgs() << "Updating base value cache"
+ << " for: " << (BDV->hasName() ? BDV->getName() : "")
+ << " from: " << fromstr
+ << " to: " << (base->hasName() ? base->getName() : "") << "\n");
+
+ if (cache.count(BDV)) {
// Once we transition from the BDV relation being store in the cache to
// the base relation being stored, it must be stable
- assert((!isKnownBaseResult(cache[v]) || cache[v] == base) &&
+ assert((!isKnownBaseResult(cache[BDV]) || cache[BDV] == base) &&
"base relation should be stable");
}
- cache[v] = base;
+ cache[BDV] = base;
}
assert(cache.find(def) != cache.end());
return cache[def];
// 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++) {
// goes through the statepoint. We might need to split an edge to make this
// possible.
static BasicBlock *
-normalizeForInvokeSafepoint(BasicBlock *BB, BasicBlock *InvokeParent, Pass *P) {
- DominatorTree *DT = nullptr;
- if (auto *DTP = P->getAnalysisIfAvailable<DominatorTreeWrapperPass>())
- DT = &DTP->getDomTree();
-
+normalizeForInvokeSafepoint(BasicBlock *BB, BasicBlock *InvokeParent,
+ 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.CreateCall3(
- GCRelocateDecl, StatepointToken, BaseIdx, LiveIdx,
+ 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);
}
}
}
+// Helper function for the "relocationViaAlloca". Similar to the
+// "insertRelocationStores" but works for rematerialized values.
+static void
+insertRematerializationStores(
+ RematerializedValueMapTy RematerializedValues,
+ DenseMap<Value *, Value *> &AllocaMap,
+ DenseSet<Value *> &VisitedLiveValues) {
+
+ for (auto RematerializedValuePair: RematerializedValues) {
+ Instruction *RematerializedValue = RematerializedValuePair.first;
+ Value *OriginalValue = RematerializedValuePair.second;
+
+ assert(AllocaMap.count(OriginalValue) &&
+ "Can not find alloca for rematerialized value");
+ Value *Alloca = AllocaMap[OriginalValue];
+
+ StoreInst *Store = new StoreInst(RematerializedValue, Alloca);
+ Store->insertAfter(RematerializedValue);
+
+#ifndef NDEBUG
+ VisitedLiveValues.insert(OriginalValue);
+#endif
+ }
+}
+
/// do all the relocation update via allocas and mem2reg
static void relocationViaAlloca(
- Function &F, DominatorTree &DT, ArrayRef<Value *> live,
- ArrayRef<struct PartiallyConstructedSafepointRecord> records) {
+ Function &F, DominatorTree &DT, ArrayRef<Value *> Live,
+ ArrayRef<struct PartiallyConstructedSafepointRecord> Records) {
#ifndef NDEBUG
// record initial number of (static) allocas; we'll check we have the same
// number when we get done.
#endif
// TODO-PERF: change data structures, reserve
- DenseMap<Value *, Value *> allocaMap;
+ DenseMap<Value *, Value *> AllocaMap;
SmallVector<AllocaInst *, 200> PromotableAllocas;
- PromotableAllocas.reserve(live.size());
+ // Used later to chack that we have enough allocas to store all values
+ std::size_t NumRematerializedValues = 0;
+ PromotableAllocas.reserve(Live.size());
+
+ // Emit alloca for "LiveValue" and record it in "allocaMap" and
+ // "PromotableAllocas"
+ auto emitAllocaFor = [&](Value *LiveValue) {
+ AllocaInst *Alloca = new AllocaInst(LiveValue->getType(), "",
+ F.getEntryBlock().getFirstNonPHI());
+ AllocaMap[LiveValue] = Alloca;
+ PromotableAllocas.push_back(Alloca);
+ };
// emit alloca for each live gc pointer
- for (unsigned i = 0; i < live.size(); i++) {
- Value *liveValue = live[i];
- AllocaInst *alloca = new AllocaInst(liveValue->getType(), "",
- F.getEntryBlock().getFirstNonPHI());
- allocaMap[liveValue] = alloca;
- PromotableAllocas.push_back(alloca);
+ for (unsigned i = 0; i < Live.size(); i++) {
+ emitAllocaFor(Live[i]);
+ }
+
+ // emit allocas for rematerialized values
+ for (size_t i = 0; i < Records.size(); i++) {
+ const struct PartiallyConstructedSafepointRecord &Info = Records[i];
+
+ for (auto RematerializedValuePair : Info.RematerializedValues) {
+ Value *OriginalValue = RematerializedValuePair.second;
+ if (AllocaMap.count(OriginalValue) != 0)
+ continue;
+
+ emitAllocaFor(OriginalValue);
+ ++NumRematerializedValues;
+ }
}
// The next two loops are part of the same conceptual operation. We need to
// this gc pointer and it is not a gc_result)
// this must happen before we update the statepoint with load of alloca
// otherwise we lose the link between statepoint and old def
- for (size_t i = 0; i < records.size(); i++) {
- const struct PartiallyConstructedSafepointRecord &info = records[i];
- Value *Statepoint = info.StatepointToken;
+ for (size_t i = 0; i < Records.size(); i++) {
+ const struct PartiallyConstructedSafepointRecord &Info = Records[i];
+ Value *Statepoint = Info.StatepointToken;
// This will be used for consistency check
- DenseSet<Value *> visitedLiveValues;
+ DenseSet<Value *> VisitedLiveValues;
// Insert stores for normal statepoint gc relocates
- insertRelocationStores(Statepoint->users(), allocaMap, visitedLiveValues);
+ insertRelocationStores(Statepoint->users(), AllocaMap, VisitedLiveValues);
// In case if it was invoke statepoint
// we will insert stores for exceptional path gc relocates.
if (isa<InvokeInst>(Statepoint)) {
- insertRelocationStores(info.UnwindToken->users(), allocaMap,
- visitedLiveValues);
+ insertRelocationStores(Info.UnwindToken->users(), AllocaMap,
+ VisitedLiveValues);
}
+ // Do similar thing with rematerialized values
+ insertRematerializationStores(Info.RematerializedValues, AllocaMap,
+ 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
// wise.
SmallVector<AllocaInst *, 64> ToClobber;
- for (auto Pair : allocaMap) {
+ for (auto Pair : AllocaMap) {
Value *Def = Pair.first;
AllocaInst *Alloca = cast<AllocaInst>(Pair.second);
// This value was relocated
- if (visitedLiveValues.count(Def)) {
+ if (VisitedLiveValues.count(Def)) {
continue;
}
ToClobber.push_back(Alloca);
auto AIType = cast<PointerType>(AI->getType());
auto PT = cast<PointerType>(AIType->getElementType());
Constant *CPN = ConstantPointerNull::get(PT);
- StoreInst *store = new StoreInst(CPN, AI);
- store->insertBefore(IP);
+ StoreInst *Store = new StoreInst(CPN, AI);
+ Store->insertBefore(IP);
}
};
}
}
// update use with load allocas and add store for gc_relocated
- for (auto Pair : allocaMap) {
- Value *def = Pair.first;
- Value *alloca = Pair.second;
+ for (auto Pair : AllocaMap) {
+ Value *Def = Pair.first;
+ Value *Alloca = Pair.second;
// we pre-record the uses of allocas so that we dont have to worry about
// later update
// that change the user information.
- SmallVector<Instruction *, 20> uses;
+ SmallVector<Instruction *, 20> Uses;
// PERF: trade a linear scan for repeated reallocation
- uses.reserve(std::distance(def->user_begin(), def->user_end()));
- for (User *U : def->users()) {
+ Uses.reserve(std::distance(Def->user_begin(), Def->user_end()));
+ for (User *U : Def->users()) {
if (!isa<ConstantExpr>(U)) {
// If the def has a ConstantExpr use, then the def is either a
// ConstantExpr use itself or null. In either case
// (recursively in the first, directly in the second), the oop
// it is ultimately dependent on is null and this particular
// use does not need to be fixed up.
- uses.push_back(cast<Instruction>(U));
+ Uses.push_back(cast<Instruction>(U));
}
}
- std::sort(uses.begin(), uses.end());
- auto last = std::unique(uses.begin(), uses.end());
- uses.erase(last, uses.end());
-
- for (Instruction *use : uses) {
- if (isa<PHINode>(use)) {
- PHINode *phi = cast<PHINode>(use);
- for (unsigned i = 0; i < phi->getNumIncomingValues(); i++) {
- if (def == phi->getIncomingValue(i)) {
- LoadInst *load = new LoadInst(
- alloca, "", phi->getIncomingBlock(i)->getTerminator());
- phi->setIncomingValue(i, load);
+ std::sort(Uses.begin(), Uses.end());
+ auto Last = std::unique(Uses.begin(), Uses.end());
+ Uses.erase(Last, Uses.end());
+
+ for (Instruction *Use : Uses) {
+ if (isa<PHINode>(Use)) {
+ PHINode *Phi = cast<PHINode>(Use);
+ for (unsigned i = 0; i < Phi->getNumIncomingValues(); i++) {
+ if (Def == Phi->getIncomingValue(i)) {
+ LoadInst *Load = new LoadInst(
+ Alloca, "", Phi->getIncomingBlock(i)->getTerminator());
+ Phi->setIncomingValue(i, Load);
}
}
} else {
- LoadInst *load = new LoadInst(alloca, "", use);
- use->replaceUsesOfWith(def, load);
+ LoadInst *Load = new LoadInst(Alloca, "", Use);
+ Use->replaceUsesOfWith(Def, Load);
}
}
// emit store for the initial gc value
// store must be inserted after load, otherwise store will be in alloca's
// use list and an extra load will be inserted before it
- StoreInst *store = new StoreInst(def, alloca);
- if (Instruction *inst = dyn_cast<Instruction>(def)) {
- if (InvokeInst *invoke = dyn_cast<InvokeInst>(inst)) {
+ StoreInst *Store = new StoreInst(Def, Alloca);
+ if (Instruction *Inst = dyn_cast<Instruction>(Def)) {
+ if (InvokeInst *Invoke = dyn_cast<InvokeInst>(Inst)) {
// InvokeInst is a TerminatorInst so the store need to be inserted
// into its normal destination block.
- BasicBlock *normalDest = invoke->getNormalDest();
- store->insertBefore(normalDest->getFirstNonPHI());
+ BasicBlock *NormalDest = Invoke->getNormalDest();
+ Store->insertBefore(NormalDest->getFirstNonPHI());
} else {
- assert(!inst->isTerminator() &&
+ assert(!Inst->isTerminator() &&
"The only TerminatorInst that can produce a value is "
"InvokeInst which is handled above.");
- store->insertAfter(inst);
+ Store->insertAfter(Inst);
}
} else {
- assert(isa<Argument>(def));
- store->insertAfter(cast<Instruction>(alloca));
+ assert(isa<Argument>(Def));
+ Store->insertAfter(cast<Instruction>(Alloca));
}
}
- assert(PromotableAllocas.size() == live.size() &&
+ assert(PromotableAllocas.size() == Live.size() + NumRematerializedValues &&
"we must have the same allocas with lives");
if (!PromotableAllocas.empty()) {
// apply mem2reg to promote alloca to SSA
/// vector. Doing so has the effect of changing the output of a couple of
/// tests in ways which make them less useful in testing fused safepoints.
template <typename T> static void unique_unsorted(SmallVectorImpl<T> &Vec) {
- DenseSet<T> Seen;
- SmallVector<T, 128> TempVec;
- TempVec.reserve(Vec.size());
- for (auto Element : Vec)
- TempVec.push_back(Element);
- Vec.clear();
- for (auto V : TempVec) {
- if (Seen.insert(V).second) {
- Vec.push_back(V);
- }
- }
+ SmallSet<T, 8> Seen;
+ Vec.erase(std::remove_if(Vec.begin(), Vec.end(), [&](const T &V) {
+ return !Seen.insert(V).second;
+ }), Vec.end());
}
/// Insert holders so that each Value is obviously live through the entire
/// 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
+// successfully reached "BaseValue" and false otherwise.
+// Fills "ChainToBase" array with all visited values. "BaseValue" is not
+// recorded.
+static bool findRematerializableChainToBasePointer(
+ SmallVectorImpl<Instruction*> &ChainToBase,
+ Value *CurrentValue, Value *BaseValue) {
+
+ // We have found a base value
+ if (CurrentValue == BaseValue) {
+ return true;
+ }
+
+ if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(CurrentValue)) {
+ ChainToBase.push_back(GEP);
+ return findRematerializableChainToBasePointer(ChainToBase,
+ GEP->getPointerOperand(),
+ BaseValue);
+ }
+
+ if (CastInst *CI = dyn_cast<CastInst>(CurrentValue)) {
+ Value *Def = CI->stripPointerCasts();
+
+ // This two checks are basically similar. First one is here for the
+ // consistency with findBasePointers logic.
+ assert(!isa<CastInst>(Def) && "not a pointer cast found");
+ if (!CI->isNoopCast(CI->getModule()->getDataLayout()))
+ return false;
+
+ ChainToBase.push_back(CI);
+ return findRematerializableChainToBasePointer(ChainToBase, Def, BaseValue);
+ }
+
+ // Not supported instruction in the chain
+ return false;
+}
+
+// Helper function for the "rematerializeLiveValues". Compute cost of the use
+// chain we are going to rematerialize.
+static unsigned
+chainToBasePointerCost(SmallVectorImpl<Instruction*> &Chain,
+ TargetTransformInfo &TTI) {
+ unsigned Cost = 0;
+
+ for (Instruction *Instr : Chain) {
+ if (CastInst *CI = dyn_cast<CastInst>(Instr)) {
+ assert(CI->isNoopCast(CI->getModule()->getDataLayout()) &&
+ "non noop cast is found during rematerialization");
+
+ Type *SrcTy = CI->getOperand(0)->getType();
+ Cost += TTI.getCastInstrCost(CI->getOpcode(), CI->getType(), SrcTy);
+
+ } else if (GetElementPtrInst *GEP = dyn_cast<GetElementPtrInst>(Instr)) {
+ // Cost of the address calculation
+ Type *ValTy = GEP->getPointerOperandType()->getPointerElementType();
+ Cost += TTI.getAddressComputationCost(ValTy);
+
+ // And cost of the GEP itself
+ // TODO: Use TTI->getGEPCost here (it exists, but appears to be not
+ // allowed for the external usage)
+ if (!GEP->hasAllConstantIndices())
+ Cost += 2;
+
+ } else {
+ llvm_unreachable("unsupported instruciton type during rematerialization");
+ }
+ }
+
+ return Cost;
+}
+
+// From the statepoint liveset pick values that are cheaper to recompute then to
+// relocate. Remove this values from the liveset, rematerialize them after
+// statepoint and record them in "Info" structure. Note that similar to
+// relocated values we don't do any user adjustments here.
+static void rematerializeLiveValues(CallSite CS,
+ PartiallyConstructedSafepointRecord &Info,
+ TargetTransformInfo &TTI) {
+ const unsigned int ChainLengthThreshold = 10;
+
+ // Record values we are going to delete from this statepoint live set.
+ // We can not di this in following loop due to iterator invalidation.
+ SmallVector<Value *, 32> LiveValuesToBeDeleted;
+
+ for (Value *LiveValue: Info.liveset) {
+ // For each live pointer find it's defining chain
+ SmallVector<Instruction *, 3> ChainToBase;
+ assert(Info.PointerToBase.count(LiveValue));
+ bool FoundChain =
+ findRematerializableChainToBasePointer(ChainToBase,
+ LiveValue,
+ Info.PointerToBase[LiveValue]);
+ // Nothing to do, or chain is too long
+ if (!FoundChain ||
+ ChainToBase.size() == 0 ||
+ ChainToBase.size() > ChainLengthThreshold)
+ continue;
+
+ // Compute cost of this chain
+ unsigned Cost = chainToBasePointerCost(ChainToBase, TTI);
+ // TODO: We can also account for cases when we will be able to remove some
+ // of the rematerialized values by later optimization passes. I.e if
+ // we rematerialized several intersecting chains. Or if original values
+ // don't have any uses besides this statepoint.
+
+ // For invokes we need to rematerialize each chain twice - for normal and
+ // for unwind basic blocks. Model this by multiplying cost by two.
+ if (CS.isInvoke()) {
+ Cost *= 2;
+ }
+ // If it's too expensive - skip it
+ if (Cost >= RematerializationThreshold)
+ continue;
+
+ // Remove value from the live set
+ LiveValuesToBeDeleted.push_back(LiveValue);
+
+ // Clone instructions and record them inside "Info" structure
+
+ // Walk backwards to visit top-most instructions first
+ std::reverse(ChainToBase.begin(), ChainToBase.end());
+
+ // Utility function which clones all instructions from "ChainToBase"
+ // and inserts them before "InsertBefore". Returns rematerialized value
+ // which should be used after statepoint.
+ auto rematerializeChain = [&ChainToBase](Instruction *InsertBefore) {
+ Instruction *LastClonedValue = nullptr;
+ Instruction *LastValue = nullptr;
+ for (Instruction *Instr: ChainToBase) {
+ // Only GEP's and casts are suported as we need to be careful to not
+ // introduce any new uses of pointers not in the liveset.
+ // Note that it's fine to introduce new uses of pointers which were
+ // otherwise not used after this statepoint.
+ assert(isa<GetElementPtrInst>(Instr) || isa<CastInst>(Instr));
+
+ Instruction *ClonedValue = Instr->clone();
+ ClonedValue->insertBefore(InsertBefore);
+ ClonedValue->setName(Instr->getName() + ".remat");
+
+ // If it is not first instruction in the chain then it uses previously
+ // cloned value. We should update it to use cloned value.
+ if (LastClonedValue) {
+ assert(LastValue);
+ ClonedValue->replaceUsesOfWith(LastValue, LastClonedValue);
+#ifndef NDEBUG
+ // Assert that cloned instruction does not use any instructions from
+ // this chain other than LastClonedValue
+ for (auto OpValue : ClonedValue->operand_values()) {
+ assert(std::find(ChainToBase.begin(), ChainToBase.end(), OpValue) ==
+ ChainToBase.end() &&
+ "incorrect use in rematerialization chain");
+ }
+#endif
+ }
+
+ LastClonedValue = ClonedValue;
+ LastValue = Instr;
+ }
+ assert(LastClonedValue);
+ return LastClonedValue;
+ };
+
+ // Different cases for calls and invokes. For invokes we need to clone
+ // instructions both on normal and unwind path.
+ if (CS.isCall()) {
+ Instruction *InsertBefore = CS.getInstruction()->getNextNode();
+ assert(InsertBefore);
+ Instruction *RematerializedValue = rematerializeChain(InsertBefore);
+ Info.RematerializedValues[RematerializedValue] = LiveValue;
+ } else {
+ InvokeInst *Invoke = cast<InvokeInst>(CS.getInstruction());
+
+ Instruction *NormalInsertBefore =
+ Invoke->getNormalDest()->getFirstInsertionPt();
+ Instruction *UnwindInsertBefore =
+ Invoke->getUnwindDest()->getFirstInsertionPt();
+
+ Instruction *NormalRematerializedValue =
+ rematerializeChain(NormalInsertBefore);
+ Instruction *UnwindRematerializedValue =
+ rematerializeChain(UnwindInsertBefore);
+
+ Info.RematerializedValues[NormalRematerializedValue] = LiveValue;
+ Info.RematerializedValues[UnwindRematerializedValue] = LiveValue;
+ }
+ }
+
+ // Remove rematerializaed values from the live set
+ for (auto LiveValue: LiveValuesToBeDeleted) {
+ Info.liveset.erase(LiveValue);
+ }
}
static bool insertParsePoints(Function &F, DominatorTree &DT, Pass *P,
continue;
InvokeInst *invoke = cast<InvokeInst>(CS.getInstruction());
normalizeForInvokeSafepoint(invoke->getNormalDest(), invoke->getParent(),
- P);
+ DT);
normalizeForInvokeSafepoint(invoke->getUnwindDest(), invoke->getParent(),
- P);
+ DT);
}
// A list of dummy calls added to the IR to keep various values obviously
}
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 purely an optimization and
+ // does not influence correctness.
+ TargetTransformInfo &TTI =
+ P->getAnalysis<TargetTransformInfoWrapperPass>().getTTI(F);
+
+ for (size_t i = 0; i < records.size(); i++) {
+ struct PartiallyConstructedSafepointRecord &info = records[i];
+ CallSite &CS = toUpdate[i];
+
+ rematerializeLiveValues(CS, info, TTI);
+ }
+
// Now run through and replace the existing statepoints with new ones with
// the live variables listed. We do not yet update uses of the values being
// relocated. We have references to live variables that need to
return !records.empty();
}
+// Handles both return values and arguments for Functions and CallSites.
+template <typename AttrHolder>
+static void RemoveDerefAttrAtIndex(LLVMContext &Ctx, AttrHolder &AH,
+ unsigned Index) {
+ AttrBuilder R;
+ if (AH.getDereferenceableBytes(Index))
+ R.addAttribute(Attribute::get(Ctx, Attribute::Dereferenceable,
+ AH.getDereferenceableBytes(Index)));
+ if (AH.getDereferenceableOrNullBytes(Index))
+ R.addAttribute(Attribute::get(Ctx, Attribute::DereferenceableOrNull,
+ AH.getDereferenceableOrNullBytes(Index)));
+
+ if (!R.empty())
+ AH.setAttributes(AH.getAttributes().removeAttributes(
+ Ctx, Index, AttributeSet::get(Ctx, Index, R)));
+}
+
+void
+RewriteStatepointsForGC::stripDereferenceabilityInfoFromPrototype(Function &F) {
+ LLVMContext &Ctx = F.getContext();
+
+ for (Argument &A : F.args())
+ if (isa<PointerType>(A.getType()))
+ RemoveDerefAttrAtIndex(Ctx, F, A.getArgNo() + 1);
+
+ if (isa<PointerType>(F.getReturnType()))
+ RemoveDerefAttrAtIndex(Ctx, F, AttributeSet::ReturnIndex);
+}
+
+void RewriteStatepointsForGC::stripDereferenceabilityInfoFromBody(Function &F) {
+ if (F.empty())
+ return;
+
+ LLVMContext &Ctx = F.getContext();
+ MDBuilder Builder(Ctx);
+
+ for (Instruction &I : instructions(F)) {
+ if (const MDNode *MD = I.getMetadata(LLVMContext::MD_tbaa)) {
+ assert(MD->getNumOperands() < 5 && "unrecognized metadata shape!");
+ bool IsImmutableTBAA =
+ MD->getNumOperands() == 4 &&
+ mdconst::extract<ConstantInt>(MD->getOperand(3))->getValue() == 1;
+
+ if (!IsImmutableTBAA)
+ continue; // no work to do, MD_tbaa is already marked mutable
+
+ MDNode *Base = cast<MDNode>(MD->getOperand(0));
+ MDNode *Access = cast<MDNode>(MD->getOperand(1));
+ uint64_t Offset =
+ mdconst::extract<ConstantInt>(MD->getOperand(2))->getZExtValue();
+
+ MDNode *MutableTBAA =
+ Builder.createTBAAStructTagNode(Base, Access, Offset);
+ I.setMetadata(LLVMContext::MD_tbaa, MutableTBAA);
+ }
+
+ if (CallSite CS = CallSite(&I)) {
+ for (int i = 0, e = CS.arg_size(); i != e; i++)
+ if (isa<PointerType>(CS.getArgument(i)->getType()))
+ RemoveDerefAttrAtIndex(Ctx, CS, i + 1);
+ if (isa<PointerType>(CS.getType()))
+ RemoveDerefAttrAtIndex(Ctx, CS, AttributeSet::ReturnIndex);
+ }
+ }
+}
+
/// Returns true if this function should be rewritten by this pass. The main
/// point of this function is as an extension point for custom logic.
static bool shouldRewriteStatepointsIn(Function &F) {
// TODO: This should check the GCStrategy
if (F.hasGC()) {
- const std::string StatepointExampleName("statepoint-example");
- return StatepointExampleName == F.getGC();
+ const char *FunctionGCName = F.getGC();
+ const StringRef StatepointExampleName("statepoint-example");
+ const StringRef CoreCLRName("coreclr");
+ return (StatepointExampleName == FunctionGCName) ||
+ (CoreCLRName == FunctionGCName);
} else
return false;
}
+void RewriteStatepointsForGC::stripDereferenceabilityInfo(Module &M) {
+#ifndef NDEBUG
+ assert(std::any_of(M.begin(), M.end(), shouldRewriteStatepointsIn) &&
+ "precondition!");
+#endif
+
+ for (Function &F : M)
+ stripDereferenceabilityInfoFromPrototype(F);
+
+ for (Function &F : M)
+ stripDereferenceabilityInfoFromBody(F);
+}
+
bool RewriteStatepointsForGC::runOnFunction(Function &F) {
// Nothing to do for declarations.
if (F.isDeclaration() || F.empty())
if (!shouldRewriteStatepointsIn(F))
return false;
- DominatorTree &DT = getAnalysis<DominatorTreeWrapperPass>().getDomTree();
+ DominatorTree &DT = getAnalysis<DominatorTreeWrapperPass>(F).getDomTree();
// Gather all the statepoints which need rewritten. Be careful to only
// consider those in reachable code since we need to ask dominance queries
// 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()))
FoldSingleEntryPHINodes(&BB);
}
+ // Before we start introducing relocations, we want to tweak the IR a bit to
+ // avoid unfortunate code generation effects. The main example is that we
+ // want to try to make sure the comparison feeding a branch is after any
+ // safepoints. Otherwise, we end up with a comparison of pre-relocation
+ // values feeding a branch after relocation. This is semantically correct,
+ // but results in extra register pressure since both the pre-relocation and
+ // post-relocation copies must be available in registers. For code without
+ // relocations this is handled elsewhere, but teaching the scheduler to
+ // reverse the transform we're about to do would be slightly complex.
+ // Note: This may extend the live range of the inputs to the icmp and thus
+ // increase the liveset of any statepoint we move over. This is profitable
+ // as long as all statepoints are in rare blocks. If we had in-register
+ // lowering for live values this would be a much safer transform.
+ auto getConditionInst = [](TerminatorInst *TI) -> Instruction* {
+ if (auto *BI = dyn_cast<BranchInst>(TI))
+ if (BI->isConditional())
+ return dyn_cast<Instruction>(BI->getCondition());
+ // TODO: Extend this to handle switches
+ return nullptr;
+ };
+ for (BasicBlock &BB : F) {
+ TerminatorInst *TI = BB.getTerminator();
+ if (auto *Cond = getConditionInst(TI))
+ // TODO: Handle more than just ICmps here. We should be able to move
+ // most instructions without side effects or memory access.
+ if (isa<ICmpInst>(Cond) && Cond->hasOneUse()) {
+ MadeChange = true;
+ Cond->moveBefore(TI);
+ }
+ }
+
MadeChange |= insertParsePoints(F, DT, this, ParsePointNeeded);
return MadeChange;
}
"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);